TERRAIN CONTROLS ON THE DISTRIBUTION OF TREE
SPECIES DIVERSITY AND STRUCTURE IN TROPICAL
LOWLAND AND TROPICAL MONTANE FOREST

Submitted by ANDREW JAMES JARVIS

A THESIS SUBMITTED TO THE UNIVERSITY OF LONDON
FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

KING’S COLLEGE LONDON

DEPARTMENT OF GEOGRAPHY

2005

ABSTRACT
This thesis examines the micro-scale variability in tree species composition, diversity
and tree structure through a combination of ground-based plot studies and computerbased analyses of terrain characteristics in two contrasting tropical forest sites; a
tropical montane cloud forest environment, and a lowland Amazonian forest
environment. The aim is to measure the micro-scale spatial variability in tree species
diversity, composition and structure in tropical forests, and to quantify the role of
topography, through both direct and indirect resource partitioning (of essential
resources such as energy and water), in controlling this variability. Ten 25m x 25m
plots have been established in each site, distributed widely around the region. Highresolution digital elevation models (DEMs) have also been generated for each site,
and a range of 12 terrain-based characteristics calculated which capture some of the
likely controls on essential resources such as light and water.
Tree diversity and composition is found to vary significantly between plots in each
study site, with species richness varying from 31 to 52 species in a montane forest,
and between 31 and 82 species in the lowland Amazonian site. Analyses of habitat
associations of tree species with respect to the terrain characteristics showed that 36%
of compositional variability in the montane site is explained by elevation, whilst just
16% of compositional variability is explained in the lowland site, also through
elevation (though this pattern is attributed to an edaphic gradient brought about by
river migration).

This indicates that some habitat association occurs, but that a

significant portion of compositional variability remains unaccounted for, at least using
the terrain characteristics used in this study. This leads to the conclusion that a
combination of equilibrium and non-equilibrium, both abiotic and biotic based
processes are maintaining diversity in tropical forests.
When diversity patterns are examined, a mid-elevation tree diversity peak is observed
in the montane site (R2 = 0.70, highest diversity at 1850 m), a pattern attributed to a
combination of the mid-domain effect and species-area relationships. Furthermore,
species richness is found to correlate significantly with slope mean curvature, with
greater richness found on convex slopes (R2 = 0.73). Fewer patterns are found in the
lowland site, with generally lower diversity in the more frequently flooded sites,
though no terrain-based patterns are found to explain variability in diversity between
non-flooded sites.

Structure is observed to vary significantly between sites, and between-plots within
each site yet surprisingly little of this variability is explained by terrain-based
characteristics. Indeed, not even a significant elevational gradient in forest structure
was observed in the montane site, though TopModel (a terrain characteristic
indicative of soil moisture) explained 67% of variability in the DBH/Height ratio.
The interpretation of this result was inconclusive. In the lowland site two marked
clusters in structure were evident, separating higher and lower elevation plots. The
shorter stature found in the low elevation plots was attributed to the disturbance effect
of regular flooding. No further terrain-based correlations with structure were found.
Finally, analysis of spatial heterogeneity of terrain and its relationship with tree
diversity indicated that regions spatially heterogeneous in elevation in the montane
environment were also more diverse (R2 = 0.53). In the lowland site no significant
correlations were observed.
Overall this thesis has found some interesting correlations between terrain
characteristics and tree species composition and diversity, and has provided some
novel methodologies for testing the processes behind the maintenance of species
diversity in tropical forests. These results lead to the conclusion that a combination of
equilibrium and non-equilibrium, abiotic- and biotic- based processes are maintaining
diversity in tropical forests.

ii

ACKNOWLEDGEMENTS
Writing this thesis has taken me to heaven then hell, and then back several times.
Heaven are the places and people I got to know along the way, hell is dealing with
myself during the painful final year. Perhaps prior to acknowledging anyone I should
dis-acknowledge my own psyche, my own worst nightmare, who persistently made
the thesis into something much harder than it really was. Were it not for my own
disservice, I’d have finished months earlier. But this is a happy time (submission),
and one for acknowledging the people (and places) that made this possible. I realise
that it is common to first acknowledge the wife (or husband), the parents, or perhaps
even the dog first, but I risk offending all these (worthy) parties by first
acknowledging Dr. Mark Mulligan, my supervisor. This is both in his supervisory
role and his role as a great friend. Throughout the arduous task of putting this thesis
together, Mark has constantly supported me both academically and personally, far
beyond the normal call of duty. Ever since he driveled on my hand whilst I held a
piece of zinc plating that he was cutting out in the field I realized that he was more
than just a scientific supervisor.

I profoundly thank Mark for his support and

friendship throughout the last 7 years.

Having already risked silent treatment from my wife for not mentioning her first, I
must acknowledge her important role in preventing me from jumping off the balcony
on days and nights of particular stress. She has endured a great deal of bad moods,
childish behaviour (yes – I did have some tantrums along the way) and long trips
away from home, yet always supported me 110%.

Her calming influence and

psychological support kept me on target (well, more or less, though that is my own
fault) and especially in the last few months repeatedly filled me with the confidence to
just finish this bl*#dy thing. And without starting a new paragraph, I must also
mention the role of my family, who gave me whatever intelligence I have (beware
examiners – you are not only judging my abilities – you’ll have my mum to deal with
too), and unfailingly pushed and supported me to continue with education. Thanks to
my mum I have always been fascinated by geography, and thanks to my dad he
instilled an inquisitive yet logical mind in me. My brother and sister have always
been there filling me with confidence and supporting me.

iii

Onto more traditional acknowledgements, a long list of people have supported me in
this thesis. Carlos “Papi Chulo” Gonzalez has provided friendship and scientific
support throughout, and this thesis would not have been possible without his input.
Fieldwork in Tambito was also greatly aided by a number of young students from
Colombia, whose energy, enthusiasm and good spirits made field data collection a
true pleasure despite the difficult conditions. These include Mike Salazar, who has a
great future and continues to work with me as a colleague and a close friend.
Fieldwork was made efficient and enjoyable by Sandra Diaz, Miguel Angel Ramos,
Juan-Carlos Garcia, Liza Gonzalez, and Tatiana Arias. I have also benefited from
strong support from institutions in Colombia, including Fundacion ProSelva (Patricia
and Alvaro Negret), the CRC (Luis Ortega, Bambam), Parques Nacionales, and the
Museo de Historia Natural (Santiago Ayerbe).

In Ecuador, I have also benefited from great support from people in the TBS field
station, including Jaime, Ramon, Santiago and all the other field guides. I also
received help from Lider and Javier Super Bien in the establishment of plots.
Logistically, all at the Universidad de San Francisco de Quito helped a great deal,
including Vladimir in the herbarium, David Romo, Kelly Swing, and Constanza.
Nigel Pitman also provided some data and help in identifying some species for the
Ecuador plots.

I must also not forget to thank fellow PhD students in TBS and Tambito, with whom I
shared good times and together we collected a lot of lovely field data. These include
Matthew Letts, Mauricio Rincon-Romero, Jorge Rubiano, Juliana Gonzalez, Pablo
Vimos, and Mauricio Larrea.

Several people have supported me in other aspects through this thesis. In CIAT these
include Simon Cook (who helped by providing scientific discussions and also showed
a great deal of patience with me towards the end), Thomas Oberthur, Peter Jones,
David Williams, Luigi Guarino and Andy Farrow.

All of these people have

unfailingly confided in my abilities, something I haven’t fully understood yet. Andy
Nelson and Matthew Letts have also helped a great deal in reading through chapters
and providing comments.

iv

Finally, this thesis was possible thanks to generous funding. My scholarship was
provided by the King’s College London Alumni, with whom I am greatly indebted.
Field research was also supported by the University of London Central Research Fund
(twice), the Royal Geographic Society (twice), the Dunsheath Awards, and the CIAT
Strategic Research Fund.

To finish off, I must acknowledge the places I have seen, and their profound role in
inspiring me to study tropical forests. The view from Munchique of the cloud forests
of the northern Andes never failed to make the hairs on the back of my neck tingle,
and these environments must be conserved.

v

Table of Contents
1

Research goals and introduction.............................................................................. 1
1.1
Hypothesis......................................................................................................1
1.2
Aim.................................................................................................................1
1.3
Objectives.......................................................................................................1
1.4
Overview of this thesis...................................................................................5
1.4.1
Chapter 1 ................................................................................................5
1.4.2
Chapter 2 ................................................................................................6
1.4.3
Chapter 3 ................................................................................................6
1.4.4
Chapter 4 ................................................................................................7
1.4.5
Chapter 5 ................................................................................................7
1.4.6
Chapter 6 ................................................................................................8
2 Literature review....................................................................................................... 9
2.1
Introduction : The biodiversity and conservation context..............................9
2.2
Diversity in tropical forests ..........................................................................13
2.2.1
Relationship between tree diversity and biodiversity ..........................14
2.2.2
Diversity and structure in the study sites .............................................17
2.3
Spatial variation of tree diversity and composition in tropical forests.........23
2.3.1
Elevational gradients ............................................................................27
2.3.2
Habitat associations among tropical forest tree species .......................30
2.3.3
Environmental controls on diversity ....................................................35
2.4
Spatial variation of structure in tropical forests ...........................................48
2.4.1
Structure – diversity interaction ...........................................................49
2.5
Spatial environmental heterogeneity and its quantification .........................51
2.6
Field sampling methods ...............................................................................55
2.7
Conclusions ..................................................................................................56
3 Research Strategy and Methods ............................................................................ 58
3.1
Overall research strategy..............................................................................58
3.2
Reserva Tambito ..........................................................................................63
3.3
Tiputini Biodiversity Station........................................................................65
3.4
Generation of biophysical data.....................................................................66
3.4.1
Reserva Tambito ..................................................................................66
3.4.2
Tiputini Biodiversity Station................................................................97
3.5
Establishment of plots ................................................................................120
3.5.1
Positioning of plots ............................................................................120
3.5.2
Plot Descriptions ................................................................................125
3.5.3
Measurement and sampling of individuals.........................................127
3.5.4
Production of plot DEMs ...................................................................129
3.6
Statistical analysis of plot data...................................................................130
3.6.1
Analysis of compositional and structural similarity...........................130
3.6.2
Diversity indices.................................................................................133
3.6.3
Structural Variables............................................................................136
3.6.4
Correlation of environmental and topographic variables with diversity
137
3.7
Examination of DEM quality with respect to ground measured slope and
aspect in TBS .........................................................................................................139
3.8
Heterogeneity Modelling............................................................................145
4 Spatial variation in diversity and composition : Environmental interactions. 156
4.1
Introduction ................................................................................................156

vi

4.2
Reserva Tambito ........................................................................................157
4.2.1
Composition .......................................................................................158
4.2.2
Diversity .............................................................................................172
4.3
Tiputini Biodiversity Station......................................................................192
4.3.1
Composition .......................................................................................192
4.3.2
Diversity .............................................................................................212
4.4
Conclusions ................................................................................................225
5 Micro-scale spatial variation in forest structure ................................................ 230
5.1
Introduction ................................................................................................230
5.2
Reserva Tambito ........................................................................................232
5.2.1
Average structure ...............................................................................232
5.2.2
Structural distribution.........................................................................243
5.3
Tiputini Biodiversity Station......................................................................252
5.3.1
Average structure ...............................................................................253
5.3.2
Distribution in structure .....................................................................262
5.4
Conclusions ................................................................................................279
6 Effect of spatial heterogeneity of environmental variables on the diversity.... 284
6.1
Research Strategy.......................................................................................284
6.2
Reserva Tambito ........................................................................................289
6.3
Tiputini Biodiversity Station......................................................................314
6.3.1
TOPO DEM........................................................................................315
6.3.2
SRTM DEM .......................................................................................326
6.4
Conclusions ................................................................................................337
7 Final Conclusions .................................................................................................. 341
References ...................................................................................................................... 348
8 Appendix 1 ............................................................................................................. 369
9 Appendix 2 ............................................................................................................. 387
10 Appendix 3 ............................................................................................................. 392
11 Appendix 4 ............................................................................................................. 395
12 Appendix 5 ............................................................................................................. 410

vii

List of Figures
Figure 1 The topographic complexity in Reserva Tambito. Photo taken at 1500m from the
main cabins at around 3pm. Cloud presence is typical of the afternoon hours, with the
late afternoon often characterised by canopy level cloud presence at all elevations. .....65
Figure 2 River Tiputini at low flow. During highflow the river exceeds the banks and floods
the adjoining forest. ................................................................................................66
Figure 3 Hydrological and climatological monitoring station at the Campo and Bosque plot,
Tambito .................................................................................................................68
Figure 4 Rainfall (mm) manually measured daily for the Tambito study site, taken at
1450masl in the region around the cabins. ................................................................69
Figure 5 Monthly distribution of rainfall measured manually on a daily basis in the Tambito
study site ................................................................................................................70
Figure 6 The cartographically derived DEM with 25m cell resolution (left), with the derived
slope (center) and aspect (right). Slope varies from 0o (light blue) to 58o (red). Aspect
is in degrees, ranging from 0o (red) to 180o (blue) and 360o (white). ...........................72
Figure 7 Total potential solar radiation receipt across the Tambito study site. Blue indicates
lower receipt (5,600 W/m2), green intermediate levels (8,400 W/m2) and yellow high
levels (9,600 W/m2). ...............................................................................................74
Figure 8 Annual variation in solar radiation receipt at different times of day, using measured
data from the Campo station and the average modelled monthly top of atmosphere solar
radiation receipt (all hours) for reference. .................................................................77
Figure 9 Variation in modelled solar radiation (top of atmosphere) for different degrees of
northness throughout the year in Tambito, averaged for all 69,678 pixels....................78
Figure 10 Variation in modelled solar radiation (top of atmosphere) for different degrees of
eastness throughout the year in Tambito, averaged for all 69,678 pixels......................79
Figure 11 Modelled and measured hourly solar radiation receipt in Tambito, averaging hourly
measurements from throughout the year ...................................................................80
Figure 12 Hourly modelled radiation receipt against measured radiation receipt for the Campo
station for four different months, January (top left), April (top right), July (bottom left)
and October (bottom right). .....................................................................................81
Figure 13 Multi-scale mean curvature for Tambito. Blue signifies concave slopes (-3), white
planar slopes with no curvature (0) and red signifies convex slopes (+3).....................85
Figure 14 Topmodel wetness index for Tambito, dark blue indicating high wetness values. .87
Figure 15 Multi-scale slope position for Tambito. Red signifies cells close to the ridges
(100), white signifies mid-slopes (50) and blue signifies cells close to the valley bottom
(0)..........................................................................................................................90
Figure 16 Feature extraction using the Landserf algorithm. The left image shows the
classification of features under standard analysis, and the right image shows the same
analysis with the addition of network rules. In both cases, the landscape is classified
into peaks (red), ridges (yellow), passes (green), planar slopes (grey) and channels
(blue). ....................................................................................................................91
Figure 17 Calculation of “topoclasses” using Zimmerman’s method. The right figure
represents the toposcale grid (red representing positive values and blue represent
negative values), and the left figure shows the reclassification of this grid into channels
(blue), toe slopes (light green), slopes (darker green) and ridges (brown). ...................93
Figure 18 Derivation of each terrain characteristic from the original DEM ..........................95
Figure 19 10-day running mean rainfall series for Tiputini Biodiversity Station 1997 – 2003,
recorded with the Boston University Campbell Scientific meteorological station installed
at the station facilities. Note the missing data from 1999 – 2001. ...............................98
Figure 20 Results of DEM generation using the cartographic maps, producing a 25m grid cell
resolution DEM (top), with the derived slope (left) and aspect (right). Slope varies from

viii

0o (light blue) to 31o (red). Aspect is in degrees, ranging from 0o (red) to 180o (blue) and
360o (white). .........................................................................................................100
Figure 21 Results of DEM generation using the SRTM satellite derived DEM producing a
100m grid cell resolution DEM (top), with the derived slope (left) and aspect (right).
Slope varies from 0o (light blue) to 12o (red). Aspect is in degrees, ranging from 0o (red)
to 180o (blue) and 360o (white)...............................................................................101
Figure 22 The VNIR ASTER image from 12th April 2003, used to extract a 15m DEM,
displayed using bands 2 (red), 3 (infra-red) and 1 (green). Whole image (left) shows
heavy cloud cover in the western and south-eastern sector of the image, but TBS is
cloud free (right). ..................................................................................................103
Figure 23 Results of DEM generation using the ASTER satellite image producing a 15m grid
cell resolution DEM (top), with the derived slope (left) and aspect (right). Slope varies
from 0o (light blue) to 68o (red). Aspect is in degrees, ranging from 0o (red) to 180o
(blue) and 360o (white). .........................................................................................104
Figure 24 Total potential solar radiation receipt across the TBS study site, using the
TOPODEM (left) and SRTM DEM (right). Blue indicates lower receipt, green
intermediate and yellow high. ................................................................................107
Figure 25 Monthly variability in solar radiation receipt for TBS, showing modelled
(BENDUM) and measured receipt for the Boston University AWS. The measured data
is averaged for data periods between 1997 – 2004. ..................................................108
Figure 26 Daily variability in solar radiation receipt for TBS, showing modelled (from the
BENDUM model) and measured receipt for the Boston University AWS. The measured
data is averaged for data periods between 1997 – 2004. ...........................................109
Figure 27 Wind speed and wind directional frequency in TBS based on 7 years (intermittent)
of measured wind data from the Boston University automated weather station in TBS.

............................................................................................................................110
Figure 28 Frequency of wind events for different wind speeds separated for each direction
TBS based on 7 years (intermittent) of measured wind data from the Boston University
automated weather station in TBS. .........................................................................110
Figure 29 Multi-scale mean curvature for TBS using the TOPODEM (left) and SRTM DEM
(right). Blue signifies concave slopes (-3), white planar slopes with no curvature (0) and
red signifies convex slopes (+3). ............................................................................111
Figure 30 Topmodel wetness index for TBS, using the TOPODEM (left) and SRTM DEM
(right), dark blue indicating high wetness values. ....................................................112
Figure 31 Multi-scale slope position for TBS for the TOPODEM (left) and SRTM DEM
(right). Blue signifies cells close to the ridges (100), yellow signifies mid-slopes (50)
and green signifies cells close to the valley bottom (0).............................................113
Figure 32 Feature extraction using the Wood’s algorithm (2004) using the TOPODEM (top),
SRTM DEM (bottom). The left image shows the classification of features under
standard analysis, and the right image shows the same analysis with the addition of
network rules. In both cases, the landscape is classified into peaks (red), ridges
(yellow), passes (green), planar slopes (grey) and channels (blue). ...........................114
Figure 33 Calculation of “topoclasses” using Zimmerman’s method, using the TOPODEM
(top), SRTM DEM (bottom). The right figure represents the toposcale grid (red
representing positive values and blue represent negative values, and grey represent
neutral values), and the left figure shows the reclassification of this grid into channels
(blue), toe slopes (light green), slopes (darker green) and ridges (brown). .................116
Figure 34 Position of plots in Tambito ............................................................................122
Figure 35 Sampling strategy used to locate ground plots in TBS ......................................124
Figure 36 Geographic distribution of the 10 25m x 25m plots established in Tiputini
Biodiversity Station. .............................................................................................125
Figure 37 High resolution (50cm) plot DEMs for the 25m x 25m diversity and structure plots
in TBS .................................................................................................................130

ix

Figure 38 Comparison of DEM derived slopes and the plot measured slopes for the two
DEMs in TBS. Slope from the plots was derived as a mean for all pixels. ................141
Figure 39 Comparison of DEM derived eastness and the plot measured slopes for the two
DEMs in TBS. Eastness of the plots was derived as a mean for all pixels. ................141
Figure 40 Comparison of DEM derived northness and the plot measured slopes for the two
DEMs in TBS. northness of the plots was derived as a mean for all pixels. ..............142
Figure 41 Scatterplot of ground-based measurement of topographic complexity (using the
coefficient of variability in slope as a surrogate) against DEM derived slope position.

............................................................................................................................144
Figure 42 Distance weighting from the Half-Cauchy curve for different median distances
from 0.01 cells (top-left) to 9 cells (bottom-right)....................................................148
Figure 43 3-dimensional representation of the half-Cauchy spatial function, with the height of
the surface representing the weighting. ...................................................................149
Figure 44 Example analysis of spatial heterogeneity for a model environment. .................153
Figure 45 Distribution of species in families in Tambito ..................................................160
Figure 46 Distribution of individuals in families in Tambito ............................................160
Figure 47 Species-area curves for Tambito, calculated by separating out the quadrants into
increasingly sized plots..........................................................................................163
Figure 48 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity) for
Tambito. ..............................................................................................................166
Figure 49 Multi-variate clustering of plots based on all 9 environmental and topographic
variables...............................................................................................................170
Figure 50 The Renyi scaleable diversity index for Tambito, indicating to what extent plot
diversities can be universally compared. .................................................................173
Figure 51 Relationship between plot diversity and elevation in Tambito. ..........................176
Figure 52 Relationship between plot richness and elevation in Tambito ............................176
Figure 53 Example of niche model, with 30 hypothetical species with randomly positioned
ranges along the elevational gradient. The thick grey line represents the respective
Simpson’s Diversity Index and the black line the Berger-Parker dominance..............180
Figure 54 Average pattern of diversity-elevation relations in the simple niche model for
Tambito. ..............................................................................................................180
Figure 55 Total area of land across the elevational gradient for the twin-catchments of
Tambito, expressed as the frequency of 25m (625m2) cells per 100m elevational band.

............................................................................................................................181
Figure 56 Land-surface area across the elevational gradient for a swath across Colombia
between 2oN and 3oN. ...........................................................................................183
Figure 57 Niche model adjusted for the area concept, with more species located in elevations
with greater land surface area, and the elevational gradient extended to range form 0m to
3000m. .................................................................................................................183
Figure 58 Significant relationships between Simpson’s diversity and mean curvature (left) and
toposcale (right) for Tambito. ................................................................................184
Figure 59 Relationship between species richness and mean curvature (left) and toposcale
(right) when plot 9 is omitted from the analysis. The linear relationship and R2
displayed corresponds to the result without plot 9, though the point is shown in the
graph for information. ...........................................................................................185
Figure 60 Residual in richness (with respect to a polynomial mid-elevational peak trendline)
compared with mean curvature, with the purpose of explaining the variation in richness
across the elevational gradient. The residual is calculated based on the actual richness
(plot measured) minus the modeled richness (polynomial trendline). Plot 9 is excluded
from the analysis for reasons explained earlier. .......................................................189

x

Figure 61 Modelled diversity (calculated using the multiple stepwise regression) correlated
against measured diversity (left), and the variation in modeled diversity across the
elevational gradient. ..............................................................................................191
Figure 62 Distribution of species in families in TBS........................................................195
Figure 63 Distribution of individuals in families in TBS ..................................................195
Figure 64 Species-area curves for TBS. ..........................................................................197
Figure 65 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity)
derived from the TOPO DEM. ...............................................................................200
Figure 66 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity)
derived from the SRTM DEM................................................................................202
Figure 67 Scatterplot of compositional similarity against the difference in distance from any
river channel, using channels identified in an IKONOS image of TBS. .....................204
Figure 68 Example of the river migration present in the region, with many oxbow lakes
forming where old meanders were cut off. This image is of an area to the south of TBS,
taken using LANDSAT TM imagery. .....................................................................205
Figure 69 Multi-variate clustering of plots based on all 9 environmental and topographic
variables for TBS using the TOPO DEM. ...............................................................208
Figure 70 Multi-variate clustering of plots based on all 9 environmental and topographic
variables for TBS using the SRTM DEM................................................................208
Figure 71 Modelled similarity against the measured similarity applying the results of the
multiple stepwise linear regression for the variables of the SRTM DEM. ..................209
Figure 72 The Renyi scaleable diversity index for TBS, indicating to what extent plot
diversities can be universally compared ..................................................................213
Figure 73 Elevational patterns in richness and diversity in TBS........................................217
Figure 74 Relationship between diversity and distance from the River Tiputini in TBS......218
Figure 75 Relationship between slope position and mean curvature and genera richness in
TBS using the SRTM DEM ...................................................................................220
Figure 76 Modelled species richness (based on multiple stepwise regression) plotted against
measured diversity in TBS using the SRTM DEM. .................................................221
Figure 77 Frequency histograms for the three major structural characteristics for all
individuals in Tambito ..........................................................................................234
Figure 78 Elevational gradient in structural parameters for Tambito .................................236
Figure 79 Significant correlation between DBH/Height and topmodel for Tambito............238
Figure 80 Modelled stem density against measured stem density in Tambito, based on the
multiple stepwise linear regression model using elevation and toposcale (right) and
modelled tree height against measured tree height in Tambito, based on the multiple
stepwise linear regression model using topmodel and solar radiation receipt (left). ....241
Figure 81 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the K-S test of the distributions) against environmental difference between plotpairs in Tambito....................................................................................................246
Figure 82 Scatterplots of similarity in distribution of tree height between plot-pairs
(calculated using the K-S test of the distributions) against environmental difference
between plot-pairs in Tambito................................................................................248
Figure 83 Scatterplots of similarity in distribution of the DBH/Height variable between plotpairs (calculated using the K-S test of the distributions) against environmental
difference between plot-pairs in Tambito. ...............................................................250
Figure 84 Scatterplots for plot-pairs of compositional similarity against structural
dissimilarity for the major structural variables in Tambito........................................252
Figure 85 Frequency histograms for the three major structural characteristics for all
individuals in TBS ................................................................................................254

xi

Figure 86 Significant correlations for plot average DBH/Height and tree height with elevation
in TBS .................................................................................................................257
Figure 87 Relationship between plot average DBH/Height and tree height and distance from
river in TBS..........................................................................................................258
Figure 88 Significant relationships between mean curvature and DBH/Height (left) and
eastness and DBH/DbaseH (right) in TBS using the TOPO DEM. ............................260
Figure 89 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the Topo DEM. ...................264
Figure 90 Scatterplots of similarity in distribution of height between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the Topo DEM. ...................266
Figure 91 Scatterplots of similarity in distribution of the DBH/Height between plot-pairs
(calculated using the Kolmogorov-Smirnov test of the distributions) against
environmental difference between plot-pairs in TBS for derivatives of the Topo DEM.

............................................................................................................................268
Figure 92 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the SRTM DEM. .................270
Figure 93 Scatterplots of similarity in distribution of height between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the SRTM DEM. .................273
Figure 94 Scatterplots of similarity in distribution of DBH/Height between plot-pairs
(calculated using the Kolmogorov-Smirnov test of the distributions) against
environmental difference between plot-pairs in TBS for derivatives of the SRTM DEM.

............................................................................................................................275
Figure 95 Scatterplots for plot-pairs of compositional similarity against structural
dissimilarity for the major structural variables in TBS. ............................................278
Figure 96 Model role of environmental heterogeneity on within-plot diversity in tropical
forests through seed dispersal. Plot A, located in a heterogeneous environment is likely
to receive seeds from more local habitats than Plot B. .............................................287
Figure 97 Cross-correlation matrix between all spatial heterogeneity variables with the
median distance set at 1 cell in Tambito (n = 69,678 in all cases).............................290
Figure 98 Control experiment of cross-scale heterogeneity for Random Grid 1, a grid with
random values. Black line represents the Pearson correlation coefficient, whilst the blue
line represents the respective p-value. The solid blue line is placed at the 95%
confidence level, meaning that a correlation is considered statistically significant when
the plot of p-values goes under the thick blue line. ..................................................292
Figure 99 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in Tambito. The black line represents the
Pearson correlation coefficient, whilst the blue line indicates the respective p-value. For
ease of interpretation, the thick blue line shows the threshold for significant correlations
at the alpha = 0.05 level.........................................................................................298
Figure 100 Elevational heterogeneity (median distance 9 cells) against species diversity for
Tambito. ..............................................................................................................302
Figure 101 Elevational gradient of heterogeneity in elevation in Tambito. ........................303
Figure 102 Scatterplot of northness heterogeneity against species diversity for Tambito using
a median distance of 0.01. .....................................................................................305
Figure 103 First multiple linear-regression of all heterogeneity variables at all scales. Left
figure compares the modeled diversity with the measured diversity in the plots, and the
right figure applies the resultant regression model to the entirety of Tambito. ...........309
Figure 104 Second multiple linear-regression of heterogeneity variables at only at the scale of
maximum correlation. Left figure compares the modeled diversity with the measured

xii

diversity in the plots, and the right figure applies the resultant regression model to the
entirety of Tambito. ..............................................................................................310
Figure 105 Third multiple linear regression combining both absolute value and
heterogeneities of each environmental and topographic variable. Left figure compares
the modeled diversity with the measured diversity in the plots, and the right figure
applies the resultant regression model to the entirety of Tambito. .............................312
Figure 106 Control experiment of cross-scale heterogeneity for a random grid 6 using the
TOPO DEM. ........................................................................................................319
Figure 107 Result of the heterogeneity model for Random Grid 6 for the TOPO DEM in TBS.
Represented here is the analysis with median distance of 1 cell (25m), where the
correlation with plot diversity was found to be highest. Red indicates low heterogeneity,
and blue high heterogeneity, with the plots displayed as the black dots. ....................320
Figure 108 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in TBS using the TOPO DEM. The black
line represents the Pearson correlation coefficient, whilst the blue line indicates the
respective p-value. For ease of interpretation, the thick blue line shows the threshold for
significant correlations at the alpha = 0.05 level. .....................................................324
Figure 109 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in TBS using the SRTM DEM. The
black line represents the Pearson correlation coefficient, whilst the blue line indicates
the respective p-value. For ease of interpretation, the thick blue line shows the threshold
for significant correlations at the alpha = 0.05 level. ................................................334
Figure 110 Scatterplot of heterogeneity in the northness variable against plot diversity, using
the SRTM DEM for TBS when plot 5 is included (left) and excluded (right).............336
Figure 111 Instrumental setup for aerial photography acquisition .....................................371
Figure 112 Ground spatial coverage of images from the three different cameras depending on
elevation of the sensor above the ground surface. ....................................................372
Figure 113 Resolution of images from the three different cameras depending on elevation of
the sensor above the ground surface. ......................................................................372
Figure 114 Required time between frames to ensure 1/3 overlap in images depending on
elevation of the sensor above the ground surface. ....................................................373
Figure 115 Elevation at which blurring is expected to occur for each of the three cameras.
This was calculated based on the spatial resolution of the pixels, the exposure of the
camera and the speed of the aeroplane (fixed at 200km/h). If during the exposure time
the aeroplane movement exceeds the equivalent of 1 ground pixel, it was assumed that
blurring would be present in the image. ..................................................................373
Figure 116 Planned and actual flight path for aerial image acquisition over TBS, March 2002.

............................................................................................................................375
Figure 117 Method for helium balloon image acquisition from forest gaps........................377
Figure 118 The helium balloon in flight..........................................................................377
Figure 119 Instrumental setup of tethered balloon photo acquisition .................................378
Figure 120 Georeferenced imagery for TBS at all resolutions ..........................................381
Figure 121 Classification hierarchy constructed in eCognition. ........................................382
Figure 122 Results of the hierarchical classification of eCognition, identifying river (blue),
gaps / shade (black) and tree crowns (green). ..........................................................383
Figure 123 Modelled canopy diversity, derived from object-oriented analysis of canopy
forms. ..................................................................................................................384
Figure 124 Plot 1 (25m x 25m) from the air, in an image taken using the helium balloon. As
can be seen, few trees penetrate the canopy, with just 5-10 of the 80 trees identified on
the ground actually visible in this high-resolution image. .........................................385

xiii

List of Tables
Table 1 Structural parameters for LRF and TMCF, from Letts (2003), based on studies by
Richards (1996), Kitayama and Aiba (2002), Hafkenscheid (2000), Heaney and Proctor
(1989), and Leigh (1999) and this study....................................................................21
Table 2 Summary values for all Tambito terrain characteristics. .........................................96
Table 3 Cross-correlation matrix between all terrain properties used in the Tambito case
study. Values represent the correlation coefficient. Correlations greater than 0.7
(positive or negative) are highlighted in dark grey, and light grey is used for correlation
of 0.3 to 0.7 (n=69, 678). In reality, most of these correlations are significant (95%
level) due to the large number of samples. ................................................................96
Table 4 TBS DEM summary statistics (above) and cross-correlation matrix (below) showing
correlation coefficients, n=10,400 (all DEMs were re-sampled to 92m to enable direct
comparison)..........................................................................................................106
Table 5 Summary values for all TBS terrain characteristics, using both DEMs ..................117
Table 6 Cross-correlation matrix of terrain characteristics for the TOPO DEM for TBS (n =
143,220) ...............................................................................................................118
Table 7 Cross-correlation matrix of terrain characteristics for the SRTM DEM for TBS (n =
10,504).................................................................................................................119
Table 8 Sampling strategy units for each of the ten established plot. .................................125
Table 9 Brief description of conditions in the Tambito plots. ...........................................126
Table 10 Brief description of conditions in the TBS plots. ..............................................127
Table 11 Error in DEM representation of plot topography for all slopes, as well as for upper
and lower slopes defined using the slope position variable. ......................................144
Table 12 Summary of species composition per family in Tambito, including the number of
plots where the family was present. ........................................................................159
Table 13 Summary results of compositional similarity analysis with distance and
environmental variables, using Mantel tests and a one-tailed Pearson for Tambito. Note
that compositional similarity was compared with environmental dissimilarity
(difference), hence negative values in most cases. The northness variable produces very
different results due to the heavily skewed distribution of plots in strongly northward
and strongly southward facing slopes. ....................................................................166
Table 14 Revised Mantel statistics for correlating variables with compositional similarity in
Tambito, applying partial Mantel tests to detect spurious correlations derived from colinearity between variables. ...................................................................................168
Table 15 Number of species associated to specific classes in the “Feature” classification for
Tambito. ..............................................................................................................171
Table 16 Number of species associated to specific classes in the “Network Feature”
classification for Tambito. .....................................................................................171
Table 17 Number of species associated to specific classes in the “topoclass” classification for
Tambito. ..............................................................................................................171
Table 18 Richness and diversity of plots in Tambito ........................................................173
Table 19 Environmental conditions at each plot in Tambito, derived from the terrain
derivatives. ...........................................................................................................174
Table 20 Pearson’s correlation coefficient and the two-tailed p-value between richness and
diversity in each plot and the environmental variables for Tambito. ..........................175
Table 21 Average richness and diversity of plots when separated into three elevational bands.
In brackets the respective coefficient of variation. ...................................................177
Table 22 Pearson correlation coefficients between plot diversity and richness and
environmental variables when plot 9 is omitted from the analysis. Significant
correlations (p<0.05) are highlighted in grey. ..........................................................185

xiv

Table 23 Pearson correlation coefficients for Simpson’s diversity and species richness
correlations with mean curvature at different scales. ................................................187
Table 24 Results of the Spearman Rank correlation tests on the categorical landscape feature
data for the topoclass, feature and network feature terrain derivatives in Tambito.Error!

Bookmark not defined.
Table 25 Summary of species composition in TBS ..........................................................194
Table 26 Summary results of compositional similarity analysis with distance and
environmental variables derived from the TOPO DEM, using Mantel tests and a onetailed Pearson. Note that compositional similarity was compared with environmental
dissimilarity (difference), hence negative values indicate a positive causal relationship.

............................................................................................................................200
Table 27 Summary results of compositional similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a onetailed Pearson. Note that compositional similarity was compared with environmental
dissimilarity (difference), hence negative values indicate a positive causal relationship.

............................................................................................................................202
Table 28 Revised Mantel statistics for correlating variables with compositional similarity in
TBS, applying partial Mantel tests to detect spurious correlations derived from colinearity between variables and the distance from river. ...........................................203
Table 29 Number of species associated to specific classes in the “Feature” classification for
TBS. ....................................................................................................................210
Table 30 Number of species associated to specific classes in the “Network Feature”
classification for TBS. ...........................................................................................210
Table 31 Number of species associated to specific classes in the “topoclass” classification for
TBS. ....................................................................................................................211
Table 32 Richness and diversity of plots in TBS .............................................................212
Table 33 Environmental and topographic conditions at each plot in TBS using each DEM
(TOPO DEM top, SRTM DEM, below)..................................................................215

Table 34 Pearson’s correlation coefficient and the two-tailed p-value between richness
and diversity in each plot and the environmental variables for TBS using the
TOPO DEM........................................................................................................216
Table 35 Pearson’s correlation coefficient and the two-tailed p-value between richness and
diversity in each plot and the environmental variables for TBS using the SRTM DEM.

............................................................................................................................216
Table 36 Landscape unit classifications for the plots in TBS using the TOPO DEM. .........224
Table 37 Landscape unit classifications for the plots in TBS using the SRTM DEM..........224
Table 38 Results of the Spearman Rank correlation tests on the categorical landscape
classifications for the topoclass, feature and network feature derivatives in TBS using
the TOPO DEM. .....................................................Error! Bookmark not defined.
Table 39 Results of the Spearman Rank correlation tests on the categorical landscape
classifications for the topoclass, feature and network feature derivatives in TBS using
the SRTM DEM. ....................................................Error! Bookmark not defined.
Table 40 Correlation coefficients between feature membership and richness and diversity for
TBS using the TOPO DEM. ..................................................................................225
Table 41 Correlation coefficients between feature membership and richness and diversity for
TBS using the SRTM DEM. ..................................................................................225
Table 42 Structural characteristics of the plots in Tambito ...............................................235
Table 43 Pearson correlation coefficients and the respective p-values between structural
characteristics of the forest and the environmental and topographic variables in Tambito.

............................................................................................................................237

xv

Table 44 Spearman Rank correlation coefficients and the respective p-values between
structural variables and the landscape classifications in Tambito. .... Error! Bookmark

not defined.
Table 45 Correlation coefficients and respective p-values for all correlations between
diversity and richness measures and structural variables in Tambito. ........................242
Table 46 Summary results of structural DBH similarity analysis with distance and
environmental variables, using Mantel tests and a one-tailed Pearson. Note that
structural dissimilarity was compared with environmental dissimilarity (difference).
The northness variable produces very different results due to the heavily skewed
distribution of plots in strongly northward and strongly southward facing slopes. ......246
Table 47 Summary results of tree height similarity analysis with distance and environmental
variables, using Mantel tests and a one-tailed Pearson. Note that structural dissimilarity
was compared with environmental dissimilarity (difference). The northness variable
produces very different results due to the heavily skewed distribution of plots in strongly
northward and strongly southward facing slopes. ....................................................248
Table 48 Summary results of structural DBH/Height variable similarity analysis with distance
and environmental variables, using Mantel tests and a one-tailed Pearson. Note that
structural dissimilarity was compared with environmental dissimilarity (difference).
The northness variable produces very different results due to the heavily skewed
distribution of plots in strongly northward and strongly southward facing slopes. ......250
Table 49 Structural characteristics of the plots in TBS .....................................................255
Table 50 Pearson correlation coefficients and the respective p-values between average
structural characteristics for each plot and the topographic characteristics derived from
the TOPO DEM in TBS ........................................................................................256
Table 51 Pearson correlation coefficients and the respective p-values between average
structural characteristics for each plot and the topographic characteristics derived from
the TOPO DEM in TBS. .......................................................................................257
Table 52 Spearman Rank correlation coefficients and the respective p-values between
structural variables and the landscape classifications in TBS using the TOPO DEM.

..............................................................................Error! Bookmark not defined.
Table 53 Spearman Rank correlation coefficients and the respective p-values between
structural variables and the landscape classifications in TBS using the SRTM DEM.

..............................................................................Error! Bookmark not defined.
Table 54 Correlation coefficients and respective p-values for all correlations between
diversity and richness measures and structural variables in TBS. ..............................261
Table 55 Summary results of structural DBH similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a onetailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................264
Table 56 Summary results of structural height similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a onetailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................266
Table 57 Summary results of structural DBH/Height similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a onetailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................268
Table 58 Summary results of structural DBH similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a onetailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................271
Table 59 Summary results of structural height similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a one-

xvi

tailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................273
Table 60 Summary results of DBH/Height similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a onetailed Pearson. Note that structural dissimilarity was compared with environmental
dissimilarity (difference). ......................................................................................275
Table 61 Results of the heterogeneity analysis for the 12 random grids in Tambito. .........292
Table 62 Summary table of results of multi-scale environmental heterogeneity modeling,
examining maximum correlations between species richness and heterogeneity for the ten
plots in Tambito....................................................................................................299
Table 63 Cross-correlation of heterogeneity values for the six correlating variables in
Tambito, using only the heterogeneity values for the ten plots (n = 10). ....................300
Table 64 Cross-correlation matrix between all spatial heterogeneity variables with the median
distance set at 1 cell in TBS using the TOPO DEM (n = 142,450). ...........................316
Table 65 Results of the heterogeneity analysis for the 12 random grids using the SRTM
DEM in TBS. .......................................................................................................317
Table 66 Results of the heterogeneity analysis for the 12 random grids when Plots 4 and 5
are excluded from the analysis using the TOPO DEM in TBS. .................................319
Table 67 Summary table of results of multi-scale environmental heterogeneity modeling,
examining maximum correlations between species richness and heterogeneity for the ten
plots in TBS using the TOPO DEM........................................................................325
Table 68 Cross-correlation matrix between all spatial heterogeneity variables with the median
distance set at 1 cell in TBS using the SRTM DEM (n = 10,400)..............................328
Table 69 Results of the heterogeneity analysis for the 12 random grids using the SRTM DEM
in TBS. ................................................................................................................329
Table 70 Results of the heterogeneity analysis for the 12 random grids when Plots 4 and 5
are excluded from the analysis using the SRTM DEM in TBS. ................................329
Table 71 Summary table of results of multi-scale environmental heterogeneity modeling,
examining maximum correlations between species richness and heterogeneity for the ten
plots in TBS using the SRTM DEM. ......................................................................335
Table 72 Summary of image acquisition and georeferencing results. ................................380

xvii

Chapter 1

1

Research goals and introduction

This opening chapter presents the aims and objectives of the study, and a brief
overview of the research strategy and structure of the thesis. This sets the context for
the review of associated literature found in Chapter 2.

1.1

Hypothesis

Tree species diversity, forest structure and species composition in tropical forests are
heterogeneously distributed in space at the micro-scale, and topography plays an
important role in generating this spatial variation.

1.2

Aim

The aim of this thesis is to measure the micro-scale spatial variability in tree species
diversity, composition and structure in tropical forests, and to quantify the role of
topography, through both direct and indirect resource partitioning (of essential
resources such as energy and water), in controlling this variability.

1.3

Objectives

Tree species diversity and structure in tropical forests are heterogeneously distributed
across the land surface, a notion strongly supported in the literature with differences
in diversity, composition and structure observed at a broad range of spatial scales (for
example Pitman, 2000; Hawkins et al., 2003; Hawkins and Diniz-Filho, 2004). A

central issue in tropical ecology is to understand the processes which drive this spatial
variability since those processes must be at least part responsible for the maintenance
of high levels of tree diversity in tropical forest systems. Since trees are the dominant
structural organisms of tropical forests, tree species diversity is also central to theories
explaining why these forests are also so diverse in other plant and animal groups.
In this thesis I endeavour to use methods of terrain analysis to quantify environmental
conditions (controlled through topography) at the micro-scale, and liken species
diversity and composition (principally) and tree structure (secondarily) to these
environmental conditions, to understand the important interaction between
environment and diversity in tropical forests. The term micro-scale is used in this
thesis to refer to spatial variability at the local scale, over distances of 20m – 100m
depending on the degree of topographic variability, though this may also be referred
to as the plot-landscape scale.

Specifically, the objectives of this thesis are:

1. To use field-collected data on tree diversity, composition and structure in
tropical forests to quantify the degree of spatial variation at the micro-scale.
2. To examine the relationship between topographic characteristics and species
composition in order to assess the degree of topographically controlled habitat
association in tropical tree species.
3. To identify the role of topographic characteristics on maintaining levels of tree
diversity, through the control of essential resources

2

4. To identify the role of topographic conditions in generating distinct structural
characteristics in the forest, and how this may control or be controlled by
variation in tree species diversity.
5. To quantify the effect of spatial environmental heterogeneity in the generation
and maintenance of tree species diversity at the micro-scale.

Using two contrasting study sites found in a tropical montane cloud forest (TMCF) in
Colombia and a tropical lowland forest (LRF) in Amazonian Ecuador, this thesis
combines intensive field-based plot measurement of diversity, composition and
structure of tropical trees, and compares the patterns of spatial variation found in the
field with the topographic conditions as measured through high resolution digital
elevation models. A number of terrain derivatives, some of which control the spatial
distribution of essential resources (energy and water), are applied for this purpose. In
addition to direct comparison of topographic conditions with the observed structure
and diversity at the site of each plot, the role of spatial heterogeneity in these
topographic conditions is also examined.

The research presented in this thesis is principally empirical and academically driven,
but these analyses are applied within the context of the current debate on theories on
the maintenance of species diversity in tropical forests (discussed in Section 2.3), and
its implications for science-based species conservation planning. For the latter point,
some effort is maintained to make the methodology used here easily applicable to
other regions, by using readily available sources of data and replicable methodologies.

This research is innovative in the field of tropical ecology in the following ways:

3

•

Though habitat associations of tropical tree species have been studied with
respect to topography (see Section 2.3.2), only basic topographic
characteristics have been used (slope, elevation and landscape unit) whilst this
thesis applies advanced methods of terrain analysis to examine habitat
associations in tree species composition for a wide variety of resource-based
terrain characteristics over a range of scales.

•

There are few examples of studies which endeavor to find relationships
between quantified topographic characteristics and tree diversity in tropical
forests (see Section 2.3.3.5), and this thesis is the first known example of
applying terrain analysis to understanding micro-scale spatial variability in
diversity.

•

Although there is a strong theoretical body of literature stating that
environmental heterogeneity is significant in shaping spatial variability in
diversity in tropical forests (see Section 2.5), this has never been rigorously
tested in tropical forests and this thesis makes a first attempt through a
combination of field data collection and analyses and the application and
validation of spatial models for diversity control.

This thesis focuses on the role of the environment on tree diversity, composition and
structure, though it is important to acknowledge the important role of biotic processes
(especially ecological interactions such as competition, allelopethy, or herbivory etc.).
Although these are not analysed here, nor thoroughly discussed in the text, biotic
interactions have also been found to be important in the maintenance and generation
of diversity (Huston, 1994).

The objectives of this thesis are only related to

4

environmental interactions, but it is important at this point to note that biotic
interactions are also of importance but not centrally addressed.

It is also important to emphasize that the direct role of soil on tree diversity,
composition and structure is not explicitly analysed in this thesis (no soil sampling
was performed), though it is regularly discussed in the text. This is in order to ensure
that the methodology applied here is readily applicable to other regions, only
requiring topographic data. However, soil attributes are often strongly controlled by
topography, and the possible significance of specific topographic characteristics on
soil type and quality is discussed both in the literature review (Section 2.3.3.4) and in
the analysis chapters.

In this sense no direct measurement of soil-diversity

relationships are made, but topography is used a possible indicator of some soil
conditions.

1.4

Overview of this thesis

This thesis is split into seven chapters. They are structured in a way to lead the reader
through the theoretical and empirical background of the research problem, explain and
justify the research design and present results in the three major thrusts of the thesis,
that of micro-scale variation in composition and diversity, micro-scale variation in
structure and the role of spatial environmental heterogeneity in species diversity.
Specifically:

1.4.1 Chapter 1

5

This short chapter has introduced the aims and objectives of the thesis, and has
outlined the general structure of the thesis.

1.4.2 Chapter 2

Some background is provided on the theoretical and empirical context of the thesis,
reviewing the state of knowledge of key themes related to the objectives. Specifically
the chapter outlines the context of biodiversity and tropical forests, stressing the
importance of conservation and the high levels of diversity found in tropical forests.
Some information is provided on the role of tree species diversity to overall
biodiversity in tropical forests. A brief introduction on spatial variation in diversity is
provided along with an introduction to the equilibrium and non-equilibrium
hypotheses for species diversity maintenance in tropical forests. Specific attention is
given to the theoretical and empirical findings in relation to elevational gradients (of
particular relevance to the TMCF site) and habitat associations (a topic under
vigorous debate in relation to LRF).

A brief summary of interactions between

environment and tree diversity and structure is presented. The chapter finishes by
introducing the theme of spatial heterogeneity, covering both theoretical and
measurement aspects.

1.4.3 Chapter 3

Chapter three presents the research strategy and methodology, starting by introducing
the study sites. The chapter then provides information on the desk-based terrain
analysis of topography and environmental data. Although terrain analysis is not

6

discussed in chapter 2, theoretical issues related to the calculation of the specific
terrain characteristics used in the thesis are analysed in this chapter.

Field

methodologies are then discussed, and the methods for integrating field-based data
with the terrain characteristics are introduced. The chapter finishes by introducing the
methodology for quantifying spatial environmental heterogeneity.

1.4.4 Chapter 4

Chapter four starts the results and analysis section of the thesis, examining the
topographic controls on tree species composition and diversity. For each site (TMCF
and LRF) a brief description of diversity, composition and structure is given, firstly at
the site-level and secondly at the plot-level through between-plot comparisons
(Objective 1). Statistical analyses are then presented to identify the degree to which
composition is controlled by terrain characteristics (Objective 2). The chapter also
examines the topographic controls on species diversity (Objective 3). The chapter is
structured firstly by site, then by compositional and diversity patterns. The chapter
concludes by examining the differences between each site, and providing some
preliminary conclusions.

1.4.5 Chapter 5

Chapter five examines the tree structural variability at the two sites, examining both
the average structural characteristics for each site and each plot within each site, but
also the distribution of structural forms. Effort is made to identify the topographic
controls on structure through statistical analyses (Objective 4). Some analysis and

7

discussion is provided on the complex relationship between diversity and structure,
though this is limited in scope due to data restrictions. As in Chapter Four, this
chapter concludes by summarizing the key results, though conclusions are left for
Chapter Seven.

1.4.6 Chapter 6

Chapter Six exclusively examines the role of spatial environmental heterogeneity on
tree species diversity (Objective 5), through presentation of the results of the
environmental modelling and comparison with the plot data, for each site separately.
Multi-variate statistics are applied to the results of previous chapters to provide a
broader vision of spatial variability in diversity taking into account both terrain
characteristics and their spatial heterogeneity.

Chapter 7

Although preliminary conclusions are provided at the end of each analysis chapter (46), this chapter attempts to coherently present the most important findings and place
them within the context of current literature and theory presented in Chapter Two.

8

Chapter 2

2

2.1

Literature review

Introduction : The biodiversity and conservation context

The world’s biodiversity is being lost at a staggering rate, and conservation of
biological resources is now a challenge that all nations must strive to achieve, and is
being pursued through international treaties such as the Convention for Biological
Diversity (CBD). Though this thesis is not directly oriented towards conservation
objectives, it is important that any study of biological diversity be placed within this
context, as science plays an important role provision of accurate biological
information and in the search for practical solutions to biodiversity loss.

The overall goal of the Convention on Biological Diversity (Article 1, CBD, 1992)) is
“….to conserve and sustainably use biological diversity for the benefit of present and
future generations”. Under this agreement, which most countries have signed to, each
country is required to provide an inventory of biological resources within the country,
and to pro-actively ensure its conservation. Parties to the Convention are committed
to develop effective and scientifically sound protocols and methodologies to conserve
biological diversity, that is ecosystem, species and genetic diversity. In the field of in
situ conservation, the CBD calls on nations to:

9

“Develop, where necessary, guidelines for the selection, establishment
and management of protected areas or areas where special measures
need to be taken to conserve biological diversity.”
Article 8 (CBD, 1992)

Biological diversity is of importance to the world for numerous reasons. Firstly,
humans throughout history have used biological resources to for food, fibre and fuel.
Alonso et al. (2001) report that close to 30 percent of pharmaceuticals on the market
today have been developed from plants and animals. Some of these uses include
“antibiotics extracted from fungi and from sources such as the African clawed frog’s
skin, an extract of the saliva of the vampire bat is used in treatment for hardening of
the arteries, wild yams contain chemicals with anti-inflammatory properties and
ovarian and breast cancer treatments have been developed from the bark of the Pacific
yew tree found in the northwestern United States” (Alonso et al., 2001). Our entire
agricultural system is built upon biodiversity. Of some 270,000 described species of
plants, 30,000 are edible, and 7,000 of these have been actively cultivated over history
(FAO, 1997). A small percentage of the world’s biological diversity has actually
been screened for potential uses, and a potentially significant percentage of the world
diversity has not even been discovered or catalogued (Erwin, 1991; Johnson, 1995).
The extent to which biodiversity will provide solutions to new threats in the future is
unknown (Aylward and Gammage, 1992), especially as society faces an uncertain
environmental future in the face of rapid land-use and climatic change.

On top of

these tangible uses, biodiversity has been shown to be of importance for ecosystem
functioning (Naeem et al., 1994; Folke, 1996; Johnson et al., 1996; Fonseca and
Ganade, 2001; Wolters, 2001; Kennedy et al., 2002; Worm et al., 2002), and amongst

10

other things maintains global biogeochemical and hydrological systems. However,
the functional role of biological diversity is still not fully understood (Bengtsson,
1998; Wardle, 1999), and is the subject of considerable discussion.

Finally,

conservation of biodiversity is also considered important for what are generally
termed “spiritual” reasons, including that humans have a responsibility to conserve
the worlds biodiversity regardless of its significance or value (Ehrenfeld, 1988;
Wilson, 1992).

Despite this economic, functional and spiritual importance, little has been achieved to
halt or reduce what is being termed the sixth great biological extinction (Pimm et al.,
1995), and continued efforts are required to halt what is believed to be an irreversible
loss (Wilson, 1992).

Though this thesis does not directly address conservation issues with respect to
biodiversity, it is hoped that accurate spatial information on where diversity is found
will help provide more informed decision-making on biodiversity conservation. This
literature review focuses on patterns of variation of species diversity, composition and
structure of trees in tropical forests. Emphasis is made to highlight studies in tropical
forest environments, though important papers from other biomes are also presented.
In most cases there is greater volume of both empirical and theoretical studies for
lowland tropical forests compared with the less-studied tropical montane forest
environments, though the same can also be said for temperate versus tropical
environments.

11

The chapter begins by reviewing diversity in tropical forests, providing the context of
this study. Some discussion is provided as to the significance of tree diversity for
total forest biodiversity, followed by a more detailed introduction of tropical forests,
and specifically the two types of tropical forest studied here. The chapter then moves
on to the central issue of the thesis, that of micro-scale spatial variation in diversity
and composition, presenting both theoretical perspectives and empirical results. Some
of the abiotic driving mechanisms of diversity are also briefly discussed, though there
is an large volume of literature in this area that cannot be covered in detail within this
chapter. The chapter then moves on to discuss spatial variation in structural patterns
in tropical forests, with special reference to the important interaction between
structure and diversity, through partitioning of resources (principally light), though
literature on this is scarce. Finally, the chapter provides some background on the
concept of spatial heterogeneity, and its significance to diversity and composition in
tropical forests, and concludes with a discussion on sampling in tropical forests, an
important bridge to the following chapter of materials and methods. What is not
discussed here are general techniques for terrain analysis and environmental
modelling issues, as this is not the focus of the thesis rather these are tools which are
used here for the study of tropical forest diversity and structure. In general terms,
terrain characteristics are chosen and used in this thesis to represent partitioning of
essential resource, principally energy and water. Though some biological literature
presented in this chapter use topographic characteristics in the study of diversity,
composition or structure, these examples are few and often limited in scope. A
specific discussion of the terrain characteristics used in this thesis is reserved to the
methods chapter (Chapter 3), where not only the terrain analysis methodology is
presented but also a detailed analysis of the reason for its use, its potential biological

12

significance in terms representing essential resources, and the issues involved with its
calculation.

2.2

Diversity in tropical forests

Tropical forests cover approximately 7% of the global terrestrial surface, across the
continents of Asia, Africa and Latin America, and are thought to contain over half of
the world’s species (Wilson, 1992). There is a great deal of variability in tropical
forests around the globe, with a broad range of structures and species compositions.
Asian forests tend to be dominated by Dipterocarps (Whitmore, 1984) whilst in the
neotropics tropical forests are not dominated by any particular family though
Leguminosae are typically most common (Gentry, 1990), though within these broad
regions there is significant compositional variation.

Tropical forests contain

overwhelming levels of diversity compared with their temperate counterparts (Huston,
1994), though biomass is comparable or even less than in temperate forests
(Whittaker and Niering, 1975). Whilst 1-Ha in a particularly diverse temperate forest
may contain 15-20 tree species (greater than 10cm diameter at breast height (DBH)), a
tropical forest may contain over 200 tree species (Gentry, 1990; Huston, 1994;
Hawkins et al., 2003). Indeed, Whitmore et al. (1985) reports 233 vascular plant
species in just 100m2 of lowland tropical forest in Costa Rica, equivalent to one-sixth
of all flora of the British Isles.

Despite being a treasure trove for biological diversity (Wilson, 1992; Crawley, 1997),
and of great importance for the global climate system (Melillo et al., 1996), tropical
forests are also undergoing rapid loss through deforestation for agricultural expansion

13

and logging (Skole and Tucker, 1993; Melillo et al., 1996; Lambin and Ehrlich, 1997;
Maio Mantovani and Setzer, 1997; Read et al., 2003).

Deforestation rates vary

between regions and continents, with Fearnside (1990) using satellite remote sensing
(LANDSAT TM images) to approximate deforestation rates of 0.5% yr-1 in the
Brazilian Amazon (21,130km2 lost per year 1978 - 1989, of the total 4,000,000 km2 of
Amazonian forest). There is considerable uncertainty in monitoring the true rates of
deforestation across regions (Grainger, 1993), but most approximations range from
0.5 – 1.3 % per year (FAO, 1990).

The term tropical forest refers to all forests within the tropical region (typically
defined as between the Tropics of Cancer and Capricorn), including tropical dry
forests, tropical montane forests, tropical montane cloud forests and lowland rain
forests, amongst others. The physical environment for each of these types of forests is
different, resulting in different levels of species diversity, different species
compositions and different forest structures (Huston, 1994). A complete review of the
different types of tropical forest is a available in Burley et al. (2004). This study
examines diversity, compositional and structural patterns in a tropical lowland rain
forest (LRF) and a tropical montane cloud forest (TMCF), which are described later in
Section 2.2.2.

2.2.1 Relationship between tree diversity and biodiversity

This study only examines tree diversity in tropical forests, but the degree to which
tree diversity is related to total biodiversity (i.e. diversity in all other groups of
organisms) has been the subject of a number of studies. According to Huston (1994),

14

organisms can be sub-divided into two classes - structural and interstitial. Structural
organisms actually form the physical environment (examples include trees and corals)
and to a certain extent alter the physical surroundings (e.g. microclimate), whilst
interstitial organisms inhabit this environment (examples are insects, birds,
vertebrates, herbaceous plants etc.). Different processes and factors affect these two
types of organism, with structural organisms directly affecting the diversity of the
interstitial organisms. Ecological theory further predicts that loss of biodiversity at
the base of the ecosystem (basal species), for example plants, will impact on the
whole system (Siemann et al., 1998; Knops et al., 1999). Though theory clearly
emphasizes the importance of plant diversity (and especially tree diversity) on ‘total’
biodiversity, and many authors cite tree diversity as being strongly correlated with
diversity in other groups (Huston, 1994), empirical studies show conflicting findings
as to the validity of this relationship (Su et al., 2004). Ingerpuu et al. (2001) show
vascular plant species richness to strongly correlate with bryophyte species richness (r
= 0.76, p = 0.007) in 10 different stands in an Estonian boreo-nemoral moist forest.
Similarly, Aptroot (2001) found that an Elaeocarpus tree in montane forest in Papua
New Guinea showed the presence of over 200 species of ascomycetes, including 173
lichenized ascomycetes, many of which are thought to be host-specific. Nieder et al.
(2001) further discuss host-specificity with respect to epiphytes, providing a strong
case for a strong (one-way) relationship between tree diversity and epiphytic and
bryophytic diversity. There are other significant correlations in the literature between
plant species richness and butterflies (Simonson et al., 2001), arthropods (Siemann et
al., 1998), tetrapods (Barthlott et al., 1999), herbivores and predators and parasites
(Knops et al., 1999), amphibians, mammals, birds and reptiles (Currie, 1991) and
belowground biota (Wardle et al., 2004), though it is important to note that the

15

majority of these studies are not all in tropical forests. However, some studies fail to
find such relationships, for example between butterfly richness and plant species
richness in Madagascar (Kremen, 1992), and in the British Isles little coincidence
between hotspots in butterflies, dragonflies, liverworts, aquatic plants and in breeding
birds are found (Prendergast et al., 1993). In tropical forests specifically, Lawton et
al. (1998) found few relationships between changes in richness in one group of taxa
with changes in an array of other groups (taxa of butterflies and insects) under some
kind of habitat modification. Unfortunately this study did not include tree (or plant)
species. These findings amongst others lead Su et al. (2004) to conclude that crosstaxon relationships are site and taxon specific and no general rule can be extracted.
However, the significant levels of host-specificity to plant species identified in
tropical forests, for example in arthropods (Erwin, 1982; Erwin, 1991) or fungi
(Aptroot, 2001), make it highly likely that greater plant diversity also signifies to a
certain extent greater diversity in other taxa.

In conclusion it cannot conclusively be stated that high tree species diversity in
tropical forests indicates high diversity in other taxa, though there are undoubtedly
important processes whereby tree diversity should theoretically generate high
diversity in other groups. These mechanisms include host-specificity, the fact that
trees are a basal species in the ecosystem, and through their structural role in the
forest filter largely homogenous physical inputs (solar radiation, temperature and
rainfall) to create a heterogeneous under-canopy. Furthermore, evidence of directcycling of nutrients by tropical trees affects edaphic conditions immediately around
the individuals (Clinebell II et al., 1995) creating a mosaic of physical and chemical
soil conditions, broadening the resource-niche space and potentially encouraging

16

greater diversity in other taxa (especially belowground biota and herbs and shrubs).
Quantification of the cross-taxon congruence in tropical forests through coordinated
plot studies would be a significant achievement for tropical ecology, but for the
purposes of this thesis it is likely that tropical tree diversity is to a certain extent
representative of biodiversity as a whole.

2.2.2 Diversity and structure in the study sites

This study examines diversity, structure and composition of trees in two tropical
forests in the Northern Andes and Choco Biogeographic Region (Gentry, 1982), and
in the western Amazon (Andean fringe), both of which represent global conservation
priorities and contain very high levels of biological diversity.

The northernmost part of the Andes is subdivided into three mountain ranges that
generate a very complex geographical pattern of exceptional biological diversity
(McNeely et al., 1990; Stattersfield et al., 1997). The Colombian Andes are also the
most densely populated areas in the country (CIESIN, 1998), hence the last remnants
of Andean montane forests are priorities in the national (IaVH, 1997), and
international conservation agenda (Barthlott et al., 1999; Myers et al., 2000).
Colombia ranks among the Earth’s highest conservation priorities due to its
exceedingly high diversity coupled with the threats posed by human development and
commercial resource exploitation (Dinerstein et al., 1995). It is one of the richest
countries in terms of species diversity per unit area comprising an estimated 14-15%
of the planet’s flora and fauna in 0.77% of its surface (International, 1997).
Furthermore, Johnson (1995) ranked Colombia first in the world in terms of species

17

richness in birds (1,752 species) and butterflies (3,500 species), and second in terms
of plants (45,000 species) and amphibians (407 species). Since many plant and
animal species have not been documented, particularly in tropical forests (Claridge,
1995), the documented diversity of Colombia is expected to increase substantially
with further biological inventory work.

The western Amazon (bordering the Andean fringe in Peru, Ecuador and Colombia) is
also identified as one of 20 global biodiversity priorities (Myers et al., 2000). Pitman
(2000) established a network of plots in Ecuador and Peru, finding extremely high
diversity in tree species throughout the region, with other plot studies in the region
further confirming this (Duivenvoorden, 1994; Duivenvoorden and Lips, 1995;
Romoleroux et al., 1995). The reasons for the high levels of plant diversity found in
the Amazon regions of Ecuador, southern Colombia and Northern Peru are the topic
of heated discussion, with Prance (1982) proposing evidence of this region being a
Pleistocene forest refuge.

The two sites studied here represent two contrasting tropical rain forest types of the
Neotropics; tropical montane cloud forest (TMCF), and lowland rain forest (LRF).

2.2.2.1 Neotropical Tropical Montane Cloud Forests (TMCF)

Cloud forests are defined according to the presence of forest in areas of frequent or
persistent ground level cloud (Bruijnzeel and Proctor, 1995), and occur in both
temperate and tropical environments. Tropical montane cloud forests (TMCF’s) are
exclusive to mountains in tropical regions. Their altitudinal distribution is highly

18

variable depending on climatic factors, with TMCF occurring as low as 350m in
coastal regions where fog forms at lower altitudes, stretching as high as 3,500m in
inland mountain systems such as the Andes (Bruijnzeel and Proctor, 1995). TMCF
occurs in three regions of the globe; SE Asia, Central Africa and Central and South
America.

In 1970 they were estimated to cover 50 million hectares, but reported deforestation
rates in excess of 1.1% per year (greater than that for lowland rainforests) has caused
a serious reduction in their coverage, putting them at serious risk (Hamilton et al.,
1995). These deforestation rates are caused by land hunger and the spatial expansion
of agriculture, and are predicted to continue under the current population explosion in
many of the areas where TMCF’s occur (CIESIN, 1998).

Castaño (1991)

approximates that just 10% of Colombia’s original TMCF now remains. Furthermore,
there is now growing concern as to the effect of climate change on these ecosystems
(Loope and Giambelluca, 1998; Pounds et al., 1999; Still et al., 1999).

The hydrological significance of TMCF has been shown to be important for
downstream water resources (Bruijnzeel and Proctor, 1995), due to their constant
provision of streamflow even during the dry season (Zadroga, 1981), brought about
by considerable volumes of cloud interception throughout the year (Gonzalez, 2000).

Tropical montane cloud forests are renowned for the large numbers of endemic
species, highly dependent on their climatic environments (Bubb et al., 2004). This is
especially the case for plants (Kappelle, 2004; Leimbeck et al., 2004), and birds with
Long (1995) reporting that ten per cent of the world’s 2,609 restricted-range bird

19

species are confined to or mainly found in cloud forests.

TMCFs also contain

exceptional levels of diversity in epiphytes (Nieder et al., 2001), bryophytes and
orchids (Muller et al., 2003), with a high concentration of threatened plant species and
some important rare-agricultural wild relatives (Bubb et al., 2004). The best plot
based study of plant diversity in TMCF in the neotropics is a study by Clinebell II et
al. (1995) presenting 69 lowland and montane 0.1Ha plots established by Alwyn
Gentry around the entire continent, with 13 plots throughout the Andes above 500m
(ranging from 560m – 3000m) having an average of 109 species in the 0.1 Ha (DBH
> 10cm).

More specifically to the region around Tambito (an Andean site on the western slope
of the western cordillera, placing it within the Choco Biogeographc Region), plot
studies of plant species diversity in the Choco Biogeographic Region on the Pacific
Coast of Colombia and Ecuador confirm exceptionally high levels of diversity
(Gentry, 1990, 1995; Galeano et al., 1998). Specifically, Galeano et al. (1998) studied
all vascular plant species in 0.9Ha of forest (in 12 transects) finding a record 970
species for that sampling scheme, with Araceae, Cyatheaceae, Piperaceae and
Rubiaceae being the most dominant families. In all, there are thought to be some
8,000 vascular plant species in the Choco Biogeographic Region (Forero and Gentry,
1989), though large tracts of the forest are unknown to science, especially in the
exceptionally wet (and remote) parts (Galeano et al., 1998).

Tree height in TMCF is generally less than in the lowlands (Table 1), with reported
canopy heights of 5-33m in TMCF compared to 25-65.4m in LRF (summarised in
Letts, 2003), though emergent trees still reach great height in TMCF, with Kappelle

20

(2004) reporting emergent trees of 33-55m height in Costa Rica. Most studies report
stem density to increase with altitude, with Kappelle (2004) reporting average stem
densities for TMCF in the region of 500-1000 stems Ha-1 for TMCF counting all stems
with DBH > 5 cm, and 1600 – 3500 stems Ha-1 when all individuals with DBH > 2.5cm
are counted.

Table 1 Structural parameters for LRF and TMCF, from Letts (2003), based on studies by

Richards (1996), Kitayama and Aiba (2002), Hafkenscheid (2000), Heaney and Proctor
(1989), and Leigh (1999) and this study.

2.2.2.2 Tropical Lowland Rainforest (LRF)

Tropical lowland (evergreen) forests occur in perhumid lowland climates where water
stress is either intermittent or non-existent, typically below 1200m elevation

21

(Whitmore, 1998). LRF is principally found in the Neotropics, in the Amazon region
and along the Pacific coast, though there are also lowland rain forest regions in
Africa. In the Amazon region, with approximately 4,000,000 km2 of LRF, the forest
is often further split into vegetation types related to flooding frequency, with terra
firme forest found on higher relief not subjected to flooding and várzea forest which is
seasonally flooded, sometimes for extended periods of time.

Based on Alwyn Gentry’s 0.1Ha plots established throughout the Neotropics, an
average of some 149 species are found per plot in the lowlands (< 500m elevation),
with all individuals with DBH > 10cm (Clinebell II et al., 1995). This is considerably
higher richness than was found in montane plots. In and around the Yasuni National
Park region in Ecuador, there is documented high diversity in a 25-Ha plot (Valencia
et al., 2004), and at the time of publication in 1994 this area had the highest recorded
plant diversity per hectare in the world (based on the sampling method), with 473
species (DBH >5cm) found in a 1-Ha square forest plot, with Fabaceae, Lauraceae
and Sapotaceae being the most dominant families (Valencia et al., 1994).

Tree height in LRF is typically greater than in TMCF, with a higher canopy, lower
stem density and greater basal area. Clark and Clark (2000) report stem densities of
462 – 504 stems Ha-1 in La Selva in Costa Rica (30m elevation) counting all
individuals DBH > 10cm. Webb et al. (1999) report average canopy height of 14.9 –
22.9m, stem densities of 550 – 945 stems Ha-1, and average DBH of 23.5 – 25.6cm in
a lowland rain forest counting all individuals DBH > 10cm in American Samoa (48m
– 339m), with Proctor et al. (1983) reporting similar densities of 615 – 778 stems Ha-1
(DBH > 10cm) for an LRF in Sarawak, with basal area ranging from 28 – 57 m3 Ha-1.

22

2.3

Spatial variation of tree diversity and composition in tropical forests

Spatial variation in species composition and diversity occurs at a range of scales, from
global to regional to the micro-scale, and the processes behind such variability are
different depending on the scale in question (Condit et al., 1992). On the global scale,
latitude is often cited as explaining the strongest global-level spatial variation in
diversity, with a general decline in diversity with greater latitude, though latitude
itself is not the physical driving mechanism, but more specifically climate is amongst
other factors (discussed in detail in Hawkins and Diniz-Filho, 2004). Within tropical
forests, that have the highest levels of diversity on the planet, there is also a
significant degree of spatial variability in diversity and composition at a range of
spatial-scales, from the continental scale (for example Gentry, 1990) down to the
regional scale (for example Gentry, 1995; Kress et al., 1998) and the local scale
(discussed in Rosenzweig, 1995). Within this section the local-scale drivers of spatial
variation of diversity and composition will be reviewed, as this is the scale relevant to
this study.

However, spatial patterns in diversity and composition cannot be discussed without
the important consideration of the temporal dimension.

The spatial pattern of

diversity observed today is at least in part a function of temporal history over a range
of timescales (Smith and Huston, 1989; Huston, 1994). In tropical forests, temporal
variation in the context of large spatial and temporal scales is discussed in depth with
respect to refuge theory (Gentry, 1982; Prance, 1982), whereby it is hypothesized that
the current distribution of plant diversity hotspots is brought about by long-term

23

climatic stability in these zones (though significant evidence is now available that
refutes this theory, but is not discussed here), both in the Amazon region (Prance,
1982) and the Choco Biogeographic Zone (Gentry, 1982). Temporal variability at
this broad-scale is of little relevance to this study, but the micro-scale spatial
variability in diversity central to this thesis is a function of shorter time-scale
dynamics (Condit et al., 1992), principally gap dynamics and succession, though
broader-scale disturbance regimes (such as drought (Condit et al., 1992), or tropical
storms) may also impact the micro-scale configuration of species composition and
diversity.

There are two over-arching hypotheses for the maintenance of high levels of species
diversity through time in tropical forests, eloquently described by Denslow (1987)
asking whether tropical forests comprise of “sets of highly coevolved nichedifferentiated tree species in stable or semi-stable floristic assemblages”, the so-called
equilibrium hypothesis, or if tropical forests consist of “diffusely coevolved, broadly
generalist species which slowly drift in relative abundance within a few large lifehistory guilds”, the so-called non-equilibrium hypothesis. Hubbell (1979; Hubbell
and Foster, 1986) has been the main proponent of non-equilibrium processes acting to
maintain species diversity. The non-equilibrium hypothesis, or "null" model of tree
dynamics holds that “species-rich tropical tree communities are maintained on a local
scale by a balance of extinction and immigration, and on a global scale by a balance
of extinction and speciation” (Yu et al., 1998). All species are considered ecologically
equivalent, such that species having equal initial abundances have equal probabilities
of extinction or fixation. Also key to Hubbell’s null model is disagreement over the
existence of density-dependence in tropical forests (Janzen, 1970; Connell, 1971), and

24

the implication that species on the whole are adaptively equivalent.

Density

dependence (the so-called Janzen-Connell model in tropical forests) dictates that
individuals of the same species are likely to perform poorly around each other, due to
higher concentrations of host-specific natural enemies (fungal pathogens and
herbivores) and as they compete for the same resource-niche, subsequently
diminishing growth and survival. Advocates of the non-equilibrium theory for the
maintenance of species diversity have provided empirical data rejecting the existence
of density-dependence in tropical forests (Hubbell, 1979; Condit et al., 1992).
Equilibrium based explanations for species diversity centre on the concept of niche
differentiation in tropical species, first advocated by Ashton (1977), and since
supported by studies such as that of Phillips et al. (2003), and support the existence of
the Janzen-Connell model of density-dependence.

The validity of the non-

equilibrium hypothesis has been questioned, with Terborgh et al. (1996) providing
evidence from Peru of strong density-dependence.

In his study, Terborgh and

colleagues test the idea that in a non-equilibrium state, a disturbed forest would not
necessarily regenerate to its former composition. Other authors have since provided
further evidence supporting a certain degree of determinism in species composition in
tropical lowland forests (Pitman et al., 1999; Pitman, 2000; Pitman et al., 2001), and
an extensive and vigorous discussion of equilibrium vs. non-equilibrium hypotheses is
present in the literature (see for example Yu et al., 1998). Currently there is strong
empirical evidence supporting both equilibrium and non-equilibrium hypotheses, and
tropical ecology continues to rigorously debate the issue, though some accept that a
combination of both mechanisms are at play in tropical forests (Brokaw and Busing,
2000). In an attempt to examine the existence of equilibrium or non-equilibrium
dynamics in a tropical rain forest in Malaysia, He et al. (1996) found no clear spatial

25

pattern in diversity but concluded that a large amount of undetermined variation may
originate from small-scale processes (<20m). This thesis in part attempts to provide
more empirical evidence to fuel the equilibrium vs. non-equilibrium discussion,
through testing of habitat associations and spatial partitioning of diversity and
composition in different topographically controlled niches at the micro-scale.

Whichever theory for the maintenance of species diversity in tropical forests is taken,
it cannot be denied that forests change considerably in both the long- and short- term.
Condit et al. (1992) provides data on short-term dynamics in species composition and
diversity in a 50-Ha plot in Barro Colorado Island, showing that although species
richness remains unchanged some 40% of the species (approx. 300 in total) changed
in abundance by more than 10% in just 3 years.

Temporal dynamics must be

carefully considered throughout this study, as measurement of diversity at any one
time is just a snapshot of a temporally dynamic system.

Specifically to lowland tropical forests, science is only beginning to understand in any
detail how variable species composition and diversity is over space (Pitman, 2000). It
has long been established that tropical tree species tend to occur in low densities in
Amazonian forests, with Black et al. (1950) stating that the average density of an
Amazonian tree species is less than 1 individual Ha-1. Over the years this figure has
not changed much despite intensive plot studies throughout the region, with Pitman et
al. (2001) concluding that the commonest species rarely exceed 90 individual Ha-1
and often account for just 3-12% of all stems. This also means that a 1-Ha plot is
likely to capture a small-subset of all species in the local pool, and through lack of

26

large enough datasets ecologists are only just beginning to understand the geographic
patterns of species distributions in the Amazon (Pitman, 2000).

Within this section, two specific themes of geographic variation in composition and
diversity are examined in detail, namely the role of elevational gradients on diversity
(particularly relevant to TMCF environments) and habitat associations of tropical tree
species (elevational gradients in composition).

2.3.1 Elevational gradients in diversity

Elevational gradients of diversity have been the subject of many studies, as they
represent a natural environmental gradient and a perfect opportunity for examining
the factors and processes that generate and maintain species diversity (Sanders, 2002).
Adiabatic-lapse rates dictate that temperature decreases with elevation through a
lowering of atmospheric pressure, but elevational gradients also often represent
gradients in wetness (through both rainfall, accumulated soil moisture, fog inputs and
reduced evapotranspiration) (Whittaker and Niering, 1975).

Elevational gradients are often seen as a mirror of the latitudinal gradient in richness
(Stevens, 1992), with global patterns in species richness often being described by a
decrease in richness with greater latitude (Hawkins and Diniz-Filho, 2004), and at any
single latitude a decrease in richness with elevation (Rahbek, 1997). This simplified
view of elevational gradients however is not supported by all studies (Lees et al.,
1999; Colwell and Lees, 2000; Sanders, 2002), and considerable literature is available

27

on both the empirical patterns (Terborgh, 1977; Vasquez and Givnish, 1998; Grytnes,
2003) and the theoretical explanation of these patterns (Stevens, 1992; Rahbek, 1997).

In general terms, two patterns in richness have been identified along the altitudinal
gradient for different habitats and for different taxa: a monotonic decrease in species
richness with elevation (Terborgh, 1977; Vasquez and Givnish, 1998) and a unimodal
hump-shaped (the shape of a normal distribution) mid-elevation peak in richness
(Whittaker and Niering, 1975; Lees et al., 1999; Colwell and Lees, 2000; Nor, 2001;
Kessler, 2002; Vetaas and Grytnes, 2002; Brehm et al., 2003), with Grytnes (2003)
finding both patterns in seven transects in Norway.

Several explanations are provided as to the mechanism behind such relationships.
The influence of area on species richness is stated to be “one of community ecology’s
few genuine laws” (Schoener, 1976).

Indeed, species-area relations have been

discussed for nearly a century in the ecological literature (Gleason, 1922), and
recently reviewed by Lomolino (2000). Many authors have examined the effect of
area specifically on species richness in elevational gradients, finding it to be important
(Whittaker and Niering, 1975; Stevens, 1992; Rahbek, 1997).

In simple terms,

assuming that an elevational gradient is representative of a floristic gradient, greater
area at any elevation would increase the probability of higher species richness.

Some authors also contend that mid-elevations are habitat-sinks due to the high
probability of range-overlap (termed the mid-domain effect), because species ranges
are bounded by the highest and lowest elevation possible in the region (Colwell and
Lees, 2000).

28

Rapoport’s rule (originally conceived in the context of the latitudinal gradient, but
adjusted to the elevational gradient) interprets elevational patterns in richness based
on the observation that there is greater temporal climatic variation in higher
elevations, Stevens (1992) and subsequently Blackburn and Gaston (1996), advocated
that species elevational ranges increase with elevation.

The result, according to

Stevens (1992) is a monotonic decrease in richness with elevation.

Others liken the elevational gradient simply to productivity, and contest that patterns
in species richness are merely a function of the productivity gradient, with different
productivity versus diversity theories explaining the monotonic (for example
Hutchinson, 1959; Macarthur, 1965) and hump-shaped patterns in species richness
(for example Tilman, 1982; Rosenzweig, 1995).

Empirical evidence of the elevational gradient in richness in tropical forests is
variable, with Kappelle (2004) reporting that the richness in flowering epiphytes
decreases with increasing altitude, whilst pteridophytic plant groups (particularly
ferns and mosses) increase in richness with altitude, whilst Vasquez and Givnish
(1998) find a gradual decrease in vascular plant richness from lowland tropical dry
forest to tropical montane cloud forest in Mexico. Based on the Clinebell et al. (1995)
compilation of 0.1Ha tree plots distributed throughout the neotropics, no clear
elevation pattern was found, although high elevation plots tended to have lower
species richness.

29

Tests of the theoretical controls on elevational gradients in richness conclude that
multiple factors determine the patterns. Sanders (2002) finds a mid-elevation peak in
diversity in ant species richness in three US States, attributing the patterns to both
species-area effects and Rapoport’s rule. Similarly, Rahbek (1997) tested four null
models (Rapoport’s, monotonic productivity, hump-shaped productivity and
area/geometric constraints) using bird diversity in South America as an example,
rejecting in this case Rapoport’s rule and the monotonic productivity model. In
reality, evidence suggests a large degree of site-specificity in richness and elevational
gradients, and patterns are likely a cause of a combination of processes, including
species-area, productivity and Rapoport’s rule. Mid-elevation peaks tend to be more
common than monotonic patterns, but there are few clear examples of either for
vascular plant species richness in elevational gradients in the tropics.

2.3.2 Habitat associations among tropical forest tree species

A central question in understanding the geographic distribution of tropical rain forest
tree species, and indirectly to the maintenance and spatial variation in species
diversity, is the degree to which abiotic or biotic factors restrict or limit fine-scale
geographic distribution.

Lowland rain forest species have historically been

considered to be generalists (Pitman, 2000), though some authors have suggested
edaphic specialization in lowland rain forest tree species (for example Clark et al.,
1998; Palmiotto et al., 2000). Central to equilibrium-based explanations of tropical
forests diversity is the concept of niche-differentiation, and this has been tested
through studies of habitat association presented in this section.

30

In LRF, there are two clear scales of habitat association in tropical tree species
(Pitman, 2000). There is little doubt of the clear compositional differences between
flooded (varzea) and non-flooded forest (terra-firme) (Phillips et al., 2003) brought
about by different degrees of flooding tolerance of the species present (Pitman et al.,
2001).

However, there is a more controversial debate on the degree of habitat

association amongst tropical tree species within these broad forest classes.

In the 1960’s and 1970’s some studies, particularly from Dipterocarp forests in the old
world, presented evidence that fine-scale species distributions in tropical forests are
driven by habitat preferences (Ashton, 1964; Wong and Whitmore, 1970; Austin et
al., 1972; Ashton, 1976).

A number of subsequent studies examined similar

relationships (Newbery et al., 1986; Baillie et al., 1987). In most of these cases parent
material was identified as the most important physical factor driving fine-scale
distribution patterns (Austin et al., 1972; Baillie et al., 1987), though not all studies
found evidence of strong habitat preferences in tropical forest species (Wong and
Whitmore, 1970; Newbery et al., 1986).

More recently a resurgence of the issue has resulted in a half dozen of focused studies
published over the past 5 years. The majority, based in Neotropical lowland forests
(except for Webb and Peart (2000; Debski et al., 2002; Potts et al., 2002; Palmiotto et
al., 2004)), have looked in detail at habitat associations within forest types in search
of explanation of micro-scale variability in tree composition and diversity. Indeed,
the concept of habitat heterogeneity and tree species specialization has become central
to theories explaining the maintenance of tropical tree species diversity (discussed in

31

detail in Pitman, 2000; Harms et al., 2001), and especially related to the nonequilibrium theory of Hubbell and Foster (1986).

Tuomisto et al. (2003) provide three useful models for evaluating micro-scale tree
species distributions and habitat associations in LRF, terming them the random-walk,
uniformity and the environmental heterogeneity model. The random-walk model is
based on a simple view of the non-equilibrium hypothesis proposed originally by
Hubbell and Foster (1986), whereby species are randomly distributed across the
landscape, but spatially-autocorrelated due to dispersal limitations.

Stochastic

processes, such as tree-fall, principally drive this model of tree species distribution.
Under this model, species presence is not indicative of presence of any other species,
and floristic similarity is expected to decrease with inter-site difference.

The

uniformity model (or more accurately “more-or-less” uniformity) is based on Pitman
et al. (1999; 2000; 2001) and Terborgh et al. (1996), with generally constant species
composition over wide-areas. Most species are expected to be widespread, and their
presence at a site is non-random, but due to a biological characteristic which makes
them compete successfully and dominate over a wide range of forest tracts (Pitman et
al., 2001). After a disturbance, a similar species composition is expected to recover
(Terborgh et al., 1996). In this sense, species composition is expected to vary little
with local site conditions, and spatial auto-correlation is not necessarily present. The
final model, environmental heterogeneity, advocated by (Poulsen and Balslev, 1991),
and the variable composition brought about by non-random, spatially auto-correlated
environmental heterogeneity. In the case of this study, Tuomisto et al. (2003) rejects
the uniformity model, lending support to a combination of the random walk model

32

and the environmental heterogeneity model, concluding that it is edaphic variability
that most strongly segregates the species.

These three models for examining habitat associations in tropical trees present a
useful means of evaluating the processes behind specialization or generalization in
tropical species, capturing contemporary theories of equilibrium, non-equilibrium and
combined dynamics.

Harms et al. (2001) found that 82% of species in a 50-Ha plot in Barro Colorado
Island showed no habitat association with basic landscape units, concluding that
specialisation plays a limited role in spatially distributing species composition and
diversity, lending support to the “uniformity” theory advocated by Pitman et al.
(1999). Valencia et al. (2004) find some habitat association related to landscape units
in a 50-Ha plot in Yasuni National Park in Ecuador with 25% of species showing
habitat specialisation, but conclude in favour of the random-walk model attributing
the environmental heterogeneity effect on tree species distributions as only minor.
Webb and Peart (2000) lend support to the environmental heterogeneity model,
through examination of habitat associations in 45 species in a mixed dipterocarp
forest in Borneo, finding 44% of species to have habitat specialization to light and
physiographic conditions, but still they fail to account for fine-scale distributional
patterns in the remaining 56% of species. Phillips et al. (2003) perform a more
rigorous analysis of habitat association amongst 849 Amazonian species across 88
plots, focusing on edaphic constraints to species distribution and finding that 77% of
species are significantly related to edaphic habitat, lending strong support to the
environmental heterogeneity model. Mantel tests are used in this case to examine

33

distance related floristic similarity (central to the random-walk model), finding only a
weak distance related effect.

Vormisto et al. (2000) also found strong edaphic

specialization in four plant groups (melastomes, palms, trees, and pteridophytes), also
using Mantel tests, though in a later paper (Vormisto et al., 2004) the degree to which
palm distributions are related to topography is questioned.

Clark et al. (1999)

approximate that 30% of species in La Selva, Costa Rica, have edaphically biased
distributional patterns, and Potts et al. (2002) find 43 out of 60 species studied to have
some degree of edaphic specialization, both of these cases providing a strong case for
the environmental heterogeneity model.

There are fewer studies and much less theoretical background available for species
habitat associations in TMCF environments, though limitations to species ranges are
likely to be far clearer due to the sharper environmental gradients. Vasquez and
Givnish (1998) examined elevational gradients in family composition, showing clear
gradients in abundances for certain families. Sugden (1982) presents similar evidence
of altitudinal zonation for some selected tree species in a Colombian cloud forest, also
showing clear patterns of elevational limits to distribution.

Furthermore, many

botanical monographs and species descriptions present elevational range as a factor in
describing species distribution. Though these studies are few, there is little doubt that
elevational gradients represent strong floristic gradients too, however no studies
examine habitat association at the fine-scale based on topographic characteristics in
TMCF environments.

In conclusion, there is no single model for habitat association in lowlands that stands
out as the dominant process, with different results for different sites and different

34

habitat measures (edaphic or landscape/topographic), though the more rigorous and
data rich studies (namely Phillips et al., 2003) find clearer patterns of specialization in
tropical tree species, especially related to edaphic conditions. Attention however is
brought to the possible role of direct-cycling by trees, whereby tree species
themselves control in some part the edaphic conditions around them through litterfall,
suggested by Clinebell et al. (1995).

However, none of the studies of habitat

association for LRF and TMCF presented in this review use DEM-derived
topographic variables beyond simple elevation, slope, or landscape unit. In most
cases, landscape unit is defined through a combination of rules of elevation, slope and
convexity, none of which are calculated in the context of a GIS (for example Valencia
et al., 2004). This lack of quantified terrain analysis is clearly a research gap in
tropical studies, despite an early paper applying GIS terrain analysis to explain
compositional variation in a temperate forest in West Virginia (Twery et al., 1991).

2.3.3 Environmental controls on diversity

Whittaker (1999) states that ‘most ecologists have to contend with systems of
bewildering complexity, in which it is hard to separate the wood from the trees’, and
nothing could be truer in the context of discussing environmental interactions with
diversity.

There is a massive amount of published material that discusses

environmental interactions with diversity, covering many different ecosystems and
many different taxa. Effort is made in this section to maintain the literature review
brief and focused on micro-scale environmental drivers of tree or plant diversity in
tropical forests. In general terms, readers particularly interested in this section are

35

strongly urged to consult Givnish (1999), who provides a comprehensive review of
environmental interactions with diversity in the context of tropical forests.

The discussion presented here is separated into diversity relations with climate and
productivity, light environment and gap dynamics, other types of disturbance, soil
quality and finally topography and topographic heterogeneity, though in reality these
distinctions are fairly arbitrary as many of these factors are related or act in
combinations to affect plant diversity (see for example discussion on the combined
effect of productivity and disturbance on diversity in Kondoh, 2000).

2.3.3.1 Climate and productivity

The productivity-diversity relationship is denied by few, but there is heated debate as
to the shape of the relationship, with a number of very detailed reviews on the subject
(Maranon and Garcia, 1997; Givnish, 1999; Waide et al., 1999; Schmid, 2002).
Productivity strictly speaking in this context is defined as the ‘rate of conversion of
resources to biomass per unit area’ (Waide et al., 1999), though many authors
examine the relationship in the context of water or energy gradients as surrogates of
productivity (see Hawkins et al., 2003 for an example).

Waide et al. (1999) present a comprehensive survey of literature on the productivitydiversity relationship, finding that 30% of 200 related articles report a unimodal
pattern, 26% a monotonic increase, 12% a monotonic decrease and 32% of articles
found no significant pattern at all.

But in general, the contemporary literature

strongly supports the unimodal relationship, with Weiher (1999) going as far as

36

suggesting it may become an “ecological law” alongside the species-area law of
Schoener (1976). However, the debate is far from over, and field-based results
showing the contrary cannot be overlooked. Chase and Leibold (2002) contest that
the different patterns found are a function of scale, and that the relationship changes
shape depending on the scale of study, with a unimodal relationship at the local scale
and a monotonic increase at the regional scale. This is also supported by Weiher
(1999), and has been strongly supported by Waide et al. (1999).

Perhaps the most comprehensive and frequently cited regional scale test of the
productivity-richness relation has been provided by Currie (1991), who found a strong
monotonic increase in tree species richness along a potential evapotranspiration (as a
surrogate of productivity) gradient across the USA. However, specifically to tropical
forests Waide et al. (1999) report that there are few examples of clear relationships
between productivity and diversity in tropical forests.

The clearest example of

climate-diversity relations in tropical forests is that of the high diversity found in
regions of high precipitation at both the local scale (Huston, 1980; Phillips et al.,
1994) and regional scale (Gentry, 1982; Clinebell II et al., 1995; Gentry, 1995).
Wright (1992) also contends that greater seasonal variability in precipitation reduces
plant diversity. No other clear patterns have been found.

There are a plethora of explanations as to why a productivity-diversity relationship
exists. Those that refer to both local scale productivity gradients and diversity include
competitive exclusion and environmental stress (Huston, 1979, 1994), changes in
environmental heterogeneity with productivity and plant abilities to resist predation
(Givnish, 1999; Waide et al., 1999), though there is no agreement on a robust

37

explanation, and patterns and processes tend to vary on a site-by-site basis (Givnish,
1999). Because of the inconclusive nature of the literature and high degree of sitespecificity evident, this is not discussed here in great detail, though readers interested
in more in depth presentation and discussion of the various hypotheses to explain the
productivity-diversity relationship are referred to Waide et al. (1999).

2.3.3.2 Light environment and gap dynamics

Disturbance in ecology is defined as ‘any relatively discrete event in time that disrupts
ecosystem, community, or population structure and changes resources, substrate
availability, or the physical environment’ (Pickett and White, 1985). Disturbance in
tropical forests includes herbivory, predation, climatic variability, but perhaps the
most common and controversial forms of short-term disturbance in tropical forests are
those of tree-fall or branch-fall, resulting in an environmental and ecological response
referred collectively as gap dynamics (Denslow, 1987).

Prior to discussing the implications of gap dynamics on tree species diversity, some
background is provided on the process of gap formation in tropical forests. In a large
census of gaps in 12-Ha of primary forest in French Guiana, Van Der Meer and
Bongers (1996) found 1.5% of trees (DBH > 10cm) fell in a single year, with 1.3% of
individuals losing a large branches. Some 4.9% of the standing basal area was found
to be damaged or felled in a single year.

Key characteristics in studying gap

dynamics include the gap size, orientation, frequency (i.e. temporal variability) and
distribution (i.e. spatial variability). Hubbell et al. (1999) contend that gaps occur
randomly in both space and time, though some empirical evidence concludes the

38

contrary. Salvador-Van Eysenrode et al. (2000) conclude for an Ecuadorian western
Amazonian rain forest that gaps form stochastically in time, but are spatially clustered
(in TBS – the same study site as the thesis). The spatial clustering contributes to
increased wind turbulence around existing gaps, soil properties, topography (also
confirmed by Poorter et al., 1994), and possibly biotic factors such as fungal attacks
(Uhl et al., 1988). Nelson et al. (1994) also finds tree falls to be non-stochastic in
space, this time using remote sensing to map the frequency of large blowdowns in the
Brazilian Amazon, though the study detects broad-scale variability in tree fall due to
storm occurrence and rainfall. Though this scale of variability is not relevant to the
micro-scale discussed here, the implication of higher frequency in areas of greater
rainfall may mean that on the local-scale tree falls are more frequent in areas of higher
soil moisture. The micro-scale variation in the spatial distribution and nature of gaps
with relation to topography has been studied by Gale (2000) in the Pacific lowlands in
Ecuador, finding higher concentrations of gaps in upper- and mid- slopes, with the
lowest occurrence of gaps on ridges.

The reasons for which were principally

attributed to wind, with leeward slopes being most exposed. Given these observed
patterns, topography is a key variable that affects (through direct and indirect
mechanisms) the probability of a tree-fall (Gale, 2000), and so some of the terrain
characteristics developed in the next chapter capture elements of gap dynamics.

The effect of a gap on the light environment is dramatic. Under closed-canopy
conditions, the understorey typically receives only 1% of full sunlight in energy
receipt, with a 200m2 small gap receiving 9% full sunlight and a large 400m2 gap
receiving 25-35% full sunlight (all in terms of photosynthetic photon flux density –
PPFD) (Chazdon and Fetcher, 1984). This increase in light intensity is crucial for

39

germination of light demanding seedlings and saplings (Denslow, 1987; Meer et al.,
1998; Dalling et al., 1999). It also provides a diverse set of light-related niches, with
both vertical and lateral penetration of light to the sub-canopy and understorey
(Terborgh and Mathews, 1999). Tomlinson (1987) describes many of the types of
tree architecture existing in tropical forests including monolayer or deep cylindrical
crowns, and leaf angles and orientations, many of which are adaptations related to
light receipt. Some species are clearly better adapted to specific light environments
than others, with some studies showing this to be the case through focused
comparisons of specific species, examining their morphological and physiological
strategies for surviving in different light environment (Dalling et al., 1999; Terborgh
and Mathews, 1999). Aside from gaps, canopy structure and topography has also
been shown to affect the light environment, with Tateno and Takeda (2003) showing
higher levels of light availability at the forest floor on steeper slopes, due to a
combined effect of vertical and lateral penetration of the canopy.

Gap dynamics are central to many contemporary theories of species diversity
maintenance in tropical forests. Recent literature on non-equilibrium dynamics in
tropical forests has focused on gap dynamics, arguing that this process, stochastic in
both space and time, results in chance occupants rather than the best adapted species
recruiting in the micro-environment (due to dispersal limitations). Hubbell et al.
(1999) first coherently presented this theory, providing compelling evidence of nonspecies specific recruitment in gaps in a 50-Ha plot on Barro Colorado Island, as well
as undifferentiated levels of vascular plant species richness in gaps and in closed
canopy forest.

40

Hubbell’s paper once again stirred an intense discussion on equilibrium vs. nonequilibrium maintenance of diversity, excellently reviewed by Brokaw and Busing
(2000). The resultant effect on species diversity if the non-equilibrium theory is
accepted is one of relatively little variability (i.e. equal levels of diversity in both
gaps, early regeneration and regenerated forest, although this is only observed when
diversity is measured as richness per stem). Conversely, the opening of a greater
diversity of niches along the light gradient from the center of the gap to the
surrounding canopy-covered understorey (often referred to as the gap-understorey
continuum) means that gaps generate higher levels of species diversity under the
equilibrium theory of niche-partitioning (Brokaw and Busing, 2000).

However, the intermediate disturbance hypothesis (IDH) is also an additional aspect
of gap dynamics central to theories of tree species diversity. The IDH was first
proposed by Connell (1978) for both tropical forests and coral reefs, and was
reviewed more recently with special reference to tropical forests by Sheil and Burslem
(2003). The hypothesis dictates a hump-shaped unimodal curve of diversity along a
disturbance gradient, measured in terms of frequency or intensity. Connell (1978)
suggests a successional process, whereby colonizing species (good dispersers, shade
intolerant and fast growing) at first establish in the gap.

The second stage of

succession involves the slower establishment of shade tolerant species, which
compete with the colonizers for energy and water in a state of coexistence. The final
stage involves the competitive exclusion of the colonizers, and the dominance of
shade tolerant species. Connell suggests that the middle successional stage is most
diverse through the existence of both colonizers and shade-tolerant species, therefore

41

sites with intermediate levels of disturbance frequency and intensity tends to maintain
the greatest number of species.

In a survey of 197 published works making empirical tests of the IDH, Mackey and
Currie (2001) find that the majority of studies find no significant relation at all of a
unimodal pattern of species richness along the disturbance gradient (36%), with just
17% confirming a mid-disturbance peak. This study included a broad range of taxa
and ecosystems, but for the tropics there are very few examples. Hubbell et al’s
(1999) model of the effect of gap dynamics on species diversity rejects the IDH,
despite both models being marked as non-equilibrium models. But a more recent
study shows very strong empirical evidence for the existence of a peak in diversity
along a disturbance gradient in tree falls in Guiana (Molino and Sabatier, 2001). In
their review, Sheil and Burslem (2003) discuss in detail these two contrasting studies
and actually suggest that Hubbell’s results are consistent with the IDH, with sitespecific conditions leading to the confused pattern. In conclusion, more large-scale
and long-term studies are required to further understand the role of the IDH, but there
is fairly compelling evidence of at least some variability in tree diversity being
explained by gap dynamics and the IDH.

Putting gap dynamics to one side, there is also the important issue of light quality in
tropical forests, especially in montane environments. Letts (2003) discusses in depth
two aspects of light quality; photosynthetically active radiation (PAR) and ultra-violet
radiation (UV). PAR refers to the frequencies of solar radiation used by plants in
photosynthetic production, and is not a constant percentage of total solar radiation,
varying depending on atmospheric attenuation through cloud cover (Letts, 2003).

42

Spatially varying levels of PAR therefore have potentially important implications for
composition, diversity and structure. For example, in terms of structural implications,
Flenley (1992) observed that light-demanding plants grown in high humidity and
lowered PAR are tall, with long internodes and large, thin, pale green leaves.

Ultra-violet radiation can cause damage to macromolecules (Letts, 2003).

For

example, Teramura (1983) showed that seedling growth is stunted following exposure
to UV-B radiation. Ultra-violet radiation is important in montane forests, as TMCF
vegetation is exposed to higher UV-B radiation than any other forest biome as a result
of the short path-length of incident radiation (Ziska, 1996) and the reflection from
clouds below, which may increase UV-B radiation by up to 70% (Flenley, 1995).
Caldwell et al. (1980) finds that ultra-violet radiation levels increase substantially
with altitude.

Numerous studies have indicated plant adaptations to high UV

radiation levels (Sullivan et al., 1992; Ziska et al., 1992; Ziska, 1996), implying that
high levels of UV radiation may have an important effect on tree composition and
structure specifically. The degree to which light quality (PAR and UV-B) affects
species diversity and composition has not been studied.

2.3.3.3 Other disturbances

Although gap dynamics are undoubtedly the dominant micro-scale disturbance
process in tropical forests (Denslow, 1987), other temporal disturbances have been
observed to affect species composition and diversity. Condit et al. (1992) provide a
detailed account of the effects of drought on Barro Colorado Island, causing
widespread mortality in species with different degrees of drought tolerance. Although

43

the disturbance was equal in severity across the whole region, spatial heterogeneity
and non-homogeneous distribution of species likely means that the drought may affect
change in the spatial distribution of diversity on the micro-scale, although this was not
studied. Another type of disturbance of great importance to lowland rain forests is
that of flooding, caused by river level fluctuation of as much as 14m, which can
submerge large tracts of forest for periods from a few days to 270 days (Junk et al.,
1989). Ferreira (1997) finds species richness to be lower in heavily flooded regions,
largely to do with dominance of a handful of species well adapted to prolonged root
submergence, though in a later study finds no clear relationship in species richness
across three landscapes with low, intermediate and high frequency of flooding
(Ferreira, 2000).

Intermediate-scale disturbances have also been found to affect forest dynamics,
including the effect of river-meandering (Salo et al., 1986). Salo and colleagues use
satellite images to conclude that 27% of lowland rain forest in the Western Amazon
exhibits evidence of recent erosional and depositional activity, suggesting that this
disturbance is another important process in maintaining high diversity in Amazonian
forest through prevention of competitive exclusion and creation of environmental
heterogeneity.

Furthermore, studies in western Amazon also indicate that river

meandering may have been stronger through enhanced flows in the recent past as a
result of tectonic activity (Rasanen et al., 1986). This intermediate scale disturbance
affects the successional stage of forest, as well as altering the substrate for soil
formation (Huston, 1994), though little other attention has been given to the role of
river-meandering in studies of tropical lowland forest diversity.

44

2.3.3.4 Soils and nutrients

Though this study does not explicitly include soil in the fieldwork nor desk-based
modelling, many topographic factors are indicative of soil texture and soil quality, and
there is an extensive body of research which examines edaphic specialization in
species (discussed previously in Section 2.3.2), and the role of edaphic conditions on
species diversity (discussed here).

There are a number of studies that have used digital elevation models (DEMs) to map
soil attributes (for example Bui et al., 1999; Ryan et al., 2000), typically texture
(Gobin et al., 2001), nutrients

(Chen et al., 1997) and organic matter content.

However, these methods often require large calibration and validation datasets,
including soil surveys or geological maps, and are not performed in this thesis for
reasons outlined in Chapter 1. However, some topographic characteristics are likely
to capture soil gradients, and these must be considered in the analysis. Topography
plays an important role in controlling the distribution of soil quality (discussed in the
next paragraph), but the distribution of soil texture is highly dependent on soil
substrate (Takyu et al., 2002) and many topographic patterns in soil texture tend to be
site-specific (Gobin et al., 2001).

In terms of plant diversity, nutrients (in much of the literature referred to simply as
soil quality) have been found to be both beneficial and detrimental to species richness,
with contrasting results of monotonic increase in richness across an increasing soil
nutrient gradient (Proctor et al., 1983; Palmiotto et al., 2000), a unimodal distribution
with species richness peaking in mid-nutrient levels (Huston, 1982), and monotonic

45

decrease in richness across the nutrient gradient (Hall and Swaine, 1976; Huston,
1982; Oliveira and Mori, 1999).

There are text-book examples of the latter

relationship, where nutrient-poor soils harbor low tree diversity in the caatinga in
Amazonia and kerangas in SE Asia (Whitmore, 1998). Huston (1994) contends that
these patterns are spurious, originating from the co-linearity between precipitation and
nutrient levels (representing a productivity gradient, producing a unimodal pattern),
and that nutrients are indeed of little importance in controlling species richness.
Clinebell et al. (1995) argue the same, using detailed soil data in 69 0.1-Ha tree plots
of Alwyn Gentry, finding species richness to be independent of soil quality, as well as
clear inverse correlations between precipitation and soil nutrient concentrations
(principally through greater leaching). However, Oliveira and Mori (1999) provide
evidence that high precipitation is not necessarily synonymous with poor soils,
providing support for the argument that low nutrient content in soils reduces growth
rates and stimulates competition, maintaining high diversity and preventing
competitive exclusion (Berendse, 1999). In conclusion, a number of studies present
different relationships between soil quality and diversity, and the nature of this
relationship may indeed be site-specific.

Many studies of edaphic specialization and soil quality - diversity relations use
topography as a factor in determining the edaphic conditions, finding that valley or
toe slope regions tend to have fertile soils and ridge landscapes have low fertile soils
(Clark et al., 1998; Clark et al., 1999), due primarily to greater degrees of leaching on
convex slopes.

46

2.3.3.5 Topography

Although topography itself is not a physical resource crucial to plant growth and
survival, it is a fundamental environmental variable which partitions critical resources
such as energy, light and nutrients. Many studies already discussed in Section 2.3.2
use topography to identify habitat specialization, based on the physical environmental
conditions found in different topographically defined landscape units (valley, slope,
ridge etc.). Although habitat association of species with relation to topography has
already been discussed, there are also a select few studies that examine microtopographic patterns of diversity (rather than composition).

Webb et al. (1999)

presents the most detailed direct analysis of topographic patterns in diversity variation
in a tropical rain forest in American Samoa, finding significantly higher tree species
richness on ridges compared to slopes and valleys.

No detailed analysis of the

reasons behind this is provided, but the authors do mention the possible role of
frequent visits by birds and large insects to the more exposed ridges, thus increasing
seed deposition. This is the only suggestion of such a process in the literature and
may be of relevance for further testing, though the high diversity on ridges might also
be due to the documented lower soil quality (Clark et al., 1998; Clark et al., 1999),
and of the negative monotonic nutrient-diversity relationship discussed in Section
2.3.3.4. Koponen et al. (2004) also studies micro-topographic patterns of tree species
richness variability in a tropical freshwater swamp forest in French Guiana, finding
greater richness in lower-sites more regularly flooded, though the extent of the
topographic analysis was in fact in the distinction of lower and upper sites. Finally,
Enoki (2003) presents analysis of species composition on different “topographic
indices” (slope and a basic measure of mean curvature termed ‘configuration’),

47

finding that species composition was somewhat related to topographic indices, though
no pattern in terms of species richness was evident. Unfortunately the steepness and
configuration variables were not measured using robust techniques of terrain analysis.

With the part exception of Enoki (2003), no studies have thoroughly analysed species
composition or diversity patterns with relation to a broad range of quantified
topographic characteristics, making this thesis the first example of this kind of
analysis applied to tropical forest composition and diversity studies.

2.4

Spatial variation of structure in tropical forests

This section provides a brief review of literature on structural variability in tropical
forests. It is less thorough than the review already presented for diversity, as it is
relevant to only one objective, though there is also less theoretical debate on the topic.

As already presented in Section 2.2.2.1 and Section 2.2.2.2 structure varies
significantly between tropical forest ecosystems, with many documented reductions in
stature and increases in stem density with greater elevation (Pendry and Proctor,
1996; Pendry and Proctor, 1997; Kappelle, 2004), but there is also micro-scale
variability aside from elevational gradients (for example Robert and Moravie, 2004).
Aubréville (1938) first coined the phrase ‘forest mosaic’ to describe the micro-scale
variability in forest structure, producing a diverse set of micro-environments in the
canopy, sub-canopy and understorey. Since then detailed studies of gap dynamics and
succession have described much of this variability (discussed in Section 2.3.3.2), but
there is still a large degree of spatial variability in structure which is not accounted for

48

by disturbances (Schaik and E, 1985) which has been overlooked in the literature
(Trichon et al., 1998).

The elevational gradient in stand height has been attributed to many factors, including
high humidity (Odum, 1970), nutrient limitation (Vitousek, 1984; Vitousek and
Sanford, 1986), low temperature (Kitayama and Aiba, 2002), soil acidity
(Hafkenscheid, 2000), leaf wetness (Letts, 2003) and cloud cover (Grubb, 1977; Letts,
2003). Once again, a certain degree of site-specificity is evident in the literature, but
all these studies only use elevational data and do not perform analyses of micro-scale
variation which may provide a clearer picture of the factors at play in reducing tree
stature.

In terms of micro-scale variability in tree structure without considering the elevational
gradient, Robert and Moravie (2004) find no patterns in diameter distributions with
respect to slope, but do find that the height of a tree per unit diameter does increase on
gentler slopes. Webb et al. (1999) analysed forest structure in different topographic
landscape units (ridge, slope and valley), and found stem densities to be highest on
ridges with a limiting maximum diameter on ridges, a low stature and a respective
low height/diameter which they attribute to windcropping. In general, there are few
studies of micro-scale forest structure variability beyond the extensively researched
role of gap dynamics.

2.4.1 Structure – diversity interaction

49

Also of relevance to this thesis is the relationship between structure and diversity,
though there is very little available in the literature about this two-way relationship.
Many of the equilibrium based theories of species diversity and gap dynamics are
based on the diversification of niches through a disturbance that drastically changes
forest structure. However, beyond these stochastic and drastic structural changes,
structural heterogeneity in a closed canopy also potentially increases the number of
niches available in the light environment (Huston, 1994), though there is little
literature which explicitly investigates this theme.
.
Givnish (1999) reports that amongst other factors that have been found to generate
high species diversity in tropical forests, there is a negative effect on diversity with
increasing diameter at breast height (DBH), and a positive effect with increased stem
density and forest stature. The reasons behind such relationships are questionable,
and may be statistical artifacts of diversity indices, or co-linearity between already
discussed drivers of forest structure (Section 2.4).

Also of importance when considering this relationship is the potential circular nature
of the comparison, with different species themselves having very different structural
forms (Tomlinson, 1987), therefore it would be expected that highly species diverse
sites would also be structurally heterogeneous. Furthermore, under Huston’s (1994)
categorizing of interstitial and structural organisms, it is also expected that a diverse
structural environment (however this be generated) provides a diverse set of niches
for organisms in other groups, such as insects, amphibians and small mammals. This
is to some degree confirmed by empirical studies (for example Tews et al., 2004). In
this sense, structural heterogeneity may in fact be a more suitable indicator of

50

diversity in other taxa than in trees (for which the lack of relationship has been
discussed in Section 2.2.1).

Though there are a number of possible theoretical linkages between structure and
diversity, there are few empirical studies that explicitly study this two-way
relationship, as it is clearly very difficult to separate cause and effect. In this thesis a
brief examination of this relationship will be made.

2.5

Spatial environmental heterogeneity and its quantification

The term environmental heterogeneity is used in a number of different senses in the
literature, making discussion of the topic confused without an accurate definition.
Environmental heterogeneity within the context of this study refers to the diversity of
environments within a certain local spatial extent, for this reason it will be referred to
as spatial environmental heterogeneity in order to stress the spatial nature of the
variable. Spatially environmentally heterogeneous sites therefore exist where there
are sharp gradients in whatever environmental factor is being examined.

Many

different terms are used to portray the same phenomena, including habitat
heterogeneity, habitat diversity, habitat complexity, structural heterogeneity, and
spatial complexity (Tews et al., 2004).

Spatial environmental heterogeneity has long been associated with diversity at both
the species level (Huston, 1994; Rosenzweig, 1995) and the genetic level (Hedrick et
al., 1976; Hedrick, 1986; Volis et al., 2004), though as Araújo et al. (2001) point out
this positive relationship has often been ‘assumed rather than measured’. The role of

51

environmental heterogeneity on species diversity is in theory strongly supported by
Macarthur (1969), stating that ‘the number of species that stably coexist in an area is
dependent on the relative abundance of limiting resources and the range of values
over which these resources are distributed’, and many authors have since supported
the positive relationship between spatial heterogeneity with related niche-based
justifications (Ricklefs, 1977; Tilman, 1982).

Despite what appears to be general acceptance of the relationship between spatial
environmental heterogeneity and diversity (Huston, 1994), there are few empirical
tests of the relationship, and in some case the methods used are less than satisfactory.
Positive relationships between spatial environmental heterogeneity and plant diversity
have been found at the local (Nilsson et al., 1989; Burnett et al., 1998; Barberis et al.,
2002; Lundholm and Larson, 2003) and regional scale (Rey Benayas and Scheiner,
2002; Pausas et al., 2003). In these studies environmental heterogeneity has been
measured based on topography (Harner and Harper, 1976; Burnett et al., 1998;
Barberis et al., 2002), soil substrate (Nilsson et al., 1989), or soil depth (Lundholm
and Larson, 2003). Only one study reports a poor relationship between richness in
vertebrates and higher plants and spatial environmental heterogeneity (Araujo et al.,
2001). None of these studies are performed in tropical forests, all originating from
temperate ecosystems.

Lundholm and Larson (2003) are the only authors to observe an important unimodal
relationship between environmental heterogeneity and the productivity gradient,
showing clearly in a limestone pavement that the greatest heterogeneity (based on
microtopography and soil depth) is found in the middle of the productivity gradient.

52

This has important implications for the many studies that have observed unimodal
diversity peaks in mid-productivity levels (discussed in Section 2.3.3.1).

However, the measurement of spatial heterogeneity in many of these studies is highly
variable, and in some cases questionable. Most apply grid or transect methods,
whereby fine-scale environmental data is degraded to a coarse-scale grid of
environmental heterogeneity through analysis of range, richness, standard deviation,
variance or evenness of environmental conditions within each coarse-scale grid
(Burnett et al., 1998; Pausas et al., 2003). In other cases, the method used to quantify
environmental heterogeneity is ambiguously described (Harner and Harper, 1976).
These methods should not be confused with measurement of habitat heterogeneity in
the landscape ecology literature, which tend to focus on factors such as habitat patch
types, patch arrangement and patch shape (Li and Reynolds, 1994), referring more to
land-use, and which are less relevant to this study as the forests studied here are nonfragmented.

To the best knowledge of the author, there have been no rigorous and quantified tests
of the role of spatial environmental heterogeneity at the local scale on tropical tree
diversity in the humid tropics. The concept of environmental heterogeneity in tropical
forests provides a degree of unification between non-equilibrium and equilibrium
based theories of species diversity maintenance. If in lowland rain forests the random
occurrence in both space and time of tree and branch fall means that recruitment in
gaps is dominated by chance occupants (Brokaw and Busing, 2000), but that there is
also some degree of niche-specialization in some tree species (for example Phillips et
al., 2003, also discussed in depth in Section 2.3.2), spatially heterogeneous

53

environments would potentially provide a diverse local seed pool for gap recruitment
because habitat preferences would mean that the pool of potential immigrants were
diverse, increasing the chances that a specialist to the conditions in the gap may arrive
through short-distance seed dispersal. The potential implication of a greater number
of specialist species finding their particular niche is likely to increase diversity,
through optimizing the resource-partitioning, potentially increasing the different
number of species that the community may support (Hill and Hill, 2001). In spatially
heterogeneous landscapes, the rapid arrival of specialists to a gap may speed up the
process of succession and increase levels of competition, resulting in there being
greater diversity in temporally dynamic regions (i.e. possibly a left-skewed IDH
curve).

In this sense, the potential role of spatial environmental heterogeneity on tropical tree
diversity is through explicitly spatial processes such as seed dispersal, predation,
herbivory and competition. With respect to seed dispersal, there is a great deal of
literature on seed dispersal mechanisms (for example Howe and Smallwood, 1982;
Chambers, 1994; Wehncke et al., 2003), and a great volume of studies which quantify
seed dispersal processes for specific taxa in tropical forests (Peres and Baider, 1997;
Wenny, 1999; Yumoto et al., 1999), many finding the majority of seed dispersal
processes acting over distances in the region of 20m – 1km, with the majority of seeds
traveling less than 200m (Wehncke et al., 2003).

At this point much of this discussion is pure conjecture with a number of different
ways of hypothesising the role of environmental heterogeneity on diversity. It is best
to test for diversity relationships with spatial heterogeneity then analyse the potential

54

mechanism behind the pattern, though within the scope of this thesis this is a large
theme and only the first step in the analysis of spatial environmental heterogeneity
can be achieved here.

2.6

Field sampling methods

A recent article by Phillips et al. (2003) reviewed different sampling methods for
studies of forest diversity and dynamics. Though these results were too late to impact
on the field-based gathering of diversity, composition and structure data for this
thesis, it is nevertheless worth a mention here. A comparison of effort (in terms of
person days) and results of floristic inventory were made for 0.1-Ha plots (made up of
ten 2m x 50m plots) and 1-Ha plots (square, 100m x 100m), finding that the smaller
0.1Ha plots were more efficient in terms of capturing floristic inventories with the
minimum of effort. However, the best choice of plot size and shape is heavily
dependent on the ecological questions being raised.

When considering field sampling methods and possible field sampling errors,
attention is especially raised to the temporal dynamism of tropical forests, explicitly
discussed in Section 2.4, and its relation to plot studies. Though long-term and largescale projects such as that undertaken in Barro Colorado Island benefit from frequent
censuses over time, and many other large-scale plot-based monitoring projects are
now underway (Burslem et al., 2001), this study is based on diversity data from a
single snap-shot in time. Small plots are used in this study to capture the micro-scale
variability central to the objectives outlined in Chapter 1, but this also makes the plot
studies particularly sensitive to stochastic processes such as tree-fall which also

55

operate at the micro-scale. It is important that these issues be considered throughout
the thesis, as sampling issues are critical both in setting out the research methodology
(Chapter 3) and in interpreting the results of the analyses (Chapters 4 – 7).

Finally, this thesis applies a simple a priori objective sampling strategy, based on
stratification of plots across a 2 dimensional environmental gradient (presented in
Section 3.5.1). After a detailed review of the literature, no other example of an
objective model for locating plot studies has been found, though Phillips and Raven
(1997) do call for stratified sampling schemes in tropical forest plot studies. Though
the sampling scheme adopted in this thesis is simple and fairly basic, it is the first
known application of this kind of method to tropical diversity plot establishment.

2.7

Conclusions

This chapter has outlined some of the theories for maintenance of species diversity at
the local scale in tropical forests, and provided results of empirical studies related to
the spatial variability in species diversity, composition and structure.

Despite

contradictory findings in many cases, likely due to site-specificity, there is strong
evidence that at least a percentage of variability in diversity is brought about by
equilibrium related processes, and that environmental factors are important in driving
this variability (along with biotic factors). In many cases it is topography that has
been attributed or used to explain the spatial variability, but terrain-based
characteristics are normally limited to simple elevation, slope and slope position (or
landscape unit).

This thesis therefore applies rigorous analysis of terrain

56

characteristics through GIS to quantify the role of topography in controlling species
diversity, composition and structure over space through niche-partitioning.

57

Chapter 3

3

3.1

Research Strategy and Methods

Overall research strategy

In order to address the hypothesis and aims outlined in the first chapter, this study
combines field-based data collection with desk-based spatial analysis and modelling.
The experimental design is to associate field measured tree diversity and structure
with carefully selected GIS-derived topographic variables, in order to understand
which of these factors, if any, are important in shaping the micro-scale variation in
tree diversity and structure.

Two study sites are chosen from proximal but very different tropical environments in
order to highlight differences in environmental controls on diversity and structure.
The first site, Reserva Tambito situated on the western cordillera of the northern
Andes in Colombia is a tropical montane cloud forest (TMCF), with steep topography
creating a highly variable environment through the site. The complex topography
creates a mosaic of micro-climates, with the physical inputs to the system (water,
radiation, nutrients) highly variable around the catchment, and in one way or another
are controlled by the topography.

Persistent canopy level cloud also provides

hydrological and nutrient inputs to the system, and also has many other indirect
effects on the ecosystem.

The second site, Tiputini Biodiversity Station (TBS)

situated in the western Amazon in lowland Ecuador, is a tropical lowland rain forest
(LRF) with gentle slopes, almost no elevational gradient but a complex network of
small streams running into a main river channel which is of 30m width. Above the

58

canopy the climate and physical inputs are homogenous, but on the ground a great
deal of micro-scale variation is found. In this sense the canopy serves as a filter of the
energy inputs, partitioning these resources unequally below the canopy.

River

channels have cut 20-30m deep valleys, and highly dynamic flow levels in the main
stream (the River Tiputini) causes widespread and frequent, though not persistent,
flooding of large tracts of low lying forest.

These two sites are designed to contrast each other, and provide additional insight into
how ecology interplays with topographic variation. In all cases throughout this thesis,
each results chapter starts by treating each site independently, but the chapters finish
by examining the similarities and dissimilarities in the patterns found at each site, and
attempt to extract generalities. All efforts are made to ensure that the same field and
desk-based methodologies are adopted in each study site, though fieldwork was not
performed at each site during the same period. Fieldwork in Tambito occurred
between 1999 and 2001, with fieldwork in TBS starting in 2000 and for the purposes
of this thesis finished in 2003. For this reason, some improved field methods were
adopted for TBS, though they do not affect the validity of the site-comparisons.

One of the important factors in the field-based experimental design was to ensure that
the field-based data is directly comparable with the GIS derived topographic data.
This basically means that square plots must be used (for comparison with grid-based
raster surfaces), and that the plots must be of a size equal to or an exact multiple of
the cell size in the raster based topographic data. Furthermore, the central hypothesis
of this study is related to micro-scale variation, and so the field data must capture
micro-scale variation.

59

In the literature a number of plot sizes have been used for ecological studies,
including rectangular 0.1Ha plots (Gentry, 1995), square 1Ha. 100m x 100m plots
(Pitman, 2000), and also rectangular 25-Ha plots (Valencia et al., 2004) and
rectangular 50-Ha plots (Condit et al., 1996). Plot size in all these cases was decided
based on the hypotheses being tested, but also depended on the resources at hand. As
with any piece of research there are limitations to the amount of data collection that
can be feasibly achieved (Phillips et al., 2003), with time and resources always being
restrictive in some sense. This thesis is no exception, and so these limitations made it
necessary to decide on the best balance between the number of plots studied and the
size of each plot. Taking into account the issues of scale and compatibility with cellbased grids, ten plots were established in each study site of size 25m x 25m. This
exactly matches one cell in the cartographically derived DEMs for each study site,
evading problems of aggregating or downscaling the topographic data for comparison
with the plot-based data. Though more plots would have been preferable to add
greater statistical significance, and also to allow split-sample validations when
comparing modelled variables with measured variables, resource, time and
accessibility restrictions have prevented the establishment of more plots.

Originally this thesis was to use aerial imagery to actually measure diversity at the
broad-scale. This was aimed at producing a result in itself as a method for monitoring
diversity, but was also designed to solve some of the problems of lack of field-based
data with which to compare with the GIS data. Despite a considerable investment in
field-data collection and image processing, this method was not sufficiently refined,
and has been dropped from this thesis as a chapter in itself, and the results are not

60

used to enhance the statistical significance of the correlations between diversity and
topographic characteristics. However, some preliminary results were achieved, which
show potential for further development. This line of research lacks the scientific rigor
for inclusion in the body of the thesis itself, but does merit some tangential mention.
For this reason, Appendix 1 presents some of these advances. Further work in this
area is being undertaken by others in the HERB research group.

Of most importance in the analysis of topographic characteristics was the digital
elevation model (DEM) itself, since all of the topographic characteristics are in some
way derived from it. It is evident later in this chapter that a great deal of effort was
invested in generating the DEMs for each of the study sites. In the case of Tiputini
two DEMs are used throughout the analysis. Despite complicating both the analysis
and the interpretation, this decision is justified based on the critical importance of
DEM quality on the accuracy of the results.

Some discussion and preliminary

analysis is presented in this chapter to understand the qualities and associated
problems with each DEM.

This becomes important in later chapters when

interpreting the results.

Also of key importance to this thesis is the selection of terrain characteristics. Given
the lack of prior analysis of terrain characteristics in the context of diversity and
structure in tropical forests that has been explored in Chapter 2, selection of these
variables is made with the objective of covering a broad range of terrain conditions
considered potentially important for both diversity and structure. Special attention is
made in selecting characteristics that directly affect essential resources for tropical
trees such as light, temperature and water, and also factors which may have some

61

indirect effect on other important factors (such as soils, micro-climate). Whilst there
is an extensive body of research attempting to map soil characteristics explicitly based
on digital elevation models (Gessler et al., 1995; Bui et al., 1999; Ryan et al., 2000;
Lookingbill and Urban, 2004), results are variable and often require large calibration
and validation datasets which are not available for this study. It is likely that some
terrain characteristics capture a degree of soil variability (type and quality), but this
indirect relationship is unknown and un-quantified.

In any case, some evidence

suggests that soil may not be all that important for tropical forest ecology (discussed
in Section 2.3.3.4), through direct cycling and the fact that most of the nutrient
dynamics take place in the litter and organic layers (i.e. a function of the vegetation
and not of the soil) (Vitousek, 1984; Clinebell II et al., 1995; Berendse, 1999). Based
on the experimental design of comparing measured diversity and structure with GISbased terrain analyses, it was deemed better to include numerous terrain
characteristics, some of which represent similar terrain conditions and contain degrees
of co-linearity. Whilst this co-linearity is problematic to results interpretation later in
the thesis, subtle differences in the potential significance of the terrain characteristics
themselves sometimes help in isolating the precise causal process. However, in order
to ease the analysis, certain terrain characteristics are discussed in greater detail than
others, and given greater importance in the interpretations.

This chapter is split into the two major methodological thrusts of this thesis, which
were the preparation of terrain-based spatial data and the field-based diversity and
structure data collection. This does not necessarily reflect the chronological order in
which this research was elaborated, and indeed little of the environmental or spatial
data was available when field-work commenced (since the collection of field data

62

began before the thesis formally commenced and required the most significant time
and effort of all the aspects of the thesis). For this reason a priori sampling strategies
were not used in Tambito, and the sampling strategy used in TBS used factors not
entirely compatible with the final environmental and terrain variables presented here.

First a brief description of each study site is given, followed by a detailed description
of the methodologies adopted to define and calculate the terrain characteristics.
Secondly, the chapter addresses the field-based data collection, providing detail on the
methodologies used in each study site. On the whole, the same methodologies were
used in each study site, but in some cases site-specific conditions necessitated slight
differences in methodologies.

For this reason, each site is discussed separately.

Finally, the methodology for the spatial heterogeneity modelling is presented, as the
method is identical for both sites.

3.2

Reserva Tambito

Reserva Tambito, Cauca, Colombia (2o 30’N, 76o 59’E) lies on the western slopes of
the Western Cordillera of the northern Andes.

It is located within the Choco

biogeographic region, identified as one of the major global biodiversity hotspots (see
for example Myers et al. (2000)). The altitudinal range is extensive, rising from
1053m to 2860m, with steep slopes (10-70 degrees), and varying altitudinally
controlled climates creating a gradient from lower montane cloud forest (LMCF) to
upper montane cloud forest (UMCF) in the upper reaches of the catchment (>1700m).
Large volumes of epiphytes are found clinging to branches and leaves throughout the
catchment (Jarvis, 2000). The total area of the two catchments covers 2150 hectares.
The current land-cover within the catchment is mainly primary forest (72% of

63

catchment), with an area of pasture around the Tambito cabin (4% of catchment) and
a larger area of surrounding secondary forest (26% of catchment). Tambito has
played an important role historically for the region, providing a stop-over for farmers
traveling from the coastal regions to the large market in El Tambo, some 30km from
Tambito. All that is evident of this history today is a deeply cut path crossing the
reserve, and the original cabin site of Tambito where researchers are now
accommodated. It is unclear to what extent Tambito was forested in the past, but
personal communications with local land-owners indicate that the majority of today’s
forest is at least 50 years old.

The northerly catchment, termed the Palo Verde, is made up of primary forest, whilst
the Tambito catchment has large tracts of primary forest with regions of secondary
growth (approximately 15-20 years old) and a small deforested area around the
accommodation cabins. The steep and variable topography and the dense understorey
make movement around the catchments difficult.

64

Figure 1 The topographic complexity in Reserva Tambito. Photo taken at 1500m from the
main cabins at around 3pm. Cloud presence is typical of the afternoon hours, with the late
afternoon often characterised by canopy level cloud presence at all elevations.

3.3

Tiputini Biodiversity Station

Tiputini Biodiversity Station (0o 33’S, 76o 09’W) is located in the Orellana Province
of Ecuador. It lies in eastern Amazon, some 110km from the edge of the Andes. The
research reserve is located directly opposite Yasuni National Park on the River
Tiputini in the important eastern Andean biological hotspot (Myers et al., 2000), with
an altitudinal range from just 200 m.a.s.l. to 270 m.a.s.l.

In total, the reserve covers an area of approximately 2400Ha. of primary lowland
rainforest. There is a network of streams within the region, ranging from small
ephemeral channels to larger permanent channels such as the River Tiputini itself.

65

Though no record of river levels exists for the region, the River Tiputini at the station
has been observed to vary in stage by as much as 15m, causing widespread flooding
within the surrounding ‘varzea’ forest. The reserve contains a range of habitats from
those that are never flooded to forest regularly found underwater.

A very detailed description of the Yasuni National Park region as a whole is provided
by Pitman (2000), and readers are directed to this work for specifics on the geology,
history, culture and biology for the region.

Figure 2 River Tiputini at low flow. During highflow the river exceeds the banks and floods
the adjoining forest.

3.4

Generation of biophysical data

3.4.1 Reserva Tambito

66

The work in Tambito contained in this thesis benefited from prior work performed by
researchers (including the author) in Reserva Tambito during the period 1997 to
present under the auspices of the HERB project (http://www.kcl.ac.uk/herb) and the
Negret Project (http://www.kcl.ac.uk/negret).

Rincon-Romero (2001) produced a

series of raster grids of biophysical variables for use in hydrological modelling, and
some of this same data has been used in this thesis.

3.4.1.1 Meteorological and hydrological monitoring

There are two fully equipped hydrological and meteorological stations at the study
site. One of the stations is located in the deforested area (the Campo Plot, 1450masl,
Figure 3), whilst the other is located in primary forest on the Cerro El Perro hillside in
the Palo Verde catchment (Bosque Plot, 1650masl). Each station is equipped with an
array of solar radiation pyranometers (upward and downward facing for determining
solar and net solar radiation), temperature probes, rainfall gauges, humidity probes,
soil moisture sensors buried at varying depths and wind sensors, in addition to
hydrological apparatus whose data are beyond the requirements of this study.

67

Figure 3 Hydrological and climatological monitoring station at the Campo and Bosque plot,
Tambito

The Campo station has provided a total of 4 years worth of data, with occasional
periods lacking data. The Bosque station has been more problematic (chiefly due to
humidity and energy problems), and the data is highly fragmented over the period
1997 – 2002.

The Campo station reports temperatures averaging 18.3 degrees

Celsius, with humidity in excess of 90% for much of the time and an annual rainfall at
the Campo plot of 3900mm. In addition to this, a six-year manual rainfall record from
Tambito has been monitored, and indicates a bi-modal seasonality with
October/November and February/March being the wettest months (Figure 4). The
dryer season (<300 mm month-1) is from June to September. 80% of the total annual
rainfall falls between October and May and only 20% falls between June and
September inclusive. The record indicates high sensitivity to El Nino and La Nina
(Figure 5). In 1997 (a strong El Nino year), just 20mm of rain fell in the 3 months
from June to September.

68

Mulligan and Jarvis (submitted) used rainfall stations distributed around the
catchment to conclude that rainfall linearly increases with elevation at 405 mm per
year per 100m elevation. Tambito is located in a highly dynamic environment where
climatic conditions change across small distances. In addition to the rainfall inputs, a
significant amount of input from cloud interception is also to be expected (Gonzalez,
2000; Jarvis, 2000). Wind speeds are generally low (with hourly averages 2000-2001
less than 0.5 ms-1 and five minute averages less than 0.66 ms-1), and only 1.2% of
five-minute average values are greater than 3 ms-1. Monthly average solar radiation
(1997-2001) varies from 101 Wm-2 (January) to 378 Wm-2 (July) Hourly average
solar radiation is generally low, especially in the afternoons (when cloud cover tends
to be more frequent). Monthly average radiation varies from 101 Wm-2 (January) to
378 Wm-2 (July). Hourly average radiation is generally lower than TBS with a peak
at 11am of 540 Wm-2. Values are below 10 Wm-2 from 7pm through to 5am.

10 Day Running Mean Rainfall Record for Tambito, 1995 2000
40

35

Rainfall (mm)

30

25

20

15

10

5

6/29/1998

5/25/1998

2/9/1998

4/20/1998

3/16/1998

1/5/1998

12/1/1997

9/22/1997

8/18/1997

10/27/1997

6/9/1997

7/14/1997

5/5/1997

3/31/1997

2/24/1997

1/20/1997

12/16/1996

9/2/1996

10/7/1996

11/11/1996

7/29/1996

6/24/1996

5/20/1996

4/15/1996

2/5/1996

3/11/1996

1/1/1996

11/27/1995

9/18/1995

10/23/1995

6/5/1995

8/14/1995

5/1/1995

7/10/1995

3/27/1995

2/20/1995

0

Figure 4 Rainfall (mm) manually measured daily for the Tambito study site, taken at
1450masl in the region around the cabins.

69

Annual Rainfall Variation for Tambito
800

1995
1996
1997
1998

700

Monthly Rainfall (mm)

600

500

400

300

200

100

0
Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

Month

Figure 5 Monthly distribution of rainfall measured manually on a daily basis in the Tambito
study site

3.4.1.2 DEM generation

The digital elevation model (DEM) for the study site is the most fundamental physical
variable, which is used as a baseline dataset to calculate all terrain characteristics. For
Tambito, two potential sources of elevation data were available.

These were

cartographic map sheets, at 1:25,000 scale and 3 arc-second elevation data from the
Shuttle Radar Topography Mission (SRTM).

ASTER satellite images were also

sought for the study area, to provide a third potential source for a high resolution
DEM, but all images as of publication of this thesis contained cloud in and around
Tambito.

Topographic data for Tambito was available from three 1:25,000 scale map sheets
(sheets 343 I a, b and c) from the Instituto Geografico Agustin Codazzi (IGAC) based

70

in Bogota. Rincon-Romero (2001) digitized contour lines and rivers from these
sheets, and made this data available. The contour data on these maps was originally
derived from aerial photos, and represents the canopy-top topography rather than the
topography at ground level. Furthermore a small patch outside of the Tambito twincatchments but nevertheless within the region for which there is data lacks contour
information due to clouds in the original image. This area is nevertheless outside of
the region under investigation in this thesis and no plots are located in this region.
This data was re-projected to decimal degrees, and used to produce a DEM with
equivalent 25m grid resolution (0.000225 decimal degrees). Arc/Info’s TOPOGRID,
based on the established algorithms of Hutchinson (1988; 1989), was used to produce
the DEM using both contour lines and rivers, using tolerance parameters of 5 for
“tolerance 1”, representing the density and accuracy of input topographic data, and a
horizontal standard error of 1 and vertical standard error of 0. The resultant DEM
contains a total of 69,678 cells, covering an area of 7.3km x 5.8km, ranging in
elevation from 1086m to 2093m. Slope and aspect were calculated using Arc/Info
(which uses the method developed by Burrough (1986)) to visually assess the quality
of the DEM, and to provide basic terrain characteristics for use in analysis and
modelling outlined later in this thesis. A single scale 3 x 3 window was used in
calculating these derivatives as there is little ecological reason why broader scale
measures of these variables might influence tree species diversity, structure and
composition except through the role of spatial heterogeneity explored in Chapter 6.
Figure 6 displays these three basic topographic properties, displayed using hillshading
in Landserf to enhance visualisation.

71

Figure 6 The cartographically derived DEM with 25m cell resolution (left), with the derived
slope (center) and aspect (right). Slope varies from 0o (cream) to 58o (red). Aspect is in
degrees, ranging from 0o (red) to 90o (white) to 180o (blue) and 270o (white).

SRTM data was also made available for the study region, with cell resolution of 3 arcseconds (approx. 90m). This was made available by NASA and the USGS late in the
process of this PhD in April of 2003. Unfortunately, Tambito contained a large
percentage of no data regions (43%) in the SRTM DEM, where steep topography
produced strong shadowing, preventing the calculation of elevation due to insufficient
textural detail in the original radar images. The SRTM DEM for Tambito is not used
in this study, as there is too much missing data (which covered central and important

72

parts of Tambito where plots had been established). Further information on SRTM
data for Tambito is also available in Jarvis et al. (2004).

3.4.1.3 Terrain characteristics

Using the cartographically derived DEM (from hereon referred to simply as DEM), a
number of terrain characteristics were calculated for comparison with measured
diversity and structure from plot studies.

These characteristics were selected to

represent a broad range of terrain properties which may control structure and diversity
through their role in shaping essential resources.

3.4.1.3.1 Potential Solar Radiation Receipt
Section 2.3.3.2 discussed in detail the potential implications of the light environment
on tree composition, diversity and structure, stressing the combined role of light
quality and quantity. Here a number of light-related factors are presented, capturing
both quantity and quality in terms of absolute solar radiation receipt and the intraannual and diurnal variation. Absolute top of atmosphere radiation is calculated (this
section), and two aspect-derived characteristics are presented and their significance on
solar radiation thoroughly analysed using measured data (Section 3.4.1.3.2).

A simple radiation model is used to calculate total potential radiation receipt, based on
the orbital parameters of the sun, the latitude of the study site and the slope and aspect
of the terrain. The solar radiation sub-model of the BENDUM hydrological model
reported in Mulligan (1999) was used for the purposes of this study. This model
calculates on an hourly basis the potential radiation receipt based on the orbit of the

73

sun at the specified latitude, and the slope and aspect of the ground surface. The
hourly time-steps are then aggregated to produce an annual total radiation receipt.
The result is a grid of potential solar radiation receipt in the absence of cloud cover or
atmospheric attenuation in Wm-2, shown in Figure 7. This “top of atmosphere”
radiation receipt should be adjusted to actual solar radiation receipt but insufficient
information is available on the spatial distribution of cloud cover in the Tambito
catchments to sufficiently grasp the spatial variation in actual radiation receipt
through cloud cover variation. For these reasons this study has used unadjusted
values of potential receipt. Total potential solar radiation receipt varies spatially in a
significant way from 5,600 to 9,600 Wm-2, with north-east facing slopes receiving the
least amount, and the greatest radiation receipt on the common gentle-sloping southwest facing hillsides.

Figure 7 Total potential solar radiation receipt across the Tambito study site. Blue indicates
lower receipt (5,600 W/m2), green intermediate levels (8,400 W/m2) and yellow high levels
(9,600 W/m2).

74

3.4.1.3.2 Quantitative Measures of Aspect
Aspect is a critical factor that controls the solar radiation receipt to the land surface.
Steep slopes of different aspects may receive quite different levels of solar radiation
and this dependence is evident in Figure 7. In Tambito, situated at just 2oN latitude,
there is relatively little intra-annual variation in top of atmosphere solar radiation
compared with higher latitudes, nevertheless the role of aspect (and slope) is
important and some annual variability in radiation receipt exists according to slope
and aspect. Section 3.4.1.3.1 applies a solar radiation model to quantify potential
light receipt around the catchment, using aspect as one of the key variables. This
model does not adjust radiation receipt for atmospheric attenuation by clouds, but
provides an overall figure of absolute annual receipt in solar radiation.

However, “top of atmosphere” solar radiation receipt fails to account for two
important aspects of solar radiation:

1. Spatial and temporal variation in cloud cover in the twin-catchments of
Tambito, which significantly reduces solar radiation receipt at the landsurface
2. Daily and annual variation in solar radiation receipt, brought about by a
combination of orbital parameters and temporal variation in cloud cover

The daily and monthly variability in solar radiation receipt is potentially just as
important as absolute potential solar radiation receipt, and so two aspect related
characteristics are calculated: eastness and northness following Zar (1999):

75

Equation 1

Eastness = sin ((aspect * π ) / 180)
Northness = cos ((aspect * π ) / 180)

For example, northness for an aspect of 360 degrees is 1, for 90 degrees is 0, and 180
degrees is -1.

Understanding the complex role of these two variables on light receipt at the land
surface is complicated by the combination of numerous factors. Measured data from
an upward pointing pyranometer in the Campo station in Tambito (aspect 7o,
northness 0.99, eastness 0.12) is used alongside the modelled solar radiation data and
the eastness and northness variables in order to shed some light on the variability of
solar radiation at the ground surface, and the significance of northness and eastness.
This detailed analysis is important for interpreting relationships found in subsequent
chapters between composition, diversity and structure and northness and eastness.
Firstly, the role of eastness and northness on top of atmosphere solar radiation is
examined, followed by a detailed analysis of their significance in seasonal and diurnal
variation in solar radiation receipt.

The section finishes by providing some

concluding remarks as to the significance of northness and eastness.

Deviation of the actual solar radiation receipt from the modelled solar radiation is not
constant, and varies during the year (Figure 8), and depending on the time of day.
The modelled solar radiation receipt for the Campo station is uni-modal, with a peak
in radiation receipt in the months of May, June and July. However, the measured
solar radiation receipt at all times of day is in fact relatively low for May and June,
peaking in July and August. Indeed, the greatest deviation from the modelled solar

76

radiation occurs in the months from March – June, coinciding with the wet season.
This indicates higher levels of atmospheric attenuation (cloud cover) during these
months.

Annual variation in solar radiation receipt in Tambito
Measured Solar Radiation
Receipt (Wm -2)

1200

8am
11am
2pm
5pm
Modelled

1000
800
600
400
200

December

November

October

September

Month

August

July

June

May

April

March

February

January

0

Figure 8 Annual variation in solar radiation receipt at different times of day, using measured
data from the Campo station and the average modelled monthly top of atmosphere solar
radiation receipt (all hours) for reference.

Relating this annual variation in cloud cover to the northness and eastness variables,
the entire Tambito twin catchments are used (i.e. all slopes) to compare northness and
eastness with modelled solar radiation receipt. A significant degree of variation in
potential solar radiation receipt exists between different degrees of northness, and the
higher levels of cloud cover during the months March – June have an important
effect. Strongly north facing slopes (positive values in northness) have the highest
peaks in potential solar radiation receipt during the summer period (May – July),
whilst south facing slopes have the opposite, with the highest potential radiation
receipt in the winter period from October – November and lowest receipt May – July
(Figure 9). Meanwhile, potential solar radiation receipt varies little on slopes with
77

different degrees of eastness. This has important implications for the northness value,
with strongly north facing slopes receiving the greatest amount of potential solar
radiation at the same time as atmospheric attenuation is lowest (Figure 8). Inversely,
south facing slopes receive the least potential solar radiation in the months with
greatest cloud cover. The resultant annual variation in solar radiation receipt is likely
to be more pronounced than that shown in Figure 9, with north facing slopes receiving
greater solar radiation (contrary to the expectation in the northern hemisphere), and
strongly north or south facing slopes also having the greatest range in solar radiation
receipt throughout the year.

Annual variation in modelled solar radiation
receipt across different degrees of northness
Average monthy
modelled solar
radiation (Wm -2)

1000

-0.9

900

-0.5

800

-0.1

700

0.1
0.5

600

0.9

500

Avg

400

Se
p
O
ct
N
ov
D
ec

Ju
l
Au
g

Ju
n

Ja
n
Fe
b
M
ar
Ap
r
M
ay

300
Month

Figure 9 Variation in modelled solar radiation (top of atmosphere) for different degrees of
northness throughout the year in Tambito, averaged for all 69,678 pixels.

78

Annual variation in modelled solar radiation
receipt across different degrees of eastness
Average monthy
modelled solar
radiation (Wm -2)

900

-0.9
-0.5
-0.1
0.1
0.5
0.9
Avg

800
700
600
500
400

Se
p
O
ct
N
ov
D
ec

Ju
l
Au
g

Ju
n

Ja
n
Fe
b
M
ar
Ap
r
M
ay

300
Month

Figure 10 Variation in modelled solar radiation (top of atmosphere) for different degrees of
eastness throughout the year in Tambito, averaged for all 69,678 pixels.

Looking at daily variation in radiation receipt, measured solar radiation at the Campo
station is compared with the modelled solar radiation receipt for the pixel representing
the Campo station (Figure 11). The measured solar radiation receipt at the Campo
station peaks at 11am, with an annual average radiation receipt of 560Wm-2 at this
hour. The afternoon hours have lower than expected solar radiation receipt as a result
of build-up of cloud in the catchment, though on average the morning build up of
radiation receipt is very well represented by the model (in comparable, not absolute
terms), indicating that cloud cover builds up in the afternoon hours.

This has

important implications for eastness, as east facing slopes (positive values for eastness)
receive the greater amount of sun in the morning hours, when atmospheric attenuation
is generally lower. Inversely, west-facing slopes tend to receive the greatest radiation
receipt in the afternoon hours when cloud cover is higher, potentially resulting in
lower radiation.

79

Modelled versus measured solar radiation at the
Campo Station in Tambito, annual average
600

1400

Modelled
Radiation
Measured
Radiation

500

1000
400
800
300
600
200
400

Measured radiation
receipt (Wm -2)

Modelled radiation receipt
(Wm -2)

1200

100

200

0

0
1

3

5

7

9

11

13

15

17

19

21

2

25

Hour

Figure 11 Modelled and measured hourly solar radiation receipt in Tambito, averaging hourly
measurements from throughout the year

Combining daily variation in radiation receipt and monthly variation, Figure 12
compares modelled with measured solar radiation for four months through the year,
once again just for the Campo plot. As can be seen, the deviation from the modelled
radiation receipt is variable throughout the year, with greatest deviation in April in the
afternoon hours. In July and especially October little deviation is evident. Examining
the absolute values rather than the relative differences, although the peak in modelled
potential radiation is generally quite even throughout the year at around 1300Wm-2
(ranging from 1210 – 1350Wm-2), the measured solar radiation at this peak varies
from 335Wm-2 in January (26% of potential) to 720Wm-2 in July (58% of potential).

80

Measured
Radiation

300

1000
250
80 0
200
60 0
150
40 0
100
20 0

0
1

3

5

7

9

11

13

Hour

15

17

19

21

2

500

400
600
300
400
200
200

0

0
7

9

11

13

Hour

15

17

19

21

2

50

0
3

5

7

9

11

13

Hour

15

17

19

21

2

25

600

Modelled
Radiation
Measured
Radiation

400
800
300
600
200
400

100

0

25

500

1000

200

100

5

100

1200

Modelled radiation receipt
(Wm -2)

600

800

3

150
400

1400

Measured radiation
receipt (Wm -2)

Modelled radiation receipt
(Wm -2)

Measured
Radiation
1000

1

200
600

M ode lled v ersus me asured solar radiation at the
Campo Station for Octobe r, Tambito

800

1200

250

0

M odelle d ve rsus measure d solar radiation at the
Campo Station for July, Tambito
700

300

800

1

Modelled
Radiation

350

Measured
Radiation
1000

25

1400

Modelled
Radiation

200

50

0

400

1200

Measured radiation
receipt (Wm -2)

350

Modelled radiation receipt
(Wm -2)

Modelled
Radiation

1400

Measured radiation
receipt (Wm -2)

1200

Modelled radiation receipt
(Wm -2)

Modelled versus measured solar radiation at the
Campo Station for April, Tambito

400

Measured radiation
receipt (Wm -2)

Modelled versus measured solar radiation at the
Campo Station for January, Tambito
1400

0
1

3

5

7

9

11

13

Hour

15

17

19

21

2

25

Figure 12 Hourly modelled radiation receipt against measured radiation receipt for the
Campo station for four different months, January (top left), April (top right), July (bottom
left) and October (bottom right).

These rather complex patterns of solar radiation receipt make the interpretation of the
variables solar radiation, northness and eastness difficult.

Whilst potential solar

radiation receipt is based on a mathematical model, combining slope, aspect, and
latitude, significant variations of cloud cover through the year shows that this is
unlikely to capture the actual radiation receipt at the ground surface. However,
without accurate spatio-temporal information as to the distribution of cloud cover
(which incidentally in the field is observed to both drop down from the top of the
catchment and flow up the catchment from the Pacific in the afternoon hours), the
solar radiation variable cannot be adjusted for actual radiation receipt. For this

81

reason, northness and eastness are used as two other light related variables, which
indicate a complex combination of factors.

Northness has been shown to indicate annual variation in radiation receipt, with
strongly north facing slopes more likely to receive the greatest amount of solar
radiation (greatest potential receipt at same time as the least atmospheric attenuation),
depending of course on slope. Strongly north- or strongly south- facing slopes also
have the greatest annual variation in radiation receipt, with neutral northness values
(approx. 0) having little annual variation.

Eastness is a more complex variable, with less pronounced annual variation in solar
radiation across different degrees of eastness, but a potentially strong daily variation.
Generally, cloud cover is shown to build up in the afternoon hours, though this
deviation of measured from modelled solar radiation is shown to vary depending on
the month. However, in general terms east facing slopes are more likely to receive
higher solar radiation (average and maximum), whilst west facing slopes likely
receive lower radiation levels (average and maximum), depending of course on the
slope.

In reality northness and eastness attempt to combine spatial and temporal variation in
radiation receipt into single indices but the result is somewhat confused. Nevertheless
they are used as terrain characteristics in this study as no other simple means of
capturing both spatio-temporal variation in cloud cover and daily or annual variation
in potential radiation receipt exists.

82

It should also be noted that northness and eastness become important in terms of
exposure, especially if there are strong and steady winds. Wind speeds for Tambito
have long term average hourly values (means of 10-minute readings at the Campo
station, intermittently 1997 -2001) of 0.48ms-1 with a range of 0 to 2.43 ms-1. Means
of five-minute averages (of 1-minutes data) give an average of 0.6 m/s and a max of
5ms-1. 0% of hourly averages are >3 ms-1 and 1.2% of 5-minute averages are >3ms-1.
In the Campo plot, wind direction was dominantly from the south-easterly direction,
though this might reflect topographic funnelling as well as the site-specific
topographic exposure and the dominant wind direction for the region (combined NE
trades and a westerly flow from the Pacific). Indeed, wind direction is likely to vary
considerable around the catchment based on topographic funneling and micrometeorology. Without distributed weather stations around the catchment measuring
wind speed and direction it is impossible to accurately map exposure to wind, but it is
likely that northness and eastness say little about exposure to wind.

3.4.1.3.3 Mean Curvature

The terrain analysis software Landserf (Wood, 2004)) was used to calculate the
topographic mean curvature. This is a derivative ranging from negative to positive
values, which in simple terms represents the trend in change of slope, negative
numbers signifying concave slopes and positive numbers convex slopes. Typically,
valleys and toe slopes are concave, and ridges are convex. Competition for light will
be more intense in concave slopes, where the topography and the forest canopy
produce less surface area for light receipt per unit area in plan form. This effect is
discussed in Gale (2000) with relation to gaps, concluding that ridges have greater

83

light receipt and more importantly greater variability in the access to that light
(through lateral penetration of the canopy, for example). Curvature is also important
for micro-climatic exposure, and especially in Tambito may be strongly related to the
exposure to cloud at the forest canopy. It is likely that convex, exposed slopes receive
greater cloud presence at the canopy level.

This may be important as cloud

interception provides an additional nutrient input to cloud forests (Vitousek and
Sanford, 1986), and epiphytes tend to concentrate in regions with persistent cloud
presence at the forest canopy epiphyte (Nieder et al., 2001). Letts (2003) also argues
that persistent leaf wetness is an important factor in the stunting of trees in cloud
forests compared to lowland rain forests (through reduced photosynthesis), so greater
exposure may also reduce productivity in this way. Finally, curvature may also be
strongly associated with soil quality, with highly convex slopes typically being the
most leached and nutrient poor (Clark et al., 1998; Clark et al., 1999; Phillips et al.,
2003).

The calculation of mean curvature is made using a moving window of user-defined
size, with different window sizes capturing variation which occurs at different spatial
scales. Nelson (2001) and Wood (1996; 1996) recommends considering multiple
scales in terrain analyses, and aggregating multi-scale results into a single map if the
most appropriate scale of analysis is not known. For Tambito, the mean curvature
was calculated using moving windows of sizes from 3 x 3 cells up to 15 x 15 cells,
with a step interval of 2 cells, and the results of each scale aggregated through equalweighted averaging. There are no precedents of using mean curvature in tropical
forest diversity studies, and for this reason the ideal analysis-scale is unknown. The
multi-scale approach used here does not limit the analysis to a single scale, though

84

later in the thesis when mean curvature is found to correlate, detailed single-scale
comparisons are made for the range of scales from 3x3 to 15x15 cell moving
windows. This strategy permits the identification of potential patterns with mean
curvature without having to analyse numerous scales, but also allows the
identification of the optimal scale when promising correlations are found.

It is

important to note that small window sizes are likely to be more sensitive to DEM
error, and there are means of quantifying this error (Albani et al., 2004), though
measurement of error bounds is not applied in this thesis. The final result of multiscale mean curvature is shown in Figure 13.

Figure 13 Multi-scale mean curvature for Tambito. Blue signifies concave slopes (-3), white
planar slopes with no curvature (0) and red signifies convex slopes (+3).

Numerous other measures of curvature exist, including profile and plan curvature
(curvature only in the direction of slope and perpendicular to it), minimum and
maximum curvature, and longitudinal and cross-sectional curvature. Rather than
complicate the analysis with different measures of curvature, only mean curvature is
85

used here (in addition to toposcale, discussed in section 3.4.1.3.6), though further in
depth studies of the role of different types of terrain curvature on composition,
diversity and structure would be worthwhile.

3.4.1.3.4 TopModel

The most commonly and popularly used terrain index to indicate soil moisture is the
TopModel wetness index (also known as the Compound Topographic Index) (Beven,
1978; Pack et al., 1988; Gessler et al., 1995). Topmodel is calculated using flow
directional paths across the landscape to calculate the upslope flow accumulation area
for any cell, and combining this with slope to indicate the likely levels of surface
overland flow and sub-surface throughflow. Note that it does not take into account
absolute nor spatial variation in levels of atmospheric water input. This terrain
characteristic is in part indicative of soil-water availability to plants.

The equation for calculating the topmodel wetness index is:

Topmodel = ln (Upslope Area)
Tan (slope)

Topmodel was calculated using PCRaster, which uses a third-order finite difference
method for flow direction and accumulation calculation. Streams and rivers are
highlighted as having very high topmodel wetness index values, and ridges and peaks
have very low wetness values (Figure 14). It is important to note that the wetness
value can change abruptly from one cell to the next, with cells adjoining rivers not

86

necessarily having high levels of wetness. This is important in terms of DEM and
GPS positional error, with small errors in positional or elevational error potentially
resulting in significant changes in wetness. Furthermore, the topmodel wetness index
may not be entirely accurate at the 25m-cell scale in Tambito as fine-scale
topographic variation at the ground surface (and sub-pixel topographic variation) is
not captured in the DEM, and may be significant in changing flow directions. Further
to this, some flow lines appear to be predominantly at 45o, 90o, 135o etc. angles, and
this occurs as a result of the D8 algorithm used to define the flow direction in
PCRaster. The combination of these potential errors and the abrupt change in wetness
that occurs between neighbouring cells urges caution when interpreting results
involving topmodel.

Figure 14 Topmodel wetness index for Tambito, dark blue indicating high wetness values.

3.4.1.3.5 Slope Position

87

Slope position is a terrain derivative that measures the relative position of a cell from
valley floor to ridge.

Independent of absolute elevation, this index provides a

measure of the position of each cell with respect to landscape features, namely ridges
and streams. Slopes close to ridges may be more exposed compared with slopes in
valleys where competition for light is likely higher (depending on the steepness of the
valley), and the sheltered topography may create warmer and temporally more
homogenous micro-climates (depending on wind direction). Proximity to streams
may also be important indirectly for presence/absence of seed dispersing species
(birds, mammals).

In the science of terrain analysis slope position is fairly

ambiguous, but numerous tropical diversity studies have used slope position (either
approximated or measured in the field) to explain variation in structure, composition
and diversity (Webb et al., 1999; Clark and Clark, 2000; Valencia et al., 2004), and
for this reason it is used here.

Slope position was calculated using an Arc/Info Arc Macro Language (AML) script
provided to the author by David Hatfield of the United States Department of
Agriculture (USDA) Forest Service. It is calculated using the following equation:

Equation 2

Slope Position = (Elevation – Elevation of nearest valley)

* 100 + 0.5

(Elevation of nearest Ridge – Elevation of nearest valley)

Values in the output grid range from 0 (valley floor) to 100 (ridgetop). Valleys and
ridges are defined through calculation of upslope and downslope flow accumulation
respectively (ridges calculated by multiplying the DEM by –1 and analyzing flow
accumulation of the resultant grid). As this variable is calculated using an Arc/Info
AML flow direction and accumulation is calculated using the basic 8 directional flow

88

method (ESRI, 2001), though it should be noted that better methods for flow
accumulation

exist

using

multi-directional

proportional

flow

algorithms

(TauDEM/TARDEM for example (Tarboton, 2004), also discussed in Mulligan
(2004)).

The AML requires user parameterisation of the flow accumulation threshold beyond
which the cell becomes a valley or ridge. In other words, when large limits are used
only large valleys and ridges will be identified as such, and small valleys and ridges
will be considered somewhere mid-slope. For the purposes of this study a multi-scale
approach was used (though the term multi-scale in this case refers to the scale of flow
accumulation threshold rather than moving window size), through calculation of slope
position using flow accumulation thresholds of 200, 300, 400 and 500. The selection
of these thresholds were based on recommendations from the valley/ridge
identification AML author (Zimmerman, personal communication).

The four

resultant maps were then aggregated through averaging, resulting in a single grid of
slope-position (Figure 15).

89

Figure 15 Multi-scale slope position for Tambito. Red signifies cells close to the ridges
(100), white signifies mid-slopes (50) and blue signifies cells close to the valley bottom (0).

3.4.1.3.6 Landscape Feature Identification

Characterising a landscape into a range of features captures various aspects of the
terrain in one single classification, and these classes may contain distinct
environmental processes generating clear patterns in structure and diversity of the
forest. Though these classifications are in essence a simplified derivative of many of
the variables already discussed (curvature, slope etc.), they capture multiple terrain
attributes creating an easily identifiable and interpretable set of landscape classes.

Two topographic feature extraction algorithms were used in this study. The software
Landserf produces a landscape feature classification using the algorithms of Wood
(1996; 1996; 1998).

This uses a moving window of defined size to classify a

landscape into ridges, passes, peaks, pits, channels and planar slopes. In this case, a

90

moving window of 5 x 5 cells was used to produce two classifications of the
landscape in Tambito; one using the standard method, and the other using an
additional network parameter to ensure connectivity of features such as channels and
ridges. These classifications are shown in Figure 16. In addition to this single-scale
classification, a multi-scale fuzzy feature classification was also used in Landserf,
using scales from 3 x 3 cells to 15 x 15 cells. Once again, this range in moving
windows is selected relatively subjectively, considered to be representative of mesoto micro- scale topographic variability. This fuzzy feature classification provides
membership probabilities to each class, ranging from 1 (signifying membership at all
scales) to 0 (no membership at any scale).

Figure 16 Feature extraction using the Landserf algorithm. The left image shows the
classification of features under standard analysis, and the right image shows the same analysis
with the addition of network rules. In both cases, the landscape is classified into peaks (red),
ridges (yellow), passes (green), planar slopes (grey) and channels (blue).

91

The second algorithm adopted for landscape characterization uses the method of
Zimmerman (2004) to classify into channels, toe slopes, slopes and ridges. This
method is based on moving windows that examine the relative elevation of
surrounding cells to the central cell (curvature). The method differs slightly from the
calculation of curvature. Integration into a single multi-scale measure is achieved by
starting with the standardized exposure values of the largest window then adding
standardized values from smaller windows only where they exceed the values of the
larger scale map.

This produces a terrain derivative in itself (termed toposcale

throughout this thesis), representing a measure of absolute exposure, closely related to
curvature.

To produce a landscape classification, the topographic exposure (toposcale) is then reclassified into four classes, with high positively exposed cells classes as ridges,
neutrally exposed cells as slopes, and high negatively exposed (sheltered) cells
classified as toe slopes and channels.
In this case, moving windows of sizes from 3 x 3 cells to 15 x 15 cells were used to
calculate the topographic exposure, producing a map of topographic exposure
between –370 and +330. Cells with values –670 to –150 were classified as channels,
-150 to –80 were classified as toe slopes, -80 to +80 as slopes, and +80 to +330 as
ridges. These classes were manually selected based on visual inspection of the result,
as recommended by Zimmermann (2004).

Though subjective, this produces a

realistic classification given sufficient field knowledge, though the result is limited in
its application to other regions, and indeed in regions containing contrasting
topography (lowlands to mountains, for example). The result of this analysis is
shown in Figure 17.

92

Figure 17 Calculation of “topoclasses” using Zimmerman’s method. The left figure
represents the toposcale grid (red representing positive values and blue represent negative
values), and the right figure shows the reclassification of this grid into channels (blue), toe
slopes (light green), slopes (darker green) and ridges (brown).

Visually assessing the topclass classification of Figure 17, it clearly fails to identify
the main river channels (Rivers Palo Verde and Tambito) in places, but accurately
captures ridges. In conclusion, greater attention should be given to the feature and
network feature classifications in the analysis rather than the topoclass variables.

3.4.1.3.7 Terrain characteristic summary

12 terrain characteristics have been produced (nine quantitative, three categorical),
and will be used in the analysis. These characteristics represent light related variables
(solar radiation, northness, eastness, slope and to some extent toposcale and mean
curvature), water availability (topmodel, slope and to some extent elevation),
93

temperature (elevation), micro-climate (toposcale, mean curvature and to some extent
slope and slope position) and landscape unit (feature, network feature and topoclass).
Some characteristics are more useful than others, and special attention in the analysis
chapters will be given to elevation, slope, mean curvature, topmodel (taking note of
the potential problems), the feature and network feature classifications and northness,
eastness and solar radiation (the latter three should really be considered together).
The potential ecological significance of slope position is less clear and for this reason
will receive less attention, despite having been used extensively in some other similar
studies (Webb et al., 1999; Clark and Clark, 2000).

Furthermore, toposcale is

considered to be very similar to mean curvature but less rigorously tested and used as
a terrain evaluation method, likewise for its landscape classification variable
topoclass.

In subsequent chapters, co-linearity between variables makes interpretation difficult.
In the following section this co-linearity is calculated and presented, but it is also
useful to show how the different variables are calculated and show in a graphic form
the path of derivation for each variable Figure 18.

94

Elevation
(DEM)

Flow
Direction

Slope

Flow
Accumulation

Toposcale

Mean
Curvature

Aspect
Eastness

Northness

Topmodel
Solar
Radiation

Slope
Position

Figure 18 Derivation of each terrain characteristic from the original DEM

3.4.1.3.8 Cross-Correlation Matrix

Many of these characteristics use similar methods and base themselves on the DEM in
order to calculate specific properties, and so it is important to examine the crosscorrelation of each pair of variables. Table 2 provides a summary of values for each
map, and Table 3 shows a cross-correlation matrix between these various terrain
properties.

In many of the analyses separations are made between continuous

variables and the categorical variables (feature, network feature and topoclass).

95

Variable (unit)
DEM (m)
Slope (degrees)
Eastness (index)
Northness (index)
Slope Position (index)
Mean Curvature (index)
Toposcale (index)
2
Solar Radiation (J/m )
TopModel Wetness Index

MIN

MAX

MEAN

1144
0.2
-1
-1
0
-2.99
-368
5598
6.1

2903
57.8
1
1
101
2.96
331
9662
20.8

2022.67
29.85
-0.40
0.04
51.08
0.01
-0.47
8996.55
9.00

Standard
Deviation
425.46
9.89
0.59
0.70
30.07
0.78
100.10
671.50
1.58

Table 2 Summary values for all Tambito continuous terrain characteristics.

Layer

DEM

Slope

Eastness

Northness

Slope
Position

Mean
Curvature

Toposcale

Solar
Radiation

Topmodel

DEM

-

-0.09

0.13

0.05

0.39

0.24

0.28

0.02

-0.08

-

0.02

0.15

-0.14

-0.09

-0.10

-0.58

-0.33

-

0.15

0.00

0.00

0.00

-0.42

-0.03

-

0.01

0.01

0.00

-0.35

-0.04

-

0.77

0.82

0.09

-0.42

-

0.94

0.06

-0.50

-

0.07

-0.46

-

0.17

Slope

Eastness

Northness
Slope
Position
Mean
Curvature
Toposcale
Solar
Radiation
Topmodel

-

Table 3 Cross-correlation matrix between all continuous terrain properties used in the
Tambito case study. Values represent the correlation coefficient. Correlation coefficients
greater than 0.7 (positive or negative) are highlighted in dark grey, and light grey is used for
correlation coefficients of 0.3 to 0.7 (n=69, 678). In reality, most of these correlations are
significant (95% level) due to the large number of samples.

The most significant correlations occur between slope position, mean curvature, and
toposcale. Toposcale logically negatively correlates with slope position indicating

96

that valleys tend to be found in areas with a low slope position. Toposcale correlates
closely with mean curvature, due to their similarity in calculation with only the
hierarchical multi-scale method used in toposcale causing a difference in the two
results.

3.4.2 Tiputini Biodiversity Station

Field studies in lowland rain forest were made in Tiputini Biodiversity Station.
During the process of this study hydrological and meteorological monitoring stations
were installed, and some of this data is available for this thesis.

3.4.2.1 Meteorological and hydrological monitoring

A Campbell Scientific automated weather station installed by Boston University has
been operating in the station since 1997, though the climate data is intermittent with
periods of no data due to technical failures. From Sept. 1997 – March 2003 the logger
has a total of 1339 days of data (67% of the time), with Figure 19 showing the period
when monitoring was made, and the daily rainfall for these days. For the period of
1997-2004, mean temperature has been 24.8oC, with a daily mean maximum and
minimum temperature of 28.3oC and 22.4oC respectively. The lowest temperatures
are in June (23.7oC), and the highest temperatures in January (25.8oC). Mean annual
rainfall is 2895 mm, with the majority of rainfall occurring between 2pm and 5pm.
The greatest rainfall occurs in July (409mm/month), and the least in January
(89mm/month). Relative humidity is on average 96.6% (highest in June/July and
lowest in December/January). Monthly average radiation varies from 80 Wm-2 in
June to 189 Wm-2 in January. Hourly average radiation is highest at 1200 hours (801

97

Wm-2) and is less than 5 Wm-2 from 8pm through to 6am. Hourly wind speed
averages 0.57 ms-1, and is highest when radiation is lowest. Hourly wind speed peaks
from 1pm-3pm at 0.81 ms-1 and follows a sinusoidal diurnal pattern (matching that of
radiation). The highest wind speeds are from the S, SE and NE. Wind direction is
usually from the SE (32%) or else NE, E or SE (18% each), 7% of winds are from the
SW and less than 5% from all other directions. More detailed analysis of wind speeds
are included in Section 3.4.1.3.2.

10 Day Running Mean Rainfall Record for TBS, 1997 - 2003
30

25

Rainfall (mm)

20

15

10

5

1998

Day of Year

2002

2003

Figure 19 10-day running mean rainfall series for Tiputini Biodiversity Station 1997 – 2003,
recorded with the Boston University Campbell Scientific meteorological station installed at
the station facilities. Note the missing data from 1999 – 2001.

3.4.2.2 DEM generation

Topography plays a fundamental role in examining relationships between biophysical
parameters and tree diversity and structure, but the generation of a DEM for TBS

98

74

55

36

17

352

333

314

294

275

173

154

135

97

116

59

40

1

347

318

299

279

252

233

213

194

175

156

137

99

118

80

60

41

3

22

349

330

311

292

273

254

235

1997

20

MISSING
DATA

0

provided a considerable challenge to this study. Prior to this research no DEM was
available for the study site, and the topographic landscape made the generation of a
representative DEM difficult. Whilst the study site is very flat on the macro-scale,
there are complex networks of streams that have created complex micro-scale
topographic variation, with some steep-sided valleys of just 10m depth and 30m
width, for example. In many cases this ground level topographic variation is not
evident in the canopy, causing cartographic maps to overly simplify the topography.
For these reasons, different sources of DEM data were sought, with the hope of
finding the best representation of small-scale topographic variation, whilst capturing
the macro-level differences.
Three potential sources of DEM data were available for the TBS study site. These
were 1:50,000 cartography (sheet P111-D4 Zamona Yuturi), the SRTM mission 3-arc
second DEM product, and ASTER stereo pair satellite images. DEMs were generated
for each of these sources.

The 1:50,000 cartography was scanned and georeferenced using ArcView. Contours
(20m interval), rivers and spot heights were then digitized on screen, and the
TOPOGRID command in Arc/Info used to produce a 25m DEM using the same
parameters as were used in Tambito. The resultant DEM can be seen in Figure 20,
along with some basic derivatives (slope and aspect) for assessing DEM quality. As
can be seen, the slope is greatest around contour lines, and the interpolation algorithm
fails to produce smooth slopes between the contour lines, especially where there are
large distances between 20m contour lines. This is a known limitation of many
interpolation techniques for DEM generation in flat regions (Hutchinson, 1989), and
one that could not be overcome despite numerous attempts using different tolerances.

99

Figure 20 Results of DEM generation using the cartographic maps, producing a 25m grid cell
resolution DEM (top), with the derived slope (left) and aspect (right). Slope varies from 0o
(light blue) to 31o (red). Aspect is in degrees, ranging from 0o (red) to 90o (white) to 180o
(blue) and 270o (white).

SRTM data was made available by NASA in March 2003, providing a 3-arc second
DEM for the study area derived from stereo radar data. TBS was extracted from this
global dataset, and is displayed in Figure 21 along with derivates (aspect and slope).

100

The spatial resolution at this equatorial latitude is 92m. None of the no-data cells in
the SRTM data which caused problems in Tambito were present in the TBS study
site.

Figure 21 Results of DEM generation using the SRTM satellite derived DEM producing a
100m grid cell resolution DEM (top), with the derived slope (left) and aspect (right). Slope
varies from 0o (cream) to 12o (red). Aspect is in degrees, ranging from 0o (red) to 90o (white)
to 180o (blue) and 270o (white).

101

The third and final source for DEM data used in this study was from the ASTER
satellite sensor. The ASTER satellite uses two telescopes to acquire stereo pair
images in the visible and near-infra-red bands (VNIR) with spatial resolution of 15m.
The primary sensor takes vertical images, whilst the backward looking telescope
compliments this image with a sideways facing image taken just a few seconds after
the vertical image. The stereo pair can then be used to produce high resolution
DEMs, with 15m resolution. It is important that no cloud is present in the original
image for DEM generation to be successful and since stereo properties are used, DEM
production is better in areas of variable topography (where vertical height differences
are greater than the horizontal grain of the ASTER image itself – 15m). This is not
always the case to TBS. Furthermore care is suggested due to the documented effect
of noise in the spectral bands of the stereo-pair relative to the cell to cell height
differences in this area (Toutin, 2002).

The ASTER Earth Observing System Data Gateway (EDG) was used to search for
cloud free ASTER images covering TBS. Only one suitable image was available with
15% cloud cover from 12th April 2003, which included some light cloud to the northwest of TBS. This ASTER image is shown in Figure 22.

102

Figure 22 The VNIR ASTER image from 12th April 2003, used to extract a 15m DEM,
displayed using bands 2 (red), 3 (infra-red) and 1 (green). Whole image (left) shows heavy
cloud cover in the western and south-eastern sector of the image, but TBS is cloud free
(right).

The ASTERDEM extension (AsterDTM-V2, 2004) to PCI image processing software
was used to automatically extract the DEM without the need for any ground control
points or user input. The result is a geopositioned DEM with relative elevation
values. These values were converted into metres above sea level using the SRTM
data as a calibration set for the entire image (vertical error in SRTM data is claimed to
be less than 16m). The resultant DEM of TBS is shown in Figure 23, along with
derived slope and aspect.

103

Figure 23 Results of DEM generation using the ASTER satellite image producing a 15m grid
cell resolution DEM (top), with the derived slope (left) and aspect (right). Slope varies from
0o (light blue) to 68o (red). Aspect is in degrees, ranging from 0o (red) to 90o (white) to 180o
(blue) and 270o (white).

Though the higher spatial resolution of the ASTER DEM is appealing, and the fact
that the ASTER DEM captures the general topographic trend of higher elevations in
the north-central zone of the DEM, and in a broad sense the river channel, serious
questions are raised about the validity of the DEM for capturing the micro-scale
topographic variation. The undulating peaks observed in Figure 23 are unlikely to be

104

representative of the true topography, though this DEM is the only one to contain
steep slopes that are observed in the field at the micro-scale. Furthermore, the DEM
contains 46,000 sinks (of the 396,000 total cells), rendering it hydrologically unsound,
and methods for pit removal (in Landserf, for example), fail to adequately correct this
problem. For these reasons, and the fact that vertical height differences are not
greater than the horizontal grain of the ASTER image in TBS, the ASTER DEM is
not used in the analysis, as little confidence can be placed on the quality of the DEM.

Excluding the ASTER DEM, the two DEMs (TOPO and SRTM) capture the general
trend of higher altitudes in the northern part of the study area, and south-west of the
reserve on the southern side of the Tiputini River. However, the DEMs differ greatly
on the fine scale, neither adequately capturing the true micro-scale topographic
variation observed in the field. The cartographically derived DEM (hereon referred to
as “TOPO DEM”) is hydrologically sound, and shows predominantly smooth
networks of ridges and valleys, but fails to capture some of the micro-scale
topographic variation observed on the ground.

The SRTM DEM is also

hydrologically sound, but the coarse resolution captures even fewer micro-scale
features, representing the study site as topographically homogenous and flat. Both the
TOPO DEM and the SRTM DEM likely underestimate the topographic variation.
Without a lengthy high precision GPS survey it is impossible to assess which DEM is
most representative of the true topography. Both DEMs will be used in the analysis.
The TOPO DEM is likely the better quality DEM due to its higher resolution.
However, the potential applicability of these analyses in other tropical forested
regions is significantly easier using the SRTM DEM, as it has global coverage and is
readily available (see for example http://srtm.csi.cgiar.org). Table 4 shows summary

105

statistics for each DEM. The cross-correlation (Pearson’s Correlation Coefficient)
between the TOPO DEM and the SRTM DEM (calculated after resampling of the
TOPO DEM to the coarser resolution of the SRTM DEM (92m) using Arc/Info
RESAMPLE) is 0.76, indicating that there are differences between the two. Section
3.7 examines these differences in greater detail.

Layer
TOPODEM
SRTM DEM

Cell Size
(m)
25.0
92.6

No. of
Cells
142450
10400

Minimum Maximum
193.7
202.0

269.0
278.0

Mean
227.6
236.5

Standard
Deviation
14.54
9.73

Table 4 TBS DEM summary statistics (above) and cross-correlation matrix (below) showing
correlation coefficients, n=10,400 (all DEMs were re-sampled to 92m to enable direct
comparison).

3.4.2.3 Terrain characteristics

As for Tambito, the same terrain characteristics were calculated for comparison with
measured diversity and structure in the plots. Many of the methods were the same as
have already been described in Section 3.4.1.3 of this chapter.

3.4.2.3.1 Solar Radiation Receipt

The solar radiation sub-model of the BENDUM hydrological model reported in
(Mulligan, 1999) was used on each DEM to calculate the total solar radiation receipt
expected for each pixel.

Once again, no correction was made for atmospheric

attenuation or cloud cover.

106

Figure 24 Total potential solar radiation receipt across the TBS study site, using the
TOPODEM (left) and SRTM DEM (right). Blue indicates lower receipt, green intermediate
and yellow high.

Both the absolute and relative values for the two DEMs vary greatly, with the TOPO
DEM being more variable (given the steeper slopes) with radiation receipt ranging
from 8657 – 9662 Wm-2 and the SRTM DEM having very little variability (9467 –
9662 Wm-2). These differences are brought about by the steeper slopes present in the
TOPO DEM.

3.4.2.3.2 Quantitative Measures of Aspect

Aspect was converted to northness and eastness for each of the two DEMs following
the same method of Zar (1999) adopted for Tambito.

Northness and eastness

variables in TBS are relevant in terms of solar radiation receipt and in terms of
exposure to wind. Given the low slopes, solar radiation receipt varies little, and
indeed northness and eastness variation are likely to have little influence on total solar
radiation receipt and the daily and annual variation in this receipt (unlike in Tambito).
Data from the Boston University automated weather station and the modelled solar

107

radiation receipt for the cell representing that station is presented for monthly
variation (Figure 25) and for daily variation (Figure 26) in solar radiation receipt.

Radiation Receipt (Wm-2)

300
250
200
150
100
Measured Solar Radiation Receipt

50

Modelled Solar Radiation Receipt

0
1

2

3

4

5

6

7

8

9 10 11 12

Month

Figure 25 Monthly variability in solar radiation receipt for TBS, showing modelled
(BENDUM) and measured receipt for the Boston University AWS. The measured data is
averaged for data periods between 1997 – 2004.

1000
Measured Solar
Radiation Receipt
Modelled Solar
Radiation Receipt

800
700
600
500
400
300
200
100

Time of Day

108

23

21

19

17

15

13

9

11

7

5

3

0
1

Radiation Receipt (Wm -2)

900

Figure 26 Daily variability in solar radiation receipt for TBS, showing modelled (from the
BENDUM model) and measured receipt for the Boston University AWS. The measured data
is averaged for data periods between 1997 – 2004.

Though the modelled solar radiation receipt is bi-modal, with greatest potential
receipt in March and September, variation in cloud cover produces a uni-modal
distribution in actual receipt, with greatest receipt in January, and least receipt in June
and July due to increased cloud cover. Daily variation follows much more closely the
modelled receipt, with slight lowering in radiation receipt during the early afternoon
due to increased cloud cover (12pm – 3pm). Despite this variation, the low slopes
and likely homogenous cloud cover over the study site mean that northness and
eastness have little relevance for light in TBS.

However, of greater significance in TBS than was the case in Tambito is wind, and
northness and eastness measure the exposure to wind. Data from 1997 to 2004
(containing some missing periods) from the Boston University weather station
installed in TBS is analysed to show that the average wind speed is 0.51ms-1. The
highest windspeeds occur from the SE direction (average 0.57ms-1), and indeed this is
the dominant wind direction for 32% of the time. 85% of the time wind blows from
the directions of NE, E, SE or S (Figure 27). The highest absolute wind speeds are
also in the SE direction, with the maximum windspeed measured at 7ms-1 in the SE
direction, and over 500 hours of wind above 2.5ms-1 in this direction, compared to
just 2 hours in the NW direction (Figure 28). After the SE direction, the easterly
direction is frequent (dominant direction 17% of the time) and has the highest
frequency of high winds (79 occurrences above 2.5ms-1).

109

Average Wind Speed (ms-1) in TBS
for each direction

Frequency of wind in each
direction for TBS

N

N

0.6

NW

5000

NE

0.4

NW

4000

NE

3000
2000

0.2

1000

W

E

0

SW

W

SE

E

0

SW

SE
S

S

Figure 27 Wind speed and wind directional frequency in TBS based on 7 years (intermittent)
of measured wind data from the Boston University automated weather station in TBS.

Frequency of wind speeds for each direction in TBS

Frequency (log scale)

10000
N
NE
E
SE
S
SW
W
NW

1000

100

10

4
4.
4
4.
9
5.
4
5.
9
6.
5
7.
2
9.
4

3
3.
5

2
2.
5

0
0.
6
1.
1
1.
6

1
-1

Wind speed (ms )

Figure 28 Frequency of wind events for different wind speeds separated for each direction
TBS based on 7 years (intermittent) of measured wind data from the Boston University
automated weather station in TBS.

110

The significance of this for eastness and northness is that regions with high –
intermediate positive values of eastness (east facing) receive the greatest winds.
Northness indicates less in terms of wind direction, though slopes with low northness
and with high eastness (i.e. SE facing) are exposed to wind.

3.4.2.3.3 Mean Curvature

Mean curvature was calculated for TBS with the same method adopted for Tambito,
using moving windows of sizes 3, 5, 7, 9, 11, 13, and 15 cells for the TOPODEM
(25m cell size like Tambito), and 3 and 5 cells for the SRTM DEM, once again to
ensure that each result is at the same geographic scale. Limitations in requiring odd
sizes of moving window greater than 3 cells explain the use of these window sizes.
The results of each scale were aggregated using equal-weighted averaging. The final
result of multi-scale mean curvature is shown in Figure 29.

Figure 29 Multi-scale mean curvature for TBS using the TOPODEM (left) and SRTM DEM
(right). Blue signifies concave slopes (-3), white planar slopes with no curvature (0) and red
signifies convex slopes (+3).

111

3.4.2.3.4 Topmodel Wetness Index

Topmodel wetness index was calculated for each DEM using the same method as was
used for Tambito. However, in the case of TBS TopModel was calculated on the
larger DEM area in order to capture more accurately the wetness including the entire
catchment.

In reality, this was only possible for the SRTM DEM, as this has

complete continental coverage, but a region including a large proportion of the lower
catchment was included in the calculation for the TOPODEM (55km x 40km). This
must be considered a limitation, and potential source of error in the topmodel
variable, especially for the TOPO DEM.

Figure 30 Topmodel wetness index for TBS, using the TOPODEM (left) and SRTM DEM
(right), dark blue indicating high wetness values.

The TOPODEM and SRTM DEM both capture the main river channel, although the
SRTM DEM loses the continuous flow direction for a short stretch of the river below
TBS. This is not important to the result, as all plots are upstream of this, and none are
located in the channel itself.

112

3.4.2.3.5 Slope Position

Slope position was calculated using the same procedure as for Tambito for both of the
DEMs (Figure 1). The flow accumulation thresholds varied between DEMs based on
the cell size in order to ensure that the results were directly comparable at the same
geographic scale, regardless of cell size. The TOPODEM used thresholds of 200,
300, 400 and 500, and the SRTM DEM used thresholds of 54, 81, 108, 135 (smaller
thresholds due to the larger cell size). The result for each threshold was aggregated
using equal-weighted averaging in all cases.

Figure 31 Multi-scale slope position for TBS for the TOPODEM (left) and SRTM DEM
(right). Blue signifies cells close to the ridges (100), yellow signifies mid-slopes (50) and
green signifies cells close to the valley bottom (0).

3.4.2.3.6 Landscape Feature Identification

113

Landscape features were identified using the same method as for Tambito, using the
algorithm described by (Wood, 2004). A moving window of 7 x 7 cells was used for
the TOPODEM and 3 x 3 cells for the SRTM DEM. These classifications are shown
in Figure 32. Additionally, multi-scale fuzzy feature memberships were calculated.

Figure 32 Feature extraction using the Wood’s algorithm (2004) using the TOPODEM (top),
SRTM DEM (bottom). The left image shows the classification of features under standard
analysis, and the right image shows the same analysis with the addition of network rules. In
both cases, the landscape is classified into peaks (red), ridges (yellow), passes (green), planar
slopes (grey) and channels (blue).

Whilst the TOPO DEM produces a feasible distribution of channels, ridges and some
peaks, especially for the network feature classification, the result of the feature and

114

network feature classification for the SRTM DEM appears to reflect little of the actual
landscape found in TBS, with not even the main River Tiputini being classified as a
channel. This is likely due to the larger cell size, and relatively minor topographic
variation present in the DEM.

Zimmerman’s (2004) method was also used to identify landscape features, using
moving windows of sizes 3, 5, 7, 9, 11, 13, and 15 cells for the TOPODEM, and 3 and
5 cells for the SRTM DEM.

Cells with “toposcale” values –670 to –150 were

classified as channels, -150 to –80 were classified as toe slopes, -80 to +80 as slopes,
and +80 to +330 as ridges. Once again, this subjective selection of limits was made
based on visual interpretation of the resultant classification. This is a strong limitation
for the quality and applicability of this terrain characteristic in other regions. The
result of this analysis is shown in Figure 33.

115

Figure 33 Calculation of “topoclasses” using Zimmerman’s method, using the TOPODEM
(top), SRTM DEM (bottom). The right figure represents the toposcale grid (red representing
positive values and blue represent negative values, and grey represent neutral values), and the
left figure shows the reclassification of this grid into channels (blue), toe slopes (light green),
slopes (darker green) and ridges (brown).

Once again, the TOPO DEM appears to produce the best landscape classification, and
it is important that this be considered in the following analysis chapters. The SRTM
DEM classification fails in some sections to capture the main channel.

3.4.2.3.7 Summary of terrain characteristics and DEMs

As for Tambito, 12 terrain characteristics have been produced (nine quantitative, three
categorical). These characteristics represent light related variables (solar radiation,
northness, eastness, slope and to some extent toposcale and mean curvature), water
availability and disturbance (topmodel, slope and to some extent elevation and slope
position), temperature (elevation), micro-climate (toposcale, mean curvature,
eastness) and landscape unit (feature, network feature and topoclass). Once again
extra attention is given to certain characteristics, whilst caution is urged in analyzing
toposcale, topoclass and slope position.

Also of importance in TBS are the two DEMs. In almost all terrain characteristics,
the TOPO DEM visually performs the best, with the SRTM data producing on the
whole realistic looking results, but with a greater cell size that results in a loss of
some topographic detail. In conclusion, the TOPO DEM is likely the best DEM in a
general sense, though correlations with the SRTM DEM are of significant interest due
to its global coverage and potential application in other regions.
116

3.4.2.3.8 Cross-Correlation Matrix

Table 5 provides a summary of values for each terrain derivative, and Table 6 and
Table 7 show cross-correlation matrices between these various terrain properties for
each DEM. The correlations are not explicitly discussed in this chapter, though these
results are relevant to some of the results interpretation in later chapters.

Variable (unit)
DEM (m)
Slope (degrees)
Eastness (index)
Northness (index)
Slope Position (index)
Mean Curvature (index)
Toposcale (index)
Solar Radiation (J/m 2 )
TopModel Wetness (index)

MIN
193.7
0.0
-1.0
-1.0
-48.0
-1.9
-661.3
8657.0
7.1

TOPODEM
MAX
MEAN
270.0
227.5
31.2
3.1
1.0
0.0
1.0
0.0
149.0
51.3
2.6
0.0
706.2
-0.9
9662.0
9632.1
25.4
11.6

STD
14.74
3.41
0.69
0.72
35.14
0.16
90.36
51.99
2.16

MIN
202.0
0.0
-1.0
-1.0
0.0
-1.0
-198.3
9457.0
8.1

SRTM
MAX
278.0
10.5
1.0
1.0
100.0
1.0
225.1
9662.0
23.4

DEM
MEAN
236.24
2.08
0.03
0.01
47.15
0.00
0.19
9646.89
11.1

Table 5 Summary values for all TBS terrain characteristics, using both DEMs

117

STD
9.9
1.4
0.7
0.7
30.0
0.2
55.1
15.6
1.91

DEM (m)

Slope (degrees)

Eastness (index)

Northness (index)

DEM (m)

Slope
(degrees)

Eastness
(index)

Northness
(index)

Slope
Position
(index)

Mean
Curvature
(index)

Toposcale
(index)

Solar
Radiation
2
(mmol/m )

TopModel

-

0.01

0.01

0.04

0.40

0.24

0.40

0.01

-0.21

-

0.03

-0.01

-0.03

0.05

0.04

-0.80

-0.55

-

0.00

0.02

0.00

0.00

0.03

-0.03

-

0.01

0.00

0.01

0.20

0.01

-

0.41

0.61

0.02

-0.32

-

0.73

-0.04

-0.33

-

-0.03

-0.32

-

0.38

Slope Position (index)
Mean Curvature
(index)
Toposcale (index)
Solar Radiation
2
(mmol/m )

-

TopModel

Table 6 Cross-correlation matrix of terrain characteristics for the TOPO DEM for TBS (n = 143,220)

118

SRTM DEM

DEM (m)

Slope
(degrees)

Eastness
(index)

Northness
(index)

Slope
Position
(index)

Mean
Curvature
(index)

Toposcale
(index)

Solar
Radiation
(J/m2)

TopModel

DEM (m)

-

0.15

0.01

0.00

0.50

0.43

0.59

-0.12

-0.40

-

-0.05

0.00

0.01

0.08

0.03

-0.77

-0.48

-

0.02

0.01

0.00

0.02

0.22

0.02

-

0.00

0.00

0.00

0.44

0.01

-

0.46

0.61

-0.01

-0.39

-

0.34

-0.05

-0.50

-

-0.04

-0.34

-

0.32

Slope (degrees)

Eastness (index)

Northness (index)

Slope Position (index)
Mean Curvature
(index)
Toposcale (index)
Solar Radiation (J/m2)

-

TopModel

Table 7 Cross-correlation matrix of terrain characteristics for the SRTM DEM for TBS (n = 10,504).

119

3.5

Establishment of plots

Tree diversity plots were established in order to examine landscape controlled
patterns in diversity and structure, and to provide validation datasets for testing some
of the modelling hypotheses.

These plots were established using methodologies

designed to permit easy comparison with spatial datasets, without compromising the
quality of the biological data gathered in the field. Transects such as those used by
Gentry (1995) would be of little use to this study, as the plots would cover numerous
incomplete pixels in the GIS data. It was important that the same method be used in
both field sites, and so given that both field sites had topographic data available at
25m resolution, field plots of 25m x 25m were used in this study. This size represents
a fair balance between having sufficient individuals in the sample to provide a
realistic figure for the diversity and structure at that point (and not being overly
sensitive to the precise positioning of the corners), without limiting the number of
different plots that could be established given the time and resource constraints. It
was important to have sufficient plots to have statistical significance in the
comparisons with environmental data, and so ten plots were established in each field
site. Whilst more plots would be preferable, limited time and resources did not permit
anymore than 10 plots.

3.5.1 Positioning of plots

Access and field conditions were markedly different between the two field sites, and
so there were some minor differences in the field sampling. Poor GPS coverage,
steep slopes and a dense understorey made conditions considerably more difficult in
Tambito. Though no formal sampling scheme was used, the plots were located in

120

regions geographically distinct, evenly covering the elevational gradient and covering
both catchments of Tambito. Existing paths were used to locate six of the ten plots,
with the location of the corner decided when a predetermined altitude was reached
(measured using an altimeter), and by walking 5m to the right of the path. In some
cases the topography made it impossible to establish a plot in that exact position, and
so the closest feasible area was used. The remaining four plots were established along
paths especially opened for this study, geographically distant from the first six plots.
Plots were marked based on tape measurement of 25m x 25m areas on the ground,
hence under steep conditions the plan area of the plots might be expected to be less
than 625m2. This is a standard means of demarcating plots in ecological studies
(Phillips et al., 2003) though attention should be noted that the plots do not therefore
necessarily completely compare to the GIS-based cell size of the DEM in steep
environments. A Garmin 12XL was used to locate the geographic position of the
center of each plot, using averaging for at least 10 minutes per plot. The dense
vegetation and steep topography often interfered with the GPS signal, resulting in
horizontal RMS errors of up to 20m. Efforts were made to use differential GPS and
take plot coordinates with better GPS equipment, but the signals were to weak for a
Trimble ProXL GPS to function.

The distribution of these plots around the

catchments is shown in Figure 34.

121

Figure 34 Position of plots in Tambito, positioned in both the Palo Verde and Tambito
catchments

Conditions in Tiputini allowed easy access to any point in the study site, due to a
thorough system of paths, a cleaner understorey and lower slopes. Under these
circumstances, a formal sampling strategy was made using the TOPO DEM (the only
DEM available early in the study). The aim of the sampling strategy was to choose a
minimum number of unrelated factors, and attempt to locate a plot in each
combination of these factors, thereby producing a stratified sampling of the landscape
characteristics. Altitude was selected as one of these factors, important for indicating
the frequency of flooding from rise in the water level in the river (discussed in Section
3.3)). The 25m TOPO DEM for TBS was classified into four equal area classes
(quartiles, Class 1 : 193 – 210m, class 2 : 210 – 220m, class 3 : 220m – 236m and
class 4 : 236 – 270m). In this case Class 1 represented areas regularly flooded, with

122

class 2 receiving occasionally floods, and classes 3 and 4 representing terra firma
forest where no flooding takes place, based on the observed rise in river level whilst
in the field. Equal area classes were used to ensure objectivity, though an alternative
method may have been to use subjectively chosen but more meaningful cut-offs
between classes.

The other factor chosen was upslope area (also known as flow accumulation),
indicative of soil wetness, proximity to river and slope position. This was also
classified into four equal area classes (Class 1 : 0 – 3 upslope cells, Class 2 : 3 – 8
upslope cells, Class 3 : 8 – 21 upslope cells, and Class 4 : > 21 upslope cells).

All permutations of these 2 factors with 4 classes were then calculated, producing 16
‘sampling units’ for TBS shown in Figure 35.

123

Figure 35 Sampling strategy used to locate ground plots in TBS

Although this sampling strategy could be improved, at the time of field sampling
many of the now available terrain characteristics had not been developed. The need
for stratified sampling schemes for plot establishment was raised by Phillips et al.
(2003) and this is the first known use of this type of a priori environmentally based
sampling site determination.

Plots were located at pre-defined locations so that 10 of the 16 sampling units were
represented (Table 8). Care was also taken to ensure that the plots covered the full
geographic coverage of the station. In each case the NW corner of each cell in the
GIS data was located using a Trimble Pro XL GPS in the field, and the plot marked
out to represent exactly one cell in the GIS. Inevitably GPS error (+/- approximately
10m) caused deviation from the pre-determined plot location, though this was actually
minimal (average deviation from the exact plot corner of 3m).

In addition to the 25m x 25m plots, an additional 1-Ha plot was studied, taking
advantage of an already established plot of Nigel Pitman (Pitman, 2000), though the
data from this plot is not presented in this thesis due to incomparability.
geographic distribution of the plots is shown in Figure 36.

124

The

Plot
1
2
3
4
5
6
7
8
9
10

DEM Class
1
3
2
1
1
2
2
4
4
4

Upslope Class
1
3
4
2
4
3
2
2
3
1

Sampling Unit
1
11
14
5
13
10
6
8
12
4

Table 8 Sampling strategy units for each of the ten established plot.

Figure 36 Geographic distribution of the 10 25m x 25m plots established in Tiputini
Biodiversity Station.

3.5.2 Plot Descriptions

125

These short descriptions of the plots are intended only to provide some background
should the reader wish to get an idea of plot conditions when individual plots are
referenced in the following chapters.

To ensure some standardization of the

descriptions, they are presented in table form with the most striking features
described. This information is based on purely subjective descriptions of conditions
observed in the field whilst establishing the plot.

3.5.2.1 Tambito

The plots in Tambito were distributed across a large elevational gradient,
encapsulating some very different environmental conditions (Table 9).

Plot

Elevation

Approx.
Slope
(degrees)

1
2
3
4
5
6
7
8
9
10

1651
1684
1449

29

Ridge

1966

24

Planar hillside

1299

31

Steep valley bottom

1749

33

Hillside w ith ridges

2253

22

Hillside below peak

1856

32

Hillside below ridge

1950

9

Flat ridge

1600

27

Planar hillside

Topography

Understorey

27

Planar hillside

34

Hillside w ith ridges

Light
Dense
Dense
Medium
Light
Medium
Dense
Dense
Light
Medium

Presence of
Epiphytism
Cloud
Medium
Medium
High
High
Low
Medium
High
High
High
Medium

Frequency Light Levels
of tree falls at ground

Low
Medium
Medium
Medium
Low
Medium
High
High
High
Medium

Low
High
High
Medium
Low
High
Medium
High
Low
Medium

Low
High
High
Medium
Low
Medium
Low
High
Low
Medium

Table 9 Brief description of conditions in the Tambito plots.

3.5.2.2 Tiputini

The plots in Tiputini were visibly less variable than those for Tambito, but still some
important characteristics are worth mentioning (Table 10).

126

Plot Elevation
1
2
3
4
5
6
7
8
9
10

208
219
210
201
200
220
219
262
238
224

Approx
Slope
(degrees)
2
20
5
1
2
5
10
5
5
8

Topography

Understorey

Flat
Steep hillside
Undulating with small river
Flat
Flat with channel in corner
Planar hillside
Undulating with small river
Planar hillside
Planar hillside
Undulating

Light
Dense
Medium
Medium
Light
Medium
Light
Light
Light
Medium

Light
Frequency
Levels at
of tree falls
ground
Low
Medium
High
Low
Low
Medium
Medium
Medium
Low
High
Medium
Medium
Medium
Medium
Low
Medium
Medium
Medium
Medium
Medium

Proximity to River

Flooding
Regularity

Directly alongside river (7m above normal flow level)
No river closeby
Small river through plot, but distant from main channel
50m from main river
Directly alongside main river (3m above normal flow level)
200m from main river
500m from main river, but alongside large stream
Distant from any river
Distant from any river
Distant from any river

Occasional
Never
Occasional
Regular
Regular
Occasional
Occasional
Never
Never
Never

Table 10 Brief description of conditions in the TBS plots.

3.5.3 Measurement and sampling of individuals

Field data collection in the 25m x 25m plots was essentially the same for both
Tiputini Biodiversity Station and Reserva Tambito. Diameter at breast height (DBH)
was used to decide which individuals would enter the data collection. All individuals
with DBH greater than 5cm were included, chosen to represent all established trees
and to exclude saplings. For each individual the following measurements were taken:

•

Diameter at breast height (DBH)

•

Diameter at base

•

X, Y and Z Position in plot

•

Height of first branch (approximated)

•

Total height of tree (approximated)

•

Botanical sample collected

These structural factors were measured based on them being simple and quick to
measure in the field. Tree girth (DBH), tree height and basal area (calculated from
diameter at base height) are the most basic structural parameters for a tree, and are
measured in all studies of structural variation in tropical forests (Federov, 1966;

127

Proctor et al., 1983; Vasquez and Givnish, 1998; Webb et al., 1999; Clark and Clark,
2000; Ferreira, 2000; Koponen et al., 2004).

Height of the first branch is also

measured as an indicator of the vertical shape of the canopy, with emergent trees
having the first branches high in the canopy, whilst sub-canopy individuals often
branch much lower in the trunk relative to the total height.

The only difference in data collection between sites was that TBS benefited from the
use of a laser rangefinder to provide precise x, y and z coordinates for each tree within
the plot (relative to the corner GPS position). The laser rangefinder was also used to
measure the height of trees above 10m in TBS. In Tambito, each plot was separated
into 25 quadrants of 5m x 5m and each individual was positioned to the quadrant. No
data were taken on exact x, y and z positions of each tree, except for 3 plots where the
laser rangefinder became available and the plots were revisited (plots 1, 2 and 3).
Unfortunately, local political instability from 2002 – present made subsequent visits
to Tambito impossible, and so accurate x, y z positions for other plots was not
possible.

All tree heights were approximated by the author in all plots and in both sites, except
that trees greater than 10m in height were measured using the laser rangefinder in
TBS. The laser rangefinder was used as a guide to make the manual approximations,
with regular checks on approximations using the rangefinder. Whilst the absolute
heights may contain some error, it is expected that the relative heights within and
between plots are representative.

128

Each individual was collected and preserved temporarily in alcohol according to the
traditional methods of botanical sample preservation. The samples were then dried
for 48 hours at 32oC, and stored as a permanent collection in the Natural History
Museum of Popayan (Tambito samples) and the University San Francisco of Quito
(TBS samples). Identification of the samples was made to species level whenever
possible, though many identifications could be no more precise than genus level.
Emphasis was made in distinguishing species, even where the species could not be
named. The full taxonomy may not have been ascertained for all individuals, but
there is accurate data on whether an individual is of the same species as any other
individual.

3.5.4 Production of plot DEMs

The tree position data taken using the laser rangefinder in TBS and for the three plots
in Tambito provided x, y and z coordinates of every individual in the plot. This data
was used to produce high resolution digital elevation models for each plot. The
elevation values in these plot DEMs represent the elevation of the ground surface, as
measured by ground based topographic survey using a rangefinder with combined
digital compass and inclinometer. Inverse distance weighted interpolation, using the
nearest 10 points was used to create 50cm resolution DEMs. Slope was calculated as
a derivative. The plot DEMs for the 25m x 25m plots in TBS are shown in Figure 37.

129

Plot 1

Plot 6

Elevational Range:

2.7m

Elevational Range:

11.2m

Mean Elevation:

201.2m

Mean Elevation:

226.0m

Mean Slope:

6.13o

Mean Slope:

22.3o

Slope coefficient of variabiliy: 0.73

Slope coefficient of variabiliy: 0.60

Plot 2

Plot 7

Elevational Range:

11.3m

Elevational Range:

8.2m

Mean Elevation:

227.7m

Mean Elevation:

222.3m

Mean Slope:

21.4o

Mean Slope:

18.0o

Slope coefficient of variabiliy: 0.80

Plot 3

Slope coefficient of variabiliy: 0.72

Plot 8

Elevational Range:

4.8m

Elevational Range:

6.2m

Mean Elevation:

208.7m

Mean Elevation:

262.4m

Mean Slope:

10.0o

Mean Slope:

10.4o

Slope coefficient of variabiliy: 0.65

Slope coefficient of variabiliy: 0.48

Plot 4

Plot 9

Elevational Range:

2.1m

Elevational Range:

6.0m

Mean Elevation:

202.1m

Mean Elevation:

220.2m

Mean Slope:

5.0o

Mean Slope:

10.9o

Slope coefficient of variabiliy: 0.82

Slope coefficient of variabiliy: 0.53

Plot 5

Plot 10

Elevational Range:

3.1m

Elevational Range:

7.9m

Mean Elevation:

202.4m

Mean Elevation:

227.4m

Mean Slope:

5.2o

Mean Slope:

13.5o

Slope coefficient of variabiliy: 0.89

Slope coefficient of variabiliy: 0.57

Figure 37 High resolution (50cm) plot DEMs for the 25m x 25m diversity and structure plots
in TBS, with the green dots representing each tree individuals in the plot (and the original
points used to generate the interpolated surfaces).

3.6

Statistical analysis of plot data

3.6.1 Analysis of compositional and structural similarity

130

The Mantel statistic can be used in its most basic form to examine the degree to which
differences in species composition between sites is controlled by distance between the
sites. Geneticists have used Mantel tests extensively to measure isolation by distance
in populations (Epperson, 2004). In tropical ecology, the Mantel test is extended
beyond comparison of composition with distance, to include environmental
interactions (Phillips et al., 2003; Almeida et al., 2004), and attempt to understand
how environmental factors may shape community composition. The Mantel test itself
is a modified version of the Pearson correlation coefficient, differing in that the
natural dependence in the distance matrix is taken into account (Potts et al., 2002).
The p-value is also calculated differently to determine whether there is significant
correlation, using a one-tailed test of a sample of permutations (10,000 in the case of
this thesis). In addition to the simple Mantel test, the partial Mantel test examines
three-way relationships, normally between geographic distance, environmental
distance and species composition or genetic distance, identifying spurious
relationships that may occur (i.e. distance only affecting species composition through
its co-linearity with environmental distance). Further discussion of the statistical
method is provided by Epperson (2004). Whilst some doubt has been expressed as to
the validity of partial Mantel tests (Raufaste and Rousset, 2001), Castellano and
Balletto (2002) used repeated simulations to show that the partial Mantel test is an
accurate means of ascertaining causal relationships. In this study partial Mantel tests
were used in addition to the standard Mantel test to search for spurious relationships
between significant correlations where variables exhibit some degree of co-linearity.

The Mantel statistic was used to examine the environmental interaction of both
composition and structure in the plots, using the software XLStat to calculate Mantel

131

statistics and the associated p-value. For composition, a Jaccard coefficient was
calculated based on the presence and absence of species for each plot pair. The
Jaccard coefficient is calculated using Equation 3.

Equation 3

Jaccard (J) =

No. Species Common to both plots
(No Species in Plot A + No Species in Plot B)

The Jaccard coefficient is the most widely used compositional similarity measure
(Magurran, 2004), and previous studies have used it alongside the Mantel statistics to
assess environmental controls on species composition (Tuomisto et al., 2003).

Geographic and environmental distance were also calculated for each plot pair using
ten quantitative variables (horizontal distance, elevation, eastness, northness, mean
curvature, slope, slope position, solar radiation, toposcale and topmodel).

The

absolute difference was calculated between the values for each pair of plots, for each
variable.

The Jaccard coefficients and geographic and environmental distances were arranged
in 10 x 10 similarity matrices, and the Mantel statistic applied. Due to the geographic
and environmental distances actually representing dissimilarity, negative Mantel
coefficients are expected in the case of positive correlations (as similarity is being
compared with dissimilarity). The p-value was also calculated in XLStat (using
10,000 random permutations), and presented alongside the Mantel coefficient. In the
case of Tiputini Biodiversity Station, the Mantel statistic was calculated for
characteristics of all both DEMs, hence two separate sets of results are given.

132

In order to apply the same analysis to the structural data, the Kolmogorov-Smirnov
(K-S) test was used to assess the similarity in the cumulative distribuion function for
each of the three major structural variables (DBH, tree height and the DBH:Height
ratio). The K-S test produces a value representing the similarity of the distributions
(based on their mean and standard deviation), that varies between 0 (equal
distribution) and 1 (entirely different distributions). For each of the 45 plot pairs, the
K-S statistic was applied for each of the three structural variables, and the resultant
coefficients arranged in a 10 x 10 dissimilarity matrices. The Mantel statistic was
then applied between these matrices and the respective distance matrix of geographic
and environmental variables. Due to both these matrices representing dissimilarity,
positive Mantel statistics are expected in the case of positive correlation. Likewise for
the composition analysis, partial Mantel test were also used to search for potentially
spurious relationships, and both DEMs were used for TBS, resulting in two sets of
results for each geographic/environmental variable.

3.6.2 Diversity indices

In 1971, Hurlburt contended that “diversity per se does not exist” (quoted in (Peet,
1974), and suggested that ecology abandons the term “diversity”. Over 30 years later,
diversity is very much a central issue in ecology, and if anything less clarity is
available on its meaning.

Magurran (2004, p.101) strongly recommends that a study does not use multiple
diversity indices blindly, but carefully selects the appropriate index based on the
research questions and the aspect of diversity that is being investigated. Whilst this
advice is applicable to many research projects, in the case of the research undertaken

133

in this thesis it would be dangerous to focus on one single aspect of diversity. Firstly,
it is unknown what aspects of diversity topography may influence, and secondly, if
significant relationships are found between topographic factors and diversity (based
on a number of different factors), understanding the type of diversity that correlates
best is a requirement in itself, and may help provide explanation as to the processes
that are involved in generating that relationship. For this reason, this thesis does not
apply Magurran’s recommendation, but bases many of the methods for quantifying
diversity on her book “Measuring Biological Diversity” (2004). Both diversity and
richness measures are used. Richness measures include species richness, genera
richness, family richness, and the richness of “plot endemic species” (number of
species that were found only in that plot).

In addition to the richness measures, the following diversity indices were calculated
for each plot and used in this study:

Shannon

H’ = – S . pi . ln (pi)

Simpson

D = 1 – ∑ (pi)

Margalef

DMg = (S – 1) / ln (N)

Menhinick

DMn = S / N0.5

Berger-Parker BP = Sd / N

Where,

S - species richness
N - total number of individuals
Ni - number of individuals in the i-th species
pi - proportional abundance of the i th species = ni / N

134

Sd – number of individuals of the most dominant species

The Shannon index is an entropy based measure of evenness, Simpson diversity index
(note that 1-D is used here) is also a measure of evenness, whilst Margalef and
Menhinick are species richness related indices adjusted for the total number of
individuals in the sample (N).

Finally, Berger-Parker is a simple and intuitive

measure of dominance.

On top of these indices, Renyi’s scaleable diversity method is used to produce
diversity plots which provide additional information as to the comparability of
diversity between plots. Renyi’s scaleable diversity index is calculated using the
DIVORD 1.5 program (Tothmeresz, 1993) and using the formula:

Equation 4

Where α is a scale parameter, which varies from 0 to 3. This index is displayed in a
plot with different scale parameters expressed along the x axis, and the diversity in the
y axis. This equation is particularly innovative in that the diversity value when the
scale parameter is 0 represents the log of species richness, with a scale parameter of 1
it represents the Shannon Index and with 2 it represents the Simpson Index.

As

α tends to infinity, it represents the inverse Berger-Parker. Typically α varies between
0 and 3. It permits easy interpretation of the diversity between different plots through
examination of the relative shapes of the curves. More importantly, in the event of

135

curves crossing it means that one plot is not more diverse than the other, but that it
depends entirely on the type of diversity being examined.

Following Magurran’s recommendation (2004), special emphasis in this thesis is
given to simple species richness and Simpson’s Index, for their robustness and
simplicity in interpretation. Whilst it would have been advantageous to also include
in the analysis a taxonomic diversity index (Desrochers and Anand, 2004) such as the
Clarke and Warwick taxonomic distinctiveness index (Clarke and Warwick, 1998), it
was deemed impractical due to the level of incompleteness and uncertainty in the
species identification.

3.6.3 Structural Variables

Eight structural parameters are analysed in this thesis. These are based on direct
measurement of some structural parameters, alongside some derivatives of these
variables. Direct measurement was made for stem density (expressed in terms of
stemsHa-1), diameter at breast height (DBH) (expressed in cm), diameter at base
height (DbaseH) (also expressed in cm), tree height (expressed in m), height to first
branch (expressed in m).

Based on these variables, three other variables were calculated; DBH/Height,
DBH/DbaseH, and basal area (m2Ha-1). Basal area is calculated based on both the
DbaseH and the stem density, and reflects the amount of ground area occupied by tree
trunk. The other two variables (DBH/Height, cm m-1, and DBH/DbaseH, cm cm-1)
are calculated in order to grasp the degree of investment of resources in different
aspects of tree structure. Bruenig and Huang (1990) first used the diameter/height

136

variable to examine resistance to wind, and since Webb et al. (1999) have used
DBH/Height to examine structural relationships with topography in American Samoa,
with the variable providing clear patterns. Individuals with relatively wide trunks
compared to their total height (i.e. a high DBH/Height) seemingly have invested
greater resources in augmentation of trunk than in height, possibly indicating amongst
other things a response to conditions for greater trunk strength, or in the case of the
inverse a search for light.

These investments are strategically different: greater

investment in height than girth leads to short term gain (of light) but long term
weakness (a pioneer strategy), greater investment in girth leads to short term loss of
light but greater strength in the long term. The balance between these strategies may
vary with the exposure of the site. The DBH/DbaseH variable is designed to measure
the degree to which the tree base branches out at the base. Trees with buttress trunks
will have significantly lower DBH/DbaseH levels than trees without, and such
individuals are likely to have greater stability against tree fall.

Both these indices are likely to reflect both composition (i.e. some species will tend to
have wider trunks relative to height than others) as well as morphological responses to
site-specific conditions.

3.6.4 Correlation of environmental and topographic variables with diversity

For each site (and each DEM in the case of TBS), the environmental or topographic
conditions are correlated with the diversity and structural measures. All diversity
measures were used in this analysis (family richness, genera richness, species
richness, endemic species richness, Simpson’s, Shannon’s, Menhinick, Margalef, and
Berger-Parker), and in the case of Chapter 5 all structural parameters are also included

137

(Stem density, DBH, DbaseH, Height, Height of first branch, basal area, DBH/Height,
DbaseH/DBH). The environmental and topographic variables are separated into two
types of data; quantitative and categorical.

Quantitative variables (elevation, northness, eastness, curvature, slope, slope position,
solar radiation, topmodel and toposcale) are subjected Pearson product-moment
correlation analyses. In some cases where the data points warrant further study, nonlinear regressions are sought, the details of which are discussed specifically in the
relevant sections. For each correlation performed, a p-value is calculated in XLStat
(using 10,000 random permutations). When significant correlations are found, further
tests are performed, the details of which are discussed in the same section.

The categorical variables (landscape classification - feature, network feature and
topoclass) could not be analysed in the same way. They were compared to diversity
and structural measures using Spearman Rank tests, and the respective p-value
calculated.

In some cases the fuzzy feature membership was also analysed,

comparing the fuzzy membership to a specific class with the diversity or structure
using standard Pearson correlations.

In addition to Pearson correlations and linear regressions, multi-variate analyses are
also performed to search for combinations of variables that may explain variability in
diversity, composition and structure. For these purposes, multiple stepwise linear
regression is applied using XLStat. Multiple stepwise linear regression has been used
in a number of similar studies to good effect (Clinebell et al., 1995; Heikkinen and
Neuvonen, 1997), though caution is taken in interpreting these results as combinations

138

of linear regressions may describe a large degree of variability in data, but reflect little
in terms of the causal relations (Fowler et al., 1998). Ideally split sample validation of
regression models should be performed, but just the ten plots of data prevent this.

3.7

Examination of DEM quality with respect to ground measured slope and
aspect in TBS

Understanding of the quality of the original DEMs and understanding the important
role of scale in capturing the topographic variability found on the ground is crucial to
permit insightful interpretation of the results in subsequent chapters. This section
examines the degree to which the coarse-scale DEMs of the region in TBS (TOPO
and SRTM) represent the topographic conditions found on the ground at the sub-plot
scales at which the trees experience them. Whilst exhaustive GPS surveys would be
one way of gathering this information, it is unfeasible due to the poor GPS satellite
coverage under canopy and the subsequent low accuracy of the GPS derived altitude
values.

Understanding the representation of the terrain at different scales is important, and a
comparison of the slope and aspect at these two scales will indicate the degree to
which scale issues are important in this analysis.

Figure 38 compares values in slope and aspect for the regional DEMs (TOPO and

SRTM) against the plot DEMs for the same areas, firstly for plot mean slope and
secondly for plot mean aspect (measured using northness and eastness, due to the
circular nature of aspect).

139

In all cases there is considerable difference between plot-based slope and aspect and
the DEM based slope and aspect. For slope the SRTM DEM correlates best but not
significantly (Pearson = 0.53, n=10), with the TOPO DEM reflecting little of the
slope measured on the ground (Pearson = 0.11, n=10). In both cases, the DEMs are
grossly underestimating the slope levels measured at the sub-pixel scale. Some
limitations in the validity of this analysis is brought about by the difference in slope
measure that is being used, as the slope in the DEMs is calculated using the elevation
of surrounding cells in a 3 x 3 window, whilst the plot DEMs only take into account
the within plot conditions. Furthermore, geopositional inaccuracy for some plots
means that part of the plot actually falls in a different cell, with a different slope
value.

Aspect derived northness (Figure 40) and eastness (Figure 39) compare even less.
Neither the SRTM DEM (eastness Pearson = 0.28, n=10) nor the TOPO DEM
(eastness Pearson = 0.14, n=10, northness Pearson = 0.30, n=10) show likeness with
ground-measured aspect.

140

Correlation between DEM derived slope and
plot measured slope in TBS

Plot measured slope
(degrees)

25
TOPO slope
20

SRTM slope

15
10
5
0
0

5

10

15

DEM derived slope (degrees)

Figure 38 Comparison of DEM derived slopes and the plot measured slopes for the two
DEMs in TBS. Slope from the plots was derived as a mean for all pixels.

Plot measured eastness

Correlation between DEM derived eastness
and plot measured slope in TBS

-1.5

-1

0.8
0.6
0.4
0.2
0
-0.5 -0.2 0
-0.4
-0.6
-0.8
-1

TOPO Eastness
SRTM Eastness

0.5

1

1.5

DEM Derived eastness

Figure 39 Comparison of DEM derived eastness and the plot measured slopes for the two
DEMs in TBS. Eastness of the plots was derived as a mean for all pixels.

141

Correlation between DEM derived northness
and plot measured slope in TBS

Plot measured northness

0.8
0.6
0.4
0.2
0
-1.5

-1

-0.5

-0.2 0
-0.4

0.5

1

1.5

TOPO Northness
SRTM Northness

-0.6
DEM derived northness

Figure 40 Comparison of DEM derived northness and the plot measured slopes for the two
DEMs in TBS. Northness of the plots was derived as a mean for all pixels.

The results of these two analyses indicate that what is measured at one scale (TOPO
or SRTM DEMs, for example) does not represent the values of the variable at a finer
scale (as measured using the plot DEMs). However, without knowing the relevant
scale for describing the variability in diversity, composition and structure of tropical
tree species it is not necessarily true that the more information rich scale (plot DEMs)
is necessarily better. This is an important point, and one which is considered later in
the thesis in the results sections.

Given these disparities between ground-measured topography and its representation in
the DEMs, some further analysis is warranted. It is reasonable to assume that each
DEM performs better in specific landscapes. Erosion and drainage are likely the most
important geomorphic processes in shaping the landscape, and so it is expected that
greater small-scale topographic complexity would occur in regions with concave

142

slopes in the lower reaches of the hillside where erosion and streams are more
prevalent compared with planar slope segments and convex slope crests. The strength
of these processes in lower reaches of the hillside may create complexity which is not
captured in DEMs of this resolution, and thus cause the disparities found above. Both
DEMs used in this study are derived from tree-top topography, and it is likely that the
canopy will smoothen complex topography found on the ground, especially small
scale complexity derived from erosion and the generation of gullies.

In order to test this hypothesis, mean profile curvature and slope position are
correlated with the coefficient of variability in slope within the plots (a surrogate for
topographic complexity). No significant correlation is found with mean curvature nor
topmodel, but slope position does show some very significant correlations (Figure
41). The TOPO DEM correlates closest CoeffVarSlopePLOT = -0.003 SlopePosTOPO +
0.86, R2 = 0.84, R2 = 0.79, n=10), followed by the SRTM DEM (CoeffVarSlopePLOT =
-0.004 SlopePosSRTM + 0.85, R2 = 0.47, n=10). In all cases the trend is for greater
ground measured complexity in topography in areas close to the valley bottom. It
would be logical therefore to conclude that the DEMs we are using are likely to
provide better representation of the ground surface in areas higher in the hillside
where erosion and water flow concentration is lower. This is, however, not supported
by the data (Table 11), with in most cases higher percentage errors in higher slopes.

143

Plot measured coefficient of
variation in slope

Correlation between DEM derived slope position
and plot derived variation in slope in TBS
1
TOPO Slope Position
0.9

SRTM Slope Position

0.8
0.7

TOPO

0.6

SRTM

0.5
0.4
0

20

40

60

80

100

120

DEM Derived slope position

Figure 41 Scatterplot of ground-based measurement of topographic complexity (using the
coefficient of variability in slope as a surrogate) against DEM derived slope position.

These findings also indicate that broader scale features such as those used to define
slope position (calculated through flow accumulation routines) are sound in the DEMs
(best represented in TOPO), but more local topographic features such as curvature,
slope and aspect are less representative (worst representation in TOPO, where the
interpolation methods produce the greatest amount of smoothing).

Table 11 Error in DEM representation of plot topography for all slopes, as well as for upper
and lower slopes defined using the slope position variable.

144

Without further field-based research it is impossible to further quantify the errors and
scale issues in the DEMs, nor understand under what conditions they are more
accurate. The conclusion of this analysis is that the DEMs are not representing very
well local conditions (sub-25m pixel) like slope and aspect (though obvious
methodological aspects in the comparison lower the statistical validity of this
observation), with different DEMs performing better for different variables. This may
be important in later in-depth analyses with respect to the diversity and structure in
the forests. It is also important to note the relationship between slope position and
topographic complexity.

3.8

Heterogeneity Modelling

There are few examples of models that quantify spatial heterogeneity of
environmental variables that are explicitly designed for application to biological
systems. Spatial environmental heterogeneity within the context of this study refers
to the diversity of environments within a certain local spatial extent. Translating this
concept into a raster grid like all the terrain variables is best done by considering it the
variability of pixel values in cells surrounding a single central cell. This type of
analysis is often called kernel estimation (Wand and Jones, 1995) moving-window
analysis (Fotheringham et al., 1996), convolution filtering (Lillesand and Kiefer,
1999), or focal calculations

(Tomlin, 1993).

In order to quantify spatial

heterogeneity, three main factors must be considered:

1. The spatial function for weighting (geographic weighting) with respect to
distance from the central cell (the shape of the kernel)

145

2. The size of the search window surrounding each cell (the radius is often called
the bandwidth)
3. The equation for actually quantifying the heterogeneity given the array of
values for surrounding cells

Given the lack of previous studies that adopt this kind of spatial analysis for tropical
ecology, there was no basis on which to select an equation to accurately represent the
average spatial function of neighbourhood interactions (such as seed dispersal,
competition etc.). Literature in the realm of genetics does provide some indication of
a likely equation based on analyses of the probability of pollen transfer. The spatial
function of pollen-transfer is likely to be similar to that of seed dispersal due to their
similarity in physical processes, both having wind and animal (and insect to a lesser
extent for seeds) dispersal mechanisms. Shaw (1995), Xu and Ridout (2001), and
Paradis et al. (2002) used the radial half-Cauchy distribution for modeling pollen
transfer, preferring this distribution to others for its characteristic long tail. The halfCauchy distribution also provides a very flexible equation that permits simple
adjustment of its shape through manipulation of the median distance variable.
Provided the lack of data on actual probabilities and distances of seed dispersal for the
average tropical tree, and a lack of data on the spatial nature of competition or other
relevant biotic processes, the median distance variable provides an easy way of
incorporating multi-scale analyses to cover all potential scales at which these
processes may occur.

The half-Cauchy curve is produced using the following

formula:

146

Equation 5

W (r) = (2 / π b) 1 / [1 + (r / b)2]

Where W(r) is the probability of interaction with a neighbouring cell, b is the median
distance of interaction and r is the distance from the central cell. When the median
distance is set to 0 the weighting for surrounding cells is also 0, increasing the median
distance variable produces higher weighting to surrounding cells. Figure 42 shows
the shape of the distance-weighting curve for different median distances, ranging from
0.01 to 9 cells. For the analysis, a total of 16 kernels were used with median distances
of 0.01, 0.05, 0.1, 0.2, 0.3, 0.5, 0.8, 1, 2, 3, 4, 5, 6, 7, 8, and 9. This multi-scale
approach is advantageous due to the lack of knowledge on the spatial function of
neighbourhood interactions.

147

Weighting against distance using Half-Cauchy
distribution - Median Distance 0.1 cells

70.0

7.0

60.0

6.0

50.0

5.0

40.0

4.0

Weight

Weight

Weighting against distance using Half-Cauchy
distribution - Median Distance 0.01 cells

30.0

3.0

20.0

2.0

10.0

1.0

0.0

0.0
0

5

10

15

20

25

30

35

40

0

5

10

15

Distance

25

30

35

40

Weighting against distance using Half-Cauchy
distribution - Median Distance 1 cell

1.4

0.7

1.2

0.6

1.0

0.5

0.8

0.4

Weight

Weight

Weighting against distance using Half-Cauchy
distribution - Median Distance 0.5 cells

0.6

0.3

0.4

0.2

0.2

0.1

0.0

0.0
0

5

10

15

20

25

30

35

40

0

5

10

15

Distance

20

25

30

35

40

Distance

Weighting against distance using Half-Cauchy
distribution - Median Distance 5 cells

Weighting against distance using Half-Cauchy
distribution - Median Distance 9 cells

0.1

0.1

0.1

0.1
0.1

0.1

0.1

0.1

Weight

Weight

20
Distance

0.1
0.0

0.0
0.0
0.0

0.0

0.0

0.0

0.0
0

5

10

15

20

25

30

35

40

0

5

10

15

20

25

30

35

Distance

Distance

Figure 42 Distance weighting from the Half-Cauchy curve for different median distances
from 0.01 cells (top-left) to 9 cells (bottom-right).

In reality the curves shown in Figure 42 are 3-dimensional, and so for some clarity
Figure 43 shows an example half-Cauchy kernel with median distance set to 1 cell.

148

40

Figure 43 3-dimensional representation of the half-Cauchy spatial function, with the height of
the surface representing the weighting.

The spatial function used here is uniform in all directions, though this could have
been adjusted to provide higher weighting in specific directions, such as in the
dominant wind direction with respect to wind-based seed dispersal. In Tambito this
would not be valid, as topographic funneling likely create different wind directions
around the catchment (and there is insufficient data to fully analyse and understand
wind patterns in the twin-catchments). In TBS a more constant wind direction is
likely, and indeed has been shown to be south-easterly. The kernel could be adjusted
to provide higher weighting in cells upwind, though this is not performed here as it
would provide an extra variable and an extra level of complexity on results
presentation and interpretation (for TBS the analysis is already complicated by two
DEMs, 12 variables and 16 different scales). This is clearly a limitation, and further
studies should incorporate wind direction into this model, though other spatial
neighbourhood processes such as competition and biotic interactions are not
necessarily affected by wind direction.

149

The size of the search window should be selected based on the distance function used,
so that the window covers all cells with a significant weighting that may affect the
final result. In reality, the size of the search window must also depend on processing
power and the size of the grid that is being analysed, due to the intensive processing
required to make a spatial heterogeneity calculation.

For this analysis a search

window of 50 x 50 cells was used. Even with the highest median distances the
potential effect of cells outside this 50 x 50 cell kernel was calculated to be
insignificant (<5% change in result when search window increased to 200 cells for
test runs in Tambito and TBS). The selection of the search window was therefore set
at of 50 x 50 cells. This avoided computational limitations, and also ensures that edge
effects do not encroach on any of the plots (in the case of Tambito, Plot 7 is just 70
cells from the border of the DEM).

Definition of heterogeneity is somewhat similar to that of diversity, with no single
number capable of describing all aspects of the phenomenon. So as to compare like
with like (i.e. diversity vs. heterogeneity) it was important to measure each variable in
a similar way. For this reason, the heterogeneity was calculated using the Simpson
diversity index. This is similar to the technique that was used effectively by Burnett
et al. (1998) to compare geomorphological heterogeneity (aspect, slope, topmodel
amongst other variables) with plant species diversity, though in this study a grid based
system was used which lacks the fine-scale capacity of the method presented here.

Simpson’s index is calculated based on the frequencies of categorical data. Of the 12
terrain characteristics studied here, nine are continuous and just three are categorical.

150

The continuous variables were therefore re-classified into equal interval classes, each
class representing environment type ‘i’, thus making it possible to calculate
Simpson’s index, where the heterogeneity for the centre cell was calculated using:

Equation 6

Heterogeneity = 1 – Σ (pi2)

The frequencies (pi) are calculated from the environmental values of all cells in the
surrounding 50 x 50 cell region. Individual contributions of each cell to the
calculation of these frequencies were weighted by the spatial function. Given this
approach, frequency pi represents the total relative probability of receiving inflowing
seeds from an environment of type ‘i’. In practical terms, the frequency (p) is the sum
of all distance-weighted cells of environment type ‘i’.

The creation of arbitrary classes creates the problem of threshold effects, where
inaccurately higher heterogeneity is found on the interface between two arbitrarily
defined classes. To eliminate this threshold effect, the re-classification was repeated
five times for each calculation of heterogeneity, splitting the topographic
characteristics into 25, 26, 27, 28, and 29 different classes of equal intervals. The final
heterogeneity was calculated from the mean of the results of these five iterations.
Changing the number of classes changes the geographic location of splits between the
classes. These classes were selected based on visual experimentation, and
computational limitations. For lower numbers of classes (i.e. 20 – 24), arbitrary
boundaries were still observed, whilst greater number of classes vastly increased the
time required to run the model.

151

In reality the calculation described above is fairly simple, and best understood by
means of an example analysis shown in Figure 44. The final result is a grid of spatial
heterogeneity of each terrain derivative for each of the 16 scales. In general terms,
the analyses with low median distances produce complex small-scale distributions of
environmental heterogeneity, while higher median distances produce smoother and
broader scale measures of heterogeneity.

152

Figure 44 Example analysis of spatial heterogeneity for a model environment.

153

Many other methods could have been used to quantify heterogeneity, including a
simple calculation of the surrounding coefficient of variation, absolute surrounding
range, non distance-weighted approaches, geostatistical methods (variograms, for
example), as well as a similar approach using other diversity indices or richness
measures (Burnett et al., 1998; Nelson, 2004). For this study a method was selected
which most accurately reflects environmental heterogeneity in a species diversity
relevant context, specifically through measuring heterogeneity with an evenness
diversity index (Simpson’s) for direct comparison with species diversity. Given the
lack of previous research determining spatial heterogeneity in terrain characteristics
for biological diversity studies, it is difficult to judge if the method adopted here is
indeed the best method. It is favourable in terms of measuring heterogeneity in terms
of evenness, and permitting the comparison of like with like, though there is also
some degree of information loss through the creation of categorical data from
continuous surfaces. For the purposes of simplicity in analysis and interpretation, this
study only uses the Simpson’s index to quantify heterogeneity. Further studies should
develop different models and compare results.

These methods for calculating heterogeneity were incorporated into an Arc Macro
Language (AML) for use in Arc/Info. Given the high level of processing that these
types of spatial analyses require, it was important to automate this analysis. Appendix
2 shows the model used for quantitative variables, and Appendix 3 shows a slightly
modified version used for the categorical variables (feature, network feature and
topoclass). The method itself is identical, with the only difference of not needing the
multiple re-classification into “i” environments.

154

The heterogeneity at each plot was then extracted from the final grids for each of the
16 scales, and for each topographic variable. This was then correlated with the
Simpson’s index based on the tree diversity measured within the plot, calculating both
the Pearson correlation coefficient and the p-value.

155

Chapter 4

4

4.1

Spatial variation in diversity and composition : Environmental interactions

Introduction

This chapter makes a basic analysis of spatial variations in diversity within the two
study sites, using the plot data along with the basic terrain derivatives described in
Chapter 3. The aim of this chapter is to examine the patterns of spatial variation in
composition and diversity, and begin to explore these patterns in the context of
environmental variation.

The analysis is split into two separate sections, firstly

examining compositional variation, and secondly examining the variation in richness
and diversity.

The central issue being tested in the compositional section is that of habitat
association for tropical trees, though attention is brought to the distinction between
species level- habitat associations and the compositional level (non-species specific)
habitat association being studied here. Numerous studies have recently searched for
habitat associations in tropical tree species in a number of different environments,
principally in tropical lowland forests (Vormisto et al., 2000; Webb and Peart, 2000;
Harms et al., 2001; Vormisto, 2002; Phillips et al., 2003; Tuomisto et al., 2003;
Valencia et al., 2004) but also in the elevational gradient of tropical montane forests
(Sugden, 1982; Vasquez and Givnish, 1998). In lowlands there is evidence that some
association occurs (approximately 20 – 70% depending on the study), but it is also
acknowledged that many species are almost randomly distributed across a landscape,
with non-equilibrium processes (such as tree fall) controlling a large proportion of the

156

spatial distribution of composition (Harms et al., 2001). In montane regions, studies
have found elevation as an important variable associated with composition (Sugden,
1982; Vasquez and Givnish, 1998), but these have generally been broad-scale studies,
which have not used micro- and meso- scale terrain variation to search for habitat
association.

This sections endeavours to statistically test if there is any habitat

association in composition present in the plot data based on habitat as measured by
the terrain derivatives.

The diversity section moves up one level (from information at the species level to
non-species specific diversity), endeavoring to look for spatial patterns in diversity,
and test the data to discover whether or not specific topographic conditions favour the
generation of high diversity. As discussed in Chapter 2, some of the key drivers of
diversity include productivity (often measured using energy or water as a surrogate),
soil nutrients (however the relationship may be), and gap dynamics.

Different

measures of diversity are examined (evenness/dominance and richness) for each plot
and compared with the range of terrain derivatives being used in this study to test for
association. Where significant relationships are found, further tests are made to
understand the physical processes that may be generating the relationship.

Each study site is treated separately at first, but the chapter concludes by integrating
the results from each site and discussing the over-arching ecological significance of
the findings at both sites. For TBS, both DEMs are used in the study, and for ease of
interpretation the results of each analysis for each DEM are presented together.

4.2

Reserva Tambito

157

4.2.1 Composition

4.2.1.1 General Site Species Composition

The ten 25m x 25m plots established in Tambito contained 1108 individuals, of which
271 different species were identified, distributed in 52 families and 111 genera.
Appendix 4 contains a full list of species found in Tambito. Taxonomic identification
was made to family level for 99.7% of species, to genus level for 87.6% of species,
and to species level for 52.0% of species. It is likely that some species as yet
unidentified to genus level will be new genera to the area, and thus the number of
different genera is likely an underestimate. Table 12 provides a summary of family
composition in Tambito, with Figure 45 and Figure 46 showing the distribution of
species and individuals within families.

158

Family
Actinidaceae
Anarcadiaceae
Annonaceae
Aquifoliaceae
Araliaceae
Arecaceae
Asteraceae
Bignoniaceae
Bombacaceae
Burseraceae
Caesalpiniaceae
Caprifoliaceae
Caricaceae
Celastraceae
Chletraceae
Chloranthaceae
Chrysobalanaceae
Clusiaceae
Cyatheaceae
Elaeocarpaceae
Euphorbiaceae
Fabaceae
Flacourtiaceae
Hippocrateaceae
Hypocastanaceae
Icacinaceae
Lauraceae
Lecythidaceae
Marcgraviaceae
Melastomataceae
Meliaceae
Mimosaceae
Monimiaceae
Moraceae
Myristicaceae
Myrsinaceae
Myrtaceae
Ochnaceae
Piperaceae
Proteaceae
Rosaceae
Rubiaceae
Sabiaceae
Sapindaceae
Sapotaceae
Simaroubaceae
Solanaceae
Staphyleaceae
Styracaceae
Theaceae
Theophrastaceae
Verbenaceae
Other
TOTAL

No. Genera
2
1
2
1
2
8
1
1
2
1
2
1
1
2
1
2
1
3
2
1
7
1
2
1
1
2
5
2
1
10
3
1
2
6
2
4
3
1
1
1
1
8
1
1
1
1
2
1
1
1
1
1
5
118

No. Species
2
1
3
4
3
15
1
1
3
6
4
1
1
2
1
2
2
16
3
2
10
2
3
1
1
2
28
5
1
33
8
4
5
15
3
7
11
1
2
1
3
31
1
2
4
1
3
1
1
1
1
1
5
271

No. Individuals No. Plots
13
7
1
1
18
5
18
6
13
3
144
9
2
1
1
1
5
3
21
5
4
4
4
2
1
1
8
4
1
1
16
4
2
2
47
10
38
8
6
4
63
9
4
3
32
5
22
5
16
4
6
3
65
9
24
6
2
1
118
9
28
9
9
4
6
5
34
7
22
6
21
9
22
9
1
1
3
3
1
1
12
4
201
10
1
1
3
3
12
4
1
1
3
3
1
1
2
1
1
1
2
1
2
2
5
5
1108
4.3 (AVG)

Table 12 Summary of species composition per family in Tambito, including the number of
plots where the family was present.

159

Figure 45 Distribution of species in families in Tambito

Figure 46 Distribution of individuals in families in Tambito

The most abundant families in terms of individuals are Rubiaceae (201 individuals),
Arecaceae (144) and Melastomataceae (118), whilst Melastomataceae is the most
species rich family (33), followed closely by Rubiaceae (31) and Lauraceae (28).
Over 50% of all species are in just 6 families, and there is only one species present for

160

18 of the 52 families. The most common species were Wettinia sp. (Arecaceae) with
31 individuals in 6 different plots, Casearia arborea (L.C. Rich.) Urban
(Flacourtiaceae) with 29 individuals in 4 plots, Cyathea sp1 (Cyatheaceae), a species
of tree fern, with 29 individuals in 5 different plots, and an unidentified palm “oja
pescado” (Arecaceae) also with 29 individuals and found in only one plot. The
average abundance per species was 4.1, and 40.6% of all species having just one
individual in the ten plots.

4.2.1.2 Between-Plot Compositional Variability

4.2.1.2.1 General discussion

There was significant variation in composition between-plots in Tambito. Just two
families were found in all ten plots (Clusiaceae and Rubiaceae) with 14 of the 52
families (27.0%) occurring in only a single plot. On average each family was present
in 4.3 plots.

At the species level, the most broadly distributed species were

Hyeronima oblonga (Tul.) Müll. Arg. (Euphorbiaceae) and Irarthea sp1 (Arecaceae),
which were found in eight plots, with another palm species Chameadora sp1
(Arecaceae) being found in seven different plots. At the other end of the spectrum,
some 188 of the 271 species (69.4%) were found in only one plot.

This does not necessarily signify that a large proportion of species are highly
restricted in their distribution, but may simply be a product of the small plot size used
here. Given the exceptionally high diversity of species in tropical forests, and the low
densities that species tend to have (for example Pitman, 2000) it is unlikely that a 25m

161

x 25m plot captures the full alpha diversity of that particular habitat. For this reason
many species “absences” in the plots may indeed be false negatives brought about by
the small plot size.

The objective made here with the plots used in this study is that the within-plot habitat
is homogenous, whilst the surrounding habitats are heterogeneous. In other words,
each plot captures the alpha diversity, whilst comparison between plots examines the
beta diversity. In reality it is near impossible to achieve this objective, because
habitats vary spatially at very different scales, and that variation is gradual so no clear
cut-off is identifiable. Perhaps in highly heterogeneous sites (like Tambito) habitats
significantly change over short distances such as 25m (making 25m x 25m plots
comply with the objective of the sampling strategy), but in more homogenous sites
(like TBS) 25m may only represent a small portion of the actual habitat. Under the
latter circumstance, the plot is not representative of the alpha diversity.

Following this logic, for the objective to be achieved it is expected that the number of
species within the plot would therefore smoothly increase as greater within-plot space
is covered, but that beyond the plot new environments are encountered, thus
increasing the accumulation of species.

Species-area curves can be used to examine whether the species richness is saturated
or still accumulating as the area inventoried increases. The shape of species-area
curves result from a number of factors, making them fairly subjective in terms of
interpretation. As greater area is included, it is expected that the number of species
will also rise because new environments (ergo niches) are being included in the

162

sample. In a heterogeneous environment like Tambito this is particularly the case.
Regardless of the inclusion of new niches, space itself is a resource which species
compete for, and in particularly diverse ecosystems like tropical forests the number of
species is likely to saturate only over very large areas because there are many species
and they can be widely distributed. Thus very large plots are usually required, though
such plots are very expensive, time-consuming and difficult to inventory in these
environments.

Despite these problems associated with the interpretation of species-area curves, they
can provide some information, which coupled with other data and analyses provide a
clearer picture. Figure 47 shows the species-area curves for Tambito, using the plot
quadrants to separate out areas of 5 x 5m, 10 x 10m, 15m x 15m, 20 x 20m and 25m x
25m.

50
Plot 1

Number of Species

45

Plot 2

40

Plot 3

35

Plot 4

30

Plot 5

25

Plot 6
Plot 7

20

Plot 8

15

Plot 9

10

Plot 10

5
0
0

200

400
Area (m2)

600

Figure 47 Species-area curves for Tambito, calculated by separating out the quadrants into
increasingly sized plots.

163

As can be seen in Figure 47, the trend of species accumulation appears to continue
beyond the 25m x 25m mark, with little evidence of saturation for any plot. Only plot
9 shows some level of saturation, with just 3 new species being found in the final
225m2 of the plot. Given the extremely high levels of diversity present in Tambito,
and the reported low densities in tropical tree species (Pitman, 2000), it is expected
that 625m2 of forest (one plot) will not be particularly representative of the total alpha
diversity. Therefore it is impossible to conclude that species are restricted in their
distributions to the extent found in the plot data (69.4% of species restricted to one
single plot), and this is in part attributable to the small plot-sizes used in the study.

4.2.1.2.2 Single-variate analysis of quantitative environmental variables

In order to understand the interaction of environment on composition, similarity
matrices and statistical analyses using the Mantel test are used. These tests compare
between-plot compositional similarity, in this case measured with the Jaccard
coefficient and presented in a matrix, with environmental “distance” between the
plots, also represented as a similarity (or in this case the inverse – a dissimilarity)
matrix. Figure 48 shows scattergrams of compositional similarity against geographic
or environment distance for each plot pair (thus 45 points in total). Trendlines are
displayed, including the coefficient of correlation. Table 13 then displays the results
of the Mantel tests, including the Mantel coefficient, and the one-tailed p-value for
10,000 random permutations.

164

Compositional Similarity and Altitude

Compositional Similarity and Distance

2.50
y = -5.0597x + 1.3479
R2 = 0.151

2.00
1.50
1.00
0.50
0.00
0.00

0.05

0.10

0.15

1200

Altitudinal Difference Between
Plots

Horizontal Distance (km)

3.00

1000

y = -3159.7x + 577.16
R2 = 0.3584

800
600
400
200
0
0.00

0.20

0.05

Compositional Similarity and Eastness
2.5

Eastness Difference

Northness Difference

y = -1.6296x + 0.6339
R2 = 0.0257

1.4
1.2
1
0.8
0.6
0.4
0.2
0
0.00

0.05

0.10

0.20

0.15

y = 5.5905x + 0.4392
R2 = 0.0762

2
1.5
1
0.5
0
0.00

0.20

0.05

Jaccard Sim ilarity

0.10

0.15

0.20

Jaccard Sim ilarity

Compositional Similarity and Slope

Compositional Similarity and Curvature
40

3
y = -1.4776x + 1.0527
R2 = 0.0086

2.5
2
1.5
1

35
Slope Difference

Curvature Difference

0.15

Compositional Similarity and Northness

1.6

0.5

y = -104.2x + 25.022
R2 = 0.1545

30
25
20
15
10
5

0
0.00

0.05

0.10

0.15

0
0.00

0.20

0.05

Compositional Similarity and Slope Position
90

Solar Radiation Difference

60
50
40
30
20
10
0.05

0.10

0.20

2500

70

0
0.00

0.15

Compositional Similarity and Solar Radiation

y = -230.15x + 55.729
R2 = 0.1478

80

0.10
Jaccard Sim ilarity

Jaccard Sim ilarity

Slope Position Difference

0.10
Jaccard Sim ilarity

Jaccard Sim ilarity

0.15

0.20

Jaccard Similarity

y = -3378.1x + 1114.7
R2 = 0.0658

2000
1500
1000
500
0
0.00

0.05

0.10
Jaccard Sim ilarity

165

0.15

0.20

Compositional Similarity and TopModel

Compositional Similarity and Toposcale
2.5

y = -596.71x + 160.25
R2 = 0.1

300

TopModel Difference

Toposcale Difference

350

250
200
150
100
50
0
0.00

0.05

0.10

0.15

0.20

y = 4.4365x + 0.3954
R2 = 0.1374

2
1.5
1
0.5
0
0.00

0.05

0.10

0.15

0.20

Jaccard Sim ilarity

Jaccard Sim ilarity

Figure 48 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity) for
Tambito.

Variable
Distance
Altitude
Eastness
Nothness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test
-0.39
-0.60
-0.16
0.28
-0.09
-0.39
-0.38
-0.26
-0.32
0.37

One-tailed p-value
0.038
0.000
0.220
0.088
0.341
0.003
0.004
0.063
0.046
0.032

Correlation
Significant
Significant
Not-significant
Not-significant
Not-significant
Significant
Significant
Not-significant
Significant
Significant

Table 13 Summary results of compositional similarity analysis with distance and
environmental variables, using Mantel tests and a one-tailed Pearson for Tambito. Note that
compositional similarity was compared with environmental dissimilarity (difference), hence
negative values in most cases. The northness variable produces very different results due to
the heavily skewed distribution of plots in strongly northward and strongly southward facing
slopes.

The Mantel tests indicate significant correlation between species composition and
topography for six of the ten variables analysed. The highest correlation is for
elevation, with slope, distance, slope position and toposcale also having significant

166

correlations to the 95% significance level.

The topmodel variable also shows

significant correlation, but in the inverse direction (i.e. the greater the compositional
difference the more similar the Topmodel values). Least significant variables were
aspect related derivatives (northness and eastness) and slope profile curvature.
There are many data points with low degrees of compositional similarity, but also low
differences in their topographic conditions, indicating that the variable is not entirely
responsible for compositional similarity, but that other factors are at play. However,
there are no data points compositionally very similar and environmentally highly
different.

Interpreting the results in more depth, it is difficult to ascertain the degree to which
distance is important over the other factors, as distance itself correlates relatively
closely with elevational difference (R2 = 0.31, p < 0.0001, n = 45). For this reason a
partial Mantel test is performed, using species composition, elevational difference and
geographic distance as the third variable.

This test concludes that the direct

relationship between distance and species composition is in fact spurious (Pearson = 0.1, p = 0.345, n = 45), and only significant due to the co-linearity with elevational
difference.

Similarly, partial Mantel tests performed on each of the correlating variables (slope,
slope position, toposcale and topmodel) using elevation as the third distance variable
show that none correlate significantly independent of the co-linearity with elevation
(Table 14). TopModel is the closest to correlating, with a Mantel statistic of 0.32
(narrowly outside the 95% confidence limit, p = 0.052, n = 45), but still negatively
(i.e. greater compositional similarity between plots with greater Topmodel difference).

167

No plausible explanation can be provided as to the processes behind such a
relationship.

Partial Correlation Correlation Correlation Correlation
Variable C
(A,B)
(B,C)
(A,C)
(A,B) | C

Variable A

Variable B

Compositional similarity
Compositional similarity
Compositional similarity
Compositional similarity
Compositional similarity

Elevation
Slope
Slope Position
Toposcale
TopModel

Distance
Elevation
Elevation
Elevation
Elevation

-0.60
-0.32
-0.39
-0.38
0.37

0.56
0.56
0.63
0.65
-0.21

-0.39
-0.60
-0.60
-0.60
-0.60

p-value

-0.60
0.03
-0.04
0.01
0.32

0.00
0.47
0.45
0.47
0.05

Table 14 Revised Mantel statistics for correlating variables with compositional similarity in
Tambito, applying partial Mantel tests to detect spurious correlations derived from colinearity between variables.

Evidence of habitat associations with elevation is not a new result for tropical
ecology, with numerous examples of species distribution restrictions to elevational
zones (Vasquez and Givnish, 1998). However, this analysis has shown a lack of
habitat association in composition in many of the other variables. The analysis
presented here indicates that habitat association in tropical tree species composition in
Tambito is restricted to broad scale factors (temperature gradients for example), rather
than local conditions (local slope, curvature, solar radiation receipt etc.), which may
still exist but not in a way that is evidenced in our data.

4.2.1.2.3 Multi-variate analysis of quantitative environmental variables

Having identified that elevation alone accounts for a large amount of compositional
difference, but that other variables present at least some influence on composition, a
multi-variate analysis is performed as a final step in this analysis. Agglomerative
Hierarchical Clustering (AHC) is performed using Ward’s Method (using XLStat Pro)
168

to produce a dendrogram of plot environmental conditions dissimilarity (Figure 49)
and an associated dissimilarity matrix. The dissimilarity matrix is then compared
with the compositional similarity matrix (Jaccard Index) using the Mantel statistic.
When all 9 environmental variables are used (geographic distance was excluded), the
correlation is insignificant at the 95% significance level, though a weak correlation is
evident (Mantel R = -0.257, p = 0.07). When the multi-variate analysis is restricted to
only those factors already found to be significant (elevation, slope, slope position,
toposcale and topmodel), the correlation improves and becomes highly significant to
the 99% confidence limit (Mantel R = -0.393, p = 0.006). This is still not as high as
the correlation found in the partial Mantel test of elevation (Mantel R = -0.60, p =
<0.000), indicating that in reality elevation (and/or distance) are the best predictors of
composition (in this case explaining some 36% of compositional variability).
Likewise, when multiple stepwise linear regression is applied to the same data, only
elevational difference is selected as an explanatory factor, with no other variable
explaining more variability in composition with the p-value limit set to 0.10. If
elevation alone explains 36% of the variability, clearly other factors are important
which are not measured in this study.

169

Tree Diagram of Multi-variate
Environmental Clustering for plots in
Tambito

54

Dissimilarity

45
36
27
18

10

6

5

3

2

1

9

8

7

0

4

9

Figure 49 Multi-variate clustering of plots based on all 9 environmental and topographic
variables

4.2.1.2.4 Analysis of categorical environmental variables

Mantel tests were only possible for continuous variables, hence the landscape
classifications (feature, network feature and topoclass) were not used in these
analyses.

The degree to which composition is controlled by these landscape

classifications is explored using simple tables showing the number of species
restricted to specific classes for each of the three variables (Table 15, Table 16, and
Table 17). The number of species found to be restricted to a single class is shown.
However, many of these species may have been found only in one plot, making it
inconclusive whether they are restricted by landscape classification, or are simply rare
and have only been captured by one single plot in this study. For this reason, the
analysis is also performed only for the 83 species found in more than a single plot.

170

Feature
Planar
Ridge
Channel
Total

No. Plots

No. Species
Restricted to
feature

2
7
1

163
26
24
213

No. species restricted to
feature only for species
distributed in more than
1 plot
24
1
0
25

Table 15 Number of species associated to specific classes in the “Feature” classification for
Tambito.

Network Feature
Planar
Ridge
Channel
Total

No. Plots

No. Species
Restricted to
feature

6
4
0

141
72
0
213

No. Species Restricted
to feature for species
distributed in more than
1 plot
21
4
0
25

Table 16 Number of species associated to specific classes in the “Network Feature”
classification for Tambito.

Topoclass
Slope
Toe Slope
Ridge
Total

No. Plots

No. Species
Restricted to
feature

5
2
3

97
35
74
206

No. Species Restricted
to feature for species
distributed in more than
1 plot
11
1
6
18

Table 17 Number of species associated to specific classes in the “topoclass” classification for
Tambito.

For all three variables a large proportion of species were restricted to one single
landscape class (>76%), but this is more an indication of the fact that 69% of species
were only encountered in a single plot. It is impossible to separate out landscape class

171

restriction from low density in these cases. More indicative of habitat association is
the proportion of the 83 species found in more than one plot that are restricted to a
single landscape class. For this the “feature” classification and the “network feature”
classification show the highest levels of habitat association (30% of the 83 species),
with the “topoclass” classification showing 22% habitat association. In none of the
landscape classes is there a family, which exhibits abnormally higher levels of habitat
association.

However these results must be taken with caution, as the few number of data points
poorly cover all landscape classes (for example only 1 plot in a channel in the feature
classification, no channels in the network feature classification, and no peaks at all in
any classification system). Perhaps a clearer way of looking at these results is to note
that 70% of the broadly distributed species (i.e. those found in more than 1 plot) were
found in different landscape classes, indicating that these terrain derivatives have little
explanatory power in terms of composition in Tambito. In any case, this analysis is
significantly weakened by the lack of data points, and more plots would greatly
enhance the validity of this conclusion.

4.2.2 Diversity

4.2.2.1.1 General discussion

The plot diversity data shows significant between-site variation in diversity in
Tambito (Table 18) with species richness varying between 30 and 52 species. On
average there were 22 families per plot (ranging from 18 to 27), and 35 genera
(ranging from 25 to 46). Prior to discussing in detail the diversity between plots using

172

different indices, the Renyi diversity plots (Figure 72) help in the interpretation of
diversity variation between plots.

No.
No. Families No. Genera No. Species
Individuals

Plot

No.
Endemic
Species

Simpson's

Shannon's

Margalef

Menhinick

BergerParker

1

97

21

34

44

17

0.96

3.45

9.40

4.47

0.12

2

107

24

33

42

16

0.97

3.44

8.77

4.06

0.11

3

118

23

31

35

11

0.93

3.05

7.13

3.22

0.20

4

137

22

34

43

21

0.95

3.31

8.54

3.67

0.16

5

112

19

35

37

23

0.93

3.04

7.63

3.50

0.19

6

87

18

37

40

8

0.97

3.44

8.73

4.29

0.09

7

176

20

40

45

29

0.94

3.24

8.51

3.39

0.16

8

106

28

46

52

26

0.98

3.68

10.94

5.05

0.10

9

75

19

25

30

13

0.96

3.16

6.72

3.46

0.12

10

93

27

34

49

24

0.98

3.64

10.59

5.08

0.08

Table 18 Richness and diversity of plots in Tambito

60

Plot 1

50

Plot 2

40

Plot 3

30

Plot 4
Plot 5

20

Plot 6
10

3

5

5
2.
7

5

2.

2

Scale Parameter

2.
2

5

5
1.
7

5

1

1.

1.
2

5
0.
7

0.

5

Plot 9

5

Plot 8

0

Plot 7

0
0.
2

Exp (Renyi) Diversity

Renyi Scaleable Diversity (Exponential) - Tambito

Plot 10

Figure 50 The Renyi scaleable diversity index for Tambito, indicating to what extent plot
diversities can be universally compared.

Plots 8 and 10 are the most diverse, with greater richness and evenness in plot 8, but
also a higher level of dominance of a single species (Berger-Parker). Although plot 9
has the lowest species richness, plot 5 has the lower “diversity” (Shannon’s and
Simpson’s), and also high levels of dominance of a single species (along with plot 3).
This simple comparison merely shows that diversity is not homogenously distributed

173

around the catchment, and that a great deal of variation occurs in species richness,
diversity and dominance. The many shapes of the Renyi plots in Figure 72 also show
that the structure of diversity is also very different between plots (i.e. non-comparable
levels of evenness vs richness). These differences are later found to be important.

4.2.2.1.2 Single-variate analysis of quantitative environmental variables

Taking this analysis forward by comparing the diversity with the environmental
variables, Table 19 presents the conditions at each plot, and Table 20 shows the
correlation coefficients (and respective p-values) between plot diversity and the
environmental variables.

In this case only continuous variables are used in the

correlation. Later in the chapter the categorical landscape features are compared with
diversity.

Plot

Eastness

Northness

Feature

Network
Feature

Mean
Curvature

Slope
(degrees)

Slope
Position

Solar
Radiation

Toposcale

Topoclass

1

-0.43

0.90

Planar

Planar

-0.13

32.38

54

8434

-12.46

Slope

7.59

2

-0.43

0.90

Ridge

Planar

0.80

46.38

48

7398

51.31

Slope

8.00

3

-0.64

-0.77

Ridge

Ridge

-1.33

50.07

13

8054

-133.38

Toe Slope

7.36

4

-0.56

0.83

Ridge

Ridge

0.37

13.46

78

9419

42.00

Slope

8.97

5

0.76

0.65

Ridge

Ridge

-0.78

43.84

27

8621

-129.46

Toe Slope

8.27

6

0.41

-0.91

Planar

Planar

-0.07

36.43

45

8354

3.23

Slope

9.38

7

-0.40

0.92

Ridge

Planar

0.67

14.10

98

9357

104.00

Ridge

7.82

8

-0.09

1.00

Ridge

Planar

1.33

20.43

98

8998

153.23

Ridge

8.12

9

-0.54

0.84

Ridge

Ridge

0.82

17.46

96

9289

119.77

Ridge

8.29

10

-0.11

-0.99

Channel

Planar

0.38

40.67

33

7706

22.54

Slope

8.79

Topmodel

Table 19 Environmental conditions at each plot in Tambito, derived from the terrain
derivatives.
Pearson's Correlation
Coefficient

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Family Richness

-0.08

-0.24

-0.07

0.36

0.11

0.00

-0.34

-0.14

Genera Richness

0.22

0.33

0.15

0.33

-0.22

0.28

0.16

0.06

0.30

Species Richness

0.22

0.05

0.06

0.47

-0.18

0.22

-0.09

0.11

0.39

Endemic Species

0.31

0.08

0.43

0.41

-0.41

0.40

0.33

-0.11

0.34

Simpson's

0.28

-0.04

-0.07

0.70

-0.19

0.31

-0.20

0.45

0.65

Shannon's

0.19

0.01

-0.07

0.62

-0.12

0.23

-0.25

0.32

0.55

Margalef

0.09

0.12

-0.04

0.49

-0.08

0.15

-0.21

0.20

0.39

Menhinick

-0.07

0.20

-0.17

0.45

0.05

0.04

-0.33

0.29

0.35

Berger-Parker

-0.18

-0.10

0.17

-0.62

0.07

-0.21

0.29

-0.49

-0.56

174

0.27

p-value

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Family Richness

0.82

0.50

0.85

0.31

0.76

0.99

0.33

0.70

0.45

Genera Richness

0.54

0.35

0.69

0.36

0.54

0.43

0.66

0.88

0.41

Species Richness

0.55

0.88

0.86

0.17

0.62

0.55

0.81

0.77

0.26

Endemic Species

0.39

0.82

0.22

0.24

0.24

0.25

0.35

0.75

0.34

Simpson's

0.44

0.91

0.86

0.03

0.60

0.38

0.58

0.20

0.04

Shannon's

0.60

0.97

0.85

0.05

0.74

0.53

0.49

0.37

0.10

Margalef

0.80

0.73

0.91

0.15

0.83

0.69

0.56

0.58

0.26

Menhinick

0.84

0.59

0.64

0.20

0.89

0.91

0.35

0.41

0.32

Berger-Parker

0.63

0.79

0.63

0.05

0.84

0.57

0.42

0.15

0.09

Table 20 Pearson’s correlation coefficient and the two-tailed p-value between richness and
diversity in each plot and the environmental variables for Tambito.

On the whole the plots cover a broad range of environmental conditions, except for
the northness variable where the plots are located only strongly north- or strongly
south- facing slopes.

Taking elevation as the primary variable (due to its strong co-linearity with many
climatic factors), there is no apparent linear relationship with any measure of richness
or diversity. However, examining the scatterplot of the data points there is some
evidence of a non-linear relationship, with the greatest diversity and lowest levels of
Berger-Parker dominance being found in mid-elevations (Figure 51). Applying a
polynomial trendline to the data points, we find that Simpson’s diversity index
correlates significantly (Simpson’s = -0.0000002 Elevation2 + 0.0006 Elevation +
0.44, R2 = 0.70, p = 0.02), with a peak in diversity at 1850m. Similarly, Shannon’s
diversity peaks at 1800m (R2 = 0.46) and Berger-Parker dominance is also lowest at
1800m (R2 = 0.54).

When richness is examined (family, genus, species and

endemics) neither linear nor polynomial relationships are evident, and the R2 are less
than 0.1 in all cases (Figure 52).

175

Diversity - Elevation relations for Tambito using
different diversity measures
4.00
Shannon's Diversity
Index

3.50

1.00

3.00
0.80

2.50

Shannon's

2.00

Simpson's

1.50

Berger-Parker

0.60
0.40

1.00
0.20

0.50
0.00
1000

1200

1400

1600 1800 2000
Elevation (masl)

2200

Simpson's Diversity
Index and BergerParker

1.20

0.00
2400

Figure 51 Relationship between plot diversity and elevation in Tambito.

Diversity - Elevation relations for Tambito using
richness measures
60

Richness

50

Genera
Families
Species
Endemics

40
30
20
10
0
1000

1200

1400

1600
1800
Elevation (masl)

2000

2200

2400

Figure 52 Relationship between plot richness and elevation in Tambito

The mid-elevation diversity peak is further evident in the data when the plots are
separated into lower-, mid- and upper- elevation bands, and the average diversity and

176

richness calculated for the plots within each elevational band (Table 21). Though the
validity of this analysis is somewhat limited by the few plots found in low and upper
elevations and the uneven distribution of plots in each elevational band, there are
some clear patterns.

No. Individuals No. Families
Lower Elevation Average (1000 1500) (2 plots)
Mid Elevation Average (1500 1900) (5 plots)
Upper Elevation Average (1900 2300) (3 plots)

No. Genera

No. Species

No. Endemic
Species

Simpson's

Shannon's

Margalef

Menhinick

115 (3.7)

21 (13.5)

33 (8.6)

36 (3.9)

17 (49.9)

0.93 (0.0)

3.04 (0.1)

7.38 (4.8)

3.36 (5.8)

Berger-Parker

0.20 (5.7)

98 (11.2)

23 (16.1)

36.8 (13.3)

45 (13.6)

18.2 (37.5)

0.97 (2.2)

3.53 (7.3)

9.69 (14.6)

4.59 (15.5)

0.10 (47.9)

129 (39.4)

20 (7.5)

33 (22.9)

39 (20.7)

21 (38.1)

0.95 (0.8)

3.24 (2.3)

7.92 (13.2)

3.51 (4.2)

0.15 (16.7)

Table 21 Average richness and diversity of plots when separated into three elevational bands.
In brackets the respective coefficient of variation.

Richness and diversity in all but one case (number of endemics) is greater in the midelevations, but there is also evidence that there is greater variability in diversity and
richness in the mid- elevations, especially across different richness measures. If this
pattern is not simply a result of sampling bias, a number of interpretations could be
offered, the theoretical background of which are discussed in Section 2.3.1 of the
literature review. Partial Mantel tests have already indicated that the compositional
similarity of the plots is strongly controlled by difference in elevation, indicating that
species have some habitat associations related to elevation.

Elevation per se is

meaningless in understanding the precise adaptations that species may have to create
that association, but it is likely that temperature is an important factor.

One potential interpretation is that mid-elevations have the greatest heterogeneity in
elevation (see Chapter 6, also discussed to be of theoretical importance in Lundholm
and Larson (2003)), especially in slope, northness and toposcale, and so there is likely
to be a greater diversity in topographically induced micro-climates.

177

If indeed

temperature is an important factor in controlling species composition, a diversity of
micro-climates may duly increase species diversity.

Alternatively, the elevational habitat association in composition identified could mean
that the mid-elevations contain the most overlaps of elevationally controlled niches
(the mid-domain effect), and thus potentially have the greatest species richness
(Colwell and Lees, 2000). More specifically, the mid-elevations are a dynamic zone
where species with adaptations to lower temperatures (higher elevations) occur in low
abundances, along with species with adaptations to higher temperatures (low
elevations). This overlap of species adaptations may create high diversity and low
dominance in the mid-elevations.

To illustrate this idea based on Tambito data, a simple exercise is used. Using Excel,
species abundances across the elevational gradient are modeled based on normal
distributions, calculated based on the mean elevation where the species are found and
the degree of adaptability the species has (amplitude). The amplitude is expressed as
a standard deviation, with species adapted to very specific elevations having a low
standard deviation, and species with wide ranging elevational distribution having high
standard deviations. The model then positions 30 hypothetical species along the
elevational gradient with mean adaptations at random points in the gradient, with
random standard deviations, and calculates for each elevational band the Simpson’s
diversity index based on abundances for each species, and the Berger-Parker
dominance index for the most dominant species. Rapoport’s rule (Stevens, 1992) is
not applied to this analysis.

178

An example result from this model is shown in Figure 53, where 30 very differently
adapted species occur across the elevational gradient. Despite some points of very
low diversity across the gradient which coincide with high dominance of a single
species, the general trend is one of less diversity in the extremes, and high diversity at
mid-elevations. When 50 random runs are made, and the average diversity per 50m
elevational interval is calculated this pattern becomes very clear (Figure 54). As can
be seen, a clear peak in diversity occurs in mid-elevations (peaking at 1750m), with
the corresponding drop in Berger-Parker dominance. The model used here is highly
simplified and fails to take into account many important factors (competition and
resources for example). However, it is aimed at providing an example of one possible
mechanism that may create a mid-elevation diversity peak, whereby the upper- and
lower- extremes of a gradient potentially have less available species to generate high
diversity.

This is in simple terms an edge effect, and only works under the

assumption that there are no other forest species above or below the elevational limits
of the analysis.

16
14
Abundance

12
10
8
6
4
2
0
1100

1600

2100
Elevation (m)

179

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
2600

Simpson's Diversity
(grey) and BergerParker Dominance
(black)

Modelled Abundance and diversity of 30 hypothetical
species along an elevational gradient

Figure 53 Example of niche model, with 30 hypothetical species with randomly positioned
ranges along the elevational gradient. The thick grey line represents the respective Simpson’s
Diversity Index and the black line the Berger-Parker dominance.

Simpson's Diversity Index (blue)
and Berger-Parker dominance
(red)

Average model diversity (50 model runs)
1

y = -2E-07x 2 + 0.0009x + 0.13
R2 = 0.8415

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2

y = 3E-07x 2 - 0.0013x + 1.3858
R2 = 0.8331

0.1
0
1000

1500

2000

2500

3000

Elevation (m)

Figure 54 Average pattern of diversity-elevation relations in the simple niche model for
Tambito.

Application of this concept to Tambito is complicated somewhat by the fact that the
reserve is only a sub-set of the full elevational gradient that stretches from the Pacific
(0m) to the high Andes (2,800m within the Tambito region). However, there is little
forest cover directly below the reserve, and little area above 2800m.

There is

contiguous forest down to the Pacific, but this is some 30km to the north of Tambito,
and if we assume low distances of seed dispersal it is fair to say that Tambito is cutoff from the lowlands to the Pacific. Therefore, strictly speaking this analysis is
covering the entire forested elevational gradient. In other words, it is feasible that

180

low-elevation edge effects on the niche model discussed above are valid, and may
explain the low diversity found in the lower plots.

An important factor which the above model fails to account for is space itself, also a
variable identified as significant in generating mid-elevation peaks in species richness
(Whittaker and Niering, 1975; Rahbek, 1997; Sanders, 2002). Having seen the steep
accumulation of species with increasing area (Figure 47), area could be seen as a
resource itself (Schoener, 1976), and one which generates higher richness through
greater probability of speciation and a larger potential seed pool. Given that elevation
is a limiting factor in terms of species range, and assuming that area is a resource, it is
important to take into account the surface area of land in elevational bands. In
Tambito, mid-elevations (in this case 1600m – 2300m) have the greatest surface area
(Figure 55).

Frequency distribution of elevation in
the Tambito twin-catchments
4000
3000
2000
1000
0
1100
1200
1300
1400
1500
1600
1700
1800
1900
2000
2100
2200
2300
2400
2500
2600
2700
2800
2900

Frequency of 25m
cells

5000

Elevation (m)

Figure 55 Total area of land across the elevational gradient for the twin-catchments of
Tambito, expressed as the frequency of 25m (625m2) cells per 100m elevational band.

181

If we assume that elevational bands with greater surface area are also likely to harbor
more species, then this further exaggerates the mid-elevational diversity peak in the
niche model presented earlier. Furthermore, this area concept also plays an important
role in explaining documented patterns of monotonic diversity loss with elevation in
long gradients from lowlands to highlands (Terborgh, 1977; Vasquez and Givnish,
1998; Givnish, 1999; Kessler et al., 2001; Grytnes, 2003). Taking a cross-section
across Colombia between 2oN and 3oN, 72% of the land surface is below 600m
elevation (Figure 56). If the niche model is then adapted to weigh the positioning of
the 30 species according to land-surface area (i.e. more species located in low
elevations), the edge effect shown in Figure 53 whereby low-elevations have less
diversity (and greater dominance) is cancelled out by area, and the trend becomes one
of monotonic loss of diversity with elevation (Figure 57). Whilst this is an obvious
observation to make (if you place more species in an elevational band then richness is
expected to be higher), it illustrates very clearly the balance between the edge effect
and land-surface area, and provides one mechanism for explaining the contrasting
results of diversity variation across elevational gradients (discussed at length in
Rahbek, 1997). In simple terms, assuming that greater area increases the potential
species pool, studies with plots located in continuous forest cover from lowlands to
highlands will find a gradual loss in diversity with elevation, whilst isolated studies in
mid-elevation catchments (like Tambito) are likely to find mid-elevation peaks in
diversity (as a result of both edge effects and greater mid-elevation land-surface area
in these kinds of catchments).

182

Area of elevational bands for a swath of
o

o

50000
45000
40000
35000
30000
25000
20000
15000
10000
5000
0
50
55
0
10
50
15
50
20
50
25
50
30
50
35
50
40
50
45
50
50
50

1000's km2

Colombia between 2 N and 3 N

Elevation

Figure 56 Land-surface area across the elevational gradient for a swath across Colombia
between 2oN and 3oN.

Average model diversity, weighted based on elevational
area (50 model runs)
Simpson's Diversity Index
(blue) and Berger-Parker
dominance (red)

1.2
1
0.8
0.6
0.4
0.2
0
0

500

1000

1500

2000

2500

3000

3500

Elevation (m)

Figure 57 Niche model adjusted for the area concept, with more species located in elevations
with greater land surface area, and the elevational gradient extended to range form 0m to
3000m.

Having examined the possible reasons for the mid-elevation peak in diversity, it is
also important to look into the high variability in diversity and richness that was found
in mid-elevations (high coefficients of variability in Table 21). As shown later in this

183

thesis (Chapter 6), environmental heterogeneity also peaks at mid- elevations, with
the greatest heterogeneity coinciding exactly at the same point where the greatest
diversity is found (1850m). This may explain the greater variability in diversity and
richness in mid-elevations. Rather than discuss and speculate further as to the driving
factors behind this mid-elevation peak in diversity (and its variation), this is left to the
conclusions where further evidence from TBS is also available for discussion.

Looking at the other environmental variables, few correlations are found, with no
visible evidence of non-linear relationships either.

Only mean curvature and

toposcale (both essentially measuring curvature) showed significant relationships with
Simpson’s index (p<0.05), where the greatest diversity was found in positively
curving convex slopes (Figure 58).

However, the regression is not entirely

convincing, essentially showing that the two least diverse plots happen to be in
concave slopes, whilst the other plots have convex slopes and seemingly show little
pattern of increased diversity with higher curvature.

Relationship between toposcale and diversity
in Tambito

0.98
y = 0.0155x + 0.9537
0.97
2
R = 0.4847
0.96

Simpson's Diversity
Index

Simpson's Diversity
Index

Relationship between mean curvature and
diversity in Tambito

0.95
0.94
0.93
0.92
-1.5

-1

-0.5

0

0.5

1

1.5

0.98
y = 0.0001x + 0.9542
0.97
2
R = 0.4178
0.96
0.95
0.94
0.93
0.92
-200

-100

Mean curvature

0

100

200

Toposcale

Figure 58 Significant relationships between Simpson’s diversity and mean curvature (left)
and toposcale (right) for Tambito.

Upon close examination of the data, plot 9 regularly appears as a strong outlier in
many relationships, and omission of this plot actually renders a number of highly

184

significant correlations (Table 12). Toposcale, mean curvature and slope position all
produce significant correlations with species and genus richness amongst other
diversity measures.

Strongest is the relationship between species richness and

curvature and toposcale (Figure 59).

Pearson's Correlation
Coefficient

Elevation

Eastness

Northness

Curvature

Slope
Position

Slope

Solar
Radiation

Topmodel

Toposcale

Family Richness

0.00

-0.36

0.00

0.49

0.00

0.16

-0.26

-0.14

Genera Richness

0.52

0.23

0.37

0.66

-0.61

0.77

0.54

0.09

0.72

Species Richness

0.51

-0.14

0.26

0.86

-0.55

0.67

0.19

0.15

0.85

Endemic Species

0.42

0.01

0.52

0.53

-0.57

0.60

0.49

-0.11

0.50

Simpson's

0.27

-0.02

-0.08

0.71

-0.18

0.32

-0.24

0.45

0.67

Shannon's

0.29

-0.06

-0.01

0.76

-0.24

0.40

-0.16

0.34

0.73

Margalef

0.27

-0.01

0.08

0.75

-0.32

0.45

-0.03

0.24

0.72

0.43

Menhinick

0.00

0.13

-0.12

0.57

-0.06

0.18

-0.26

0.31

0.50

Berger-Parker

-0.15

-0.13

0.20

-0.62

0.03

-0.17

0.36

-0.49

-0.56

Table 22 Pearson correlation coefficients between plot diversity and richness and
environmental variables when plot 9 is omitted from the analysis. Significant correlations
(p<0.05) are highlighted in grey.

60

60

Simpson's Diversity
Index

Relationship between toposcale and species
richness in Tambito (excluding plot 9)

Simpson's Diversity
Index

Relationship between mean curvature and
species richness in Tambito (excluding plot 9)
50
40
30

Plot 9

y = 5.6187x + 42.225
20
R2 = 0.732
10
0
-1.5

-1

-0.5

0

0.5

1

1.5

50
40
30
y = 0.048x + 42.461
20
R2 = 0.7224
10

Plot 9

0
-200

Mean curvature

-100

0

100

200

Toposcale

Figure 59 Relationship between species richness and mean curvature (left) and toposcale
(right) when plot 9 is omitted from the analysis. The linear relationship and R2 displayed
corresponds to the result without plot 9, though the point is shown in the graph for
information.

Omitting plot 9 from the analysis as an outlier is justified based on the conditions
experienced in the field. Plot 9 is positioned almost directly on a major ridge (and at
185

the catchment boundary), it was almost flat (slope approximated in the field at 9
degrees) and certainly not convex in terms of curvature (at small spatial scales, though
the larger-scale curvature is definitely convex). The widening of the ridge and the
small flat and sheltered region in which the plot is located is not identified by the
DEM, and therefore the curvature and toposcale values derived from the DEM are not
actually representative of the reality found in the field. It is not however justifiable
omitting plot 9 from the slope position regression because the slope position is
accurately identified as 96, i.e. very close to the ridge.

Having justified the omission of plot 9 from the curvature and toposcale calculations,
there is a very strong relationship which has some interesting implications.
Furthermore, these patterns are independent of the elevational relationships already
discussed. When correlation coefficients are applied to the data for mean curvature at
each individual scale (windows of 3, 5, 7, 9, 11, 13 and 15 cells), in all cases,
including or omitting plot 9, the highest degree of correlation occurs at mean
curvature with a window size of 7 or 9 cells, i.e. a mid-scale measure of mean
curvature. The lower correlation at small window sizes may be due to deficiencies in
DEM quality at the micro-scale rather than have biological significance, but the effect
of curvature on diversity and richness is clearly lower for broader scale measures of
curvature (i.e. high window sizes).

186

Mean
Curvature
Scale

All Plots
Pearson
Richness

Omitting Plot 9

Pearson
Diversity

Pearson
Richness

Pearson
Diversity

3

0.41

0.45

0.64

0.44

5

0.55

0.63

0.80

0.62

7

0.58

0.69

0.87

0.69

9

0.57

0.72

0.89

0.73

11

0.51

0.72

0.88

0.73

13

0.44

0.70

0.87

0.72

15

0.37

0.67

0.84

0.70

Table 23 Pearson correlation coefficients for Simpson’s diversity and species richness
correlations with mean curvature at different scales.

The fact that there is greater diversity on convex slopes could be explained by two
possible reasons. One is the topographic exposure, particularly important in cloud
forests due to potentially higher levels of cloud presence in exposed regions.
Presence of cloud at the canopy provides an extra input of water to the system through
cloud interception, which is not necessarily an insignificant amount compared to
water input from rainfall (Jarvis, 2000; Mulligan and Jarvis, submitted), but becomes
significant in dry months, when cloud interception prevails despite low rainfall. The
water contained in clouds has also been found to be richer in nutrients, and so exposed
sites may also be richer in nutrients though no studies have made direct measurements
of this. However, there is also an argument against this explanation. Increased cloud
interception due to exposure also reduces incoming light and temperature,
encouraging

waterlogging,

and

photosynthesis and respiration.

increases

leaf

wetting,

thus

constricting

These conditions may favour only a handful of

species better adapted to these conditions. The likely effect might be one of high
dominance and low richness.

187

The second possible reason regards access to light, whereby convex slopes potentially
have greater light receipt surface area at the canopy level, but also greater probability
of direct radiation reaching the ground surface through lateral penetration of the
canopy. The variety of angles from which light may be received by the canopy and
the sub-canopies and understorey beneath provides a greater number of potential
niches for different light-related adaptations. This combination of greater surface area
for light receipt per unit area of land, and the resultant diversity in light niches may
generate species richness.

There is insufficient data to ascertain which of the two reasons generates the higher
diversity observed on convex slopes, though it may indeed be a combination of both
factors.

Having isolated this important correlation between richness and curvature, and
equally the lack of clear correlation between diversity measures and curvature, this
may shed further light on the mid-elevation diversity peaks that have been discussed
earlier.

Statistically significant mid-elevation diversity peaks were found using

diversity measures, but richness measures failed to show the same relationship. Given
the strong correlation with curvature, this co-variable may be creating a more
complex and variable pattern in richness with elevation, and this effect can be isolated
by fitting a polynomial mid-elevation peak trendline to the richness data, and
comparing the residual in richness with the curvature for each plot (Figure 60).

188

Species Richness
Residual

Residual in richness (assuming polynomial midelevation diversity peak) against the mean
curvature
y = 2.4812x - 0.3423
R2 = 0.2372

8
6
4
2
0

-1.5

-1

-0.5

-2 0

0.5

1

1.5

-4
-6

Mean Curvature

Figure 60 Residual in richness (with respect to a polynomial mid-elevational peak trendline)
compared with mean curvature, with the purpose of explaining the variation in richness across
the elevational gradient. The residual is calculated based on the actual richness (plot
measured) minus the modeled richness (polynomial trendline). Plot 9 is excluded from the
analysis for reasons explained earlier.

The relationship between richness residual and curvature is not statistically significant
(Pearson = 0.49, p-value 0.18). A weak trend is evident, whereby plots with a
positive residual (i.e. greater diversity than expected with the polynomial function)
also have convex curvature, and those with a negative residual have concave
curvature. However, this is not significant and it cannot be concluded that richness is
generated through a simple combination of linearly increasing richness on convex
slopes, and a polynomial peak in richness at mid-elevations. Other processes appear
to be at play, that are not evident in the data studied here.

4.2.2.1.3 Multi-variate analysis of quantitative environmental variables

189

It is clear that elevation is important in explaining diversity, but equally curvature and
some other factors appear to be important. Analyses are therefore applied that take
into account all variables at once to examine to what extent each variable contribute to
explaining the variability in diversity.

For this, multiple stepwise regression is

applied for Simpson’s Diversity and richness as the dependent variables, and all nine
terrain derivatives as the independent variables.

For Simpson’s index, just two

variables explain 71% of the variability in diversity (D = 0.0215 Mean Curvature –
0.0114 Northness + 0.95, p = 0.01, n = 10). Mean curvature explains 48% of the
variability and northness explains a further 22% of the variability. Applying this
model to the topographic data produces a clear relationship between modeled and
measured diversity (Figure 61), and interestingly the modeled diversity captures the
mid-elevation peak already observed (Figure 51). Curvature has already been shown
to be important in defining diversity, and some possible reasons for this have also
been presented. However, northness alone shows little correlation with diversity
when treated alone (Pearson = -0.07), but in combination with curvature does explain
some of the variability. This further suggests that solar radiation is indeed important,
northness being important in defining the time of year of maximum radiation receipt
and the degree of annual fluctuation in radiation receipt. The correlation and the
elevational gradient in modeled diversity shown in Figure 61 is expected as the same
data went into the model as it is compared against, and this result would be greatly
enhanced by a split-sample regression and validation, but with just 10 plots to work
with this was impossible.

190

Relationship between modelled diversity and
measured diversity in Tambito
0.97

0.98

y = 0.9999x + 9E-05
R2 = 0.7074

Modelled Simpson's
Diversity

Measured Simpson's
Diversity

0.98

Relationship between modelled diversity and
elevation in Tambito

0.96
0.95
0.94
0.93
0.92
0.92

0.93

0.94

0.95

0.96

0.97

0.98

0.97
0.96
0.95
0.94
0.93
0.92
1000

Modelled Simpson's Diversity

y = -9E-08x 2 + 0.0003x + 0.6408
R2 = 0.5471
1500

2000

2500

Elevation (m)

Figure 61 Modelled diversity (calculated using the multiple stepwise regression) correlated
against measured diversity (left), and the variation in modeled diversity across the elevational
gradient.

When species richness is examined in the multiple stepwise regression, no significant
multiple regressions are found on top of the relationships already discussed (Figure
59, for example), both including and excluding plot 9. This in itself is interesting, as
strong correlations are found with diversity but few patterns in richness. The Renyi
diversity plots have already shown that richness is non-comparable with diversity in
most cases, but are to some degree correlated (Pearson = 0.54, p = 0.10, n = 10).
These results do show that different interactions occur between environment and
richness and environment and diversity, with the topography and environment
variables measured in this study affecting evenness/dominance more than they affect
absolute species richness.

4.2.2.1.4 Analysis of categorical environmental variables

It would be beneficial to also use the categorical variables to examine possible
relationships between the diversity measured in the field and the categorical landscape
feature variables.

However, there is not a technique available that provides a
191

statistically sound results given the type of data and number of data points.
Spearman’s rank test is not valid as the landscape features are nominal and not ordinal
categories (whilst you can order some variables like valley -> slope, it is questionable
as to where a pass might fit in the order). Alternatively, chi-square contingency tests
could be applied, but the number of data points (10) means that some boxes in the
contingency table will have missing values, invalidating the result. For these reasons
the categorical data are not statistically analysed in terms of relationships with
diversity.

The multi-scale fuzzy feature classification does however provide a quantitative
variable of ridge, channel and planar membership for each plot. Correlations between
richness and diversity measures and membership to channel, planar and ridge provide
no clear pattern. The fact that greater ridge membership does not necessarily mean
greater diversity provides evidence that the categorical variables provide little
explanation of diversity patterns.

Greater depth and detail as to the possible explanation of these patterns are left to the
conclusions, benefiting from results and discussion from the TBS data.

4.3

Tiputini Biodiversity Station

Having completed the analysis of diversity and compositional patterns in Tambito, the
data from TBS is now examined in much the same way.

4.3.1 Composition

192

4.3.1.1 General Site Species Composition

The ten 25m x 25m plots established in TBS contained 937 individuals, of which 455
different species were identified, distributed in 53 families and 181 genera. Appendix
5 contains a full list of species found in TBS. Taxonomic identification was made to
family level for 98.7% of species, to genus level for 76.6% of species, and to species
level for 23.1% of species. This significantly lower success rate in terms of species
identification is due to the TBS plots having been established after Tambito, leaving
less time for post field-work taxonomic identification in the herbarium. As stated
before, it is likely that some species as yet unidentified to genus level will be new
genera for the plots, and thus the number of different genera is likely an
underestimate. Table 24 provides a summary of family composition in Tambito, with
Figure 62 and Figure 63 showing the distribution of species and individuals within
families.

193

Family
Anacardiaceae
Annonaceae
Apocynaceae
Aquifoliaceae
Araliaceae
Arecaceae
Bignoniaceae
Bixaceae
Bombacaceae
Boraginaceae
Burseraceae
Capparaceae
Caricaceae
Cecropiaceae
Celastraceae
Chrysobalanaceae
Clusiaceae
Dichapetalaceae
Ebenaceae
Elaeocarpaceae
Euphorbiaceae
Fabaceae
Flacourteaceae
Humiriaceae
Icacinaceae
Lauraceae
Lecythidaceae
Malpighiaceae
Melastomataceae
Meliaceae
Menispermaceae
Moraceae
Myristicaceae
Myrsinaceae
Myrtaceae
Nyctaginaceae
Ochnaceae
Olacaceae
Piperaceae
Polygonaceae
Proteaceae
Rubiaceae
Sapindaceae
Sapotaceae
Simaroubaceae
Staphyliaceae
Sterculiaceae
Styracaceae
Theoprastinaceae
Tiliaceae
Ulmaceae
Violaceae
Vochysiaceae
Other
Total

No. Genera

No. Individuals

No. Species

No. Plots

4
7
4
1
1
7
3
1
4
1
3
1
1
3
1
3
3
1
1
1
14
15
9
1
1
7
4
1
3
4
1
12
4
3
3
1
1
1
1
2
1
13
2
5
1
1
3
1
1
3
2
3
2
4
181

6
19
4
2
4
50
7
1
33
2
22
1
2
28
2
26
6
2
1
10
24
173
13
2
2
46
22
1
16
42
3
111
22
6
25
10
2
4
3
8
1
38
4
30
1
1
7
1
1
3
4
20
3
60
937

6
16
4
2
2
9
4
1
10
2
16
1
1
15
2
11
5
2
1
6
17
79
12
2
2
33
9
1
11
16
2
30
14
3
24
3
1
3
1
3
1
24
4
19
1
1
4
1
1
3
3
4
2
5
455

5
9
3
1
3
9
4
1
10
2
9
1
2
9
2
9
4
2
1
6
9
10
8
2
2
10
8
1
5
7
2
10
10
4
9
4
2
3
1
5
1
9
3
9
1
1
2
1
1
1
3
8
2
3
AVG (4.6)

Table 24 Summary of species composition in TBS

194

Figure 62 Distribution of species in families in TBS

Figure 63 Distribution of individuals in families in TBS

The most abundant families in terms of individuals are Fabaceae (173), Moraceae
(111), Arecaceae (50), and Lauraceae (46), with Fabaceae also being the most species
rich family (79 species), followed by Lauraceae (33), Moraceae (30) and Myrtaceae
(24). Fifty percent of all species are found in seven families, whilst 12 families
195

encountered had just one species present in the plots. The most abundant species
were Sorocea sp. (Moraceae) with 56 individuals found in just 3 plots, Browneopsis

ucayalina Huber (Fabaceae) with 37 individuals in 6 plots, Rinorea viridifolia Rusby
(Violaceae) with 32 individuals found in seven different plots, and Iriarthea deltoidea
Ruiz & Pavon (Arecaceae) with 25 individuals distributed in seven different plots.
Valencia et al. (2004) reports similar family level composition and dominance in
these species in a nearby 25-Ha plot in Yasuni (approximately 30km from TBS). The
average abundance per species was just 2.1 individuals, and some 309 of the 455
species (68%) having only one individual in the plots.

The taxonomic identification for the TBS collections is less advanced than that of
Tambito, and for this reason less species have been classified to date. It is also
possible that some collections currently identified as different species will be
consolidated into the same species upon further study, so the total number of species
may fall. Nevertheless, the plots in TBS contain remarkable levels of diversity.

4.3.1.2 Between-Plot Compositional Variability

4.3.1.2.1 General discussion

Five families were broadly distributed across all plots (Bombacaceae, Fabaceae,
Lauraceae, Moraceae, and Myristicaceae), while 12 of the 53 families (22.6%) were
found in only one plot. On average a particular family is found in 4.1 plots. At the
species level, the most broadly distributed species were Brownea grandiceps
(Fabaceae), Rinorea viridifolia, and Iriarthea deltoidea Ruiz & Pavon (Arecaceae)

196

which were found in seven plots, and Browneopsis ucayalina (Fabaceae) and Perebea

sp. (Moraceae) which were found in six different plots. Some 363 of the 455 species
(79.8%) were found in only one plot, likely indicating that the 25m x 25m plots only
capture a tiny sub-sample of the total diversity to be found in the region.

This is strongly reflected in the species-area curves (Figure 64), where the trend of
species accumulation with increasing area is practically linear in all plots. This
indicates that the plots are not saturating at a representative value of alpha diversity,
and thus the plot richness and diversity is just a sub-sample of the likely alpha
diversity in the region.

Number of species

Species -Area curves for TBS 25m plots
Plot 1

90
80
70
60
50
40
30
20
10
0

Plot 2
Plot 3
Plot 4
Plot 5
Plot 6
Plot 7
Plot 8
Plot 9
Plot 10

0

200

400
Area (m2)

600

800

Figure 64 Species-area curves for TBS.

4.3.1.2.2 Single-variate analysis of quantitative environmental variables

As performed in Tambito, plot compositional similarity is compared with
environmental difference using scattergrams and Mantel statistics. This is done using

197

each DEM, so two sets of results are shown (for the TOPODEM Figure 65 and Table
25, and for the SRTM DEM Figure 66 and Table 26). The average elevation value for
each plot was collected using GPS, rather than extracted from the DEM. For this
reason the same result is achieved regardless of the DEM used. In addition to the
terrain derivatives, distance from river was also used as an indicator of potential flood
frequency, and also to a certain extent the age of forest. Satellite images over the
region show some evidence of past river courses, and indicate that river channels may
migrate rapidly in the region.

An approximation of the speed of lateral river

migration was made based on comparison of historic air photography and recent
satellite imagery and approximated at 1-4m per year in flat areas (Mark Mulligan,
personal communication). Under this logic, it is more probable that regions far from
the main Tiputini River contain older forest and soils.

198

Com positional Sim ilarity and Elevation

Compositional Similarity and Distance
Elevational Difference (m)

Horizontal Distance
Between Plots (m)

3000
y = -5397.1x + 1587.3
R2 = 0.0726

2500
2000
1500
1000
500
0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

70
y = -27.096x + 21.591
R2 = 0.0023

60
50
40
30
20
10
0
0.00

0.02

Jaccard Sim ilarity

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Compositional Similarity and Distance from
River

Compositional Similarity and Eastness

Eastness Difference

Difference in distance
from river(m)

2.5
2500
y = -8108.3x + 1120.4
R2 = 0.1622

2000
1500
1000
500
0
0.00

0.05

0.10

0.15

y = 1.4685x + 0.8736
R2 = 0.0048

2.0
1.5
1.0
0.5
0.0
0.00

0.02

Jaccard Sim ilarity

Curvature Difference

1.2
y = -2.372x + 0.9051
R2 = 0.0157

2.0
1.5
1.0
0.5
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.8
0.6
0.4
0.2
0.0
0.00

0.02

6.0
4.0
2.0
0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Slope Position Difference

8.0

0.0
0.00

0.04

0.06

0.08

0.10

0.12

Compositional Similarity and Slope Position

y = 11.549x + 3.9741
R2 = 0.0086

10.0

0.12

Jaccard Sim ilarity

Compositional Similarity and Slope

12.0

0.10

y = 1.0679x + 0.3008
R2 = 0.0082

1.0

Jaccard Sim ilarity

14.0

0.08

Compositional Similarity and Curvature

2.5
Northness Difference

0.06

Jaccard Sim ilarity

Compositional Similarity and Northness

Slope Difference

0.04

100.0
y = -311.84x + 56.678
R2 = 0.0962

80.0
60.0
40.0
20.0
0.0
0.00

0.02

0.04

0.06

0.08

Jaccard Similarity

199

0.10

0.12

Compositional Similarity and Toposcale
500.0

70.0
Toposcale Difference

Solar Radiation Difference

Compositional Similarity and Solar Radiation

y = -70.293x + 25.411
R2 = 0.0122

60.0
50.0
40.0
30.0
20.0
10.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

y = -154.32x + 144.61
R2 = 0.0014

400.0
300.0
200.0
100.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Jaccard Sim ilarity

Compositional Similarity and TopModel

TopModel Difference

5.0

y = 2.473x + 1.4517
R2 = 0.0047

4.0
3.0
2.0
1.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Figure 65 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity) derived
from the TOPO DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

-0.27
-0.40
-0.05
0.07
-0.13
0.09
0.09
-0.31
-0.11
-0.04
0.07

0.079
0.041
0.408
0.370
0.210
0.376
0.374
0.017
0.248
0.400
0.407

Not-significant
Significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Significant
Not-significant
Not-significant
Not-significant

Table 25 Summary results of compositional similarity analysis with distance and
environmental variables derived from the TOPO DEM, using Mantel tests and a one-tailed

200

Pearson. Note that compositional similarity was compared with environmental dissimilarity
(difference), hence negative values indicate a positive causal relationship.

Com positional Sim ilarity and Elevation
Elevational Difference (m)

Compositional Similarity and Distance

Horizontal Distance
Between Plots (m)

3000
y = -5397.1x + 1587.3
R2 = 0.0726

2500
2000
1500
1000
500
0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

70
y = -27.096x + 21.591
R2 = 0.0023

60
50
40
30
20
10
0
0.00

0.02

Jaccard Sim ilarity

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Compositional Similarity and Distance from
River

Compositional Similarity and Eastness

Eastness Difference

Difference in distance
from river(m)

2.5
2500
y = -8108.3x + 1120.4
R2 = 0.1622

2000
1500
1000
500
0
0.00

0.05

0.10

0.15

2.0
y = -3.5248x + 0.799
R2 = 0.0245

1.5
1.0
0.5
0.0
0.00

0.02

Jaccard Sim ilarity

Curvature Difference

Northness Difference

1.2

1.5
y = -7.7205x + 0.9617
R2 = 0.273

1.0
0.5

0.02

0.04

0.06

0.08

0.10

0.12

0.8
0.6
0.4
0.2
0.0
0.00

0.02

Slope Position Difference

Slope Difference

y = 6.7642x + 1.4886
4.0
R2 = 0.0295
3.0
2.0
1.0

0.08

0.04

0.06

0.08

0.10

0.12

Compositional Similarity and Slope Position

5.0

0.06

0.12

Jaccard Sim ilarity

Compositional Similarity and Slope

0.04

0.10

y = 0.7488x + 0.3549
R2 = 0.007

1.0

Jaccard Sim ilarity

0.02

0.08

Compositional Similarity and Curvature

2.0

0.0
0.00

0.06

Jaccard Sim ilarity

Compositional Similarity and Northness

0.0
0.00

0.04

0.10

0.12

Jaccard Sim ilarity

80.0
70.0
60.0
50.0
40.0
30.0
20.0
10.0
0.0
0.00

y = -165.31x + 36.462
R2 = 0.0584

0.02

0.04

0.06

0.08

Jaccard Sim ilarity

201

0.10

0.12

Compositional Similarity and Toposcale

60.0

250.0
Toposcale Difference

Solar Radiation Difference

Compositional Similarity and Solar Radiation

y = 30.27x + 17.847
R2 = 0.0038

50.0
40.0
30.0
20.0
10.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

y = -880.3x + 112.41
R2 = 0.2454

200.0
150.0
100.0
50.0
0.0
0.00

0.12

0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Jaccard Sim ilarity

Compositional Similarity and TopModel

TopModel Difference

10.0
y = -20.76x + 3.8874
R2 = 0.048

8.0
6.0
4.0
2.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Similarity

Figure 66 Scatterplots for plot-pairs of compositional similarity (measured with the Jaccard
coefficient) and environmental distance (in this case the difference, i.e. dissimilarity) derived
from the SRTM DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

-0.27
-0.40
-0.05
-0.16
-0.52
0.08
0.17
-0.24
0.06
-0.50
-0.22

0.092
0.039
0.379
0.219
0.007
0.312
0.130
0.067
0.417
0.010
0.160

Not-Significant
Significant
Not-Significant
Not-Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Significant
Not-Significant

Table 26 Summary results of compositional similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a one-tailed

202

Pearson. Note that compositional similarity was compared with environmental dissimilarity
(difference), hence negative values indicate a positive causal relationship.

The Mantel tests for TBS show significant correlations (p<0.05) between environment
and composition for distance to river, slope position for the TOPO DEM, and
northness and toposcale for the SRTM DEM. Given that no single variable correlates
for more than one DEM makes the interpretation of these results problematic and very
much dependent upon the data source (see for example the differences in DEM values
in Section 3.7 in Chapter 3). Out of 19 correlations between terrain derivatives and
composition, four correlate to the 95% confidence limit. At least one of these is
expected in a random set of data.

Applying partial Mantel tests on these correlating variables, using distance from river
as the third variable indicates that the correlation with slope difference (TOPO DEM)
is spurious, but the other two correlations hold (Table 27).

Variable A
Compositional similarity
Compositional similarity
Compositional similarity

Partial
Correlation Correlation Correlation
Variable C
(A,B)
(B,C)
(A,C)
Slope difference (TOPO)
Distance
-0.31
0.49
-0.40
Northness (Northness)
Distance
-0.48
0.29
-0.40
Toposcale (ASTER)
Distance
-0.47
0.17
-0.40
Variable B

Correlation
(A,B) | C
-0.16
-0.44
-0.45

p-value
0.14
0.02
0.01

Table 27 Revised Mantel statistics for correlating variables with compositional similarity in
TBS, applying partial Mantel tests to detect spurious correlations derived from co-linearity
between variables and the distance from river.

It is interesting to note that Euclidean distance between the plots does not correlate
significantly (although a weak trend is evident), but that the difference in the distance
to river does correlate. This indicates that broadly-speaking the composition follows

203

a steady gradient from the river-bank to the more “inland” regions. Interpreting this
relation is complicated by a number of factors. As already discussed, distance from
river is in one sense an indicator of flooding frequency, along with elevation. Regions
close to the river are more likely to flood than regions of the same elevation further
from the main river channel, although this generalization is complicated by the effect
of the network of smaller channels (which during times of high flow become flooded
themselves). Though subjective and likely incomplete, an IKONOS image was used
to identify the sub-channels within the TBS region (Mark Mulligan, personal
communication). The distance of each plot from any type of channel (including the
River Tiputini itself) was calculated, and a Mantel test performed on the
compositional similarity and the difference in distance from any of these channels
(Figure 67).

Difference in distance
from any river channel

Compositional Similarity and distance from any
river channel
200.0

y = -234.41x + 63.501
R2 = 0.0237

150.0
100.0
50.0
0.0
0.00

0.02

0.04

0.06

0.08

0.10

0.12

Jaccard Sim ilarity

Figure 67 Scatterplot of compositional similarity against the difference in distance from any
river channel, using channels identified in an IKONOS image of TBS.

No significant correlation was found between the distance from any river channel
(large or small) and compositional similarity (Mantel statistic = -0.154, p = 0.25).

204

Furthermore, elevation does not correlate significantly with compositional similarity,
leading to the conclusion that flooding frequency is not a plausible explanation for the
significant correlation between distance from the main river and composition.

In this case, the alternative explanation involves forest age, under the idea that the
conditions and forest in those regions further away from the river are in fact “older”.
This is plausible, as satellite imagery from the region shows past river courses in the
form of oxbow lakes (Figure 68), as well as broader scale re-routing of rivers. If the
rivers migrate significantly (supported by Rasanen et al., 1986; Salo et al., 1986), it is
probable that the regions closer to the river were more recently river channel than
more distant regions.

Figure 68 Example of the river migration present in the region, with many oxbow lakes
forming where old meanders were cut off. This image is of an area to the south of TBS, taken
using LANDSAT TM imagery.

205

If this is the case, it is unlikely that the compositional pattern is brought about by
successional processes, as the time frame involved in river migration is likely to have
operated on a longer time-scale, though this cannot be completely excluded from the
explanation.

However, other indirect and slower effects are likely to be more

important. Soils, for instance, are likely to be strongly affected by the time since the
site was occupied by a river channel (Huston, 1994). The river is likely to wash away
many soil horizons, and subsequently deposit sediments. Once the river moves away
from the site, pedogenic processes will commence and the soil will slowly develop.
This process of soil evolution will result in differing edaphic conditions (biological,
chemical and physical) depending on the time elapsed since the original horizons
were washed away by river flow. Given that several studies have shown lowland
tropical tree species to have edaphic preferences (for example Clark et al., 1998;
Palmiotto et al., 2000; Phillips et al., 2003), the compositional similarity between sites
with similar distances from the river may provide indirect evidence that edaphic
conditions are important in defining species composition in TBS. Unfortunately no
soil data is available for testing this hypothesis, as this thesis is attempting to find
readily measurable variables (or surrogates) for explaining the spatial variability in
diversity.

Distance to river was not the only variable that correlated with composition. Table 27
confirms that northness (SRTM) and toposcale (SRTM) correlate independently of
distance from river.

The significant correlation between the difference in northness (SRTM) and toposcale
(SRTM) are only significant for the SRTM DEM, where the cell size limits the

206

accuracy of the terrain characteristic, especially for toposcale.

The micro-scale

topographic variation is not picked up by SRTM, meaning that the toposcale pixel
value over the plot is unlikely to reflect the actual plot conditions. The significance of
northness in TBS is somewhat weak, with the low slopes unlikely to produce much
variation in solar radiation around the study site, and the dominant wind direction
form the south-east and east means that northness is not the strongest indicator of
exposure to wind. Furthermore, the correlation with northness shows no correlation at
all for other DEMs, providing little confidence in the significance of this result.

4.3.1.2.3 Multi-variate analysis of quantitative environmental variables

Taking a multi-variate approach once more, all nine quantitative terrain and
environment variables are subjected to an agglomerative hierarchical clustering
analysis for each of the three DEMs, producing a dendrogram of plot similarity
(Figure 69, and Figure 70), and an associated multi-variate dissimilarity matrix. The
difference in the dendrograms for each DEM is alarming, and is yet another indication
that the DEMs themselves are very different and that the analyses for TBS are
significantly more subjective due to problems with DEM quality.

When the

dissimilarity matrix is compared with compositional similarity using the Mantel
statistic, no significant correlations are found for either of the DEMs (MantelTOPO =
0.18, and MantelSRTM = 0.22, p>0.05 in both cases).

207

Tree diagram of multi-variate environmental clustering for
plots in TBS using the TOPO DEM

30

Dissimilarity

25
20
15
10
5
0
Plot 8

Plot 1

Plot 5

Plot 4

Plot 7

Plot 2

Plot 3

Plot 9

Plot 6 Plot 10

Figure 69 Multi-variate clustering of plots based on all 9 environmental and topographic
variables for TBS using the TOPO DEM.

Tree diagram of multi-variate environmental clustering for
plots in TBS using the SRTM DEM

42

Dissimilarity

35
28
21
14
7
0
Plot 4

Plot 1

Plot 3

Plot 5

Plot 2

Plot 6

Plot 9

Plot 8

Plot 7 Plot 10

Figure 70 Multi-variate clustering of plots based on all 9 environmental and topographic
variables for TBS using the SRTM DEM.

Multiple stepwise linear regressions do however produce some interesting results.
When applied to the compositional similarity and environmental difference data for

208

each plot pair, no additional variables outside of distance from river explain the
compositional variability when the TOPO DEM is used. However, for the SRTM
DEM four variables together explain 52% of the variability in composition, starting
with northness (23%), toposcale (a further 13%), distance from river (a further 7%)
and elevation (a further 9%) (Jaccard Similarity = 0.076 –0.0189Northness
DifferenceSRTM –0.00026Toposcale DifferenceSRTM –0.00002Difference in Distance
from River + 0.0006Altitudinal difference, R2 = 0.52, n = 45). When the modeled
similarity is plotted against the measured similarity (Figure 71), the greatest errors in
predicted compositional similarity occur in modelled high levels of similarity, where
there are cases of vast over- and under- estimate of the measured similarity.

Modelled compositional similarity against
measured compositional similarity for TBS using
the SRTM DEM
Measured Jaccard
Similarity

0.12
0.10
0.08

y = x - 1E-16
R2 = 0.5157

0.06
0.04
0.02
0.00
-0.020 -0.020.000

0.020

0.040

0.060

0.080

Modelled Jaccard Sim ilarity

Figure 71 Modelled similarity against the measured similarity applying the results of the
multiple stepwise linear regression for the variables of the SRTM DEM.

Extreme caution is urged in trusting the causal significance of the multiple linear
regressions, and no concrete conclusions can be derived of these without validation
plots.

4.3.1.2.4 Analysis of categorical environmental variables

209

Finally for the analysis of composition, the restriction of species to landscape classes
is analysed using the same methodology presented for Tambito using species
distribution restrictions to individual landscape classes (Table 28,Table 29 and Table
30).

Feature
Planar (TOPO)
Planar (SRTM)
Ridge (TOPO)
Ridge (SRTM)
Channel (TOPO)
Channel (SRTM)

No. Plots

No. Species
Restricted to
feature

6
5
3
3
1
2

243
192
125
133
27
60

No. species restricted
to feature only for
species distributed in
more than 1 plot
30
11
3
9
0
1

Table 28 Number of species associated to specific classes in the “Feature” classification for
TBS.

Network Feature
Planar (TOPO)
Pass (SRTM)
Peak (SRTM)
Ridge (TOPO)
Ridge (SRTM)
Channel (TOPO)
Channel (SRTM)
Pit (SRTM)

No. Plots

No. Species
Restricted to
feature

3
1
2
3
4
4
2
1

119
44
58
125
184
133
53
45

No. Species Restricted
to feature for species
distributed in more
than 1 plot
7
0
1
3
19
3
0
0

Table 29 Number of species associated to specific classes in the “Network Feature”
classification for TBS.

210

Topoclass
Valley Bottom (TOPO)
Valley Bottom (SRTM)
Toe Slope (TOPO)
Toe Slope (SRTM)
Slope (TOPO)
Slope (SRTM)
Ridge (TOPO)

No. Plots

No. Species
Restricted to
feature

1
2
4
5
4
3
1

45
50
136
217
155
125
47

No. Species Restricted
to feature for species
distributed in more
than 1 plot
0
5
12
16
7
7
0

Table 30 Number of species associated to specific classes in the “topoclass” classification for
TBS.

Some 363 of the 455 species were found only in one plot, leaving just 92 species for
examination of restriction to a single landscape class. As already discovered for
Tambito, very few of these species are shown to be restricted to particular landscape
classes.

The greatest restriction occurs using the TOPO DEM and the feature

classification, with 36% of the 92 species restricted to landscape classes, and the
majority are restricted to planar landscapes. The network feature classification shows
the least landscape feature association, with the topoclass variable (a derivative of
toposcale which has been shown to be significant in defining compositional similarity
between plots) showing moderate habitat association in species composition using the
SRTM DEM, with 28 of the 92 species showing restrictions in their distributions
(30%).

In none of these cases was there a single family showing more habitat

association than any other.

Though some habitat association is evident, this cannot be considered significant and
may simply be a product of the low densities in the species distributions and the
related problems associated with the small size of plots used in this study.

211

4.3.2 Diversity

4.3.2.1.1 General discussion

Once again, significant variation in diversity is found between-plots, with species
richness ranging from 31 (plot 5) to an astonishing 82 (plots 2 and 10) (Table 31). On
average there were 24 families per plot (ranging from 14 to 28), and 35 genera
(ranging from 12 to 49). Similarly for Tambito, Renyi diversity plots are used to aid
the discussion on diversity patterns between plots in TBS (Figure 72).

Plot
1
2
3
4
5
6
7
8
9
10

No. of
No.
No.
Individuals Families Genera
80
132
95
101
88
90
66
103
95
97

26
24
24
25
14
25
21
28
27
26

36
36
35
29
12
35
33
49
49
34

No.
No.
Endemic Simpson's Shannon's Margalef Menhinick
Species
Species
62
44
0.99
4.01
13.96
6.98
82
44
0.98
4.09
16.75
7.31
77
40
0.99
4.24
16.69
7.90
59
29
0.95
3.60
12.59
5.90
31
15
0.84
2.58
6.70
3.30
61
25
0.98
3.87
13.33
6.43
43
28
0.98
3.56
10.06
5.33
79
49
0.99
4.25
16.86
7.82
75
40
0.99
4.11
16.25
7.69
82
49
0.99
4.31
17.75
8.37

Table 31 Richness and diversity of plots in TBS

212

BergerParker
0.06
0.14
0.05
0.21
0.38
0.12
0.09
0.06
0.09
0.06

90
80
70
60
50
40
30
20
10
0

Plot 1
Plot 2
Plot 3
Plot 4
Plot 5
Plot 6
Plot 7

Scale Parameter

3

5

5
2.
7

5

2

2.

2.
2

5

5
1.
7

5

1

1.

1.
2

5

5
0.
7

0.

5

Plot 8
0.
2

0

Exp (Renyi) Diversity

Scaleable Diversity - TBS - Renyi Exponential

Plot 9
Plot 10

Figure 72 The Renyi scaleable diversity index for TBS, indicating to what extent plot
diversities can be universally compared

Plot 5 is the least diverse plot in all aspects by a considerable margin, having high
levels of dominance (33 individuals of Sorocea sp. (Moraceae)), and very low species
richness (31 species). This plot was located immediately next to the main river
channel, with strong evidence of regular flooding. Plot 4 is also low in diversity (and
is also close to the river where flooding is likely to occur frequently), though it is
comparatively more species rich (59 species) than plot 7 (just 43 species). Plot 8 is
both the most species rich and the most diverse plot (and is also positioned at the
highest elevation), whilst plot 2 has high species richness but relatively low diversity
due to high levels of dominance of Browneopsis ucayalina (18 individuals). Also
worthy of note is plot 9, with high species richness (75 species) but comparably low
diversity due to some dominance of Licania glablanca (Chrysobolanaceae) (9
individuals). The many shapes of the curves in the Renyi diversity plot shows that not
only are there very different levels of diversity around the TBS region, but that also

213

the structure of diversity is also variable (i.e. species richness vs. evenness/dominance
is variable).

4.3.2.1.2 Single-variate analysis of quantitative environmental variables

Exploring the potential drivers of this variation in richness and diversity, Table 32
presents the environmental conditions at each plot using each of the DEMs. On the
whole the plots cover a complete range of environmental conditions, with a good
distribution of points in all the variables, as a result of the more formal sampling
strategy that was adopted compared to Tambito (see Chapter 3, Section 3.5.1). There
is a great deal of variation between the topographic conditions for each plot
depending on the DEM used, so interpretation of the patterns is somewhat confused
by DEM quality, and any results must be analysed with caution.

Plot

Altitude
(m )

Mean
Curvature

Slope

Slope
Position

Solar
Radiation

TopMod

Toposcale

1

203

-0.95

2

219

-1.00

-0.31

-0.18

6.66

18

9592

11.76

-142.92

-0.03

-0.06

3.62

35

9629

11.40

3

210

-50.00

0.08

1.00

-0.10

4.50

39

9648

11.28

4

-81.46

201

0.99

0.11

0.09

2.49

16

9652

9.57

10.62

5

198

-0.41

-0.91

-0.09

3.49

0

9621

16.26

-75.85

6

220

-0.99

-0.12

0.05

1.06

96

9651

12.37

20.85

7

219

0.96

0.28

-0.11

12.90

58

9639

10.40

-23.85

8

262

0.99

0.13

0.42

11.92

100

9638

9.07

257.46

9

238

-0.27

0.96

0.02

2.38

85

9659

12.47

6.23

10

224

-0.48

-0.87

0.10

1.15

100

9646

11.45

54.92

Eastness Northness

214

Plot

Altitude
(m )

Mean
Curvature

Slope

Slope
Position

Solar
Radiation

TopMod

Toposcale

1

203

-0.99

2

219

-1.00

-0.13

-0.12

1.25

22

9652

12.75

-120

0.05

0.05

4.10

63

9624

9.07

3

210

19

-1.00

0.00

-0.09

0.46

46

9654

14.55

-185

4
5

201

-0.51

-0.86

-0.08

3.60

19

9620

9.20

-60

198

0.23

-0.97

-0.15

2.06

15

9641

17.47

-149

6

220

-0.95

-0.30

0.09

4.69

69

9608

8.94

158

7

219

0.96

0.26

0.04

2.64

38

9656

9.51

79

8

262

-0.39

-0.92

0.10

1.77

81

9642

9.92

119

Eastness Northness

9

238

-0.65

0.76

0.05

4.05

75

9639

9.08

59

10

224

-0.94

0.34

0.06

1.81

26

9650

9.89

39

Table 32 Environmental and topographic conditions at each plot in TBS using each DEM
(TOPO DEM top, SRTM DEM, below).

Similarly for Tambito, the various richness and diversity measures are correlated with
each environmental and topographic variable in the search for patterns in the data.
Each analysis is separated for each DEM (Table 33, and Table 34).

Pearson's
Correlation
Coefficient

Elevation

Eastness Northness Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel Toposcale

Fam ily Richness

0.57

0.05

0.40

0.47

0.00

0.60

0.26

-0.73

Genera Richness

0.80

0.13

0.59

0.47

0.25

0.68

0.29

-0.63

0.51

Species Richness

0.54

-0.15

0.33

0.40

-0.18

0.53

0.28

-0.52

0.36

Endemic Species

0.55

-0.05

0.19

0.33

0.14

0.44

-0.05

-0.57

0.33

Simpson's

0.49

0.04

0.48

0.21

0.18

0.60

0.22

-0.74

0.25

Shannon's

0.55

-0.07

0.42

0.32

0.01

0.61

0.24

-0.65

0.32

Margalef

0.55

-0.13

0.36

0.38

-0.15

0.56

0.28

-0.54

0.35

Menhinick

0.55
-0.51

-0.10
-0.05

0.38
-0.46

0.34
-0.18

-0.08
-0.28

0.60
-0.61

0.26
-0.13

-0.57
0.68

0.33
-0.25

Berger-Parker

p-value

Elevation

Eastness Northness Curvature

Slope

Slope
Position

Solar
Radiation

0.44

Topm odel Toposcale

Fam ily Richness

0.08

0.89

0.25

0.17

0.99

0.07

0.47

0.02

0.20

Genera Richness

0.01

0.72

0.07

0.18

0.49

0.03

0.41

0.05

0.13

Species Richness

0.11

0.68

0.34

0.26

0.61

0.12

0.43

0.13

0.30

Endem ic Species

0.10

0.88

0.59

0.35

0.71

0.21

0.89

0.09

0.35

Sim pson's

0.15

0.92

0.16

0.57

0.63

0.07

0.55

0.01

0.48

Shannon's

0.10

0.85

0.22

0.37

0.98

0.06

0.51

0.04

0.36

Margalef

0.10

0.72

0.31

0.28

0.69

0.09

0.44

0.11

0.31

Menhinick

0.10
0.13

0.78
0.88

0.27
0.18

0.34
0.62

0.82
0.43

0.07
0.06

0.47
0.72

0.09
0.03

0.35
0.49

Berger-Parker

215

Table 33 Pearson’s correlation coefficient and the two-tailed p-value between
richness and diversity in each plot and the environmental variables for TBS using the
TOPO DEM.

Pearson's
Correlation
Coefficient

Elevation

Eastness Northness Curvature

Slope

Slope
Position

Solar
Radiation

Topm odel Toposcale

Fam ily Richness

0.57

-0.60

0.32

0.54

0.10

0.52

-0.10

-0.69

Genera Richness

0.80

-0.34

0.44

0.66

0.12

0.77

0.04

-0.65

0.54

Species Richness

0.54

-0.74

0.39

0.50

-0.02

0.54

-0.03

-0.47

0.24

Endem ic Species

0.55

-0.55

0.38

0.39

-0.29

0.35

0.31

-0.37

0.16

Sim pson's

0.49

-0.44

0.55

0.56

-0.03

0.48

0.12

-0.65

0.41

Shannon's

0.55

-0.64

0.50

0.54

-0.08

0.53

0.10

-0.56

0.32

Margalef

0.55

-0.72

0.43

0.51

-0.06

0.53

0.03

-0.48

0.25

Menhinick

0.55
-0.51

-0.68
0.37

0.50
-0.56

0.50
-0.50

-0.13
0.21

0.50
-0.44

0.12
-0.32

-0.48
0.50

0.26
-0.34

Berger-Parker

p-value

Elevation

Eastness Northness Curvature

Slope

Slope
Position

Solar
Radiation

0.43

Topm odel Toposcale

Fam ily Richness

0.08

0.07

0.36

0.11

0.78

0.12

0.77

0.03

0.21

Genera Richness

0.01

0.33

0.20

0.04

0.74

0.01

0.91

0.04

0.10

Species Richness

0.11

0.01

0.27

0.14

0.96

0.11

0.92

0.17

0.50

Endem ic Species

0.10

0.10

0.27

0.26

0.42

0.32

0.38

0.29

0.66

Sim pson's

0.15

0.20

0.10

0.09

0.94

0.16

0.73

0.04

0.24

Shannon's

0.10

0.05

0.14

0.10

0.82

0.12

0.79

0.09

0.36

Margalef

0.10

0.02

0.21

0.13

0.87

0.12

0.94

0.16

0.49

Menhinick

0.10
0.13

0.03
0.30

0.14
0.09

0.14
0.14

0.73
0.55

0.15
0.20

0.73
0.37

0.16
0.14

0.47
0.33

Berger-Parker

Table 34 Pearson’s correlation coefficient and the two-tailed p-value between richness and
diversity in each plot and the environmental variables for TBS using the SRTM DEM.

A number of interesting correlations are found between richness and diversity and the
topographic variables. Elevation itself correlates with a high significance level (99%)
for genera richness, and also non-significant but general trends of an increase in
diversity with greater elevation for all other richness and diversity measures (p values
in all cases < 0.15) (Figure 73).

216

Relationship between elevation and
diversity in TBS

90
80
70
60
50
40
30
20
10
0
190

1.20
Simpson's Diversity
Index and BergerParker

Richness

Relationship between elevation and
genera richness in TBS

No. Genera
No. Species

210

230

250

1.00
0.80
Berger-Parker

0.40
0.20
0.00
190

270

Simpson's

0.60

210

230

250

270

Elevation (m)

Elevation (m)

Figure 73 Elevational patterns in richness and diversity in TBS.

In many cases, the pattern of diversity and richness variation with elevation is nonlinear, with the two lowest elevation plots (plot 5 and plot 4) having significantly
lower diversity (Simpson’s of 0.84 and 0.95 respectively) and high levels of
dominance (Berger-Parker of 0.38 and 0.21 respectively). Plots above the 210 m
elevation mark (the remaining plots except plot 1) on the whole have greater diversity
(Simpson’s diversity average of 0.99), though it is highly variable and not correlated
with elevation (Simpson’s diversity correlation R2 = 0.02). The elevational gradient
in TBS is so small (60 vertical metres) that climatic differences will be insignificant in
this case. It is more likely that this pattern occurs as a result of flooding frequency or
forest and soil age.

Plots below 210m are frequently flooded, interrupting the

successional process and favouring a few species well adapted to regular flooding.
The high levels of dominance found in plots 4 and 5 provide further evidence that this
is the case, with prevalence of Sorocea sp. in both plots (21 and 33 individuals
respectively). This species was only found in one other plot (plot 6, 220m), with just
2 individuals.

217

If flooding is an important factor in defining the diversity, it is also important for
examining the relationship between diversity and richness and the distance from the
River Tiputini (Figure 74), which is another important factor when examining
flooding frequency. Plots with low diversity are also close to the river, with the
exception of plot 1 with mid-levels of diversity and which is just 46m from the river.
However, this plot is some 10m above the normal river level, and so floods less
frequently. Eliminating the plots closest to the river (plots 1, 4, and 5), there is no
relationship between distance from river and diversity (Simpson’s Index correlation
R2 = 0.15). These patterns in richness and diversity with respect to elevation and
distance from river indicate that low diversity (specifically high dominance) is
expected in regions frequently flooded. However, there is no clear pattern (linear or
non-linear) between diversity and richness with elevation and distance from river for
plots that are not frequently flooded. This indicates that other factors are likely
responsible for shaping the variation in diversity observed in the plot data.

Relationship between distance from river
and diversity in TBS
Simpson's Diversity
Index

1.00
y = 3E-05x + 0.9468
R2 = 0.2159

0.95
0.90
0.85
0.80
0

500

1000

1500

2000

2500

Distance from river (m)

Figure 74 Relationship between diversity and distance from the River Tiputini in TBS

218

In general terms, the SRTM DEM provides the greatest number of correlations,
despite having a much larger cell size which is unlikely to capture meso-scale
variability.

In terms of variables, TopModel correlates best with diversity and richness (less
wetness, greater diversity), though in all cases this relationship is highly dependent on
the high wetness and low diversity found in plot 5. Exclusion of plot 5 from the
correlation renders the correlation insignificant for all richness and diversity measures
in both DEMs. After TopModel, the slope position variable correlates positively with
genera diversity (i.e. higher up the slope higher the richness), though this is likely due
to co-linearity with elevation (Pearson = 0.81, p = 0.004, n = 10), and likewise
probably an effect of flooding frequency in low slope positions. Indeed, with the
exclusion of plot 5 once more (identified as having low diversity due to frequent
flooding), the correlation falls apart for the TOPO DEM, but is maintained for the
SRTM DEM (Pearson and corresponding p-value unchanged) (Figure 75). Though
the relationship appears solid, there is actually very little variation in genera richness
in mid-slopes (slope position 20 – 70) where all plots have 34 – 37 genera present.
The only other correlating variable is mean curvature using the SRTM DEM, which
correlates positively with genera diversity (i.e. convex slopes have higher richness).
Once again, mean curvature correlates highly significantly with elevation (Pearson =
0.79, p = 0.006, n = 10), and so this is not a new relationship but one brought about by
the same pattern identified between elevation and slope position.

219

Relationship between mean curvature and
richness in TBS using SRTM DEM

60

60

50

50

Genera Richness

Genera Richness

Relationship between slope position and
richness in TBS using SRTM DEM

40
30

y = 0.3206x + 20.244
R2 = 0.5985

20
10

40
30
20

0
0

20

40

60

80

100

y = 72.27x + 35.309
R2 = 0.432

10
-0.20 -0.15 -0.10

Slope Position

-0.05

0
0.00

0.05

0.10

0.15

Slope Position

Figure 75 Relationship between slope position and mean curvature and genera richness in
TBS using the SRTM DEM

All the correlations identified between genera richness and elevation, mean curvature
and slope position appear to originate from the same pattern, but it is unclear where
the causal relationship lies. In any case, these correlations only exist for genera
richness, where in the mid-elevations, mid slope positions and mid-curvature levels
no clear pattern is evident (only in the extremes for plots 4 and 5 and plots 8 and 9).

4.3.2.1.3 Multi-variate analysis of quantitative environmental variables

Multiple stepwise regression is once again applied to the data in order to identify
multiple environmental interactions, and quantify the variability in diversity that is
explainable through the topographic and environmental variables used in this study.
For the TOPO DEM, topmodel and eastness combine to predict 78% of the variability
in Simpson’s diversity (D = -0.025 TopModTOPO –0.033 EastnessTOPO +1.26, R2 =
0.78, n = 10), but this relationship is highly dependent on the high wetness found in
plot 5, and indeed the relationship collapses if plot 5 is excluded from the regression
(R2 = 0.001, n = 9). When plot 5 is excluded from the regression, no significant

220

multiple regressions are found. It is a similar story when the variables from the
SRTM DEM are analysed, with TopModel, slope and eastness explaining some 85%
of

the

variability

(D

=

1.20

-1.679TopModelSRTM

–0.023SlopeSRTM

–

0.022EastnessSRTM, R2 = 0.85, n =10), but this relationship once again collapses with
the exclusion of plot 5 (R2 = 0.17, n = 9), and when the analysis is preformed once
more excluding plot 5, no significant multiple stepwise regressions are found.

When species richness is examined using multiple stepwise regression, no patterns are
found using the TOPO DEM, but the SRTM DEM does predict species richness,
accounting for 82% of the variability using eastness (55%) and elevation (a further
27%) (Species Richness = -48.35 - 19.32EastnessSRTM + 0.47ElevationSRTM, R2 =
0.82, n = 10). This relationship is not particularly dependent on any single plot
richness, and is shown in Figure 76.

Measured Species
Richness

Relationship between modelled species richness
and measured species richness in TBS using the
SRTM DEM
100
80
60
40
y=x
R = 0.8167

20

2

0
0

20

40

60

80

100

Modelled Species RIchness

Figure 76 Modelled species richness (based on multiple stepwise regression) plotted against
measured diversity in TBS using the SRTM DEM.

221

Under this model, highest species richness is found in west facing higher elevation
slopes. Higher elevation slopes may be important in generating diversity due to lower
frequency of flooding. Plots 4 and 5 (the two regularly flooded plots) showed high
levels of dominance and resultant low diversity, and this is attributed to flooding
earlier in this chapter. Furthermore some evidence exists that the distance from the
river (co-linear with elevation) is important in defining composition, and it is
hypothesized that this may in reality represent an edaphic gradient (section 4.3.1.2).
These two factors may be at play in generating the trend of higher diversity in high
elevation slopes.

The significance of eastness in generating diversity in TBS is attributable to a number
of factors. Firstly, the wind direction is predominantly easterly in TBS, leading to
believe that wind may be an important factor to consider, although the mechanism for
such a relationship is unclear. Exposure to wind generates a higher probability of tree
falls, though slope is also important in generating higher levels of tree falls and no
relationship is found at all between diversity and slope. Wind is also important in
terms of evaporation, and slopes exposed to the wind are likely to experience harsher
evaporation regimes than more sheltered slopes, and this has important implications
for water retention strategies, especially for high canopy trees. Finally, eastness is
indicative of the time of day of the most direct sunlight, with east facing slopes
receiving the strongest sun in the morning hours, and west facing slopes (in this case
where greater diversity is found) receiving the strongest sun in the afternoon hours.
Meterological station data from TBS indicates that solar radiation is in fact higher in
the afternoons due to less atmospheric attenuation, although the difference is fairly
minimal. It is unlikely that the solar radiation significance of eastness is controlling

222

this relationship, as the low slopes render aspect relatively insignificant in varying
solar radiation across the region. It is impossible to know which of these plausible
explanations is the direct cause of the higher diversity on west facing slopes, if any,
and this should be the subject of further investigation.

In conclusion, no clear correlation between environment and diversity is found using
Simpson’s Index (the most robust diversity measure) nor single variable correlations
with species richness outside of the observed low richness and low diversity in the
two regularly flooded plots (plots 4 and 5). However, multiple stepwise regression
does provide a strong explanation of variability in species richness based on eastness
and elevation using the SRTM DEM, with west facing high elevation sites having the
highest richness. Ideally this model should be validated using a split sample but this
was impossible with just 10 plots.

4.3.2.1.4 Analysis of categorical environmental variables

Having examined in detail the quantitative variables and identified few correlations, it
is unlikely that the landscape classifications provide many significant correlations, as
they are derivatives of the characteristics already examined. Nevertheless, statistical
analyses are made (Pearson correlation coefficients) to examine if the diversity and
richness is indeed distinct in different landscape classes based on the fuzzy
membership to the feature and network feature classifications. The landscape class of
each plot, and the respective fuzzy feature membership to the feature classification are
first presented (Table 35 and Table 36).

223

Feature

Netw ork
Feature

Pit
Channel
Pass
Ridge
Peak
Planar
Membership Mem bership Mem bership Mem bership Mem bership Mem bership

Plot

Topoclass

1

Valley Bottom

Ridge

Ridge

0.00

0.00

0.00

0.14

0.00

2

Toe Slope

Planar

Planar

0.00

0.00

0.00

0.00

0.00

1.00

3

Toe Slope

Planar

Channel

0.00

0.57

0.00

0.00

0.00

0.57

0.86

4

Slope

Ridge

Ridge

0.00

0.00

0.00

0.14

0.00

0.86

5

Toe Slope

Planar

Channel

0.00

0.00

0.00

0.00

0.00

1.00

6

Slope

Planar

Planar

0.00

0.00

0.00

0.00

0.00

1.00

7

Toe Slope

Channel

Channel

0.00

0.71

0.00

0.00

0.00

0.57

8

Ridge

Ridge

Ridge

0.00

0.00

0.00

1.00

0.00

0.00

9

Slope

Planar

Planar

0.00

0.00

0.00

0.00

0.00

1.00

10

Slope

Planar

Channel

0.00

0.00

0.00

0.14

0.14

0.71

Table 35 Landscape unit classifications for the plots in TBS using the TOPO DEM.

Feature

Netw ork
Feature

Pit
Channel
Pass
Ridge
Peak
Planar
Mem bership Mem bership Mem bership Mem bership Mem bership Mem bership

Plot

Topoclass

1

Toe Slope

Planar

Pit

0.00

0.86

0.00

0.00

0.00

0.14

2

Slope

Channel

Pass

0.00

0.14

0.00

0.00

0.00

0.86

3

Toe Slope

Planar

Channel

0.29

0.14

0.14

0.14

0.00

0.29

4

Valley Bottom

Planar

Peak

0.00

0.71

0.00

0.00

0.00

0.29

5

Valley Bottom

Channel

Channel

0.00

0.86

0.00

0.00

0.00

0.14

6

Slope

Ridge

Peak

0.00

0.00

0.00

0.14

0.00

0.86

7

Toe Slope

Planar

Ridge

0.00

0.00

0.00

0.14

0.00

0.86
0.00

8

Slope

Ridge

Ridge

0.00

0.00

0.00

1.00

0.00

9

Toe Slope

Planar

Ridge

0.00

0.00

0.00

0.14

0.00

0.86

10

Toe Slope

Ridge

Ridge

0.00

0.00

0.00

0.57

0.00

0.43

Table 36 Landscape unit classifications for the plots in TBS using the SRTM DEM.

When multi-scale fuzzy feature classifications (explained in Chapter 3) are used
(Table 37 and Table 38), there is no relationship between diversity and membership to
any landscape class.
Pearson's
Correlation
Coefficient

Pit
Channel
Pass
Ridge
Peak
Planar
Mem bership Mem bership Mem bership Mem bership Mem bership Mem bership

Fam ily Richness

N/A

-0.22

N/A

0.42

0.18

-0.33

Genera Richness

N/A

-0.05

N/A

0.47

-0.03

-0.43

Species Richness

N/A

-0.21

N/A

0.31

0.34

-0.26

Endem ic Species

N/A

-0.14

N/A

0.46

0.39

-0.44

Sim pson's

N/A

0.19

N/A

0.22

0.20

-0.36

Shannon's

N/A

0.00

N/A

0.30

0.30

-0.36

Margalef

N/A

-0.16

N/A

0.31

0.36

-0.29

Menhinick

N/A
N/A

-0.08
-0.28

N/A
N/A

0.30
-0.27

0.38
-0.23

-0.33
0.47

Berger-Parker

224

Table 37 Correlation coefficients between feature membership and richness and diversity for
TBS using the TOPO DEM.

Pearson's
Correlation
Coefficient

Pit
Channel
Pass
Ridge
Peak
Planar
Mem bership Mem bership Mem bership Mem bership Mem bership Mem bership

Fam ily Richness

0.00

-0.45

0.00

0.45

N/A

0.07

Genera Richness

0.01

-0.64

0.01

0.51

N/A

0.21

Species Richness

0.24

-0.54

0.24

0.45

N/A

0.07

Endem ic Species

0.11

-0.39

0.11

0.55

N/A

-0.13

Sim pson's

0.19

-0.62

0.19

0.36

N/A

0.26

Shannon's

0.25

-0.61

0.25

0.46

N/A

0.13

Margalef

0.26

-0.56

0.26

0.47

N/A

0.07

Menhinick

0.28
-0.26

-0.58
0.61

0.28
-0.26

0.48
-0.44

N/A
N/A

0.07
-0.15

Berger-Parker

Table 38 Correlation coefficients between feature membership and richness and diversity for
TBS using the SRTM DEM.

4.4

Conclusions

In terms of species composition and diversity, TBS was found to have greater richness
at the species level (average of 57 species per plot against 42 in Tambito), but similar
richness in families.

In TBS the study found 53 families compared with 52 in

Tambito. Some 28 of these families were common to both sites. Notable differences
in composition were in the low abundance of Melastomataceae and Rubiaceae in TBS
in comparison to high abundances in Tambito. TBS was found to have much higher
abundances in Moraceae and Mimosaceae. Fabaceae was the most abundant family
found in TBS, whilst just four individuals were recorded in Tambito.

The terrain characteristics found to correlate with the species composition in the two
sites were few, but clearly reflect the effect of very different processes in each site. In
Tambito, only elevation was found to control composition, but this in itself explains
225

only 36% of the variability according to the Jaccard coefficient.

Furthermore,

elevation fails to accurately reflect compositional similarity between sites of similar
elevations, though in all cases sites elevationally distant exhibited clear compositional
differences, so only significant differences in elevation (and the factors that it controls) affect
species composition. Clearly other processes that are not captured with the terrain

characteristics presented here are important in controlling composition at the microscale in Tambito.

In TBS, composition was found to be controlled in part by distance to the main river
channel (just 16% of variability), and a hypothesis is presented that isolates this
correlation to being caused by forest age, with distance to river acting as a surrogate
of time since a region was occupied by river channel. It is hypothesized that this
represents a gradient of soil development and forest succession (to a lesser extent),
though no soil data is available to validate this supposition. Similarly to the case with
elevation in Tambito, sites equally distant from the river tended to have high
compositional differences (indicating that other factors are also important), but sites
with large differences in their distance from river do have significantly different
composition in all cases.

Distance to river was not the only factor exhibiting

significant correlation with composition in TBS, toposcale and northness for the
SRTM DEM both show significant correlations independently of distance to river
(that is, they are not co-linear with distance to river), though the validity of these
correlations is questioned and no biological interpretation can be offered at this time.

In conclusion, at neither site is composition convincingly correlated with the terrain
characteristics, and the majority of variability remains to be unexplained. This may

226

be due to the DEMs poor representation of the small-scale topography, and this error in the
terrain characteristics calculated from them. In Tambito the DEM is likely a good
reflection of the real topography, but the 25m resolution may fail to capture terrain

variability at a relevant scale for composition, though this is unlikely. In TBS neither
of the DEMs sufficiently capture the topographic variability observed in the field, and
they are likely both inaccurate to some degree and, more importantly, are not at the
best scale for identifying compositional differences. Also of importance to note is the
plot size used in this study. 25m x 25m plots may be highly sensitive to successional
dynamics compared with larger plots, and though some terrain characteristics may

indicate the frequency of processes that interrupt succession (slope indicating likely
frequency of tree fall, for example), they are stochastic processes (in both space and
time (Salvador-Van Eysenrode et al., 2000)) that cannot be properly identified in
relation to terrain characteristics with so few plots. In order to get to grips with this

interplay of deterministic and stochastic processes, more plots would be required,
lowering the dependence of the result on each single data point.

In terms of patterns in diversity, clear correlations between terrain characteristics and
diversity are found in Tambito, but in TBS patterns appeared to be less clearly
defined. A clear peak in diversity is found for Tambito in mid-elevations, though this
only applied to diversity measures, and not richness measures.

Two potential

explanations of this are presented, including a mid-domain effect and an area-based
hypothesis.

Whatever the cause, 70% of variation in diversity is explained by

elevation through a polynomial regression, with a diversity peak at 1800m. This
macro-scale observation is complemented by a strong micro-scale relationship
between curvature and richness and diversity, with convex slopes having higher

227

species diversity. This may be due to a greater diversity in light environments
through lateral penetration of the canopy, but may also be due to greater exposure to
cloud inputs (water and nutrients), and indirectly to soil quality with greater leaching
leading to lower soil nutrient levels. The latter explanation has been observed by
other authors, showing monotonic decrease in diversity with soil quality (Hall and
Swaine, 1976; Huston, 1982; Oliveira and Mori, 1999). Without further investigation
it is impossible to ascertain exactly which of these is the dominant process in
explaining diversity, but a considerable amount of variability can be explained
through terrain characteristics.

In TBS fewer and weaker patterns are found, though analysis is complicated by the
strong co-linearity between many characteristics, and the poor quality of the DEM.
Low richness and diversity is found in the two sites most regularly flooded (both low
in elevation and close to the river channel), but no other patterns in diversity variation
are found. The low diversity in the regularly flooded sites is principally due to high
dominance of a single species (Sorocea sp.) found in these two plots, but which is rare
in other plots.

Though multi-variate analyses using multiple stepwise linear-

regressions manage to account for 82% of variability in species richness, no
biological interpretation can be provided, and further data collection and analysis is
required to confirm the validity of this. The lack of clear result in TBS may be due to
deficiencies in the quality of the terrain information, brought about by coarse
resolution DEMs, which fail to capture the all important micro-scale topographic
variability central to this thesis. Section 3.7 in Chapter 3 shows that the DEMs reflect
little of the on-the-plot conditions, and so there is little wonder that patterns have not
been found. Alternatively, there simply may not be clear terrain controls on richness

228

and diversity in a tropical lowland rain forest such as TBS (thus supporting nonequilibrium theories of species diversity maintenance), though further investigation
with better quality DEMs is required.

Despite using 12 different terrain characteristics in this analysis, just two variables
provide robust explanations of composition and diversity; elevation and mean
curvature.

Further investigation should focus on these relationships in order to

broaden the understanding of the specific biological and ecological processes that
these variables are capturing.

229

Chapter 5

5

5.1

Micro-scale spatial variation in forest structure

Introduction

In this chapter the spatial variation in forest structure is briefly explored, once again
using the environmental and topographic variables to identify correlations and discuss
the possible mechanisms behind such relationships. It is well documented that forest
height tends to diminish, the density of individuals increases and DBH reduces with
increasing elevation (Huston, 1994; Pendry and Proctor, 1996; Pendry and Proctor,
1997; Kappelle, 2004). The reasons behind tree stunting, for example, with greater
elevation are the subject of many studies, and many different hypotheses have been
suggested to explain this phenomenon. Overall, stuntedness is thought to be an
indicator of low primary productivity. This in turn may be the result of one or many
factors which include high humidity (Odum, 1970), nutrient limitation (Vitousek,
1984; Vitousek and Sanford, 1986), low temperature (Kitayama and Aiba, 2002), soil
acidity (Hafkenscheid, 2000), leaf wetness (Letts, 2003) and cloud cover (Grubb,
1977; Letts, 2003)

There is still significant debate as to the factor, or indeed the combination of factors
behind such a relationship. This chapter examines micro-scale variation in structure,
examining both elevational gradients (specifically for Tambito) and the role of
topography (at both sites) in generating structural variation. It is hoped that in
Tambito examination of the fine-scale variation may shed some light on the
mechanisms behind tree stunting along elevational gradients, whilst in lowland rain

230

forests the effects of micro-topography and flooding frequency on structure are
examined in the hope that this will also shed some light on the plant-significant
environmental gradients that exist in this environment.

In addition to using the environmental and topographic variables in the analysis,
diversity itself is also incorporated and the role of diversity on forest structure (and
vice-versa) is examined. The relationship between diversity and structure is likely to
be complex and bi-directional. Diversity of tree forms is one factor behind the
complex structures of trees in a forest, with different species having different
branching patterns, sizes, and shapes. Meanwhile, the structure of the forest also
filters energy inputs (light, water, temperature) changing the physical conditions from
the upper canopy to the forest floor, which in turn favours certain species adapted to a
specific conditions and creates a diverse sub-canopy light environment. Chapter 4
identified some topographic factors that, through control on energy inputs, explain
some of the variability in diversity and composition. Therefore this complex two-way
relationship between diversity and structure may be significant, and is included in the
analysis.

This chapter is structured similarly to Chapter 4, with the two study sites being treated
separately at first, and site comparisons made at the end of the chapter. Two aspects
of structure are examined at each study site, each of which is analysed in a similar
way to that of composition and diversity in Chapter 4.

Analogous with the

composition analysis, the similarity in the frequency distribution of tree forms at a
plot (height, DBH, DBH/Height) is plotted against environmental difference between
plots, and Mantel tests used to test for correlation.

231

Analogous to the diversity

analysis, the average values of structural parameters are compared with the
environmental and topographic variables at each plot. For easier interpretation, the
sections are presented starting with between-plot variability in average structural
values, and moving onto a more detailed examination of the distribution of structural
parameters in each plot. In both cases, correlations are made on a variable-byvariable basis at first, and then multi-variate analyses are applied to search for how
combinations of factors may explain the forest structure.

5.2

Reserva Tambito

In Tambito the plots cover a large elevational gradient of nearly 1000m, and based on
other studies in similar environments, should show clear patterns of structural change
along this gradient. Furthermore the rugged topography and almost constant wetness
of soil and trunks make treefall a common occurrence, with a high density of fallen
trunks visible throughout the twin catchments. This temporally stochastic process
drives much of the variability in forest structure (Kappelle, 2004), but the spatial
distribution of treefall occurrence may be tightly controlled by physical factors such
as slope and soil stability.

But it is important to reiterate that the plot scale

measurements represent only a snapshot of the highly dynamic system that results.

5.2.1 Average structure

5.2.1.1 Forest Structure in general

232

In general, Tambito has high stem densities (average 1773 stems Ha-1 for the 10
plots), with relatively low diameters at breast height (DBH) (average 11.3cm,
maximum 79.6cm) and few tall trees (average tree height 7.84m, maximum 30m).
The frequency distributions for the three major structural characteristics (height, DBH
and DBH/Height) are shown in Figure 77. Some 98.4% of individuals were less than
20m tall, with 62.6% of trees measuring 4-8m tall (itself possibly an artifact of only
measuring tree with DBH>5cm). The greatest frequency for DBH occurred between
7-9cm, with 96.2% of individuals having a DBH less than 30cm. It is interesting that
there are more stems at 7-9cm than at 5-7cm, possibly indicating that there are a
number of species which reach maturity at 7-9cm DBH and cease to continue to grow.
The DBH/Height ratio is a measure of the stem girth per unit height, and is shown to
be important when comparing plots and sites later in the chapter. However, it is
important to note that the majority (57.0%) have ratio values between 1-1.6, with just
9.1% of individuals having ratio values below 1 (indicating relatively low DBH
compared with height) and a longer tail towards higher ratio values with the
remaining 33.9% of individuals having ratio values greater than 1.6 (reaching a
maximum of 7.6). High ratio values represent individuals with a relatively high DBH
compared to height. This may indicate an investment in stem strength (to combat tree
fall) over height (to search for light). This kind of investment may be particularly
necessary in situations where steep slopes and exposure to wind and fog potentially
destabilize a tree leading to the potential for fall.

233

DBH distribution for all individuals in
Tambito

Tree height distribution for all individuals in
Tambito
350

160

300

140

250

Frequency

120

200

100
80

150

60

82

76

70

64

58

52

46

40

34

30

27

21

24

18

15

12

6

9

3

0

28

0
22

50

0

4

20

16

100

40

10

Frequency

180

DBH (cm)

Tree Height (m)

7.8

7.2

6.6

6

5.4

4.8

4.2

3

3.6

2.4

1.8

1.2

0

200
180
160
140
120
100
80
60
40
20
0
0.6

Frequency

DBH:Height Ratio distribution for all
individuals in Tambito

DBH:Height Ratio

Figure 77 Frequency histograms for the three major structural characteristics for all
individuals in Tambito

In terms of variability, the greatest variability between individuals occurred in DBH
(coefficient of variability 69.6%), followed by tree height (53.9%) and finally for the
DBH/Height variable (48.7%). At this stage no more discussion is provided on the
general patterns of structure. These figures are discussed further in Section 5.4 at the
end of the chapter in direct comparison with TBS.

5.2.1.2 Between-Plot Structural Variability
5.2.1.2.1 General discussion

234

Having identified significant range and variability in structural forms in Tambito, this
section addresses the between-plot variability of the average structural parameters,
using the environmental and topographic variables to explain this variation. The
structural conditions at each plot are shown in Table 39.

Plot

Elevation
(m)

Stem
Density
(stem s/ha)

DBaseH
(cm )

DBH (cm )

Height of
DBH/DBase
Basal
DBH/Height
Height (m ) 1st Branch
H
Area (m 2)
(m )

1

1651

1552

13.36

15.55

0.83

1.54

3.21

8.90

6.12

2

1684

1712

11.85

13.92

0.85

1.24

2.44

10.19

7.47

3

1449

1888

10.87

12.24

0.89

1.35

1.70

9.02

6.86

4

1966

2192

10.89

12.69

0.87

1.76

2.62

6.39

4.22

5

1299

1792

9.92

11.55

0.87

1.58

1.49

6.67

4.10

6

1749

1392

11.57

13.73

0.83

1.81

1.70

6.97

4.20

7

2253

2816

9.61

11.27

0.87

1.42

2.15

7.06

4.63

8

1856

1696

10.71

13.60

0.82

1.46

2.15

7.62

3.98

9

1950

1200

13.92

16.91

0.82

1.48

2.72

8.99

5.84

10

1600

1488

12.70

15.22

0.83

1.65

2.33

7.79

4.84

Table 39 Structural characteristics of the plots in Tambito

Plots 7 and 4 have the highest stem densities (2816 and 2192 stems Ha-1 respectively)
and also happen to be the two highest elevation plots. However, the lowest stem
density is found in plot 9 (less than half the density with 1200 stems Ha-1), which is
the third highest elevation plot (though this site has already been identified in Chapter
4 as an outlier in diversity patterns). Plot 9 also has the highest DBH (average
13.9cm), with Plot 7 having the lowest (average 9.6cm). Plot 2 has the greatest
average tree height (10.2m), with Plot 4 having the lowest (6.4m). The highest trees
were found in Plot 1 (Elaeagia sp. - 30m, and Protium sp. – 28m) and Plot 2
(Clusiaceae spp. – 30m). Plot 2 had the lowest value for DBH/Height (1.24) and plot
6 has the highest (1.81).

5.2.1.2.2 Single-variate analysis of quantitative environmental variables

235

Prior to examining in detail the correlations between structural parameters and all
environmental and topographic variables, the elevational gradient in structure is
examined first (Figure 78).

18.00

2.00

16.00

1.80

14.00

1.60
1.40

12.00

1.20

10.00

1.00

8.00

0.80

6.00
4.00
2.00
0.00
1000

0.60

DBaseH (cm)
DBH (cm)
Height (m)
DBH/Height Ratio

1200

1400

DBH:Height Ratio

Structure

Structure - Elevation relations for Tambito

0.40
0.20
1600 1800
2000
Elevation (masl)

2200

0.00
2400

Figure 78 Elevational gradient in structural parameters for Tambito

Despite numerous accounts in the literature of a clear change in structure across the
elevational gradient, this is clearly not the case in the data collected in Tambito. No
relationship is evident in any of the major structural parameters shown in Figure 78,
nor in any of the other variables measured. Though Plot 7 (the highest elevation plot,
2253m) has relatively low stature (7.1m) and the lowest DBH (9.6cm), Plot 5 at the
other end of the elevational gradient (1299m) has an even lower average stature
(6.7m) and comparably low DBH (9.9cm). If anything, there is evidence of the midelevations having the highest stature and greatest average DBH.

236

This goes against the findings of many studies in similar environments, but analysis of
the environmental and topographic variables may explain some of this variability that
is clearly not explained by elevation alone. Pearson correlation coefficients and their
respective p-values are calculated between all structural parameters and the 9
quantitative environmental and topographic variables (Table 40).

Pearson's Correlation
Coefficient

Elevation

Eastness

Northness Curvature

Slope

Slope
Position

Solar
Topm odel Toposcale
Radiation

Stem Density

0.46

-0.21

0.30

0.00

-0.35

0.25

0.38

-0.26

DBH

-0.05

-0.32

-0.08

0.19

0.03

0.02

-0.23

0.06

0.20

DBaseH

0.04

-0.26

-0.02

0.36

-0.08

0.17

-0.15

0.12

0.37

DBH/DBaseH

-0.16

-0.12

-0.10

-0.62

0.24

-0.39

-0.01

-0.29

-0.60

DBH/Height

0.02

0.44

-0.35

-0.10

-0.24

-0.03

0.28

0.82

-0.08

Basal Area

0.38

-0.62

0.47

0.46

-0.43

0.44

0.14

-0.15

0.47

Height

-0.19
-0.22

-0.52
-0.59

0.07
0.01

0.05
-0.18

0.41
0.47

-0.17
-0.31

-0.56
-0.56

-0.56
-0.60

0.03
-0.18

Height of 1st Branch

p-value

Elevation

Eastness

Northness Curvature

Slope

Slope
Position

0.05

Solar
Topm odel Toposcale
Radiation

Stem Density

0.18

0.56

0.41

0.99

0.33

0.48

0.28

0.46

0.89

DBH

0.90

0.37

0.82

0.61

0.93

0.96

0.53

0.86

0.58

DBaseH

0.92

0.46

0.95

0.31

0.83

0.65

0.69

0.73

0.30

DBH/DBaseH

0.67

0.75

0.77

0.05

0.50

0.26

0.99

0.42

0.07

DBH/Height

0.95

0.21

0.32

0.79

0.51

0.93

0.44

0.00

0.83

Basal Area

0.27

0.06

0.17

0.18

0.22

0.21

0.69

0.68

0.17

Height

0.61

0.12

0.85

0.89

0.24

0.65

0.09

0.09

0.94

Height of 1st Branch

0.54

0.07

0.98

0.62

0.17

0.38

0.09

0.07

0.61

Table 40 Pearson correlation coefficients and the respective p-values between structural
characteristics of the forest and the environmental and topographic variables in Tambito.

As can be seen, only one significant correlation exists. This is found between the
DBH/Height variable and topmodel (Figure 79), with a very high level of significance
(Pearson = 0.83, p = 0.004). The nature of this relationship is such that the greater the
surface soil moisture (as defined by TopModel) the higher the DBH relative to height
(in other words the greater the investment in trunk girth per unit height).

237

Relationship between topmodel and
structure in Tambito
DBH:Height Ratio

2.00
y = 0.2337x - 0.4002
R2 = 0.6723

1.80
1.60
1.40
1.20
1.00
7

7.5

8

8.5

9

9.5

TopModel Wetness Index

Figure 79 Significant correlation between DBH/Height and topmodel for Tambito

Topmodel (an indicator of soil moisture convergence) could be seen as a resource
permitting trees to uptake water and augment trunk diameter, but this explanation fails
to account for the relative large trunk diameter compared to height that the
relationship is showing and in any case, water is unlikely to be limiting in these
environments. Alternatively, surface soil moisture could be seen as an indicator of
soil stability, with wetter soils having less stability. Given lower levels of stability, it
is preferable for trees not to risk mortality from tree fall through excessive top heavy
branch and leaf growth in the search of light, but instead to invest in solid lower trunk
and root growth. However, soil moisture is just one mechanism that lowers stability
for trees, with slope also being important, along with wind speed (more specifically
gust velocity), wind direction and associated topographic exposure. Neither slope nor
exposure or aspect related variables such as mean curvature, toposcale, northness or
eastness show any relationship, somewhat weakening the argument that stability is an
important variable in determining the DBH/height value. Furthermore, topmodel is
calculated based on both upslope area and slope, and tends to be highest on low slopes
with large contributing upslope area and so is not a good indicator of stability.

238

Treefall is more probable on steeper slopes (Gale, 2000), and also more likely on
unstable soils, so to test if there is a relation between DBH/Height and stability an
index is produced based on upslope area multiplied by slope, with the least stability
found on steep slopes with a large contributing area (i.e. greater soil moisture). No
significant relation is found, linear (DBH/Height = 7 e-7 Stability Index + 1.45,
Pearson = 0.41, p = 0.24) nor non-linear. The lack of correlations with any other
topographic variables makes the interpretation of this relationship between topmodel
and DBH/Height difficult, and no explanation can be provided.

The lack of correlations between topography and tree structure is of interest in itself,
with the basic structural variables (DBH and tree height) not only failing to show any
elevational pattern, but also showing no relationship with the other 9 quantitative
terrain derivatives. This is surprising, and leads to the conclusion that structure is
principally controlled by stochastic processes, and little relationship exists between
environment and structure in Tambito, at least at the micro-scale as measured in the
plots. If stochastic processes are prominent at this scale in defining tree structure, the
distinct structures observed in the plots in Tambito may be reflecting different age
profiles since a disturbance drastically altered the structure.

The results presented here are in contrast to many prior studies showing clear
elevational gradients, but may simply be an artifact of the small plot sizes used here
capturing successional stage after specific recent tree falls rather than the average
regional structure.

Before this is discussed in depth, multi-variate analyses are

applied to look for combinations of factors that may explain structure.

239

5.2.1.2.3 Multi-variate analysis of quantitative environmental variables

Multiple linear regressions are applied for the 5 major structural variables (stem
density, DBH, height, DBH/height and basal area), using all 9 quantitative
environmental and topographic variables. No significant multiple linear regressions
were found for DBH, the DBH/Height value (on top of the already discussed singlevariate relationship with topmodel) nor basal area.

For stem density, combinations of two variables jointly explain 53% of variability
(Figure 80). These are elevation (explaining 21%) and toposcale (explaining a further
31%) (Stem Density = -1816 + 2.12Elevation - 4.72Toposcale, R2 = 0.53, p = 0.02).
Based on this equation the highest stem density is found in high elevations with
exposed convex slopes, though there is a significant amount of variation (47%) not
explained by these factors. Indeed, examination of the scatterplot of modelled against
measured stem density shows that the model poorly explains low stem densities
(Figure 80). However, this result agrees with patterns reported by Webb et al. (1999),
where exposed ridges in American Samoa were found to have low stature and low
DBH but high stem densities compared to slopes and valleys, with the authors
attributing this pattern to wind-cropping and higher levels of disturbance.

For tree height a combination of two variables jointly explain 58% of the variability
(Figure 80). These are topmodel (explaining 32%) and solar radiation (explaining the
remaining 26%) (Height = 24.17 - 1.02Topmod – 0.0009Solar, R2 = 0.58, p = 0.01).
This relationship is interesting in that it once again highlights surface soil moisture
(topmodel) as potentially important. Based on this relationship, the highest trees are

240

found in areas with dry surface soils and low solar radiation receipt. These are areas
where there is lower danger of tree fall due to soil instability, but where lower solar
radiation resources makes the investment of energy into reaching higher into the
canopy important. This way the trees in the community tend to grow higher as they
compete for greater light resource at the upper canopy. Equally this relationship
could be argued as questionable and possibly spurious, based on the argument that
forest stature may be expected to be lower in areas with lower resources (less soil
moisture and less light receipt). Once again, there is no way of showing (without the
collection of substantially more field data) whether this is a truly causal relationship
due to the process of stability and competition for light, or whether this is a spurious
result with no causal relation.

Modelled stem density against
measured stem density in Tambito
12

y = 0.5305x + 832.35
R2 = 0.5305

Measured Tree Height (m)

2500
(stems Ha -1)

Measured Stem Density

3000

Modelled tree height against
measured tree height in Tambito

2000
1500
1000
1000

y = 0.5772x + 3.3658
R2 = 0.5772

11
10
9
8
7
6
5

1500

2000

2500

3000

5

6

7

8

9

10

11

12

Modelled Tree Height (m)

Modelled Stem Density (stems Ha-1)

Figure 80 Modelled stem density against measured stem density in Tambito, based on the
multiple stepwise linear regression model using elevation and toposcale (right) and modelled
tree height against measured tree height in Tambito, based on the multiple stepwise linear
regression model using topmodel and solar radiation receipt (left).

5.2.1.2.4 Average structure – diversity relationships

241

The final analysis using the average of structural variables for the plots looks at the
relationship between diversity and structure. All structural variables are correlated
with all diversity and richness measures, calculating the Pearson correlation
coefficient and the respective p-value (Table 41).

Pearson's Correlation
Coefficient

Stem
Density

DBH

DBaseH

DBH/DBaseH DBH/Height Basal Area

Height

Height of
1st Branch

Fam ily Richness

-0.04

0.01

0.10

-0.23

-0.27

0.12

0.22

0.07

Genera Richness

0.40

-0.61

-0.47

-0.11

0.08

-0.26

-0.46

-0.57

Species Richness

0.26

-0.22

-0.12

-0.27

0.11

0.15

-0.25

-0.37

Endem ic Species

0.59

-0.46

-0.36

0.04

-0.08

0.05

-0.42

-0.48

Sim pson's

-0.48

0.54

0.65

-0.84

0.23

0.46

0.15

-0.09

Shannon's

-0.25

0.27

0.39

-0.69

0.18

0.34

0.03

-0.19

Margalef

-0.04

0.01

0.13

-0.50

0.17

0.20

-0.14

-0.33

Menhinick

-0.41

0.31

0.42

-0.72

0.24

0.24

0.01

-0.23

Berger-Parker

0.56

-0.57

-0.68

0.86

-0.25

-0.35

-0.16

0.09

Stem
Density

p-value

DBH

DBaseH

DBH/DBaseH DBH/Height Basal Area

Height

Height of
1st Branch

Fam ily Richness

0.91

0.98

0.78

0.52

0.44

0.97

0.55

0.85

Genera Richness

0.25

0.06

0.17

0.77

0.83

0.22

0.18

0.09

Species Richness

0.46

0.53

0.74

0.45

0.77

0.95

0.49

0.29

Endem ic Species

0.07

0.18

0.31

0.91

0.84

0.80

0.22

0.17

Sim pson's

0.16

0.11

0.04

0.00

0.52

0.38

0.67

0.80

Shannon's

0.48

0.46

0.27

0.03

0.62

0.60

0.93

0.59

Margalef

0.92

0.98

0.73

0.14

0.63

0.86

0.70

0.35

Menhinick

0.23

0.39

0.22

0.02

0.51

0.75

0.98

0.52

Berger-Parker

0.09

0.08

0.03

0.00

0.49

0.51

0.66

0.80

Table 41 Correlation coefficients and respective p-values for all correlations between
diversity and richness measures and structural variables in Tambito.

Some correlations are found between diversity indices and the diameter at the base of
trees (DbaseH), and also the DBH/DbaseH variable. With DbaseH, Simpson’s and
Berger-Parker correlate to the 95% significance level, where highest diversity is found
in plots with a high average DbaseH. In the case of DBH/DbaseH, Simpson’s,
Shannon’s, Menhinick and Berger Parker all correlate such that highest diversity
242

(ergo low dominance) are found in low index values, where the DBH is significantly
smaller than the DbaseH. It is possible that the relationships with DBH/DbaseH are a
result of some co-linearity with curvature, which has already been presented in
Chapter 4 as a key driver of diversity in Tambito. Curvature narrowly fails to
correlate significantly to the 95% confidence level with DBH/DbaseH (Table 40), but
this relationship may be behind the significant correlation presented here. In the case
of Simpson and Berger-Parker correlating with the average DbaseH, this appears to
have no co-linearity with any of the environmental variables (Table 40). Though not
significant at the 95% significance level, there is also a weak positive correlation
between DBH and Simpson’s (Pearson = 0.54, p = 0.11, n = 10) and negative
correlation with Berger-Parker (Pearson = -0.57, p = 0.08, n = 10). It seems that in
general, plots with greater tree girths tend to have lower dominance and higher
diversity, contrary to a pattern reported by Givnish (1999). However, the mechanism
behind such a relationship is not clear, but analysis of the frequency distribution of
tree girths might shed further light. This is analysed in section 5.2.2.1.3.

5.2.2 Structural distribution

5.2.2.1.1 General discussion

Having examined the environmental and topographic relationships with average
structural parameters for each plot, the frequency distribution of structural parameters
within each plot are examined here. Two plots may have the same average tree
height, for example, but the distribution of tree heights for all the individuals may
indeed be very different. This section examines the stratification of structure through

243

analysis of frequency histograms for the three principal structural variables (DBH,
height and DBH/height), and through correlating the similarity in the distribution of
structural forms between plot pairs with the environmental and topographic similarity
between the plot pairs themselves.

5.2.2.1.2 Single-variate analysis of quantitative environmental variables

The Kolmogorov-Smirnov (K-S) test is used on the structural data for DBH, tree
height and DBH/height for each plot pair to examine the difference in their
distributions. The K-S test uses the mean and standard deviation of the distribution to
indicate similarity between samples. The higher the K-S statistic, the more different
the distributions, so the matrix between plot pairs is in fact a dissimilarity matrix of
the structural variable (the method is discussed in greater detail in Section 3.6.4).
Mantel tests are then used on the dissimilarity matrix of structural distribution and the
dissimilarity matrix of environmental difference, in the same way as in Chapter 4. In
each case, the scatterplots are shown (Figure 81, Figure 82 and Figure 83), along with
a summary table indicating whether significant correlation was found (Table 42,
Table 43 and Table 44).

244

DBH Similarity and Altitude

DBH Similarity and Distance
1200

y = 2.1961x + 0.6575
R2 = 0.0216

2.50

Altitudinal Difference (m)

Horizontal Distance (km)

3.00

2.00
1.50
1.00
0.50
0.00
0.00

0.05

0.10

0.15

0.20

800
600
400
200
0
0.00

0.25

DBH Similarity and Eastness

Northness Difference

Eastness Difference

1.2
1
0.8
0.6
0.4
0.2
0.05

0.10

0.15

0.25

0.20

1.5
1
0.5
0
0.00

0.25

0.05

0.10

0.15

0.20

0.25

Kolm ogorov-Sm irnov DBH Sim ilarity

DBH Similarity and Curvature

DBH Similarity and Slope
40

3
y = 0.464x + 0.8742
R2 = 0.0006

2.5

Slope Difference (degrees)

Curvature Difference

0.20

y = 3.39x + 0.4421
R2 = 0.0213

2

Kolm ogorov-Sm irnov DBH Sim ilarity

2
1.5
1
0.5

0.05

0.10

0.15

0.20

35
y = 30.867x + 12.68
R2 = 0.0103

30
25
20
15
10
5
0
0.00

0.25

Kolm ogorov-Sm irnov DBH Sim ilarity

0.05

0.10

0.15

0.20

0.25

Kolm ogorov-Sm irnov DBH Sim ilarity

DBH Similarity and Slope Position

DBH Similarity and Solar Radiation

90

2500

y = 87.115x + 25.996
R2 = 0.0161

80
70
60
50
40
30
20
10
0.05

0.10

0.15

0.20

0.25

Kolm ogorov-Sm irnov DBH Sim ilarity

Solar Radiation Difference

Slope Position Difference

0.15

2.5

y = -0.5898x + 0.581
R2 = 0.0026

1.4

0
0.00

0.10

DBH Similarity and Northness

1.6

0
0.00

0.05

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

0
0.00

y = 1397.9x + 142.57
R2 = 0.0533

1000

y = 853.44x + 733.83
R2 = 0.0032

2000
1500
1000
500
0
0.00

0.05

0.10

0.15

0.20

Kolm ogorov-Smirnov DBH Sim ilarity

245

0.25

DBH Similarity and Toposcale

DBH Similarity and TopModel

350
Toposcale Difference

250
200
150
100
50
0
0.00

0.05

0.10

0.15

0.20

TopModel Difference

2.5

y = 138.13x + 94.622
R2 = 0.0041

300

0.25

y = -1.8147x + 0.9862
R2 = 0.0175

2
1.5
1
0.5
0
0.00

Kolmogorov-Smirnov DBH Sim ilarity

0.05

0.10

0.15

0.20

0.25

Kolm ogorov-Sm irnov DBH Sim ilarity

Figure 81 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the K-S test of the distributions) against environmental difference between plot-pairs in
Tambito.

Variable
Distance
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test
0.15
0.23
-0.05
0.15
0.03
0.10
0.13
0.06
0.06
-0.13

One-tailed p-value
0.250
0.104
0.421
0.202
0.448
0.185
0.171
0.344
0.354
0.240

Correlation (alpha = 0.05)
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant

Table 42 Summary results of structural DBH similarity analysis with distance and
environmental variables, using Mantel tests and a one-tailed Pearson. Note that structural
dissimilarity was compared with environmental dissimilarity (difference). The northness
variable produces very different results due to the heavily skewed distribution of plots in
strongly northward and strongly southward facing slopes.

246

Height Similarity and Distance

Height Similarity and Altitude
1200

y = -1.1184x + 1.1759
R2 = 0.0362

2.50

Altitudinal Difference (m)

Horizontal Distance (km)

3.00

2.00
1.50
1.00
0.50
0.00
0.00

0.10

0.20

0.30

0.40

y = -353.2x + 398.31
R2 = 0.022

1000
800
600
400
200
0
0.00

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Eastness Difference

Northness Difference

y = 0.2451x + 0.4532
R2 = 0.0029

1.4
1.2
1
0.8
0.6
0.4
0.2
0.10

0.20

0.30

0.40

1
0.5
0
0.00

0.50

0.10

0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Height Similarity and Slope
40

y = -0.9554x + 1.1325
R2 = 0.0177

2.5
2
1.5
1
0.5

0.10

0.20

0.30

0.40

Slope Difference (degrees)

3
Curvature Difference

0.50

1.5

Height Similarity and Curvature

y = 13.759x + 13.866
R2 = 0.0132

35
30
25
20
15
10
5
0
0.00

0.50

0.10

Kolm ogorov-Sm irnov Height Sim ilarity

0.20

0.30

60
50
40
30
20
10
0.10

0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Solar Radiation Difference

2500

y = -21.254x + 41.776
R2 = 0.0062

70

0.50

Height Similarity and Solar Radiation

90
80

0.40

Kolm ogorov-Sm irnov Height Sim ilarity

Height Similarity and Slope Position

Slope Position Difference

0.40

y = -0.1852x + 0.9233
R2 = 0.0004

2

Kolm ogorov-Sm irnov Height Sim ilarity

0
0.00

0.30

2.5

1.6

0
0.00

0.20

Height Similarity and Northness

Height Similarity and Eastness

0
0.00

0.10

Kolm ogorov-Sm irnov Height Sim ilarity

y = 1793.3x + 474.24
R2 = 0.0911

2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov Height Sim ilarity

247

0.50

Height Similarity and Toposcale

Height Similarity and TopM odel
2.5

350
Toposcale Difference

TopModel Difference

y = -182.21x + 150.38
R2 = 0.0458

300
250
200
150
100
50
0
0.00

0.10

0.20

0.30

0.40

0.50

2

y = 1.2126x + 0.4982
R2 = 0.0504

1.5
1
0.5
0
0.00

Kolm ogorov-Sm irnov Height Sim ilarity

0.10

0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Similarity

Figure 82 Scatterplots of similarity in distribution of tree height between plot-pairs
(calculated using the K-S test of the distributions) against environmental difference between
plot-pairs in Tambito.

Variable
Distance
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test
-0.22
-0.17
0.07
-0.12
-0.08
0.07
-0.18
0.21
-0.22
0.29

One-tailed p-value
0.162
0.200
0.285
0.228
0.382
0.253
0.089
0.103
0.100
0.071

Correlation (alpha = 0.05)
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant

Table 43 Summary results of tree height similarity analysis with distance and environmental
variables, using Mantel tests and a one-tailed Pearson. Note that structural dissimilarity was
compared with environmental dissimilarity (difference). The northness variable produces
very different results due to the heavily skewed distribution of plots in strongly northward and
strongly southward facing slopes.

248

DBH:Height Ratio Similarity and Distance

DBH:Height Ratio Similarity and Altitude
1200
Altitudinal Difference (m)

Horizontal Distance (km)

3.00
2.50
y = -2.1009x + 1.3456
R2 = 0.0941

2.00
1.50
1.00
0.50
0.00
0.00

0.10

0.20

0.30

y = -471.36x + 415.22
R2 = 0.0289

1000
800
600
400
200
0
0.00

0.40

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

Northness Difference

y = -0.1293x + 0.5286
R2 = 0.0006

1.4
Eastness Difference

0.30

0.40

2.5

1.6

1.2
1
0.8
0.6
0.4
0.2
0
0.00

0.10

0.20

0.30

y = 0.8732x + 0.7183
R2 = 0.0067

2
1.5
1
0.5
0
0.00

0.40

0.10

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

40
Slope Difference (degrees)

y = 0.2664x + 0.884
R2 = 0.001

2
1.5
1
0.5
0
0.00

0.10

0.20

0.30

30
25
20
15
10
5
0
0.00

0.40

0.10

2500
Solar Radiation Difference

y = -11.687x + 39.609
R2 = 0.0014

60
50
40
30
20
10
0.20

0.30

0.30

0.40

DBH:Height Ratio Similarity and Solar
Radiation

90

0.10

0.20

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

DBH:Height Ratio Similarity and Slope
Position

0
0.00

0.40

y = 20.087x + 12.878
R2 = 0.0208

35

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

80
70

0.30

DBH:Height Ratio Similarity and Slope

3
2.5

0.20

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

DBH:Height Ratio Similarity and Curvature

Curvature Difference

0.20

DBH:Height Ratio Similarity and Northness

DBH:Height Ratio Similarity and Eastness

Slope Position Difference

0.10

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

0.40

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

2000

y = 1613.6x + 537.24
R2 = 0.0543

1500
1000
500
0
0.00

0.10

0.20

0.30

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

249

0.40

DBH:Height Ratio Similarity and TopModel

DBH:Height Ratio Similarity and Toposcale
2.5

350

250

y = -81.254x + 128.18
R2 = 0.0067

TopModel Difference

Toposcale Difference

300

200
150
100

2

y = 3.3915x + 0.1015
R2 = 0.2905

1.5
1
0.5

50
0
0.00

0.10

0.20

0.30

0.40

0
0.00

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov DBH:Height Ratio Similarity

Kolm ogorov-Sm irnov DBH:Height Ratio Sim ilarity

Figure 83 Scatterplots of similarity in distribution of the DBH/Height variable between plotpairs (calculated using the K-S test of the distributions) against environmental difference
between plot-pairs in Tambito.

Variable
Distance
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test
-0.31
0.22
-0.03
0.11
0.01
0.16
-0.01
0.27
-0.09
0.53

One-tailed p-value
0.046
0.224
0.502
0.162
0.435
0.140
0.536
0.060
0.347
0.005

Correlation (alpha = 0.05)
Significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Significant

Table 44 Summary results of structural DBH/Height variable similarity analysis with distance
and environmental variables, using Mantel tests and a one-tailed Pearson. Note that structural
dissimilarity was compared with environmental dissimilarity (difference). The northness
variable produces very different results due to the heavily skewed distribution of plots in
strongly northward and strongly southward facing slopes.

No significant relationships were found between similarity in the distribution of DBH
or tree height and difference in any of the 9 environmental and topographic variables
between plot-pairs. This further confirms the findings in section 5.2.1.2.2, indicating
that no relationship exists between environment and average DBH or height nor in the
250

distribution of these structural forms. However, the significant correlation between
structural dissimilarity and environmental difference in TopModel for the
DBH/Height variable provides even greater evidence that the correlation already
found in section 5.2.1.2.2 (Figure 79) is worthy of further discussion. This indicates
that not only does the average value for DBH/Height correlate with topmodel
wetness, but also the distribution of DBH/Height for individuals within each plot
does. There are no plot-pairs that are similar in DBH/Height distribution with very
different topmodel wetness indices, though there is greater variation in topmodel
wetness index difference between plot-pairs with highly dissimilar DBH/Height
distributions. This scale dependence in the relationship significantly lowers the level
of correlation, though it is still significant to the 99% confidence level.

5.2.2.1.3 Structural distribution – species composition relationships

Finally for Tambito, the relationship between structural dissimilarity and
compositional similarity is examined for the three major structural variables (DBH,
Height and DBH/Height), also through K-S analyses and Mantel tests (Figure 84).

Scattergram of structural similarity (DBH) against
compositional similarity

Scattergram of structural similarity (Height) against
compositional similarity

0.2

y = -0.3702x + 0.1281
R2 = 0.1041

0.18

0.2

y = 0.0382x + 0.0721
R2 = 0.006

0.18

0.16
0.16
0.14
0.14
0.12
0.12
0.1

0.1

0.08

0.08

0.06

0.06

0.04

0.04

0.02

0.02

0

0
0

0.05

0.1

0.15

0.2

0

0.25

0.1

0.2

0.3

H e ight S t ruc tura l D is s im ila rit y

D B H S t ruc tura l D is s im ila rit y

251

0.4

0.5

Scattergram of structural similarity (DBH/Height)
against compositional similarity
0.2

y = 0.0331x + 0.0734
R2 = 0.0038

0.18
0.16
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

D B H / H e ight S truc t ura l D is s im ila rit y

Figure 84 Scatterplots for plot-pairs of compositional similarity against structural
dissimilarity for the major structural variables in Tambito.

There is a significant relationship between the similarity in DBH distribution and
composition at the 95% significance level (Mantel = -0.32, p = 0.05), whereby
compositionally similar plots are also similar in the distribution of DBH values,
though as can be seen in the scatterplot just 10% of the variability is explained by this
relationship. It is expected that individual species have specific structural forms, and
that compositionally similar plots would therefore have similar distributions in
structural forms, though clearly a great deal of variability is not explained in these
relationships. This is likely due to the successional stage in each plot, whereby
compositionally similar plots are likely to have significant variation in structural
characteristics depending on the successional stage, but also due to environmental
interactions causing further variability in the structural form of each species.

5.3

Tiputini Biodiversity Station

252

Surprisingly few patterns have been found between topographic characteristics and
structure in Tambito, and for TBS (LRF) it would be expected that structural
variability be much less pronounced. Furthermore, quality issues with the DEMs for
TBS add a further problem in attributing confidence to the results of these analyses.

5.3.1 Average structure

5.3.1.1 Forest Structure in general

TBS has lower stem densities (average 1515 stems Ha-1 for the 10 plots), with average
DBH of 12.8cm (maximum 92.3cm) and taller trees (average 10.4m, maximum 38m).
The frequency distributions for the three major structural characteristics (height, DBH
and DBH/Height) are shown in Figure 85. 92.8% of individuals are less than 20m
tall, with 49.5% of individuals 4-8m tall. The greatest frequency of DBH occurs
between 7-9cm (24.7% of all individuals, similar to Tambito in that there are more
stems in this class than the 5-7cm class), and 93.2% of individuals have a DBH less
than 30cm. 16.8% of individuals have a DBH/Height value below 1 (a greater
percentage than in Tambito), and 67.4% of individuals with DBH/Height between 11.6. The highest DBH/Height value is 9.4, though this is a clear outlier with just 6
individuals with DBH/Height greater than 5. This distribution of DBH/Height values
for TBS is significantly skewed towards lower values than found in Tambito (despite
a higher maximum in TBS), indicating greater investment of resources in height
relative to trunk diameter.

253

DBH distribution for all individuals in TBS

Tree height distribution for all individuals in TBS
250

180
200

140
120

Frequency

Frequency

160

100
80
60
40

150
100
50

20
0
10

84

92

68

76

52

60

36

Tree Height (m)

44

20

28

4

40

36

32

28

20

24

16

12

4

8

0

12

0

0

DBH (cm )

DBH/height distribution for all individuals in TBS
250

Frequency

200
150
100
50

8.
8
9.
6

8

7.
2

6.
4

4.
8

5.
6

4

3.
2

2.
4

1.
6

0
0.
8

0

DBH/Height

Figure 85 Frequency histograms for the three major structural characteristics for all

individuals in TBS

In terms of variability, the greatest variability between individuals occurred in DBH
(coefficient of variability 82.1%), followed by DBH/Height (57.2%) and finally tree
height (56.8%).

Further discussion of these patterns in direct comparison with

Tambito is presented in Section 5.4.

5.3.1.2 Between-Plot Structural Variability
5.3.1.2.1 General discussion

254

This section addresses the between-plot variability of the average structural
parameters, using the topographic characteristics to explain this variation.

The

structural conditions at each plot are shown in Table 45.

Plot

Elevation
(m)

Stem
Density
(stems/ha)

DBH (cm)

DBaseH
(cm)

1

199

1280

15.66

17.09

0.90

1.38

2

219

2112

11.39

13.14

0.91

1.54

3

210

1520

11.31

13.30

0.87

4

201

1616

10.46

12.12

5

211

1408

13.28

6

220

1440

7

219

8

262

Height (m)

Height of
1st Branch
(m)

3.26

10.86

7.48

2.25

7.59

5.26

1.28

1.82

9.59

6.70

0.86

1.22

1.74

8.28

5.48

14.89

0.87

1.35

3.10

9.27

5.88

14.57

16.06

0.90

1.49

2.57

10.03

7.30

1056

11.93

13.67

0.86

1.29

1.35

9.53

6.27

1648

12.36

14.66

0.83

0.95

2.62

12.95

8.98

Basal Area
DBH/DBase
DBH/Height
2
H
(m )

9

238

1520

13.21

15.62

0.84

1.00

3.12

12.51

8.19

10

224

1552

14.44

16.51

0.86

1.05

3.22

13.22

8.68

Table 45 Structural characteristics of the plots in TBS

Plot 2 has a significantly higher stem density (2112 stems Ha-1) than the rest of the
plots (average 1449 stems Ha-1), with Plots 1 and 7 have the lowest stem density.
Plots 8, 9 and 10 have the greatest stature with over 12m average tree height in all
three plots, but these plots do not have the highest average tree girths, with Plot 1
having the highest average DBH (15.7cm) and highest basal area (3.26m2). As well
as the high stem density, Plot 2 is also distinct from the other plots in terms of average
tree height with an average of just 7.6m. There is significant variability in the average
DBH/Height between plots, with Plot 2 having an average value of 1.54, whilst Plot 8
has an average DBH/Height value of just 0.95. The tallest trees were found in Plot 9
(38m, Micropholis sp. (Sapotaceae)), Plot 8 (37.3m, Jacaranda copaia Aublet.) and
Plot 10 (34m, Ardisia sp. (Myrsinaceae)).

5.3.1.2.2 Single-variate analysis of quantitative environmental variables

255

As in Section 5.2.1.2.2 for Tambito, variation in average structural characteristics are
compared with the topographic characteristics through Pearson correlation
coefficients, in the case of TBS applied for the TOPO DEM (Table 46) and the SRTM
DEM (Table 47).

Pearson's Correlation
Coefficient

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Stem Density

0.21

-0.24

0.03

0.31

-0.37

0.02

0.16

-0.15

0.22

DBH

-0.01

-0.63

-0.48

-0.15

-0.18

0.26

-0.44

0.40

-0.15

DBaseH

0.15

-0.58

-0.39

-0.04

-0.16

0.38

-0.34

0.35

-0.03

DBH/DBaseH

-0.53

-0.72

-0.28

-0.58

-0.28

-0.34

-0.46

0.26

-0.61

DBH/Height

-0.64

-0.54

-0.20

-0.64

-0.21

-0.51

-0.40

0.34

-0.66

Basal Area

0.13

-0.60

-0.47

0.08

-0.38

0.17

-0.34

0.51

0.01

Height

0.66
0.71

0.05
0.01

-0.03
0.02

0.51
0.55

0.11
0.12

0.72
0.78

0.14
0.13

-0.12
-0.20

0.55
0.58

Height of 1st Branch

p-value

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Stem Density

0.57

0.51

0.94

0.38

0.29

0.95

0.66

0.69

0.54

DBH

0.99

0.05

0.16

0.68

0.61

0.47

0.20

0.25

0.69

DBaseH

0.67

0.08

0.26

0.92

0.65

0.27

0.33

0.32

0.94

DBH/DBaseH

0.12

0.02

0.43

0.08

0.44

0.34

0.18

0.47

0.06

DBH/Height

0.05

0.10

0.58

0.05

0.57

0.13

0.26

0.34

0.04

Basal Area

0.73

0.07

0.17

0.83

0.28

0.64

0.33

0.13

0.97

Height

0.04
0.02

0.89
0.99

0.93
0.96

0.13
0.10

0.77
0.74

0.02
0.01

0.71
0.72

0.73
0.58

0.10
0.08

Height of 1st Branch

Table 46 Pearson correlation coefficients and the respective p-values between average

structural characteristics for each plot and the topographic characteristics derived from the
TOPO DEM in TBS

Pearson's Correlation
Coefficient

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Stem Density

0.21

-0.56

-0.09

0.25

0.31

0.37

-0.47

-0.25

0.06

DBH

-0.01

-0.25

0.20

0.03

-0.10

-0.09

0.12

0.12

0.13

DBaseH

0.15

-0.27

0.27

0.13

-0.12

0.02

0.18

0.07

0.19

DBH/DBaseH

-0.53

-0.32

0.05

-0.14

0.22

-0.14

-0.35

0.06

-0.13

DBH/Height

-0.64

-0.09

-0.10

-0.28

0.26

-0.20

-0.38

0.18

-0.22

Basal Area

0.13

-0.39

0.07

-0.04

-0.08

-0.01

0.06

0.23

-0.02

Height

0.66
0.71

-0.14
-0.25

0.27
0.24

0.41
0.47

-0.27
-0.25

0.29
0.40

0.38
0.30

-0.16
-0.21

0.36
0.42

Height of 1st Branch

256

p-value

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar
Radiation

Topmodel

Toposcale

Stem Density

0.57

0.09

0.81

0.49

0.39

0.30

0.17

0.49

0.86

DBH

0.99

0.48

0.59

0.93

0.78

0.81

0.74

0.74

0.73

DBaseH

0.67

0.45

0.45

0.72

0.74

0.95

0.62

0.85

0.60

DBH/DBaseH

0.12

0.36

0.90

0.69

0.55

0.71

0.33

0.86

0.73

DBH/Height

0.05

0.80

0.78

0.43

0.48

0.58

0.27

0.61

0.55

Basal Area

0.73

0.27

0.85

0.92

0.82

0.97

0.87

0.53

0.95

Height

0.04
0.02

0.70
0.49

0.45
0.51

0.24
0.17

0.45
0.48

0.42
0.26

0.27
0.40

0.67
0.57

0.31
0.23

Height of 1st Branch

Table 47 Pearson correlation coefficients and the respective p-values between average

structural characteristics for each plot and the topographic characteristics derived from the
TOPO DEM in TBS.

The greatest number of correlations occur with elevation, with DBH/Height, tree
height and height of first branch all correlating significantly. It is important to note
that elevation in the plots is measured by GPS and not derived from the DEMs, so
issues with DEM quality are not relevant in these correlations with elevation. The
correlations are such that greater tree height and height of first branch are found at
higher elevations, with the DBH/Height lower in higher elevations (i.e. greater height
relative to DBH) (Figure 86).

1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
190

Relationship between tree height and
elevation in TBS
14.00

y = -0.0067x + 2.7246
R2 = 0.4083

12.00
Tree height (m)

DBH/Height

Relationship between DBH/Height and
elevation in TBS

10.00
8.00

y = 0.0674x - 4.3953
R2 = 0.44

6.00
4.00
2.00

210

230
Elevation (m)

250

0.00
190

270

210

230
Elevation (m)

250

270

Figure 86 Significant correlations for plot average DBH/Height and tree height with elevation
in TBS

257

Both the correlations shown in Figure 86 appear to consist of high tree height and low
DBH/Height in the three high-elevation plots (8, 9 and 10), and respectively low tree
height and high DBH/Height values for the remaining seven lower elevation plots.
The significant correlations are due to this general pattern, and indeed no significant
correlation is found between elevation and tree height or DBH/Height when only the
seven lower elevation plots are included (for example for tree height, Pearson = -0.11,
p = 0.81). As discussed in depth in Chapter 4, there is a great deal of co-linearity
between elevation, distance from river and slope position (also correlates significantly
for the TOPO DEM with tree height and height of first branch), and it appears that all
these correlations are in fact related. The correlation between tree height and distance
from river is also highly significant (Figure 87), though once again there is a clear
difference between the seven low elevation plots and the three high elevation plots.

1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00

Relationship between tree height and
distance from river in TBS
16.00

y = -0.0002x + 1.4011
R2 = 0.5231

14.00
Tree height (m)

DBH/height

Relationship between DBH/height and
distance from river in TBS

12.00
10.00
8.00
y = 0.0022x + 8.8479
R2 = 0.6164

6.00
4.00
2.00
0.00

0

500

1000
1500
2000
Distance from river (m)

2500

0

500

1000
1500
2000
Distance from river (m)

2500

Figure 87 Relationship between plot average DBH/Height and tree height and distance from
river in TBS

These relationships with tree height and DBH/height, whereby the three higher
elevation plots have higher average tree heights and a respective lower DBH/height
value could be for a number of reasons, many of which are similar to those discussed

258

in Chapter 4, Section 4.3.2.1.2. These include flooding frequency (higher frequency
at lower elevations), forest age (possibly greater forest age with greater distance from
river) and an edaphic gradient with greater distance from the river (associated with
forest age). The fact that there is greater average tree height in Plots 8, 9 and 10, and
that the tallest trees recorded in the plots are also found in these plots is an indication
that either there is less disturbance in these plots, permitting the survival of large
individuals, or that the forest is older (in the sense of river migration), that has
permitted that some trees reach very large sizes. However, the abrupt change in tree
height evident in the scatterplots above 220m elevation also indicates that flooding
may be a factor which limits tree height, with all plots that are at least occasionally
flooded containing lower forest stature. This is further supported by the low stature
found in the two regularly flooded plots (4 and 5). Indeed this may be evidence of the
distinct structural forms found between varzea forest (Plots 1-7) and terra firma forest
(Plots 8-10).

In addition to the elevational pattern (and the correlation with slope position/distance
from river which is closely related), there is a significant correlation between mean
curvature and toposcale and DBH/Height when the TOPO DEM is used, and eastness
and DBH/DbaseH also for the TOPO DEM, though neither of these correlations hold
for the SRTM DEM. However, the relationship between curvature and toposcale and
DBH/Height is heavily dependent on the high curvature/exposure (i.e. convex slope)
and the low DBH/Height found in Plot 8 (Figure 88), and so these cannot be
considered a robust relationship.

259

Relationship between mean curvature and
DBH/Height in TBS using TOPO DEM

Relationship between eastness and
DBH/DbaseH in TBS using TOPO DEM

2.00
DBH/Height

1.50

0.92
0.91
0.90
0.89
0.88
0.87
0.86
0.85
0.84
0.83
0.82

DBH/DbaseH

y = -0.76x + 1.27
2
R = 0.41

1.00
0.50
0.00
-0.4

-0.2

0

0.2

0.4

0.6

-1.5

Mean Curvature

-1

-0.5

y = -0.02x + 0.87
R2 = 0.51

0

0.5

1

1.5

Eastness

Figure 88 Significant relationships between mean curvature and DBH/Height (left) and
eastness and DBH/DbaseH (right) in TBS using the TOPO DEM.

The relationship between eastness and DBH/DbaseH is however more robust based
on eastness measured by the TOPO DEM (Figure 88), though the relationship does not
hold for eastness derived from the high-resolution plot DEMs (DBH/DbaseH = -0.02
Eastness + 0.87, Pearson = -0.33, p = 0.36). The significance of eastness in TBS is
related to exposure to wind from the south-east and east, meaning that strongly east
facing slopes (i.e. high eastness) are more exposed to winds. It may be that eastness
as measured at the ground level is not significant in indicating exposure to wind at the
canopy level, whilst the TOPO DEM captures canopy level exposure better. Under
these circumstances it may be that trees exposed to wind at the canopy level invest
greater in basal area (including buttresses) in order to have greater stability, a pattern
supported by the low DBH/DbaseH found in strongly east facing slopes.

5.3.1.2.3 Multi-variate analysis of quantitative environmental variables

260

Multiple stepwise linear regressions are applied to explain the variation in the major
structural characteristics (DBH, tree height, DBH/Height).

No significant multi-

variate correlations are found to explain the variability in any of these structural
variables for either of the DEMs on top of the already discussed single-variate
relationships.

5.3.1.2.4 Average structure – diversity relationships

Finally, average structural characteristics are compared with diversity measures for
each plot, in order to search for structural-diversity relations (Table 48).

Pearson's Correlation
Coefficient

Stem
Density

DBH

DBaseH

DBH/DBaseH DBH/Height

Basal Area

Height

Height of 1st
Branch

Family Richness

0.27

0.09

0.19

-0.19

-0.42

0.08

0.49

Genera Richness

0.20

0.03

0.16

-0.33

-0.49

0.05

0.55

0.63

Species Richness

0.63

-0.06

0.06

-0.08

-0.33

0.14

0.38

0.46

Endemic Species

0.40

0.10

0.21

-0.15

-0.43

0.23

0.53

0.57

Simpson's

0.09

0.07

0.14

-0.02

-0.24

-0.15

0.38

0.49

Shannon's

0.33

0.06

0.16

-0.07

-0.32

0.03

0.46

0.55

Margalef

0.54

0.00

0.11

-0.10

-0.36

0.14

0.44

0.52

Menhinick

0.39
0.05

0.08
-0.16

0.20
-0.24

-0.14
0.09

-0.40
0.31

0.14
0.08

0.52
-0.49

0.60
-0.59

Berger-Parker

Stem
Density

p-value

DBH

DBaseH

DBH/DBaseH DBH/Height

Basal Area

Height

0.59

Height of 1st
Branch

Family Richness

0.46

0.79

0.60

0.60

0.23

0.82

0.15

0.07

Genera Richness

0.58

0.94

0.65

0.36

0.15

0.89

0.10

0.05

Species Richness

0.05

0.88

0.87

0.82

0.35

0.71

0.27

0.18

Endemic Species

0.25

0.79

0.56

0.67

0.21

0.53

0.12

0.08

Simpson's

0.81

0.85

0.70

0.95

0.50

0.69

0.28

0.16

Shannon's

0.35

0.87

0.65

0.85

0.36

0.93

0.19

0.10

Margalef

0.11

0.99

0.75

0.77

0.31

0.70

0.20

0.12

Menhinick

0.27
0.90

0.84
0.67

0.59
0.50

0.71
0.81

0.25
0.39

0.70
0.82

0.12
0.15

0.07
0.07

Berger-Parker

Table 48 Correlation coefficients and respective p-values for all correlations between
diversity and richness measures and structural variables in TBS.

261

As can be seen, no significant relationships exist, with structure completely
independent of diversity for all variables, unlike the situation in Tambito, though there
are few theoretical reasons as to why such a relationship should occur. It is far more
likely that correlations exist between structural distribution and composition, as is
examined later in Section 5.3.2.1.3.

5.3.2 Distribution in structure

5.3.2.1.1 General discussion

Having examined the patterns of variability in average structural characteristics for
each plot, this section examines the distribution in structural forms for each major
structural variable (DBH, Height, and DBH/Height), comparing it with topographic
characteristics. The same method adopted for Tambito is used here for TBS.

5.3.2.1.2 Single-variate analysis of quantitative environmental variables

The single-variate analysis of structural distribution in the plots involves the
application of Mantel tests between similarity matrices for each structural parameter
(calculated using Kolmogorov-Smirnov tests between plot pairs for each structural
variable, as explained in Section 3.6.4) along with the difference in topographic
conditions for plot-pairs. This is performed for the TOPO DEM (Figure 89, Figure 90
and Figure 91), with associated tables summarizing the results (Table 49, Table 50 and

262

Table 51), and for the SRTM DEM (Figure 92, Figure 93 and Figure 94), also with

summary tables of the results (Table 52, Table 53 and Table 54).

Structural Sim ilarity and Elevation

Horizontal Distance
Between Plots (m)

3000
y = -318.08x + 1424.2
R2 = 0.0011

2500
2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

Elevational Difference (m)

Structural Similarity and Distance
70

y = -58.778x + 30.55
R2 = 0.048

60
50
40
30
20
10
0
0.00

0.10

Kolm ogorov-Sm irnov DBH Sim ilarity

Eastness Difference

Difference in distance
from river(m)

2000
y = 169.7x + 764.65
R2 = 0.0003

1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

2.0
y = 0.4456x + 0.8566
R2 = 0.0019

1.5
1.0
0.5
0.0
0.00

0.10

Curvature Difference

Northness Difference

1.5
1.0
0.5

0.10

0.20

0.30

0.40

1.0
0.8
y = -1.8849x + 0.6662
R2 = 0.1118

0.6
0.4
0.2
0.0
0.00

Structural Similarity and Slope

y = -13.397x + 6.7305
R2 = 0.0504

6.0
4.0
2.0
0.20

0.30

0.40

Kolm ogorov-Sm irnov DBH Sim ilarity

Slope Position Difference

12.0

0.10

0.20

0.30

0.40

Structural Similarity and Slope Position

14.0

8.0

0.10

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

10.0

0.40

1.2

y = 0.7089x + 0.6883
R2 = 0.0061

2.0

0.30

Structural Similarity and Curvature

Structural Similarity and Northness
2.5

0.20

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

Slope Difference

0.40

2.5

2500

0.0
0.00

0.30

Structural Similarity and Eastness

Structural Similarity and Distance from River

0.0
0.00

0.20

Kolm ogorov-Sm irnov DBH Sim ilarity

100.0
80.0
60.0
y = 160.94x + 16.589
R2 = 0.1116

40.0
20.0
0.0
0.00

0.10

0.20

0.30

Kolm ogorov-Sm irnov DBH Sim ilarity

263

0.40

70.0

Structural Similarity and Toposcale
500.0

y = 19.441x + 19.253
R2 = 0.0041

60.0
50.0

Toposcale Difference

Solar Radiation Difference

Structural Similarity and Solar Radiation

40.0
30.0
20.0
10.0
0.0
0.00

0.10

0.20

0.30

0.40

400.0
y = -651.78x + 249.85
R2 = 0.1057

300.0
200.0
100.0
0.0
0.00

0.10

0.20

0.30

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

Structural Similarity and TopModel

TopModel Difference

5.0
y = -0.6956x + 1.6703
R2 = 0.0016

4.0
3.0
2.0
1.0
0.0
0.00

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov DBH Sim ilarity

Figure 89 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental difference
between plot-pairs in TBS for derivatives of the Topo DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

-0.03
0.02
-0.22
0.04
0.08
-0.33
-0.22
0.33
0.06
-0.33
-0.04

0.490
0.383
0.185
0.338
0.347
0.030
0.156
0.012
0.357
0.040
0.454

Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Significant
Not-significant
Significant
Not-significant
Significant
Not-significant

Table 49 Summary results of structural DBH similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a one-tailed

264

0.40

Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

Structural Sim ilarity and Elevation
Elevational Difference (m)

Structural Similarity and Distance

Horizontal Distance
Between Plots (m)

3000
y = 1554.8x + 978.87
R2 = 0.0704

2500
2000
1500
1000
500

0
0.00

0.10

0.20

0.30

0.40

0.50

70
60 y = 45.864x + 8.9663
R2 = 0.0782
50
40
30
20
10
0
0.00

0.10

Kolm ogorov-Sm irnov Height Sim ilarity

2500

1000
500
0
0.00

0.10

0.20

0.30

0.40

0.50

1.5
1.0
0.5
0.0
0.00

0.10

Curvature Difference

Northness Difference

1.0
0.5

0.30

0.40

0.50

0.8
0.6
0.4
0.2
0.0
0.00

0.10

8.0
6.0
4.0
2.0
0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Slope Position Difference

Slope Difference

y = -6.1634x + 5.9893
R2 = 0.0286

0.10

0.30

0.40

0.50

Structural Similarity and Slope Position

14.0

0.0
0.00

0.20

Kolm ogorov-Sm irnov Height Sim ilarity

Structural Similarity and Slope

10.0

0.50

y = -0.4143x + 0.448
R2 = 0.0145

1.0

Kolm ogorov-Sm irnov Height Sim ilarity

12.0

0.40

1.2

1.5

0.20

0.30

Structural Similarity and Curvature

y = 0.6355x + 0.6497
R2 = 0.0131

0.10

0.20

Kolm ogorov-Sm irnov Height Sim ilarity

Structural Similarity and Northness

0.0
0.00

0.50

y = -1.0527x + 1.1975
R2 = 0.0291

2.0

Kolm ogorov-Sm irnov Height Sim ilarity

2.0

0.40

2.5
y = 3431.2x - 69.129
R2 = 0.339

1500

2.5

0.30

Structural Similarity and Eastness

Eastness Difference

Difference in distance
from river(m)

Structural Similarity and Distance from River

2000

0.20

Kolm ogorov-Sm irnov Height Sim ilarity

100.0
80.0

y = 143.13x + 8.1205
R2 = 0.2364

60.0
40.0
20.0
0.0
0.00

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov Hight Sim ilarity

265

0.50

Structural Similarity and Toposcale
500.0

70.0
y = -8.1585x + 24.629
R2 = 0.0019

60.0
50.0

Toposcale Difference

Solar Radiation Difference

Structural Similarity and Solar Radiation

40.0
30.0
20.0
10.0
0.0
0.00

0.10

0.20

0.30

0.40

400.0

y = 127.33x + 106.37
R2 = 0.0108

300.0
200.0
100.0

0.50

0.0
0.00

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov Height Sim ilarity

Kolm ogorov-Sm irnov Height Sim ilarity

Structural Similarity and TopModel

TopModel Difference

5.0
y = -2.1972x + 2.1039
R2 = 0.043

4.0
3.0
2.0
1.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Figure 90 Scatterplots of similarity in distribution of height between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental difference
between plot-pairs in TBS for derivatives of the Topo DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

0.27
0.58
0.28
-0.17
0.11
-0.12
-0.17
0.49
-0.04
0.10
-0.21

0.063
0.002
0.034
0.085
0.203
0.224
0.192
0.012
0.423
0.281
0.077

Not-Significant
Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Significant
Not-Significant
Not-Significant
Not-Significant

Table 50 Summary results of structural height similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a one-tailed

266

0.50

Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

Structural Sim ilarity and Elevation
Elevational Difference (m)

Structural Similarity and Distance

Horizontal Distance
Between Plots (m)

3000
y = 865.51x + 1147.4
R2 = 0.0319

2500
2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

0.50

70
60 y = 67.051x + 3.2656
R2 = 0.2444
50
40
30
20
10
0
0.00

0.60

0.10

Kolm ogorov-Sm irnov Ratio Sim ilarity

2500
y = 1917.4x + 300.86
R2 = 0.1547

1500
1000
500
0
0.00

0.20

0.40

1.0
0.5
0.0
0.00

0.10

0.60

0.30

1.2
Mean Curvature
difference

Northness difference

0.20

1.5
1.0
0.5
0.20

0.30

0.40

0.50

0.6
0.4
0.2
0.0
0.00

0.60

4.0
2.0
0.20

0.30

0.40

0.50

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

267

Slope position difference

6.0

0.10

0.40

0.60

Structural Similarity and Slope Position

8.0

0.0
0.00

0.20

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

y = -2.8213x + 5.1646
R2 = 0.0088

10.0

0.60

0.8

Structural Similarity and Slope

12.0

0.50

y = 0.1382x + 0.3083
R2 = 0.0024

1.0

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

14.0

0.40

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

Structural Similarity and Curvature

y = 0.0471x + 0.7974
R2 = 0.0001

0.10

0.60

1.5

2.5

0.0
0.00

0.50

y = 0.3087x + 0.8535
R2 = 0.0037

2.0

Structural Similarity and Northness

2.0

0.40

2.5

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

Slope difference
(degrees)

0.30

Structural Similarity and Eastness

Eastness difference

Difference in distance
from river (km)

Structural Similarity and Distance from River

2000

0.20

Kolm ogorov-Sm irnov Ratio Sim ilarity

100.0

y = 44.003x + 32.801
R2 = 0.0327

80.0
60.0
40.0
20.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Structural Similarity and Toposcale
500.0

70.0
y = -30.005x + 30.29
R2 = 0.038

60.0
50.0

Toposcale difference

Solar radiation difference

Structural Similarity and Solar Radiation

40.0
30.0
20.0
10.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

y = 425.37x + 29.057
R2 = 0.1763

400.0
300.0
200.0
100.0
0.0
0.00

0.60

0.10

0.20

0.30

0.40

0.50

Structural Similarity and TopModel

TopModel Difference

5.0

y = 0.7609x + 1.3558
R2 = 0.0075

4.0
3.0
2.0
1.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Figure 91 Scatterplots of similarity in distribution of the DBH/Height between plot-pairs
(calculated using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the Topo DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

0.18
0.39
0.49
0.06
0.01
0.05
-0.09
0.18
-0.19
0.42
0.09

0.161
0.029
0.010
0.280
0.426
0.373
0.439
0.102
0.134
0.014
0.300

Not-Significant
Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Significant
Not-Significant

Table 51 Summary results of structural DBH/Height similarity analysis with distance and
environmental variables derived from the Topo DEM, using Mantel tests and a one-tailed

268

Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

Structural Sim ilarity and Elevation

Horizontal Distance
Between Plots (m)

3000
y = -318.08x + 1424.2
R2 = 0.0011

2500
2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

Elevational Difference (m)

Structural Similarity and Distance
70
50
40
30
20
10
0
0.00

Kolm ogorov-Sm irnov DBH Sim ilarity

1000
500

0.10

0.20

0.30

Eastness Difference

Difference in distance
from river(m)

1500

0.40

1.5
1.0
0.5
0.0
0.00

0.20

0.30

0.40

Structural Similarity and Curvature
1.2
Curvature Difference

y = 2.4612x + 0.2524
R2 = 0.1256

1.0
0.5

0.10

0.10

Kolm ogorov-Sm irnov DBH Sim ilarity

Structural Similarity and Northness

Northness Difference

0.40

y = 0.3184x + 0.6032
R2 = 0.0009

2.0

Kolm ogorov-Sm irnov DBH Sim ilarity

0.0
0.00

0.30

2.5
y = 169.7x + 764.65
R2 = 0.0003

2000

1.5

0.20

Structural Similarity and Eastness

2500

2.0

0.10

Kolm ogorov-Sm irnov DBH Sim ilarity

Structural Similarity and Distance from River

0
0.00

y = -58.778x + 30.55
R2 = 0.048

60

0.20

0.30

0.40

y = -0.0959x + 0.4168
R2 = 0.0005

1.0
0.8
0.6
0.4
0.2
0.0
0.00

0.10

0.20

0.30

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

269

0.40

Structural Similarity and Slope

Structural Similarity and Slope Position
Slope Position Difference

Slope Difference

5.0
y = -1.6184x + 2.065
R2 = 0.0077

4.0
3.0
2.0
1.0
0.0
0.00

0.10

0.20

0.30

0.40

80.0
70.0
60.0
50.0
40.0
30.0
20.0
10.0
0.0
0.00

Kolm ogorov-Sm irnov DBH Sim ilarity

0.20

60.0

250.0
y = 46.906x + 11.068
R2 = 0.0402

50.0
40.0
30.0
20.0
10.0
0.10

0.20

0.30

0.30

0.40

y = 230.23x + 40.103
R2 = 0.0762

200.0
150.0
100.0
50.0
0.0
0.00

0.10

0.20

0.30

Kolm ogorov-Sm irnov DBH Sim ilarity

Kolm ogorov-Sm irnov DBH Sim ilarity

Structural Similarity and TopModel

TopModel Difference

10.0
8.0

y = 11.177x + 1.1601
R2 = 0.0614

6.0
4.0
2.0
0.0
0.00

0.10

0.40

Structural Similarity and Toposcale

Toposcale Difference

Solar Radiation Difference

0.10

Kolm ogorov-Sm irnov DBH Sim ilarity

Structural Similarity and Solar Radiation

0.0
0.00

y = 36.969x + 24.856
R2 = 0.0127

0.20

0.30

0.40

Kolm ogorov-Sm irnov DBH Sim ilarity

Figure 92 Scatterplots of similarity in distribution of DBH between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental difference
between plot-pairs in TBS for derivatives of the SRTM DEM.

270

0.40

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

-0.03
0.02
-0.22
0.03
0.35
-0.02
-0.09
0.11
0.20
0.28
0.25

0.455
0.400
0.175
0.397
0.038
0.481
0.307
0.212
0.184
0.068
0.158

Not-Significant
Not-Significant
Not-Significant
Not-Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant

Table 52 Summary results of structural DBH similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a one-tailed
Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

271

Structural Sim ilarity and Elevation
Elevational Difference (m)

Structural Similarity and Distance

Horizontal Distance
Between Plots (m)

3000
2500

y = 1554.8x + 978.87
R2 = 0.0704

2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

0.50

70
60 y = 45.864x + 8.9663
R2 = 0.0782
50
40
30
20
10
0
0.00

0.10

Kolm ogorov-Sm irnov Height Sim ilarity

2500

1000
500
0
0.00

0.10

0.20

0.30

0.40

0.50

1.5
1.0
0.5
0.0
0.00

0.10

Curvature Difference

0.5

0.20

0.30

0.40

0.50

0.8
0.6
0.4
0.2
0.0
0.00

Slope Position Difference

y = -2.4512x + 2.4046
R2 = 0.0476

2.0
1.0

0.10

0.20

0.30

0.20

0.30

0.40

0.50

Structural Similarity and Slope Position

3.0

0.0
0.00

0.10

Kolm ogorov-Sm irnov Height Sim ilarity

Structural Similarity and Slope

4.0

0.50

y = -0.365x + 0.4922
R2 = 0.0199

1.0

Kolm ogorov-Sm irnov Height Sim ilarity

5.0

0.40

1.2

1.0

0.10

0.30

Structural Similarity and Curvature

y = 1.2472x + 0.3597
R2 = 0.0864

0.0
0.00

0.20

Kolm ogorov-Sm irnov Height Sim ilarity

Structural Similarity and Northness

Northness Difference

0.50

y = 0.2439x + 0.5963
R2 = 0.0014

2.0

Kolm ogorov-Sm irnov Height Sim ilarity

Slope Difference
(degrees)

0.40

2.5

y = 3431.2x - 69.129
2000
R2 = 0.339
1500

1.5

0.30

Structural Similarity and Eastness

Eastness Difference

Difference in distance to
river (km)

Structural Similarity and Distance from River

2.0

0.20

Kolm ogorov-Sm irnov Height Sim ilarity

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

80.0
70.0
60.0
50.0
40.0
30.0
20.0
10.0
0.0
0.00

y = 20.769x + 25.955
R2 = 0.0107

0.10

0.20

0.30

0.40

Kolm ogorov-Sm irnov Height Sim ilarity

272

0.50

Structural Similarity and Toposcale

60.0

250.0
y = -15.151x + 22.899
R2 = 0.0112

50.0

Toposcale Difference

Solar Radiation DIfference

Structural Similarity and Solar Radiation

40.0
30.0
20.0
10.0
0.0
0.00

0.10

0.20

0.30

0.40

200.0

y = 36.45x + 70.308
R2 = 0.0051

150.0
100.0

0.50

50.0
0.0
0.00

Kolm ogorov-Sm irnov Height Sim ilarity

0.10

0.20

0.30

0.40

Kolm ogorov-Smirnov Height Sim ilarity

Structural Similarity and TopModel

TopModel Difference

10.0
8.0

y = -0.247x + 3.1335
R2 = 8E-05

6.0
4.0
2.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

Kolm ogorov-Sm irnov Height Sim ilarity

Figure 93 Scatterplots of similarity in distribution of height between plot-pairs (calculated
using the Kolmogorov-Smirnov test of the distributions) against environmental difference
between plot-pairs in TBS for derivatives of the SRTM DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

0.27
0.58
0.28
0.04
0.29
-0.14
-0.22
0.10
-0.11
0.07
-0.01

0.053
0.004
0.037
0.428
0.040
0.156
0.025
0.178
0.258
0.266
0.498

Not-Significant
Significant
Significant
Not-Significant
Significant
Not-Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant

Table 53 Summary results of structural height similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a one-tailed

273

0.50

Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

Structural Sim ilarity and Elevation

Horizontal Distance
Between Plots (m)

3000
y = 865.51x + 1147.4
R2 = 0.0319

2500
2000
1500
1000
500
0
0.00

0.10

0.20

0.30

0.40

0.50

0.60

Elevational Difference (m)

Structural Similarity and Distance
70

60 y = 67.051x + 3.2656
R2 = 0.2444
50
40
30
20
10
0
0.00

0.10

Kolm ogorov-Sm irnov Ratio Sim ilarity

y = 1917.4x + 300.86
R2 = 0.1547

1500
1000
500

0.20

0.40

Eastness difference

Difference in distance
from river (km)

2500

0
0.00

1.0
0.5
0.0
0.00

0.60

0.20

0.30

1.2
Mean Curvature
difference

Northness difference

0.10

1.0
0.5

0.20

0.30

0.40

0.50

0.6
0.4
0.2

2.0
1.0

0.40

0.50

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

274

Slope position difference

3.0

0.30

0.40

0.60

Structural Similarity and Slope Position

4.0

0.20

0.20

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

y = 0.9104x + 1.5543
R2 = 0.0096

0.10

0.60

0.8

0.0
0.00

0.60

Structural Similarity and Slope

0.0
0.00

0.50

y = 0.0367x + 0.391
R2 = 0.0003

1.0

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

5.0

0.40

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

Structural Similarity and Curvature

y = -0.1199x + 0.7041
R2 = 0.0012

0.10

0.60

y = -1.1522x + 0.9538
R2 = 0.0448

1.5

2.0

0.0
0.00

0.50

2.0

Structural Similarity and Northness

1.5

0.40

2.5

Kolm ogorov - Sm irnoff DBH:Height Ratio
Sim ilarity

Slope difference
(degrees)

0.30

Structural Similarity and Eastness

Structural Similarity and Distance from River

2000

0.20

Kolm ogorov-Sm irnov Ratio Sim ilarity

80.0
70.0
60.0
50.0
40.0
30.0
20.0
10.0
0.0
0.00

y = 42.231x + 20.323
R2 = 0.0648

0.10

0.20

0.30

0.40

0.50

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Structural Similarity and Toposcale

y = 16.545x + 14.836
R2 = 0.0196

60.0

250.0
Toposcale difference

Solar radiation difference

Structural Similarity and Solar Radiation

50.0
40.0
30.0
20.0
10.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

200.0
y = 11.826x + 76.434
R2 = 0.0008

150.0
100.0

0.60

50.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

Structural Similarity and TopModel

TopModel Difference

10.0
8.0

y = -4.2245x + 4.1572
R2 = 0.0344

6.0
4.0
2.0
0.0
0.00

0.10

0.20

0.30

0.40

0.50

0.60

Kolm ogorov - Smirnoff DBH:Height Ratio Sim ilarity

Figure 94 Scatterplots of similarity in distribution of DBH/Height between plot-pairs
(calculated using the Kolmogorov-Smirnov test of the distributions) against environmental
difference between plot-pairs in TBS for derivatives of the SRTM DEM.

Variable
Distance
Distance from River
Altitude
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Toposcale
TopModel

0.60

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Kolm ogorov - Sm irnoff DBH:Height Ratio Sim ilarity

Mantel Test

One-tailed p-value

Correlation (alpha = 0.05)

0.18
0.39
0.49
-0.21
-0.03
0.02
0.10
0.25
0.14
0.03
-0.19

0.164
0.031
0.008
0.149
0.440
0.397
0.231
0.066
0.191
0.362
0.173

Not-Significant
Significant
Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant
Not-Significant

Table 54 Summary results of DBH/Height similarity analysis with distance and
environmental variables derived from the SRTM DEM, using Mantel tests and a one-tailed

275

Pearson. Note that structural dissimilarity was compared with environmental dissimilarity
(difference).

A number of correlations are found between structural frequency distribution
similarity and topographic similarity between plot pairs. Independent of the DEMs,
the structural frequency distributions of tree height and DBH/Height correlate
significantly with both distance from river and elevation (both variables are strongly
co-linear, and likely correlating as a result of the same causal process, Pearson = 0.40,
p = 0.01), whereby greater structural dissimilarity occurs in plots elevationally highly
different or highly different in their distances from river (both of which amount to the
same pattern due to co-correlation).

In terms of topographic characteristics derived from each DEM, the TOPO DEM
provides the greatest number of correlations with DBH and tree height similarity
correlating positively with slope position similarity (co-linear with elevational
similarity, Pearson = 0.35, p = 0.02), and DBH similarity inversely correlating with
toposcale and mean curvature similarity (with these two variables co-linear, Pearson =
0.38, p = 0.01). The latter inverse correlation is likely spurious or due to inverse colinearity as there is no reasonable explanation for significantly differently “curved”
sites having similar structural distributions. DBH/Height similarity also positively
correlates with toposcale similarity.

There are three significant correlations for the SRTM DEM, including positive
correlation between similarity in DBH and northness, and between tree height
similarity and northness similarity (with northness difference for the SRTM DEM

276

being significantly co-linear with distance from river difference, Pearson = 0.29, p =
0.05).

Finally an inverse correlation between tree height similarity and slope

similarity is found (and slope difference is weakly inversely correlated with northness
difference, Pearson = -0.27, p = 0.07). No significant correlations are found for the
DBH/Height variable using the SRTM DEM.

After unraveling many of the co-linearities in the data, only one correlation is worth
highlighting and merits further discussion. This is the correlations between tree
height and DBH/Height similarity and difference in distance from river / elevation
(which are strongly co-linear).

This pattern provides stronger evidence of the

variation in structural characteristics across the elevational gradient and with greater
distance from the main river channel. However, as shown in Section 5.2.1.2.2, this
correlation is not consistent when only low elevation plots are included in the
analysis. When a Mantel test is performed for tree height similarity and difference in
distance from river for Plots 1 – 7 only, the correlation is no longer significant
(Mantel = 0.07, p = 0.30, n = 21), similarly for the DBH/Height variable (Mantel =
0.16, p = 0.20, n = 21).

Once again, the conclusion is that there are two distinct forests in TBS in terms of
structural tree height and DBH/Height, with upland sites distant from the river (i.e.

terra firme forest) having significantly greater tree height and lower DBH/Height
values. The exact reason for this pattern cannot be identified with the data available.

5.3.2.1.3 Structural distribution – diversity relationships

277

For each of the three major structural variables (DBH, Height and DBH/Height), K-S
analyses and Mantel tests are applied between structural similarity and compositional
similarity in order to quantify the relationships between structure and species
composition (Figure 95).

Scattergram of structural similarity (DBH)
against compositional similarity for the TOPO
DEM

Scattergram of structural similarity (tree
height) against compositional similarity for the
TOPO DEM

5

0.3

Compositional Similarity
(Jaccard)

4

Compositional Similarity
(Jaccard)

y = -0.6956x + 1.6703
R2 = 0.0016

4.5
3.5
3
2.5
2
1.5
1

y = -0.2002x + 0.1219
R2 = 0.1289

0.25
0.2
0.15
0.1
0.05

0.5
0

0
0

0.1

0.2

0.3

0

0.4

DBH Structural dissimilarity

0.1

0.2

0.3

0.4

0.5

Tree height structural dissimilarity

Scattergram of structural similarity
(DBH/Height) against compositional similarity
for the TOPO DEM

Compositional Similarity
(Jaccard)

0.3

y = -0.0136x + 0.0751
R2 = 0.0009

0.25
0.2
0.15
0.1
0.05
0
0

0.1

0.2

0.3

0.4

0.5

0.6

DBH/height structural dissimilarity

Figure 95 Scatterplots for plot-pairs of compositional similarity against structural
dissimilarity for the major structural variables in TBS.

As can be seen, significant correlation is found between tree height similarity and
species composition (Mantel = -0.359, p = 0.02, n = 45), whereby structurally similar

278

plots also tend to be compositionally similar, though just 13% of variability is
explained through this relationship. The lack of correlation in DBH and DBH/Height
and the 87% of unexplained variability in the tree height similarity may be due to a
combination of two factors. First, compositionally identical plots with individuals of
different ages would be expected to have different structural characteristics.
Secondly, it is also likely that the poor correlations are the result of varying
environment, whereby tree structure for any species is strongly dependent on the
environment where it is found. In this sense compositionally identical plots may have
very different structural characteristics if found in different environments.

5.4

Conclusions

This chapter has described the structure in a tropical montane (cloud) forest (Tambito)
and in a lowland Amazonian forest (TBS), with marked differences between the sites.
Stem density in Tambito is higher (average 13.5% higher), with an associated lower
average diameter at breast height (13.3% higher in TBS).

Basal area is also

significantly greater in TBS (49% greater in TBS than in Tambito), possibly as a
result of greater density of individuals with buttress roots (reported by Richards,
1996).

However, conversely the DBH/Height in TBS is higher, indicating that

relative to height trees in Tambito invest greater in trunk growth, possibly as a
response to lower stability due to high slopes and unstable soils, and high wetness
increasing the chances of trunk rot. Average tree height is the most contrasting
between the sites, with TBS having an average tree height which is 36.5% greater
than in Tambito, with 7.2% of individuals being over 20m tall in TBS compared to
just 1.6% in Tambito. This difference in tree height between sites is well documented
in the literature (Whitmore, 1998), and can also be attributed to the lower stability of
279

trees in montane sites due to the steep slopes and high wetness, which make the
establishment of large individuals unlikely due to frequent tree fall.

When micro-scale variability is examined within each site, surprisingly few patterns
emerge in the structural data. In Tambito no elevational patterns are evident, despite
numerous reports in the literature of increasing stem densities and decreasing tree
stature with increasing elevation (Huston, 1994; Pendry and Proctor, 1996; Pendry
and Proctor, 1997; Kappelle, 2004). However, a strong significant correlation is
found between DBH/Height (average and the distribution of values) and topmodel (an
indicator of surface soil moisture, also an indicator of fluvial incision into the
landscape), whereby individuals make greater investment in trunk diameter relative to
their height on modelled more moist (incised) slopes. Whilst sites with potentially
high soil moisture may have low levels of soil stability indicating a higher risk of tree
fall, soil moisture is also highest on flatter slopes where water tends to accumulate,
rather than on steeper slopes. This renders topmodel a poor indicator of risk of tree
fall, and indeed when a model of tree fall risk is produced based on cumulative
upslope area and local slope, no correlations are found. The significant correlation
between DBH/Height and topmodel may be either spurious/non-causal, or indicate a
physiological response of trees to invest in trunk diameter in sites with higher soil
moisture.

No reports of this are available in the literature to provide further

discussion on the topic.

When multi-variate statistics are applied to explain the micro-scale variability in
structure in Tambito, interesting patterns are found with stem density and tree height.
Micro-scale variability in stem density is explained through a combination of

280

elevation and curvature/exposure (measured with the toposcale variable), whereby
stem density increase with elevation, and is higher on more exposed sites. This may
be explained by the potential stunting of trees on exposed ridges due to wind-cropping
and prolonged leaf wetness (supported by Letts, 2003) meaning that the site may
support a greater number of smaller tree individuals. This is a finding supported in
the literature by a similar study by Webb et al. (1999). Tree height is also found to be
highest in sites with low topmodel values (i.e. dry soils) and in areas of low solar
radiation, though the validity of this relationship is more questionable as few
ecologically interpretations are available and solar radiation alone does not capture
many important aspects of solar radiation receipt in Tambito (discussed in depth in
Section 3.4.1.3.1). More plot points would be required to conclude that this be a
causal relationship.

In TBS, a number of significant correlations are found between topographic
characteristics and structure (in terms of both average and distribution), though strong
co-linearity between variables means that it is likely that only one clear pattern
emerges in the data. Significant difference is found in tree height between the three
highest elevation plots and the lower-elevation plots, possibly different in terms of
flooding regime with plots below 220m elevation being subjected to occasional
flooding and having significantly lower average tree stature than the three higher
elevation plots (high elevation plots have 28% greater average tree height). However,
significant correlation is also found between distance from river and tree height,
meaning that the observed difference in tree heights between Plot 8-10 and Plots 1-7
may also be as a result of a long-term gradient in edaphic and successional conditions
brought about by river migration (a theory also presented and discussed in Section

281

4.3.2.1.2 with respect to an observed diversity gradient). It is important to note that a
gradient in tree height is not observed with greater distance from river or elevation
(i.e. the correlation does not stand for Plots 1-7 only), but that rather two unique
classes are observed between terra firme and varzea forest. No other correlations are
found to explain the remaining variability, though DEM quality issues in TBS also
limit the degree to which it can be concluded that environment has no role in
explaining variability in structure.

Finally, this chapter has briefly examined relationships between diversity and
structure, and composition and structure. In Tambito a significant correlation is found
between plot average DBaseH and DBH/DbaseH and diversity, with plots with higher
average DbaseH and lower DBH/DbaseH ratio values having the higher diversity.
The latter correlation is likely due to co-linearity with curvature and therefore
spurious (found in Chapter 4, Section 4.2.2.1.2 to be significant in explaining
diversity), and few ecological interpretations are available for the former correlation.
Compositional similarity is also shown to explain some variability in DBH structural
similarity between plots, though a remaining 90% of variability (the vast majority) is
unexplained. No such pattern was found for tree height or the DBH/Height variable.
In TBS even fewer relationships between structure and diversity were found, except
for composition explaining 13% of variability in tree height. In both sites, a large
degree of structural variability is not explained by composition, indicating that tree
age and stochastic processes such as tree fall and the subsequent successional
processes prevent many species from reaching full maturity.

However, the

DBH/Height variable is less sensitive to tree age (though DBH/Height is likely to
increase slightly over time as the tree reaches the canopy, but nevertheless the value is

282

far more stable with age than basic DBH or tree height), and the abject lack of
correlations between compositional similarity and DBH/Height similarity betweenplots also leads to believe that environment significantly affects the structural forms
of individuals of the same species. More in depth studies could be made to quantify
this ‘plasticity’ in structural form of species across environmental gradients.

In conclusion, a great deal of variability in structure is not accounted for using the
topographic variables adopted in this study. This signifies that either the variables
themselves are not capturing environmental conditions that are significant in driving
structural variability, or a large degree of micro-scale variability in structure is
actually controlled by processes such as tree fall, predominantly stochastic in both
space and time. These stochastic processes result in different age distributions of
individuals, and resultant differences in structural characteristics especially when
examined at the 25 m plot-scale. Furthermore, forest structure itself is a significant
factor in further partitioning landscape-controlled resources, resulting in a feedback
mechanism which changes alongside the dynamic in tree age distribution. Instead of
considering how landscape controls the partitioning of resources and thus the
structure of the forest (as studied here), it may be more important to consider how a
combination of landscape and chance disturbances control forest age distributions,
which alongside landscape controlled resources affect forest structure through a
feedback mechanism. Temporal studies of forest structural change across a landscape
are required to understand these complex interactions.

283

Chapter 6

6

6.1

Effect of spatial heterogeneity of environmental variables on the diversity

Research Strategy

Previous chapters have examined the conditions found at each plot, and examined the
relationship between topography and the structure and diversity of the forest. Some
significant relationships have been found, and some variability can be explained
through within-plot conditions.

This chapter endeavours to explain more of the

variability in composition and diversity through examining how the topographic
conditions surrounding each plot may shape the within-plot diversity and composition
that has been measured within-plot. The rationale for this analysis is that many

within-plot processes are affected by local effects through inherently spatial
(neighbourhood) ecological processes such as seed dispersal and competition, which
permit the movement and interaction of species across a landscape, maintaining
diversity and driving evolution. Ecological research has often cited “environmental
heterogeneity” as an important driver of diversity (starting with Ricklefs, 1977), under
the hypothesis that regions with abrupt environmental gradients contain a high
number of different species because of their greater availability of different niches,
which creates a rich local seed pool. Central to this theory is the notion that species
distributions are associated with their environment, so that a diverse surrounding
environment does indeed signify a diverse local pool of species. Recent studies have
questioned the degree to which habitat associations occur in tropical lowland forests
(discussed in Section 2.3.2) with 20% - 80% of species found to have habitat
association depending on the study and the site (Webb and Peart, 2000; Harms et al.,

284

2001; Phillips et al., 2003; St-Louis et al., 2004). In montane forests there are many
studies that show strong associations between elevation (representing an
environmental gradient in many climatic factors) and species composition (discussed
in Section 2.3.1), though none of these focus on micro-scale environmental
variability. The analysis presented in this thesis has shown strong evidence of habitat
association in species composition in Tambito (association between composition and
elevation), along with some degree of habitat association in TBS (association between
composition and northness, toposcale and elevation, though the validity of this has
been questioned).

Given this evidence, it is reasonable to expect that spatial

heterogeneity is an important factor in shaping the diversity in a montane forest, and
to a lesser extent in a lowland forest.

As already introduced in Chapter 2, the concept of environmental heterogeneity in
tropical forests may provide a degree of unification between non-equilibrium and
equilibrium based theories of species diversity maintenance. If in tropical forests the
random occurrence in both space and time of tree and branch fall means that
recruitment in gaps is dominated by chance occupants (Brokaw and Busing, 2000),
but that there is also some degree of niche-specialization in some tree species (for
example Phillips et al., 2003, also discussed in depth in Section 2.3.2), spatially
heterogeneous environments would potentially provide a diverse local seed pool for
gap recruitment because habitat preferences would mean that the pool of potential
immigrants were diverse, increasing the chances that a specialist to the conditions in
the gap may arrive through short-distance seed dispersal. This argument is centered
on the process of seed dispersal, though spatial heterogeneity is also of importance for
biotic interactions, specifically herbivory and competition Spatially heterogeneous

285

landscapes are considered to be favourable for diversity through the prevention of
competitive exclusion (discussed in detail by Huston (1994)) assuming species to
have some niche specialisation. The resultant increase in diversity of host species is
also argued to intensify herbivory, further preventing competitive exclusion and thus
promoting higher species diversity through what has been termed a positive feedback
(DeAngelis et al., 1986).

The significance of spatial heterogeneity on tropical tree diversity in the context of
this thesis is outlined in Figure 96, showing how the environmental conditions in
neighbourhood regions may be relevant to the species composition and diversity
within the plots. The central assumption of the model is that habitat association
occurs at least to some extent in tropical forests. In this example, Plot A is located
where the neighbouring regions are environmentally heterogeneous, therefore through
a degree of habitat association there is also a diverse surrounding local seed pool.
This increases the chances of immigration of many different species into the plot
through seed dispersal. For these reasons, Plot A is expected to be more diverse than
Plot B on the basis of heterogeneity alone. In the real system there will also be
interactions of the effect of the heterogeneity on other important properties such as
total resources (light would be less in plot A than plot B here).

286

Majority seed
dispersal distance

Heterogeneous
environment

Plot B
Homogenous
environment

Plot A
Majority seed
dispersal distance
Figure 96 Model role of environmental heterogeneity on within-plot diversity in tropical
forests through seed dispersal. Plot A, located in a heterogeneous environment is likely to
receive seeds from more local habitats than Plot B.

There are numerous examples of studies which examine the role of spatial
heterogeneity on ecological processes (Harner and Harper, 1976; Bell et al., 2000;
Barberis et al., 2002; Lundholm and Larson, 2003; Pausas et al., 2003). Few of these
actually quantify heterogeneity numerically (though there are notable exception, for
example Burnett et al., 1998), and none examine the role of spatial heterogeneity in
shaping diversity in tropical forests.

This chapter aims to quantify the spatial

heterogeneity for the terrain characteristics used in previous chapters, and examine
how heterogeneity relates to the tree diversity measured in the plots in Tambito and in
TBS. The approach here is to take each variable, quantify heterogeneity at multiple
scales and compare for each scale the relationship between topographic heterogeneity

287

and tree diversity. Multiple-scales are used because there is insufficient literature
quantifying the probability of seed dispersal against distance for the “average”
tropical species, and indeed, since much seed dispersal is through animal movements,
the patterns of dispersal are likely to be complex. There are numerous examples that
quantify the distance-function of seed dispersal for specific species (for example
Peres and Baider, 1997; Wenny, 1999), but insufficient information is available to
restrict this analysis to one single scale representing the average distance function for
the community as a whole.

As described in Chapter 3, 16 scales are used on 12 terrain derivatives. This means
that a total of 192 correlations are performed per site, and there are two sites with two
different DEMs used for one of them (TOPO and SRTM DEMs in TBS). This results
in a total of 576 correlations presented in this chapter, and for this reason not all
scattergrams are presented. Instead, graphs of the correlation coefficients across all
scales for each variable are presented, and summary tables provide a summary of the
most significant results for each variable. Graphs are plotted for each variable across
all scales, with a line (black) plotting the Pearson correlation, and a second line (blue)
which plots the respective p-value. For ease of interpretation, a solid blue line is also
shown representing the threshold at which point a correlation becomes significant (pvalue < 0.05). This 95% confidence limit means that one in twenty correlations
would be significant in a randomly generated set of numbers, so if we are to conclude
any significance in the relationship between heterogeneity and diversity in this case
the number of significant correlations per DEM should far exceed 9.6 (5% of the 192
correlations). As spatial heterogeneity for any one variable between different scales is
likely co-linear, significant correlations should also occur across multiple scales for

288

any concrete conclusion to be reached. A random control is used in the analysis to
further examine this problem of non-causal random relationships being found.

As in the previous chapters each site is treated separately, and in this case the results
are presented separately for each DEM in the case of TBS. The chapter concludes by
discussing in detail the results from both sites, attempting to extract generalities in the
findings and to link these to the biological processes that may be creating the patterns.

6.2

Reserva Tambito

The Tambito site is a tropical montane forest with dramatic environmental gradients.
The environmental heterogeneity is apparent in any photograph of the landscape, with
strong changes in elevation across short distances which closely control a number of
associated environmental variables). In the field, this is noticeable simply by walking
along the network of paths, with rapid changes in micro-climates depending on
exposure, aspect, and slope amongst other factors. Slopes reach 60 degrees in the
steepest parts of the catchments, meaning that in just 100m of horizontal distance the
elevation may change by 173m, signifying a change in temperature of 1.0oC. Slope is
a key variable, where steep slopes are likely to imply rapid change in many other
variables, including elevation, wetness (TopModel), and slope position. However,
steep slopes also imply higher levels of disturbance through tree fall (Gale, 2000), and
favour the establishment and dominance of species adapted to gaps with rapid growth
under high light levels. In this respect high heterogeneity in some variables could
also have a detrimental affect on diversity.

289

6.2.1.1 Spatial heterogeneity results
6.2.1.2 Cross-correlation matrix
Having run the heterogeneity model, the resultant grids of heterogeneity for each
variable are cross-correlated to examine their degree of co-linearity (Figure 97). As
all terrain derivatives originate from the same DEM, it is expected that heterogeneity
in many of these derivatives will be somewhat cross-correlated.

Elevation

Eastness

Northness

Curvature

Slope
Slope
Position

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar

Toposcale

Topmodel

Topoclass

Feature

Feature
(network)

-

-0.11

-0.21

0.57

0.04

0.28

0.55

0.52

-0.01

0.44

0.19

-0.36

-

-0.40

0.07

0.13

-0.16

0.67

0.07

-0.01

0.21

-0.06

0.18

-

0.06

0.17

-0.10

-0.38

0.01

0.17

0.06

-0.07

0.29

-

0.53

-0.09

0.52

0.51

0.23

0.68

0.33

0.30

-

-0.16

0.22

0.36

0.17

0.44

0.26

0.31

-

0.04

0.24

-0.37

-0.15

0.09

-0.45

-

0.34

-0.06

0.49

0.14

-0.05

-

0.18

0.65

0.23

0.03

-

0.19

0.06

0.29

-

0.06

0.26

-

-0.11

Solar

Toposcale

Topmodel

Topoclass

Feature
Feature
(network)

-

Figure 97 Cross-correlation matrix between all spatial heterogeneity variables with the
median distance set at 1 cell in Tambito (n = 69,678 in all cases).

Greatest cross correlation exists between topoclass and curvature (Pearson = 0.68),
eastness and solar radiation (Pearson = 0.67), and toposcale and topoclass (for
obvious reasons as topoclass is simply a reclassification of toposcale, Pearson = 0.65).
The heterogeneity in the feature classification is only slightly negatively correlated
with heterogeneity in the network feature classification (Pearson = -0.11). Also worth
mentioning is that heterogeneity in northness inversely correlates with heterogeneity

290

in eastness (Pearson = -0.40), not surprising as they are both directly derived from
aspect. This results from the complex relationship between eastness and northness
(when northness is plotted against eastness a perfect circle of radius 1 index unit is
formed). The inverse correlation originates with this relationship, where moderatelow heterogeneity is actually impossible in both variables simultaneously.
Heterogeneity in mean curvature is on average the most cross-correlated variable, and
the heterogeneity of the feature classification and topmodel are the most independent.

On the whole, cross-correlations of heterogeneity in the terrain derivatives correlate
slightly more than the cross-correlation of the derivatives themselves (average
Pearson = 0.24 for heterogeneity cross-correlations compared with average Pearson =
0.23 for the cross-correlations of the derivatives themselves). This is important, and
signifies that a specific site cannot be described as universally heterogeneous.
Heterogeneity in one variable does not necessarily signify heterogeneity in other
variables, and indeed in some cases may signify homogeneity.

6.2.1.3 Comparison with plot data

6.2.1.3.1 Control Experiment

Prior to performing detailed analyses of relationships between environmental and
topographic heterogeneity and plot diversity, it is necessary to make a control
experiment using pseudo-random ‘environmental’ data to understand to what extent
correlations may arise through chance rather than causal relationship. Twelve random
grids were produced using Arc/Info (RAND command), with values ranging from 0 -

291

1, and the heterogeneity model applied to these grids using all 16 scales. The Pearson
correlation coefficients between the plot diversity data and the resultant random grid
heterogeneity results for each scale are then calculated (Table 55), and an example
multi-scale correlation graph is shown for one of the random grids (Figure 98), in the
same way as the results of the terrain derivatives are to be presented.

Random Run

Maximum
Pearson

Minimum
Pearson

No. Scales with
significant correlation

Random grid 1
Random grid 2
Random grid 3
Random grid 4
Random grid 5
Random grid 6
Random grid 7
Random grid 8
Random grid 9
Random grid 10
Random grid 11
Random grid 12

0.29
0.26
0.25
0.25
0.25
0.30
0.44
0.25
0.25
0.29
0.25
0.54

0.10
0.08
-0.24
0.12
0.02
0.24
0.25
0.07
0.06
0.00
0.11
0.25

0
0
0
0
0
0
0
0
0
0
0
0

Table 55 Results of the heterogeneity analysis for the 12 random grids in Tambito.

Pearson correlation
coefficient

0.30
0.25
0.20
0.15
0.10
0.05
0.00
0

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
2
4
6
8
10
Median Distance (cells) in Cauchy Kernel

Two-Tailed p-value

Cross-Scale Heterogeneity - Random grid 1

Figure 98 Control experiment of cross-scale heterogeneity for Random Grid 1, a grid with
random values. Black line represents the Pearson correlation coefficient, whilst the blue line

292

represents the respective p-value. The solid blue line is placed at the 95% confidence level,
meaning that a correlation is considered statistically significant when the plot of p-values goes
under the thick blue line.

No significant correlations are found between heterogeneity in any of the random
grids and the diversity found in the plots, though steady movement of the points in the
analysis for each scale produce smooth cross-scale correlation curves (as evident in
Figure 98). This gradual movement of points between scales arises from the effect of

heterogeneity at one scale being somewhat correlated with heterogeneity at a similar
scale.

This produces smooth peaks in the correlation at specific scales, but it is

questionable the depth to which these curves can be used to ascertain if there is a
scale-dependent relationship; with only 10 points in the statistical analysis, there is
likely heavy reliance of each correlation on the position of one single point.

The co-correlation between heterogeneity at similar scales for each grid (average
Pearson correlation coefficient between adjoining scales is 0.99, n = 180) signifies
that the 192 correlations being calculated for the twelve variables are not independent
of each other, and thus less than the 5% of correlations would be expected to correlate
significantly as a result of chance. This is reflected in these results, with no spurious
correlations being found.

Whilst it would be preferable to repeat this analysis 100 times to gain a better insight
into the average number of expected spurious correlations in this analysis,
computational (amongst other) limitations make this impossible.

293

6.2.1.3.2 Single-variate comparison

Firstly, the relationship between each environmental and topographic variable is
examined separately, with multi-variate analyses being produced later in the chapter.
For each plot in Tambito, the heterogeneity for each variable is extracted, and
Simpson’s diversity index is then correlated with heterogeneity at each individual
scale. Graphs of correlation coefficient and p-value are plotted against the scale
(Figure 99), and a summary table provided indicating whether significant correlations
were found for each variable, and the scale (median distance) at which the most
significant correlation was found (Table 56).

294

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00

0.80
Pearson correlation
coefficient

0.70
0.60
0.50
0.40
0.30
0.20
0.10

alpha = 0.05

0.00
0

2
4
6
8
Median Distance (cells) in Cauchy Kernel

Two-Tailed p-value

Cross-Scale Heterogeneity - Elevation

10

Cross-Scale Heterogeneity - Eastness
1.00

0.10

0.80

0.00
-0.10

0

2

4

6

8

10 0.60

-0.20

0.40

-0.30
0.20

-0.40

Two-Tailed p-value

Pearson correlation
coefficient

0.20

alpha = 0.05

-0.50

0.00
Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Northness

-0.10

0

2

4

6

-0.20
-0.30
-0.40
-0.50
-0.60
-0.70

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Median Distance (cells) in Cauchy Kernel

295

Two-Tailed p-value

Pearson correlation
coefficient

0.00

Cross-Scale Heterogeneity - Curvature

-0.10

0

2

4

6

-0.20
-0.30
-0.40
-0.50
-0.60

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

0.00

Median Distance (cells) in Cauchy Kernel

0.00
-0.05 0
-0.10
-0.15
-0.20
-0.25
-0.30
-0.35
-0.40
-0.45
-0.50

2

4

6

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Slope

Median Distance (cells) in Cauchy Kernel

Pearson correlation
coefficient

0.60
0.40
0.20
0.00
-0.20

0

2

4

6

-0.40

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
8
10 0.20
alpha = 0.05 0.10
0.00

Median Distance (cells) in Cauchy Kernel

296

Two-Tailed p-value

Cross-Scale Heterogeneity - Slope Position
0.80

Cross-Scale Heterogeneity - Solar
0.30

1.00
0.80

0.20
0.15

0.60

0.10
0.05

0.40

0.00
-0.05 0

2

4

6

8

10 0.20

Two-Tailed p-value

Pearson correlation
coefficient

0.25

alpha = 0.05

-0.10
-0.15

0.00
Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - TopModel

Pearson correlation
coefficient

-0.05 0

2

4

6

-0.10
-0.15
-0.20
-0.25
-0.30
-0.35
-0.40

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

0.00

Median Distance (cells) in Cauchy Kernel

0.25
Pearson correlation
coefficient

0.20
0.15
0.10
0.05
0.00
-0.05 0

2

4

6

-0.10
-0.15
-0.20

1.00
0.90
0.80
0.70
0.60
0.50
0.40
8
10
0.30
0.20
alpha = 0.05 0.10
0.00

Median Distance (cells) in Cauchy Kernel

297

Two-Tailed p-value

Cross-Scale Heterogeneity - Toposcale

Cross-Scale Heterogeneity - Topoclass

Pearson correlation
coefficient

-0.10 0

2

4

6

8

1.00
10 0.90

-0.20

0.80
0.70

-0.30

0.60

-0.40

0.50

-0.50

0.40

-0.60

0.30
0.20

-0.70

Two-Tailed p-value

0.00

alpha = 0.05 0.10
0.00
Median Distance (cells) in Cauchy Kernel

-0.80

0.70
Pearson correlation
coefficient

0.60
0.50
0.40
0.30
0.20
0.10
0.00
0

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
2 Distance
4 (cells) in6Cauchy Kernel
8
10
Median

Two-Tailed p-value

Cross-Scale Heterogeneity - Feature

Cross-Scale Heterogeneity - Feature Network

Pearson correlation
coefficient

-0.10 0

2

4

6

-0.20
-0.30
-0.40
-0.50
-0.60
-0.70
-0.80

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

0.00

Median Distance (cells) in Cauchy Kernel

Figure 99 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in Tambito. The black line represents the
Pearson correlation coefficient, whilst the blue line indicates the respective p-value. For ease

298

of interpretation, the thick blue line shows the threshold for significant correlations at the
alpha = 0.05 level.

Variable
Elevation
Eastness
Northess
Curvature
Slope
Slope Position
Solar Radiation
Topmodel
Toposcale
Topoclass
Feature
Network Feature

Minimum Pearson
Correlation
Coeffcient
0.20
-0.44
-0.67
-0.52
-0.44
-0.16
-0.08
-0.38
-0.14
-0.72
0.21
-0.68

Maximum
Pearson
Correlation
Coeffcient
0.74
0.16
-0.18
-0.04
-0.02
0.75
0.25
-0.02
0.22
-0.03
0.64
-0.34

Minimum pvalue

Significance
(alpha=0.05)

Scale of lowest pvalue (median
distance in cells)

0.02
0.21
0.04
0.12
0.20
0.01
0.49
0.29
0.55
0.02
0.05
0.03

Significant
Not-significant
Significant (inverse)
Not-significant
Not-significant
Significant
Not-significant
Not-significant
Not-significant
Significant (inverse)
Significant
Significant (inverse)

9
1
0.01
0.01
0.8
2
0.8
1
9
0.01
3
0.01

Table 56 Summary table of results of multi-scale environmental heterogeneity modeling,
examining maximum correlations between species richness and heterogeneity for the ten plots
in Tambito.

Six variables correlate (at least at one scale) between heterogeneity and Simpson’s
diversity, three of which do so inversely (i.e. in these cases greater homogeneity =
greater diversity). Some 27 significant correlations are found when counting all
scales, the most of which are for heterogeneity in elevation (correlates at six of the 16
scales), slope position (correlates at six of the 16 scales), northness and network
feature (inversely correlate at five of the 16 scales). No variable correlated with
diversity independently of scale. These figures are above the expected number of
correlations at the 95% confidence limit (if 5% of the 192 correlations performed are
randomly expected, we would expect 9.6 correlations, furthermore in the random
control no spurious correlations were found, see Section 6.2.1.3.1).

299

Given that we find more than the expected number of random correlations, it suggests
that heterogeneity for some variables is indeed important in shaping diversity (or
some covariate of it, e.g. heterogeneity in altitude signifying high slopes, or that the
same processes which generate spatial heterogeneity also generate high species
diversity). However, it is important that we consider co-linearity in the discussion of
these correlations, as some of the correlations may indeed be spurious and correlate
only due to co-linearity with other causal variables. Whilst Table 56 shows crosscorrelations between variables for the whole of Tambito, it is important to re-analyse
the cross-correlations only for the variables found to correlate, using just the ten
points where the plots are located (Table 57).

Elevation
(median
distance 9)

Northness
(median
distance 0.01)

Slope
Position
(median
distance 2)

Topoclass
(median
distance 0.01)

Feature
(median
distance 3)

Network
Feature
(median
distance 0.01)

Elevation
(median
distance 9)

-

-0.60

0.71

-0.40

0.67

-0.47

Northness
(median
distance 0.01)

-0.60

-

-0.39

0.44

-0.75

0.48

Slope Position
(median
distance 2)

0.71

-0.39

-

-0.84

0.51

-0.48

Topoclass
(median
distance 0.01)

-0.40

0.44

-0.84

-

-0.46

0.57

Feature
(median
distance 3)

0.67

-0.75

0.51

-0.46

-

-0.29

Network
Feature
(median
distance 0.01)

-0.47

0.48

-0.48

0.57

-0.29

-

Table 57 Cross-correlation of heterogeneity values for the six correlating variables in
Tambito, using only the heterogeneity values for the ten plots (n = 10). Cells are shaded dark
grey when the correlation coefficient is above 0.80 and light grey when the correlation
coefficient is between 0.60 – 0.80 in order to highlight the most significant correlations.

300

As suspected, strong and significant correlations are evident between the variables,
with all presenting some degree of positive or negative co-linearity. Deciphering the
direct cause and effect from this mix of non-independent variables is difficult,
especially considering that some are positively correlated with heterogeneity and
others negatively. For example, if we assume that heterogeneity in elevation is indeed
the true direct driver of diversity, this renders the slope position correlation as indirect, along with a large proportion of the inverse correlation with northness and
feature classification. Only topoclass and the network feature classifications are left
accounting for some of the remaining unexplained variability in diversity. This is just
one scenario, and a number of different scenarios exist. In reality, heterogeneity in
slope position actually correlates best (Pearson = 0.75, p = 0.01), leading one to
believe that this is the direct driver and that many of the others are in fact spurious.
However, the correlation with heterogeneity in slope position is only marginally better
than elevational heterogeneity (Pearson = 0.74, p = 0.01), and indeed topoclass
(Pearson = -0.72, p = 0.02). Heterogeneity in northness and the feature classification
provide significant correlations at the 95% confidence level, but have the weakest
correlations of the six variables (Pearson = -0.67 and 0.64 respectively).

Though it is impossible to statistically separate out the direct and indirect linkages
behind the correlations without establishing more plots, some informal speculation in
the context of ecological theory can provide some insight as to the likely causal
factors.

Having already established that composition is strongly influenced by

elevation (and no other factors measured in this study), it is expected that spatial
heterogeneity in elevation would likewise be important in generating diversity
through overlapping distributions and seed sources.

301

Furthermore the ecological

literature has long attributed high spatial heterogeneity to generating high diversity,
therefore we would expect elevationally heterogeneous regions to have the highest
diversity. This is essentially found here (Figure 100), though plots 3 and 5 have lower
than expected diversity.

Also significant in this argument is the scale dependence of this relationship, with a
correlation only occurring with median distances above 3 cells (75 metres, which can
represent a change of as much as 100m of altitude or 0.6oC temperature in Tambito).
In other words, heterogeneity only becomes significant in shaping diversity when
broader scale patterns are examined, rather than the small-scale heterogeneity

Simpson's Species
Diversity (D)

represented in small-median distances.

Relationship between elevational
heterogeneity (median distance of 9) and
species diversity
0.98
y = 0.3084x + 0.6828
0.97
R2 = 0.5348
0.96
0.95
0.94
0.93
0.92
0.80

0.85

0.90

0.95

Elevational heterogeneity (D)

Figure 100 Elevational heterogeneity (median distance 9 cells) against species diversity for
Tambito.

This correlation between elevational heterogeneity and diversity once again provides
another possible explanation of the mid-elevation diversity peak found in Chapter 4,

302

as elevational heterogeneity also peaks in mid-elevations, in this case at 2100m
(Figure 101).

Average Heterogeneity

Altitudinal variation in elevational heterogeneity
(median distance 9 cells)

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1000

1500

2000

2500

3000

Elevation

Figure 101 Elevational gradient of heterogeneity in elevation in Tambito.

It is important to also consider that elevation is not the only likely causal relationship,
with 5 other variables providing significant relationships. Heterogeneity in slope
position itself is likely to be very similar to that of elevation, and is unlikely to be the
causal relationship (slope position alone has already shown little explanatory power in
terms of composition and diversity in Tambito).

Likewise, heterogeneity of the

feature classification is unlikely to be providing the causal relationship due to its clear
derivation from other factors and its abstraction of gradients into specific classes,
though this is an interesting and potentially useful surrogate. The inverse correlation
between diversity and heterogeneity in northness, topoclass and network feature
classification is a little more interesting though, and one which deserves some further
attention.

303

The correlation with heterogeneity in topoclass may simply be an artifact, as
toposcale itself provides no significant correlation, and topoclass is merely a
reclassification of toposcale.

Furthermore, significant edge effects occur around

topoclass boundaries (clearly more heterogeneity around the boundaries than in the
centre of a particular class), and so the result is highly dependent on the subjectivity
of the reclassified boundaries. The heterogeneity in the network feature classification
also has the problem of strong boundary effects, though these boundaries are less
generalized and less subjective in their definition. It is indeed surprising that the
heterogeneity in the network feature classification inversely correlates with the
network feature classification, but no plausible explanation for this can be provided.

However, the inverse correlation with heterogeneity of northness (though weak) must
be taken seriously, as the multiple stepwise linear-regressions performed on diversity
in Chapter 4 highlighted northness as accounting for some 22% of the variability in
diversity, with south facing slopes containing greater diversity (though northness was
not found to be significant in terms of composition). The correlation is borderline
significant (p = 0.04), and dependent on the low heterogeneity but high diversity
found in plots 6 and 10 (both strongly south facing slopes). This may be attributable
to the mid-elevation diversity peak already found and discussed in Chapter 4, as
heterogeneity in northness is actually lowest in mid-elevations due to the more planar
nature of these slopes (the five mid-elevation plots, 1600m – 1900m, have the five
lowest values of heterogeneity in northness). Given that no correlation is found
between composition and northness, it is likely that the correlation between diversity
and heterogeneity in northness is principally driven by the mid-elevation diversity
peak rather than the heterogeneity in northness.

304

Scatterplot of northness heterogeneity (median
distance = 0.01) against plot diversity
Simpson's Diversity Index)

0.99
0.98
0.97
0.96
0.95
0.94
0.93
0.92
0.00

y = -0.0567x + 0.9981
R2 = 0.4434
0.20

0.40

0.60

0.80

1.00

Northness heterogeneity

Figure 102 Scatterplot of northness heterogeneity against species diversity for Tambito using
a median distance of 0.01.

With no solid data or statistics telling us which variable is the true causal relationship,
if there is indeed only one, this section is heavily inconclusive. The significant
elevational heterogeneity correlation is however both statistically strong and
ecologically understandable given the previously outlined compositional relationship
with elevation.

To truly understand the causal linkages in the six significant

correlations that have been found, more plots are required in strategically located
sites, where there are lower levels of co-linearity between the individual heterogeneity
variables.

6.2.1.3.3 Multi-variate data analysis

305

Whilst some heterogeneity variables singularly correlate with diversity, it is likely that
a number and combination of factors are indeed important in shaping diversity, both
in terms of heterogeneity derivatives themselves and the original terrain derivatives
that have already been examined in Chapter 4. Furthermore, confusion over the
separation of causal relationships in non-independent variables may also be examined
using multi-variate methods.

Prior to performing multi-variate analyses on all variables, attention is paid to the
important distinction between slope position heterogeneity and elevational
heterogeneity. Multiple stepwise regression is applied to just two variables; slope
position (median distance 2) and elevation (median distance 8), in order to examine
the extra variability in diversity that is explained by elevation after slope position is
taken into account.

Elevational heterogeneity is found to explain just 8% extra

variability in diversity (despite explaining 53% in the single-variate correlation),
indicating that these two variables are to a large extent measuring the same variability.
It is important that this is quantified, indicating that slope position and elevation are
not only largely co-linear, but co-linear in their explanatory power of diversity.

Having clarified this important issue, three sets of multiple stepwise linear-regressions
are performed to fully understand the interactions between heterogeneity in all
variables and diversity:

1. Analysis using all heterogeneity variables at all scales
2. Analysis of all heterogeneity variables only at the scale where the greatest
correlation with diversity was found

306

3. Analysis of all heterogeneity variables at the scale where maximum
correlation was found alongside the original terrain derivatives already
analysed in Chapter 4.

In all three cases the heterogeneity values for the random grid are also included in the
analysis in order to check for chance relationships. Given the multi-variate nature of
this statistical procedure, there is danger of non-causal relationships by chance
providing some explanation of the diversity. Inclusion of any of the random grids in
the multiple linear regression models will clearly indicate that non-causal
relationships are being selected.

Taking the first set of variables (1), multiple stepwise linear regression is applied to
all heterogeneity variables at all scales. The resultant model extracts three significant
variables that are significant in explaining diversity. These are slope position (median
distance 2 cells) (explaining some 56% of variability), then elevation (median
distance 0.2 cells) (explaining a further 23% of variability, and cumulatively covering
78% of all variation), and finally topmodel (median distance 1 cell) (explaining a
further 13% of variability), with all three variables combined explaining some 92% of
variability (D = 0.570 + 0.479Slope Position(HET

- 2)

+ 0.141 Elevation(HET

- 0.2)

-

0.193Topmodel(HET - 1)).

Prior to performing detailed discussion of these results, the other two multiple
stepwise linear-regressions are analysed. The second model using heterogeneity data
only at the scale of maximum correlation for each variable pulls out two variables that
significantly contribute to explaining 72% of the variability in diversity.

307

Heterogeneity in slope position with median distance of 2 cells explains 56% of
variability and heterogeneity in northness at a median distance of 0.01 cells (explains
a further 17% of variability) (D = 0.669 + 0.332Slope Position(HET

- 2)

-

0.038Northness(HET – 0.01)).

Finally, the third model combines the environmental and topographic variables
(studied in detail in Chapter 4) with their respective heterogeneities in a multiplestepwise linear regression (total of 21 variables + 1 random variable). Four variables
explain 99% of the variability in this analysis, with heterogeneity in slope position
accounting for 56% of the variability in diversity, followed by solar radiation
accounting for a further 28% (84% cumulative), heterogeneity in slope position
accounting for a further 10% (94% cumulative) and heterogeneity in solar radiation
with median distance 0.8 accounting for the final 5% (D = 0.596 + 0.522Slope
Position(HET - 2) – 0.00002Solar + 0.00037Slope Position + 0.101Solar(HET - 0.8)).

Depending on the variables used in the multiple stepwise linear-regression different
results are achieved, making interpretation difficult. Firstly it is important to note that
despite their inclusion, none of the random heterogeneity variables provided any
explanation of the diversity. This is not to say that non-causal relationships have not
been selected in the three models presented above, but provides a little more
confidence in the results during their interpretation.

Looking at the first multiple regression (Figure 103), with all heterogeneity variables
at all scales included in the analysis, the additional variability that elevational
heterogeneity at short median distances provides is interesting. Despite elevational

308

heterogeneity at median distances of 0.2 cells explaining just 6% when correlated
directly with diversity, it explains some 23% of variability in the residual after the
multivariate regression with slope position heterogeneity. This may be due to slope
position being a more generalized variable, whilst elevation contains more detail. In
other words, diversity is generated through a multi-scale combination of small-scale
and broader-scale heterogeneity in slope position/elevation, if we assume that their
co-linearity is capturing the same variability in diversity.

However, multiple-

regression on all scales of elevational heterogeneity alone provides no evidence of
this, with only heterogeneity at a median distance of 8 cells providing a significant
correlation. No other scales explain any of the remaining variability. This draws us
to the conclusion that slope position is explaining some important variability
independently of elevation. This may be an artifact of the few points used in this
analysis (n = 10), and a greater number of plots may clarify these complex
interactions between non-independent variables.

Relation between modelled diversity and
measured diversity in Tambito
Measured Diversity (D)

0.99
0.98
0.97
0.96

y = 0.9163x + 0.0801
R2 = 0.9163

0.95
0.94
0.93
0.92
0.92

0.93

0.94
0.95
0.96
Modelled diversity (D)

0.97

0.98

Figure 103 First multiple linear-regression of all heterogeneity variables at all scales. Left
figure compares the modeled diversity with the measured diversity in the plots, and the right
figure applies the resultant regression model to the entirety of Tambito. Areas in white and

309

blue are beyond the range of the input data to the regression, hence are based on extrapolation
and should be considered potentially less trustworthy in their model fit.

In the second regression (Figure 104), the role of small-scale homogeneity (median
distance 0.01) of northness is once again observed, with it explaining some 17% extra
variability after the already discussed importance of slope position heterogeneity.
This indicates that the role of northness in these correlations is fairly independent of
that of heterogeneity of slope position/elevation (17% explained in the residual
compared to 22% explained singularly). As discussed, this relationship is somewhat
dependent on the high diversity, low heterogeneity found in the south-facing plots 6
and 10, and may simply be a result of the mid-elevation diversity peak.

Measured Diversity (D)

Relation between modelled diversity and
measured diversity in Tambito
0.99
0.98
0.97
0.96

y = 0.7229x + 0.2652
R2 = 0.7229

0.95
0.94
0.93
0.92

0.93

0.94
0.95
0.96
Modelled diversity (D)

0.97

0.98

Figure 104 Second multiple linear-regression of heterogeneity variables at only at the scale of
maximum correlation. Left figure compares the modeled diversity with the measured
diversity in the plots, and the right figure applies the resultant regression model to the entirety
of Tambito. Areas in white and blue are beyond the range of the input data to the regression,
hence are based on extrapolation and should be considered potentially less trustworthy in their
model fit.

310

Areas heterogeneous in northness are likely to be concentrated around areas with high
curvature (already identified as an important driver of diversity in Chapter 4), though
no relationship is present (Pearson = -0.09, p = 0.80, n = 10) meaning that the
importance of northness is independent of curvature. There is likely to be a clearer
relationship with plan curvature, but as mean curvature mixes both plan and profile
curvature the relationship is lost. These types of curvature are not studied in this
thesis, and therefore are not discussed further.

If the correlation between small-scale northness heterogeneity (median distances < 0.5
cells) is independent of other variables, and a truly causal relationship with diversity it
means that a locally diverse set of conditions in northness generates to some extent
low diversity.

In other words, relatively planar hillsides have greater diversity,

though this is more likely due to the aforementioned relationship between
heterogeneity in northness and the mid-elevation diversity peak.

Finally, the third multiple-regression analysis combines absolute values of
environmental and topographic variables and their respective spatial heterogeneities.
Interestingly, it was a combination of the absolute value for slope position and solar
radiation receipt and their respective heterogeneities that were pulled out in the
regression to explain an impressive 99% of variability in diversity (Pearson = 0.999, p
= 0.0001, n = 10) (Figure 105).

311

Relation between modelled diversity and
measured diversity in Tambito
Measured Diversity (D)

0.98
0.97

y = 0.9983x + 0.0016
R2 = 0.9983

0.96
0.95
0.94
0.93
0.92
0.92

0.93

0.94
0.95
0.96
Modelled diversity (D)

0.97

0.98

Figure 105 Third multiple linear regression combining both absolute value and
heterogeneities of each environmental and topographic variable. Left figure compares the
modeled diversity with the measured diversity in the plots, and the right figure applies the
resultant regression model to the entirety of Tambito. Areas in white and blue are beyond the
range of the input data to the regression, hence are based on extrapolation and should be
considered potentially less trustworthy in their model fit.

Interestingly neither slope position nor solar radiation receipt demonstrate any
explanatory power in Chapter 4, neither singularly nor in the multi-variate analyses.
However, when combined in multiple linear-regressions with their respective
heterogeneity values they are shown to be important. Once again slope position is by
far the most important variable, but some 28% of the residual is explained by
homogeneity in solar radiation receipt. Heterogeneity in solar radiation, despite being
strongly controlled by aspect, shows no clear relationship across the elevational
gradient (i.e. mid-elevations are not more homogenous in solar radiation, despite the
already discussed homogeneity in northness found in the mid-elevation planar slopes).
This goes against many ecological concepts about heterogeneity generating diversity,
and once again no interpretation of this pattern can be provided.

312

These analyses provide a confusing set of results, and it is almost certain that many of
these are not causal relationships but chance relationships achieved through multiple
regression of many variables with so few field plots. Four of the twelve variables
exhibit some type of control on diversity (slope position, elevation, northness, solar
radiation) on top of the three classification variables (Feature, Network Feature and
Topoclass) that show some correlation with diversity when treated on their own.
Interpretation is difficult and fairly arbitrary depending on the variables used in the
regression analysis. However, generalities are visually evident in the resultant maps
of each model presented in Figure 103, Figure 104 and Figure 105. The general
pattern of each of the three models is somewhat similar, with low diversity in the
valley bottoms, low-moderate diversity in peaks and high ridges, and high diversity in
mid-elevations, especially in north, north-east and south-west facing slopes. Under
these circumstances, it is difficult to explicitly quantify the degree to which spatial
heterogeneity is affecting the diversity in Tambito, with complex combinations of
variables all producing a similar mid-elevation diversity peak which could be entirely
attributed to a number of other more established ecological processes (the middomain effect or the productivity-diversity gradient, for example).

Perhaps as a conclusion to this section few concrete findings can be reported, except
that heterogeneity in slope position and elevation are significantly explaining some
50% of the variability in diversity, and that both these variables are co-linear and
explaining the same variability. Furthermore, highest heterogeneity in these variables
is found in mid-elevations, which have already been found to harbor the highest
diversity, potentially due to a number of different explanations outlined in Chapter 4.
If elevation is assumed to represent a productivity gradient in Tambito (a notion

313

supported by Letts, 2003), this shows that environmental heterogeneity in Tambito is
greatest in mid-productivity levels, a pattern also found in limestone pavements
(Lundholm and Larson, 2003). This coincidence of high spatial heterogeneity in midelevations and mid-productivity regions adds another factor which may be significant
in explaining mid-elevation diversity peaks.

Although complex combinations of variables manage to explain as much as 99% of
the variability in diversity, the confidence of these findings is questionable as there
are few clear ecological interpretations of the patterns and too few data points. It
would be preferable to have a larger number of plots to permit split sample validation
of these multiple stepwise linear regressions.

6.3

Tiputini Biodiversity Station

In TBS the extent to which spatial heterogeneity is significant in the maintenance of
species diversity is unknown, but theoretically it may be important. Firstly, the extent
of habitat association in LRF tree species is questionable and still under debate,
though the literature reports 20-80% of species have some kind of association (Webb
and Peart, 2000; Harms et al., 2001; Phillips et al., 2003; St-Louis et al., 2004).
Furthermore, gap dynamics play an important role in maintenance of forest diversity
in LRF, and non-equilibrium theories predict that gaps are recruited principally by
chance occupants through low densities in tropical tree species distributions and
dispersal limitations (Hubbell et al., 1999).

This has important implications for

spatial heterogeneity, as the probability of a specialist species recruiting in the
successional process after gap formation is enhanced if a diverse local seed pool is

314

available. Under these circumstances there is potentially a strong role for spatial
heterogeneity shaping species diversity at least to some extent in LRF. However,
practical limitations limit the degree to which this can be tested in TBS due to quality
issues in the DEMs. Section 3.7 has already shown that the DEMs poorly represent
the conditions on the ground, with neither DEM capturing the more complex
topography around the small streams and 20-30m deep valleys.

Under these

circumstances it is questionable the degree to which this analysis can conclude
whether spatial heterogeneity is significant or not in shaping tree diversity, but
nevertheless the analysis is presented in detail in a similar fashion to that of Tambito.
Unlike in Chapters 4 and 5, the results here are presented for each DEM separately as
the different spatial resolutions merit more specific discussion.

6.3.1 TOPO DEM

The spatial heterogeneity analysis for TBS is first performed on the TOPO DEM.
The TOPO DEM is a smooth representation of the topography in TBS, in many cases
failing to account for the micro-scale topographic variation observed in the field.
Under these circumstances it is likely that the model will underestimate the true
heterogeneity across the region, but also fail to accurately quantify the spatial
distribution. These concerns must be considered throughout the analysis.

6.3.1.1 Spatial heterogeneity results

315

6.3.1.2 Cross-correlation matrix

As was done for Tambito, a cross-correlation is performed on the resultant
heterogeneity grids for median distances of 1 cell (Table 58).

Elevation

Eastness

Northness

Curvature

Slope
Slope
Position

Slope

Slope
Position

Solar

0.66

0.84

0.65

0.69

0.75

-0.02

0.01

0.36

0.41

0.02

0.29

0.43

-

0.21

0.14

0.00

0.08

-

0.87

0.30

-

Elevation

Eastness

Northness Curvature

-

0.14

-0.05

-

Feature

Feature
(network)

0.72

0.54

0.44

0.36

0.42

0.33

0.53

0.23

0.36

0.21

0.24

0.33

0.90

0.90

0.37

0.83

0.92

0.82

0.46

0.80

0.93

0.24

0.87

0.75

0.79

-

0.29

0.39

0.01

0.39

0.22

0.21

-

0.82

0.23

0.77

0.83

0.62

-

0.35

0.90

0.78

0.82

-

0.26

0.32

0.48

-

0.72

0.75

-

0.75

Solar

Toposcale

Topmodel

Topoclass

Feature
Feature
(network)

Toposcale Topmodel Topoclass

-

Table 58 Cross-correlation matrix between all spatial heterogeneity variables with the median
distance set at 1 cell in TBS using the TOPO DEM (n = 142,450). Cells are shaded when the
correlation coefficient is above 0.80 in order to highlight the most significant correlations.

In general the correlation coefficients between heterogeneity variables are higher than
for Tambito. Furthermore, all heterogeneity variables are positively co-correlated
using the TOPO DEM in TBS, meaning that high heterogeneity in one variable to
some extent is also indicative of high heterogeneity in another variable. This may be
due to TBS being an essentially flat landscape with valleys cut through it whereas the
topography of Tambito varies at multiple scales because of changes in geology,

316

tectonic features and fluvial features of various river orders (a dendritic scheme,
whereas TBS is more trellised in drainage pattern).

Northness, topmodel, slope

position and eastness are the least cross-correlated heterogeneity variables (in that
order), whilst toposcale, curvature and slope are the most cross-correlated variables
(also in that order). Specific discussion of these cross-correlations is not provided
here, but referred to later in the text.

6.3.1.3 Comparison with plot data
6.3.1.3.1 Control Experiment

As performed for Tambito, twelve random grids are created in Arc/Info (using the
RAND command), and the spatial heterogeneity model applied to these grids for all
16 scales. The results of which are compared with the Simpson’s diversity measured
in the plots (Table 59).

Random Run

Maximum
Pearson

Minimum
Pearson

No. Scales with
significant correlation

Random grid 1
Random grid 2
Random grid 3
Random grid 4
Random grid 5
Random grid 6
Random grid 7
Random grid 8
Random grid 9
Random grid 10
Random grid 11
Random grid 12

0.70
0.09
0.10
0.84
-0.26
0.51
0.15
0.43
-0.31
0.34
0.86
0.41

-0.39
-0.54
-0.27
0.09
-0.69
-0.17
-0.09
-0.34
-0.69
-0.13
-0.11
-0.19

4
0
0
8
4
0
0
0
1
0
8
0

Table 59 Results of the heterogeneity analysis for the 12 random grids using the SRTM DEM

in TBS.

317

Significant correlations are found between heterogeneity in the random grid and
Simpson’s diversity for five of the 12 random grids for at least one scale per grid,
with a total of 25 correlations occurring (on average five scales with significant
correlation per correlating random grid). This is an alarming number of correlations,
all of which must be assumed to be spurious but nevertheless an artifact of either the
heterogeneity model or the TBS diversity dataset.

As discussed in Chapter 4, the diversity found in the plots in TBS is exceptionally
high in eight plots, with almost indistinguishable differences between plots
(Simpson’s Index between 0.976 and 0.996), and the remaining two plots having
significantly lower levels of diversity (Plot 4 and 5 with Simpson’s diversity indices
of 0.95 and 0.84 respectively). This non-normal irregular distribution of diversity
values for the ten plots produced a number of significant correlations with terrain
characteristics that did not hold true when only the high diversity plots were included
in the correlation. In other words, the significant correlations were not robust, and
heavily sensitive to the low diversity in two points. This is also somewhat the case in
the correlations shown in Table 59 with the random grid heterogeneity analyses,
whereby the omission of plots 4 and 5 from the analysis significantly lowers the
number of correlations to just one grid at three different scales (Table 60, Figure 106).
At least in this case it can be concluded that the high number of potentially spurious
correlations found with the random grids are caused by the poor distribution of points
rather than problems with the heterogeneity model itself.

318

Random Run

Maximum
Pearson

Minimum
Pearson

No. Scales with
significant correlation

Random grid 1
Random grid 2
Random grid 3
Random grid 4
Random grid 5
Random grid 6
Random grid 7
Random grid 8
Random grid 9
Random grid 10
Random grid 11
Random grid 12

0.11
0.09
0.22
0.48
0.02
0.78
0.38
0.22
0.68
0.12
-0.21
0.34

-0.09
-0.57
0.02
-0.65
-0.67
0.03
0.05
-0.56
-0.58
-0.49
-0.59
-0.27

0
0
0
0
0
3
0
0
0
0
0
0

Table 60 Results of the heterogeneity analysis for the 12 random grids when Plots 4 and 5 are

excluded from the analysis using the TOPO DEM in TBS.

Cross-Scale Heterogeneity - Random Grid 1
1.00

0.80

Pearson correlation
coefficient

0.80

0.40

0.70

0.20

0.60
0.50

0.00
-0.20

0.40
0

2

4

6

10

0.30

alpha = 0.05 0.20
0.10

-0.40
-0.60

8

Median Distance (cells) in Cauchy Kernel

Two-Tailed p-value

0.90

0.60

0.00

Figure 106 Control experiment of cross-scale heterogeneity for a random grid 6 using the
TOPO DEM.

However, one random grid still correlates, despite the exclusion of Plots 4 and 5,
meaning that spurious correlations can occur at multiple spatial scales, and must be
carefully considered in the analysis and discussion. Though some non-equilibrium
based theories of species diversity maintenance predict that stochastic processes
319

generate an auto-correlated but relatively random distribution in diversity, there is no
causal reason that heterogeneity in a particular random grid should correlate with plot
diversity (for info, the heterogeneity in random grid 6 with a median distance of 1 cell
is displayed in Figure 107).

Figure 107 Result of the heterogeneity model for Random Grid 6 for the TOPO DEM in
TBS. Represented here is the analysis with median distance of 1 cell (25m), where the
correlation with plot diversity was found to be highest. Red indicates low heterogeneity, and
blue high heterogeneity, with the plots displayed as the black dots.

As stated before, it would be beneficial to apply this control experiment multiple
times for different sets of random grids, but computational limitations prevent this.

6.3.1.3.2 Individual variables

320

Having highlighted the danger of non-causal relationships occurring in this analysis,
the results of the heterogeneity model for each terrain characteristic using the TOPO
DEM is presented variable by variable (Figure 107), and the results summarized in a
single table (Table 61).

0.10

1

0.00

0.8
0

2

4

6

8

10

-0.10

0.6

-0.20

0.4
0.2

-0.30

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Elevation

alpha = 0.05
0

-0.40
Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Eastness
1.00
2

4

6

8

10
0.80

-0.10
-0.15

0.60

-0.20
0.40

-0.25
-0.30

0.20

-0.35

alpha = 0.05

-0.40

Two-Tailed p-value

Pearson correlation
coefficient

0.00
-0.05 0

0.00

Median Distance (cells) in Cauchy Kernel

0.80
Pearson correlation
coefficient

0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
0

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
alpha = 0.05
0.00
2
4
6
8
10
Median Distance (cells) in Cauchy Kernel

321

Two-Tailed p-value

Cross-Scale Heterogeneity - Northness

Cross-Scale Heterogeneity - Curvature
2

4

6

-0.10
-0.15
-0.20
-0.25
-0.30
-0.35
-0.40

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

0.00
-0.05 0

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Slope
0.00

1.00
2

4

6

8

10

-0.10

0.80

-0.20

0.60

-0.30

0.40

-0.40

0.20

Two-Tailed p-value

Pearson correlation
coefficient

0

alpha = 0.05
-0.50

0.00
Median Distance (cells) in Cauchy Kernel

Pearson correlation
coefficient

0.05
0.00
-0.05 0

2

4

6

-0.10
-0.15
-0.20
-0.25
-0.30

1.00
0.90
0.80
0.70
8
10
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00

Median Distance (cells) in Cauchy Kernel

322

Two-Tailed p-value

Cross-Scale Heterogeneity - Slope Position
0.10

Cross-Scale Heterogeneity - Solar
0

-0.10

2

4

6

-0.20
-0.30
-0.40
-0.50
-0.60

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05
0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

0.00

Median Distance (cells) in Cauchy Kernel

0.00
-0.05 0
-0.10
-0.15
-0.20
-0.25
-0.30
-0.35
-0.40
-0.45
-0.50

2

4

6

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - TopModel

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Toposcale

-0.05

0

2

4

6

-0.10
-0.15
-0.20
-0.25
-0.30

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Median Distance (cells) in Cauchy Kernel

323

Two-Tailed p-value

Pearson correlation
coefficient

0.00

0.00
Pearson correlation
coefficient

0

2

4

6

8

-0.05

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30

-0.10
-0.15
-0.20
-0.25

alpha = 0.05

-0.30

0.20
0.10
0.00

Two-Tailed p-value

Cross-Scale Heterogeneity - Topoclass

Median Distance (cells) in Cauchy Kernel

0.10

1.00

0.08

0.80

0.06

0.60

0.04
0.40

0.02

0.20

0.00
-0.02

0

2

4

6

alpha = 0.05
8
10

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Feature

0.00

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Feature Network
1.20

0.00
-0.05

1.00
0

2

4

6

8

10

0.80

-0.10

0.60

-0.15

0.40

-0.20

alpha = 0.05

-0.25

0.20

Two-Tailed p-value

Pearson correlation
coefficient

0.05

0.00
Median Distance (cells) in Cauchy Kernel

Figure 108 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in TBS using the TOPO DEM. The black line
represents the Pearson correlation coefficient, whilst the blue line indicates the respective pvalue. For ease of interpretation, the thick blue line shows the threshold for significant
correlations at the alpha = 0.05 level.

324

Variable

Elevation
Eastness
Northness
Curvature
Slope
Slope Position
Solar Radiation
Topmodel
Toposcale
Topoclass
Feature
Network Feature

Minimum
Pearson
Correlation
Coeffcient

Maximum
Pearson
Correlation
Coeffcient

Minimum
p-value

Significance
(alpha=0.05)

Scale of lowest pvalue (median
distance in cells)

-0.36
-0.35
0.28
-0.36
-0.44
-0.26
-0.55
-0.46
-0.28
-0.28
-0.01
-0.22

0.06
-0.25
0.71
-0.12
-0.20
0.08
-0.39
-0.12
-0.14
-0.04
0.08
0.03

0.31
0.32
0.02
0.31
0.21
0.47
0.10
0.18
0.43
0.44
0.83
0.55

Not-significant
Not-significant
Significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant
Not-significant

4
9
9
0.01
9
9
0.01
3
9
0.01
0.01
1

Table 61 Summary table of results of multi-scale environmental heterogeneity modeling,

examining maximum correlations between species richness and heterogeneity for the ten plots
in TBS using the TOPO DEM.

Few significant results are evident, with only heterogeneity in northness correlating
for large median distances. However, the results of all these correlations are heavily
affected by the low Simpson’s diversity of plot 5 (0.84), which is considerably more
dominant than the other plots (with Simpson’s diversity of 0.97-0.99).

The

correlations are therefore highly sensitive to this point and when plot 5 is discounted
from the analysis, all correlations (including northness) are insignificant. This shows
that the correlation with northness is not sufficiently robust, and should not be
considered as a concrete result.

Given the lack of correlations for single variables, multiple linear regressions are not
performed for this data.

325

The abject lack of correlations using the TOPO DEM is a result in itself, but the role
of spatial environmental heterogeneity on tree species diversity in LRF cannot be
rejected due to quality and scale issues in the DEM used to calculate the terrain
characteristics and subsequent heterogeneity. As stated many times before in this
thesis, the TOPO DEM does not capture the intricate network of small streams found
in steep sided but small valleys. These characteristics, likely the principal cause of
environmental heterogeneity in most variables in the site, are not captured as the
aerial photographs used to produce the cartography only map the canopy level
topography, which is significantly smoother than the underlying terrain. Under these
circumstances, many topographic features are not captured, and the measurement of
heterogeneity in terrain characteristics using this DEM is not sufficiently reliable to
make a concrete conclusion.

6.3.2 SRTM DEM

Having examined the results of the heterogeneity modelling using the TOPO DEM,
finding no significant correlation but data quality issues hampering this analysis, this
section performs the same analysis using the SRTM DEM. Though the SRTM DEM
likely contains greater precision than the cartographic DEM, the coarse cell size
(92m) renders it even less representative of the micro-scale environmental
heterogeneity found in TBS.

Nevertheless, the scale at which environmental

326

heterogeneity may be significant in shaping the tree species diversity is unknown and
coarse-scale patterns may prove to be significant.

6.3.2.1 Spatial heterogeneity results

6.3.2.2 Cross-correlation matrix

There is much more cross-correlation between heterogeneity variables in TBS using
the SRTM DEM expressed using the Pearson correlation coefficient, though this is
likely due to the few number of cells used in the analysis (Table 62). It is also
important to note that all variables are positively correlated, indicating once more that
in TBS an area that is heterogeneous in one variable is also likely to be heterogeneous
in others (this was not the case in Tambito), again likely a function of the small and
uniform scale of topography variation in TBS compared with the more multi-scale
topographic variation found in Tambito. Heterogeneity in the feature classification,
solar radiation and slope position variables were the least cross-correlated (in that
order), and heterogeneity in curvature, toposcale and eastness were the most crosscorrelated variables.

327

Elevation

Eastness

Northness

Curvature

Slope

Slope
Position

Solar

0.96

0.98

0.77

0.89

0.85

0.75

0.83

0.89

0.82

0.78

0.63

0.93

-

0.82

0.74

0.68

0.53

-

0.96

0.82

-

Elevation

Eastness

Northness Curvature

-

0.81

0.72

-

Slope
Slope
Position

Feature

Feature
(network)

0.85

0.49

0.77

0.91

0.86

0.71

0.93

0.89

0.86

0.79

0.76

0.90

0.83

0.92

0.83

0.88

0.65

0.86

0.78

0.90

0.86

0.78

0.88

0.51

0.79

-

0.64

0.80

0.71

0.76

0.55

0.75

-

0.67

0.56

0.75

0.34

0.57

-

0.91

0.93

0.70

0.94

-

0.84

0.72

0.92

-

0.59

0.86

-

0.73

Solar

Toposcale

Topmodel

Topoclass

Toposcale Topmodel Topoclass

Feature
Feature
(network)

-

Table 62 Cross-correlation matrix between all spatial heterogeneity variables with the median
distance set at 1 cell in TBS using the SRTM DEM (n = 10,400). Cells are shaded when the
correlation coefficient is above 0.80 in order to highlight the most significant correlations.

6.3.2.3 Comparison with plot data
6.3.2.3.1 Control Experiment

The application of the heterogeneity model to twelve random grids with the
dimensions and resolution of the SRTM DEM also provides an alarming result, with
50% of the random grids (six) producing a significant correlation (95% significance
level) with Simpson’s diversity in the plots for at least one scale (Table 63). For these
six correlations, an average of 4.5 scales correlated for each grid (partly as a result of
the co-correlation between heterogeneity grids of adjoining scale, Pearson = 0.97, n =
180), involving a total of 27 significant correlations. Once again this highlights the
danger of non-causal relationships in the multi-scale correlation analyses.
328

Random Run

Maximum
Pearson

Minimum
Pearson

No. Scales with
significant correlation

Random grid 1
Random grid 2
Random grid 3
Random grid 4
Random grid 5
Random grid 6
Random grid 7
Random grid 8
Random grid 9
Random grid 10
Random grid 11
Random grid 12

0.13
0.29
0.39
-0.15
0.88
0.72
-0.08
0.40
0.69
0.08
0.65
0.82

-0.14
0.07
-0.21
-0.50
-0.03
0.00
-0.27
-0.02
0.18
-0.74
-0.54
0.33

0
0
0
0
8
3
0
0
4
2
1
9

Table 63 Results of the heterogeneity analysis for the 12 random grids using the SRTM DEM
in TBS.

When Plots 4 and 5 are excluded from the correlations for reasons outlined in Section
6.3.1.3.1, four of the 12 random grids still provide a correlation (Table 64), with an
average of six different scales correlating significantly per random grid.

Random Run

Maximum
Pearson

Minimum
Pearson

No. Scales with
significant correlation

Random grid 1
Random grid 2
Random grid 3
Random grid 4
Random grid 5
Random grid 6
Random grid 7
Random grid 8
Random grid 9
Random grid 10
Random grid 11
Random grid 12

0.65
0.70
0.04
-0.05
0.59
-0.02
0.32
0.43
0.92
0.52
0.87
0.58

-0.30
0.04
-0.55
-0.46
-0.14
-0.17
-0.32
0.14
0.16
-0.55
-0.42
0.17

1
4
0
0
0
0
0
0
10
0
8
0

Table 64 Results of the heterogeneity analysis for the 12 random grids when Plots 4 and 5 are

excluded from the analysis using the SRTM DEM in TBS.

329

This calls for extreme caution in interpreting the results of the heterogeneity
modelling as numerous spurious correlations may be expected, though the reasons for
so many significant correlations are unknown. This is a strong limitation to the
interpretation of these heterogeneity results in TBS, and is something that requires
further study.

6.3.2.3.2 Individual variables

The heterogeneity model is applied for all terrain characteristics from the SRTM
DEM, and the multi-scale correlation graphs shown in Figure 109, and a summary of
the results is shown in Table 65.

330

Cross-Scale Heterogeneity - Elevation
1.00

0.40

0.80

0.30
0.20

0.60

0.10

0.40

0.00
-0.10 0

2

4

6

8
10 0.20
alpha = 0.05

-0.20

Two-Tailed p-value

Pearson correlation
coefficient

0.50

0.00
Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Eastness
-0.10

0

2

4

6

-0.20
-0.30
-0.40
-0.50
-0.60
-0.70

1.00
10 0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00
8

Two-Tailed p-value

Pearson correlation
coefficient

0.00

Median Distance (cells) in Cauchy Kernel

1.00

1.00

0.80

0.80

0.60

0.60

0.40

0.40

0.20

alpha = 0.05

0.00

0.20
0.00

0

2

4

6

8

Median Distance (cells) in Cauchy Kernel

331

10

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Northness

0.45
0.40
0.35
0.30
0.25
0.20
0.15
0.10
0.05
0.00

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00

alpha = 0.05
0

2

4

6

8

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Curvature

10

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Slope
1.00

0.06

0.80

0.05
0.04

0.60

0.03

0.40

0.02
0.20

0.01

Two-Tailed p-value

Pearson correlation
coefficient

0.07

alpha = 0.05

0.00

0.00
0

2

4

6

8

10

Median Distance (cells) in Cauchy Kernel

1.00

0.05

0.80

0.00

0.60
0

2

4

6

8

10

-0.05

0.40

-0.10

0.20
alpha = 0.05

-0.15

0.00
Median Distance (cells) in Cauchy Kernel

332

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Slope Position
0.10

Pearson correlation
coefficient

0.30
0.20
0.10
0.00
-0.10

0

2

4

6

-0.20
-0.30
-0.40

1.00
0.90
0.80
0.70
0.60
8
10 0.50
0.40
0.30
0.20
alpha = 0.05 0.10
0.00

Two-Tailed p-value

Cross-Scale Heterogeneity - Solar

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - TopModel
0.00

1.00
2

4

6

8

10
0.80

-0.20
-0.30

0.60

-0.40
0.40

-0.50
-0.60

0.20

-0.70

Two-Tailed p-value

Pearson correlation
coefficient

-0.10 0

alpha = 0.05

-0.80

0.00
Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Toposcale
1.00

0.10

0.80

0.00
-0.10

0

2

4

6

8

10 0.60

-0.20

0.40

-0.30
0.20

-0.40

alpha = 0.05

-0.50

0.00
Median Distance (cells) in Cauchy Kernel

333

Two-Tailed p-value

Pearson correlation
coefficient

0.20

Cross-Scale Heterogeneity - Topoclass
-0.10

1.00
0

2

4

6

8

10
0.80

-0.20
-0.30

0.60

-0.40

0.40

-0.50
0.20

-0.60

Two-Tailed p-value

Pearson correlation
coefficient

0.00

alpha = 0.05

-0.70

0.00
Median Distance (cells) in Cauchy Kernel

0.50

1.00

0.40

0.80

0.30

0.60

0.20
0.40

0.10

0.20

0.00
-0.10

0

2

4

6

alpha = 0.05
8
10

Two-Tailed p-value

Pearson correlation
coefficient

Cross-Scale Heterogeneity - Feature

0.00

Median Distance (cells) in Cauchy Kernel

Cross-Scale Heterogeneity - Feature Network
0.20

1.00
0.80

0.10
0.05

0.60

0.00
-0.05 0

2

4

6

8

10 0.40

-0.10

0.20

-0.15

Two-Tailed p-value

Pearson correlation
coefficient

0.15

alpha = 0.05

-0.20

0.00
Median Distance (cells) in Cauchy Kernel

Figure 109 Cross-scale correlations between species diversity (Simpson’s) and spatial
heterogeneity for each environmental variable in TBS using the SRTM DEM. The black line
represents the Pearson correlation coefficient, whilst the blue line indicates the respective p-

334

value. For ease of interpretation, the thick blue line shows the threshold for significant
correlations at the alpha = 0.05 level.

Variable

Minimum
Pearson
Correlation
Coeffcient

Maximum
Pearson
Correlation
Coeffcient

Minimum
p-value

Significance
(alpha=0.05)

Scale of lowest pvalue (median
distance in cells)

-0.08
-0.64
0.53
0.07
0.04
-0.13
-0.36
-0.70
-0.43
-0.62
-0.02
-0.18

0.40
-0.38
0.95
0.40
0.07
0.05
0.18
-0.24
0.12
-0.31
0.41
0.16

0.25
0.04
0.00
0.25
0.85
0.71
0.30
0.03
0.21
0.06
0.23
0.62

Not-significant
Significant (inverse)
Significant
Not-significant
Not-significant
Not-significant
Not-significant
Significant (inverse)
Not-significant
Not-significant
Not-significant
Not-significant

9
2
0.3
3
0.8
9
0.01
1
3
0.01
9
2

Elevation
Eastness
Northess
Curvature
Slope
Slope Position
Solar Radiation
Topmodel
Toposcale
Topoclass
Feature
Network Feature

Table 65 Summary table of results of multi-scale environmental heterogeneity modeling,
examining maximum correlations between species richness and heterogeneity for the ten plots
in TBS using the SRTM DEM.

Significant correlations are found between plot species diversity and spatial
environmental heterogeneity in three variables for the SRTM DEM, including
eastness (inversely correlated at 1 scale), northness (positively correlated at 13
scales), and topmodel (inversely correlated at 2 scales). As discussed earlier, many of
these correlations are heavily affected by the low Simpson’s diversity in Plot 5 and to
a lesser extent in Plot 4. When these plots are excluded from the correlation, eastness
and topmodel no longer provide any significant correlation (p>0.05 in both cases),
whilst northness still correlates but with a reduced Pearson coefficient of 0.77 (p =
0.01) (Figure 110).

335

Scatterplot of average heterogeneity
(median distance = 0.05) against plot
diversity (Simpson's Diversity Index),
excluding Plot 5

1.02
1
0.98
0.96 y = 0.8037x + 0.2115
R2 = 0.8963
0.94
0.92
0.9
0.88
0.86
0.84
0.82
0.60
0.70
0.80

1
Simpson's Diversity
Index)

Simpson's Diversity
Index)

Scatterplot of average heterogeneity
(median distance = 0.03) against plot
diversity (Simpson's Diversity Index),
including Plot 5

0.90

1.00

1.10

0.99
0.98

y = 0.4169x + 0.5833
R2 = 0.6005

0.97
0.96
0.95
0.94
0.93
0.60

0.70

0.80

0.90

1.00

1.10

Northness heterogeneity (SRTM)

Northness heterogeneity (SRTM)

Figure 110 Scatterplot of heterogeneity in the northness variable against plot diversity, using
the SRTM DEM for TBS when plot 5 is included (left) and excluded (right).

Lowest species diversity is found in areas homogenous in northness, with low
diversity Plots 4 and 5 both having markedly low levels of heterogeneity in northness
at low median distances. However, these two plots are in close proximity to the river,
within the flat flood plain that neighbours the main River Tiputini channel with
frequent flooding, and it is unlikely that the low diversity in these plots is brought
about by homogeneity in northness. Rather the low diversity in these plots has
already been attributed to the frequent flooding outlined in Chapter 4.

Beyond the correlation with northness, the significant correlations with topmodel and
eastness are not sufficiently robust to merit specific discussion. Given the lack of a
clear result in the single variate analysis presented here, the multiple linear regression
is not performed as the interpretation of results is difficult and would be highly
inconclusive. The lack of correlation between heterogeneity and diversity for the
SRTM DEM may originate from the resolution of the original DEM, which captures

336

general topographic patterns but fails to identify much of the micro-scale
heterogeneity present in TBS.

6.4

Conclusions

This chapter has examined the role of spatial environmental heterogeneity on tree
species diversity in a montane environment (Tambito) with extensive environmental
gradients over short distances and in a lowland environment (TBS) with less extensive
gradients but nevertheless a degree of environmental heterogeneity at the micro-scale.

In Tambito, heterogeneity in elevation is concluded to be of importance in shaping
tree species diversity with greater diversity found in more spatially heterogeneous
sites, though the precise cause of this is open to discussion. This observation is
supported in theory by the habitat association that has already been identified in tree
species composition across the elevational gradient (Chapter 4).

If species are

associated to specific altitudinally controlled habitats, regions with a great
neighbouring diversity in these habitats are likely to contain more species, due to
them representing a mixing-zone with a diverse pool of seeds arriving to the site
through distance-dependent dispersal.

There are likely also important biotic

interactions in heterogeneous sites, including the important effects of herbivory (and
associated host-specificity) that have not been discussed explicitly here, but
nevertheless important to acknowledge.

However, issues of co-linearity between variables complicate greatly the analysis.
These problems occur not only between heterogeneity variables (i.e. heterogeneity in

337

elevation is co-linear with heterogeneity in slope position) but also between
heterogeneity and other basic factors such as elevation. An important finding of this
chapter has been the identification of a peak in elevational heterogeneity in midelevations. Given that Letts (2003) has found productivity to decrease with greater
elevation this also represents a productivity gradient, indicating that highest
heterogeneity is found in mid-productivity levels, a finding reported only once in the
literature across the productivity gradient in limestone pavements (Lundholm and
Larson, 2003). In this case study it makes it impossible to isolate the causal factor for
the mid-diversity peak observed in the plot data, with plausible explanations including
the mid-domain effect, species-area relationships, productivity-diversity relationships
and the role of spatial environmental heterogeneity. Indeed, further testing of the
heterogeneity model and associated analyses in other regions may indicate that this
peak in heterogeneity in mid-elevations (or mid-catchment more precisely) is a
property shared by many landscapes, brought about by geomorphological processes
originating from the erosional processes of rivers especially in tectonic environments.

Finally, it is also important to note that heterogeneity using the method applied here is
not a universal concept for this kind of landscape, with numerous examples of
heterogeneity in one variable signifying homogeneity in another. This means that the
selection of environmental variable for quantifying heterogeneity is of great
importance, and must be carefully considered.

In TBS the results of the heterogeneity modelling were not satisfactory, and no
significant correlations can be concluded using either of the DEMs. Quality issues
with the topographic data means that spatial heterogeneity cannot be discounted as a

338

driver of tree species diversity maintenance, as the DEMs may not be capturing the
topographic variation relevant to spatially distributing species composition and
diversity. Neither DEM accurately captures the intricate network of valleys and
ridges, and it may be this environmental variation and the spatial heterogeneity of this
variation that to some extent shapes tree species diversity. However, perhaps it can be
conclusively stated that broad scale heterogeneity (which is suitably captured in both
DEMs) does not appear to shape tree species diversity at the micro-scale.

However, contrary to the findings in Tambito, regions heterogeneous in one
topographic variable are also likely to be heterogeneous in another variable, with high
levels of positive co-linearity in heterogeneity for most variables. This indicates that
heterogeneity in lowland rain forests at the scale used in this study is a more universal
phenomenon in the landscape, and is less sensitive to the type of variable used.

Though the results presented here show little evidence of significant patterns between
spatial environmental heterogeneity and diversity in lowland rain forests, further
studies should apply the models over high resolution and more precise DEMs in wellstudied diversity plots.

Furthermore the worrying number of correlations found

between diversity and heterogeneity of random grids must be further examined,
especially in the case of the SRTM DEM. The uncertainty in the validity of the
correlations with terrain characteristics that the control experiments highlighted
significantly weakened the degree to which concrete conclusions could be achieved in
TBS.

339

In general, the analyses shown here represents the first application of a model that
quantifies spatial environmental heterogeneity in tropical forests, and compares it to
tree species diversity.

The results are far from conclusive, yet some important

patterns have been identified and further study is undeniably warranted. Firstly,
spatial environmental heterogeneity could be quantified in a broad range of ways, and
different models should be developed and tested to examine the degree to which the
results are dependent on the statistical method used to quantify heterogeneity, as well
as the distance function applied to define the spatial weighting. Secondly, more
concrete conclusions could be achieved with a greater number of diversity plots,
minimizing the potential effect of single data points which significantly affected the
results in TBS, for example.

340

Chapter 7

7

Final Conclusions

This thesis has attempted to understand micro-scale spatial variation in tree diversity,
composition and structure in two contrasting sites using terrain-based characteristics
that through different mechanisms determine some of the essential resources for
plants (light, water and temperature principally). The approach adopted is centered on
testing equilibrium-based theories for diversity maintenance in tropical forests.

The results have been mixed.

Firstly, plot-based measurement of tree diversity,

composition and structure showed significant between-plot variability in each of the
two study sites. In the steep topographic gradients of the tropical montane (cloud)
forest at Tambito, plot richness varied from 30 to 52 species, with an average
abundance of 4.1 individuals per species. Similarly, for the topographically more
subtle lowland rainforest TBS site, plot richness was found to vary between 31 and 82
species, with an average abundance of just 2.1 individuals per species. These results
alone present sufficient evidence to state that tree diversity and composition is
significantly variable over space in both of these environments (Objective 1), and also
that both sites studied here have remarkable levels of diversity.

When spatial patterns in composition are examined in detail, a very clear elevational
gradient is evident in the TMCF site in Tambito, with 36% of variability in
composition explained by elevation. Although elevation itself is not a variable that is
likely to affect composition, it represents a gradient in temperature, wetness, cloud

341

cover and likely other variables. This compositional gradient is attributed to species
adaptations to specific climatic bands. Similarly in TBS, a compositional gradient is
also identified along the short elevational gradient (explaining just 16%), though this
pattern is also observed with greater distance from the main river channel (which
correlates with altitude since the river occupies the lowest land), and with greater
geographic (Cartesian) distance between plots (all three variables co-linear). The
elevational gradient itself is too small to represent a temperature gradient in TBS, but
is indicative of flooding frequency, also partly captured with the distance from river
variable. However, these factors may also represent a gradient of forest and soil age,
driven by long-term migration of the river channel across its floodplain. Ultimately
the compositional gradient observed may indeed be a combination of these factors.

These results indicate that physical (as opposed to biotic) habitat association in
composition does occur in both sites to different extents, but that the majority of
compositional variability between plots remains unexplained by the terrain-based
variables used here (Objective 2).

The remaining unexplained variability in

composition may be driven by a combination of abiotic factors not measured here,
biotic factors as well as non-equilibrium based processes of forest dynamics. This is
an interesting result, and indeed the first example of the identification of micro-scale
habitat associations in composition using quantified terrain-based analysis in tropical
forests. The levels of habitat associations observed here may be sensitive to sampling
strategy, and indeed the use of different sized plots (with associated different scale
terrain-based characteristics) will likely cause variation in the habitat associations
found here (35% in Tambito and 16% in TBS). Further research with larger plots,
and multi-scale terrain analyses would provide better results.

342

When patterns in diversity are examined (Objective 3), this study has found a midelevation peak in diversity in Tambito at around 1850m, and also a strong pattern
between richness and mean topographic curvature, with greatest richness being found
on convex slopes.

The peak in diversity in mid-elevations is attributed to a

combination of the mid-domain effect and an area-based hypothesis, and a simple
model is produced to illustrate this (Section 4.2.2.1.2). The identification of higher
richness on convex slopes is principally attributed to a greater diversity of light
environments, brought about by a combination of vertical and lateral light penetration
of the canopy, though this pattern may also be explained by greater exposure to cloudbased water and nutrient inputs on convex slopes, or the poor soils expected on
leached convex slopes. In the lowland forest site at TBS, few richness or diversity
patterns were found except for significantly lower diversity in two sites alongside the
river and which are thus regularly subjected to flooding (sometimes for prolonged
periods of time). The remaining variability in diversity in the TBS site is unexplained
by the terrain characteristics used in this study, though issues with DEM quality mean
that the existence of a link between tree diversity and terrain characteristics cannot be
discounted.

Tree structure has been shown to vary significantly between sites in Tambito, yet
surprisingly little of this variability is explained by terrain characteristics (Objective
5). Particularly surprising is the lack of patterns across the elevational gradient, with
no evidence of a progressive increase in stem density, decrease in DBH or decrease in
tree height with greater elevation. These results are contradictory to many empirical
studies found in the literature (presented in Section 2.4).

343

Just one significant

correlation was found in Tambito, with 67% of variability in the DBH/Height ratio
being explained by the topmodel wetness index variable, indicative of soil moisture.
This pattern is attributed to an investment by trees in increased girth relative to height
in order to achieve greater stability and lessen the likelihood of treefall through
uprooting, which is more likely on wetter, therefore less stable soils. In TBS, a
significant difference is found in tree height between the three highest elevation plots
and the lower-elevation plots, with these groups possibly differentiated in terms of
flooding regime with plots below 220m elevation being subjected to occasional
flooding and having significantly lower average tree stature than the three higher
elevation plots. This pattern may indeed be indicating two unique forest structures,
no doubt reflecting the two types of Amazonian lowland forest discussed in the
literature; terra firme and varzea forest.

Finally, models of spatial environmental heterogeneity have been applied to examine
the role of characteristics in surrounding areas on diversity and composition at each of
the plots, as brought about by spatial interactions such as seed dispersal (Objective 4).
In Tambito, spatial heterogeneity in elevation is shown to explain some variation in
diversity (explaining 53% of variability). This is an important finding, showing how
both absolute elevation combined with spatial heterogeneity in elevation both
contribute significantly to micro-scale diversity patterns in Tambito. In TBS, many
complications contributed to a confusing set of results, for which no concrete
conclusion can be reached. These problems included questionable quality in the
original DEMs for the study site, many surprising and spurious correlations with
control (random) analyses, and the poor distribution of diversity levels in the ten
plots. DEM quality could be addressed through identifying alternative topographic

344

data sources with higher resolution, which may capture topographic variation at a
more relevant scale for the processes which drive spatial variability in tree diversity.
The heterogeneity model seems to work where the terrain gradient (heterogeneity) is
high, but fall subject to statistical constraints (inclusion of random effects) where the
gradients are low (as in TBS). In general, the spatial heterogeneity analyses have
shown promise in explaining some variability in diversity, and indeed furthering our
understanding of the processes behind diversity maintenance, but fall short of
providing solid results and merit further study.

Amongst these results are some important findings that further our understanding of
micro-scale variation in diversity in tropical forests.

Whilst terrain-based

characteristics that determine the spatial variation in essential resources in tropical
forests have explained some variability in composition, diversity and structure, a great
deal of variability remains unexplained.

This leads to the possibility that a

combination of equilibrium and non-equilibrium (physical) processes (such as gap
dynamics) is shaping the micro-scale variability in diversity alongside a suite of biotic
(competition etc) interactions that also determine resource availability. This thesis
has also presented innovative methodologies for examining the relationship between
environment and diversity at the micro-scale, and the application of these methods in
other sites with greater volume of ground-based diversity data may yield further
insights.

There have been significant complications in reaching concrete conclusions in this
research due to the low number of ground-based plots used to examine the
relationships between topography and composition, diversity and structure.

345

The

original experimental design was to use high-resolution aerial imagery to monitor the
spatial distribution of diversity, and use this larger dataset to compare canopy
diversity with topography, at least for TBS where there was available imagery.
Unfortunately the analysis of high-resolution imagery was complicated by many
factors (shading, image coverage, suitable methods) and was not included in this PhD
(though Appendix 1 provides an overview of what was achieved in this line of
research). Within the time available for this thesis just ten plots could be generated
for each study site, and making concrete conclusions with this data has been difficult,
especially in TBS where two extremely low diversity plots alongside the river had
great influence on the statistical significance of the linear regressions.

The analysis has also been complicated by strong co-linearity between some of the
different terrain characteristics, making it difficult to separate true correlations from
spurious relationships. Whenever possible, ecological interpretation of the results has
been used to explain the most likely process behind such relationships (e.g.
understanding the relationship between composition and elevation / geographic
distance / distance from river in TBS), but in some cases this has not been possible.
More data points (plots) would help illustrate more clearly the true correlations,
alongside some more advanced statistical methods, though this is left for further
study.

Whilst this research has provided some interesting results and conclusions, as in most
research studies it has opened up many more questions. Specifically, the following
topics should be further investigated in future research studies:

346

1. Full testing of different spatial heterogeneity measures as potential
explanations of spatial variation in tree species diversity, and analysis of the
sensitivity of the results to the different methods
2. Detailed analysis of the diversity gradient with greater distance from the main
river channel, including soil surveying to confirm if this is in reality a pattern
brought about by a disturbance from river flooding (short-term) or the result of
longer term non-equilibrium adjustment along a primary successional gradient
3. Repetition of the analyses in other sites, and with greater number of groundbased diversity plots
4. Repetition of the analyses in lowland forest using higher-quality DEMs,
possibly derived from ground based surveying or LIDAR / RADAR.

Perhaps the best strategy for following-on this research would be to apply the models
to large, long-term plots such as the 50-Ha plot found in Yasuni National Park near
the TBS site presented in this thesis, where abundant tree diversity data is available.

347

References
Albani, M., B. Klinkenberg, D. Andison and J. Kimmins (2004). "The choice of
window size in approximating topographic surfaces from Digital Elevation
Models." International Journal of Geographical Information Science 18(6):
577-593.
Almeida, A. M., P. I. Prado and T. M. Lewinsohn (2004). "Geographical distribution
of Eupatorieae (Asteraceae) in South-eastern and South Brazilian Mountain
Ranges." Plant Ecology 174(1): 163.
Alonso, A., F. Dal lmeier, E. Granek and P. Raven (2001). Biodiversity - Connecting
with the Tapestry of Life. Washington DC, Smithsonian
Institution/Monitoring and Assessment of Biodiversit y Program and President
's Committee of Advisors on Science and Technology.
Aptroot, A. (2001). "Lichenized and saprobic fungal biodiversity of a single
Elaeocarpus tree in Papua New Guinea, with the report of 200 species of
ascomycetes associated with one tree." Fungal Diversity 6: 1-11.
Araujo, M. B., C. J. Humphries, P. J. Densham, R. Lampinen, W. J. M. Hagemeijer,
A. J. Mitchell-Jones and J. P. Gase (2001). "Would environmental diversity be
a good surrogate for species diversity." Ecography 24(1): 103-110.
Ashton, P. (1976). "Mixed dipterocarp forest and its variation with habitat in the
Malayan lowlands: a re-evaluation at Pasoh." Malayan Forester 39: 56-72.
Ashton, P. S. (1964). "Ecological studies in the mixed Dipterocarp forests of Brunei
state." Oxford Forest Memoirs Number 25.
Ashton, P. S. (1977). "A contribution of rain forest research to evolutionary theory."
Annals of the Missouri Botanical Garden 64(4): 694-705.
AsterDTM-V2 (2004). "Asterdtm plugin for ENVI". available online at
http://www.creaso.com/english/12_swvis/13_envi/AsterDTM/asterdtm.htm.
Aubréville, A. (1938). "La forêt coloniale: les forêts de l'Afrique Occidentale
Francaise." Annales de l'Acedémie des Sciences Coloniales 9.
Austin, M., P. S. Ashton and P. Greig-Smith (1972). "The application of quantitative
methods to vegetation survey. III. A re-examination of rain forest data from
Brunei." Journal of Ecology 60: 305-324.
Aylward, B. and S. Gammage (1992). Preserving future options. Global biodiversity:
status of the earth's living resources. in B. Groombridge. London, UK,
Chapman and Hall.
Baillie, I., P. Ashton, M. Court, J. Anderson, E. Fitzpatrick and J. Tinsley (1987).
"Site characteristics and the distribution of tree species in Mixed Dipterocarp
Forest on tertiary sediments in central Sarawak, Malaysia." Journal of
Tropical Ecology 3: 201-220.
348

Barberis, I. M., W. B. Batista, E. F. Pire, J.-P. Lewis and R. J. Leon (2002). "Woody
population distribution and environmental heterogeneity in a Chaco forest,
Argentina." Journal of Vegetation Science 13: 607-614.
Barthlott, W., G. Kier and J. Mutke (1999). Biodiversity - The Uneven Distribution of
a Treasure.
Bell, G., M. J. Lechowicz and M. J. Waterway (2000). "Environmental heterogeneity
and species diversity of forest sedges." Journal of Ecology 88: 67-87.
Bengtsson, J. (1998). "Which species? What kind of diversity? Which ecosystem
function? Some problems in studies of relations between biodiversity and
ecosystem function." Applied Soil Ecology 10: 191-199.
Berendse, F. (1999). "Implications of increased litter production for plant
biodiversity." Trends in Ecology and Evolution 14(1): 4-5.
Beven, K. J. (1978). "The hydrological response of headwater and sideslope areas."
Hydrological Sciences Bulletin 23(4): 419-437.
Black, G., T. Dobzhansky and C. Pavan (1950). "Some attempts to estimate species
diversity and population density of trees in Amazonian forests." Botanical
Gazette 111: 413-425.
Blackburn, T. and K. Gaston (1996). "Spatial patterns in the geographic range sizes of
bird species in the New World." Philosophical Transaction of the Royal
Society London B 351: 897-912.
Brehm, G., D. Sussenbach and K. Fiedler (2003). "Unique elevational diversity
patterns of geometrid moths in an Andean montane rainforest." Ecography 26:
456-466.
Brokaw, N. and R. T. Busing (2000). "Niche versus chance and tree diversity in forest
gaps." Trends in Ecology and Evolution 15(5): 183-188.
Bruenig, E. and Y. Huang (1990). Patterns of tree species diversity and canopy
structure and dynamics in humid tropical evergreen forests on Borneo and in
China. Tropical forests. in L. Holm-Nielson, I. Neilson and H. Balslev.
London, Academic Press: 75-88.
Bruijnzeel, L. and J. Proctor (1995). Hydrology and biochemistry of tropical montane
cloud forests: What do we really know? Tropical Montane Cloud Forests. in
L. Hamilton, J. Juvik and F. Scatena. New York, Springer-Verlag.
Bubb, P., I. May, L. Mile and J. Sayer (2004). "Cloud Forest Agenda". available
online at http://www.unepwcmc.org/resources/publications/UNEP_WCMC_bio_series/20.htm.
Bui, E., A. Loughhead and R. Corner (1999). "Extracting soil-landscape rules from
previous soil surveys." Australian Journal of Soil Research 37(3): 495-508.

349

Burley, J., J. Evans and J. Younquist (2004). Encyclopedia of Forest Sciences Vol. 4.
Oxford, Elsevier.
Burnett, M. R., P. V. August, J. H. Brown and K. T. Killingbeck (1998). "The
influence of geomorphological heterogeneity on biodiversity. I. A patch-scale
perspective." Conservation Biology 12: 363-370.
Burrough, P. (1986). Principles of Geographic Information Systems for Land Surface
Assessments. New York, Oxford University Press.
Burslem, D. F., N. C. Garwood and S. C. Thomas (2001). "Enhanced: Tropical forest
diversity - The plot thickens." Science 291(5504): 606-607.
Caldwell, M., R. Robberecht and W. Billings (1980). "A steep latitudinal gradient of
solar ultraviolet-B radiation in the arctic-alpine life zone." Ecology 6: 600-611.
Castano, C. (1991). Bosque de niebla y hombre tradicional. Bosques de niebla de
Colombia. Bogota, Banco de Occidente: 151-170.
Castellano, S. and E. Balletto (2002). "Is the partial Mantel test inadequate?"
Evolution 56(9): 1871-1873.
CBD (1992). "Convention on Biological Diversity". available online at available at
http://www.biodiv.org/convention/articles.asp.
Chambers, J. C. (1994). "A day in the life of a seed: Movements and fates of seeds
and their tests implications for natural and managed systems." Annual Review
of Ecology and Systematics 25: 263-292.
Chase, J. M. and M. A. Leibold (2002). "Spatial scale dictates the productivity biodiversity relationship." Nature 416(427-430).
Chazdon, R. and N. Fetcher (1984). "Photosynthetic light environments in a lowland
tropical rain forest in Costa Rica." Journal of Ecology 72: 553-564.
Chen, Z.-S., C.-F. Hsieh, F.-Y. Jiang, T.-H. Hsieh and I. F. Sun (1997). "Relations of
soil properties to topography and vegetation in a subtropical rain forest in
southern Taiwan." Plant Ecology 132(2): 229.
CIESIN (1998). Unsmoothed map of world population density.
Claridge, M. F. (1995). "Introducing Systematics Agenda 2000." Biodiversity and
Conservation 4(5): 451-454.
Clark, D. B. and D. A. Clark (2000). "Landscape-scale variation in forest structure
and biomass in a tropical rain forest." Forest Ecology and Management 137:
185-198.
Clark, D. B., D. A. Clark and J. M. Read (1998). "Edaphic variation and the
mesoscale distribution of tree species in a neotropical rain forest." Journal of
Ecology 86: 101-112.

350

Clark, D. B., M. W. Palmer and D. A. Clark (1999). "Edaphic factors and the
landscape-scale distributions of tropical rain forest trees." Ecology 80(8):
2662-2675.
Clarke, K. and R. Warwick (1998). "A taxonomic distinctness index and its statistical
properties." Journal of Applied Ecology 35: 523-531.
Clinebell II, R. R., O. L. Phillips, A. H. Gentry, N. Stark and H. Zuuring (1995).
"Prediction of neotropical tree and liana species richness from soil and climate
data." Biodiversity and Conservation 4: 56-90.
Clinebell, R. I., O. L. Phillips, A. H. Gentry, N. Stark and H. Zuuring (1995).
"Prediction of neotropical tree and liana species richness from soil and climate
data." Biodiversity and Conservation 4: 56-90.
Colwell, R. and D. Lees (2000). "The mid-domain effect: Geometric constraints on
the geography of species richness." Trends in Ecology and Evolution: 70-76.
Colwell, R. K. and D. C. Lees (2000). "The mid-domain effect: geometric constraints
on the geography of species richness." Trends in Ecology and Evolution 15(2):
70-76.
Condit, R., S. P. Hubbell and R. B. Foster (1992). "Short-term dynamics of a
neotropical forest." BioScience 42(11): 822-828.
Condit, R., S. P. Hubbell, J. V. LaFrankie, R. Sukumar, N. Manokaran, R. B. Foster
and P. S. Ashton (1996). "Species-area and species-individual relationships for
tropical trees: a comparison of three 50-ha plots." Journal of Ecology 84: 549562.
Connell, J. (1971). On the role of natural enemies in preventing competitive exclusion
in some marine animals and in rain forest trees. Dynamics of populations. in P.
den Boer and G. Gradwell. Wageningen, Centre for Agricultural Publishing
and Documentation.
Connell, J. H. (1978). "Diversity in tropical rainforests and coral reefs." Science 199:
1302-1309.
Crawley, M. J. (1997). Biodiversity. Plant Ecology. in M. J. Crawley. Oxford,
Blackwell.
Currie, D. J. (1991). "Energy and large-scale patterns of animal- and plant-species
richness." American Naturalist 137: 27-49.
Dalling, J. W., C. E. Lovelock and S. P. Hubbell (1999). "Growth responses of
seedlings of two neotropical pioneer species to simulated forest gap
environments." Journal of Tropical Ecology 15: 827-839.
DeAngelis, D., W. Post and C. Travis (1986). Positive feedback in natural systems.
Heidelberg, Springer-Verlag.

351

Debski, I., D. F. Burslem, P. Palmiotto, J. LaFrankie, H. Lee and N. Manokaran
(2002). "Habitat preferences of Aporosa in two Malaysian forests:
Implications for abundance and coexistence." Ecology 83(7): 2005-2018.
Denslow, J. S. (1987). "Tropical rainforest gaps and tree species diversity." Annual
Review of Ecology and Systematics 18: 431-451.
Desrochers, R. E. and M. Anand (2004). "From traditional diversity indices to
taxonomic diversity indices." International Journal of Ecology and
Environmental Sciences 30: 85-92.
Dinerstein, E., D. M. Olsen, D. J. Graham, A. L. Webster, S. A. Primm, M. P.
Bookbinder and G. Ledec (1995). A conservation assessment of the terrestrial
ecoregions of Latin America and the Caribbean. Washington DC, The World
Bank.
Duivenvoorden, J. and J. Lips (1995). A land-ecological study of soils, vegetation and
plant diversity in Colombian Amazonia. Wageningen, The Netherlands,
Tropenbos Foundation.
Duivenvoorden, J. F. (1994). "Vascular plant species counts in the rain forests of the
middle Caqueta area, Colombian Amazonia." Biodiversity and Conservation
3: 685-715.
Ehrenfeld, D. (1988). Why put a value on biodiversity? Biodiversity. in E. O. Wilson.
Washington, U.S.A., National Academy Press.
Enoki, T. (2003). "Microtopography and distribution of canopy trees in a subtropical
evergreen broad-leaved forest in the northern part of Okinawa Island, Japan."
Ecological Research 18: 103-113.
Epperson, B. (2004). Geographic Genetics, Princeton University Press.
Erwin, T. L. (1982). "Tropical forests: Their richness in Coleoptera and other
arthropod species." The Coleopterists Bulletin 36: 74-75.
Erwin, T. L. (1991). "How many species are there - revisited." Conservation Biology
5(3): 330-333.
ESRI (2001). Arc/Info 8.1 Software. Redlands, USA, Environmental Systems
Research Institute.
FAO (1990). Forest Resources Assessment 1990, Tropical Countries. Forestry Paper
112. Rome, Food and Agriculture Organisation of the United Nations.
FAO (1997). State of the World's Plant Genetic Resources. F. a. A. Organisation.
Rome.
Fearnside, P. (1990). "The rate and extent of deforestation in Brazilian Amazonia."
Environmental Conservation 17: 213-225.

352

Federov, A. A. (1966). "The structure of the tropical rain forest and speciation in the
humid tropics." Journal of Ecology 54(1): 1-11.
Ferreira, L. V. (1997). "Effects of the duration of flooding on species richness and
floristic composition in three hectares in the Jau National Park in floodplain
forests in central Amazonia." Biodiversity and Conservation 6: 1353-1363.
Ferreira, L. V. (2000). "Effects of flooding duration on species richness, floristic
composition and forest structure in river margin habitat in Amazonian
blackwater floodplain forests: implications for future design of protected
areas." Biodiversity and Conservation 9: 1-14.
Flenley, J. (1992). UV-B Insolation and the Altitudinal Forest Limit. Nature and
Dynamics of Forest-Savanna Boundaries. in P. Furley, J. Proctor and J. Ratter.
London, Chapman and Hall: 616.
Flenley, J. (1995). Cloud forest, the Massenerhebung Effect, and ultraviolet
insolation. Tropical Montane Cloud Forests. in L. Hamilton, J. Juvik and F.
Scatena. New York, Springer-Verlag.
Folke, C. (1996). "Biological diversity, ecosystems, and the human scale." Ecological
Applications 6(4): 1018-1024.
Fonseca, C. R. and G. Ganade (2001). "Species functional redundancy, random
extinctions and the stability of ecosystems." Journal of Ecology 89: 118-125.
Forero, E. and A. Gentry (1989). Lista Anotado de las plantas del Departamento del
Choco, Colombia. Biblioteca Jose Jeronimo Triana No. 10. Bogota, Instituto
de Ciencias Naturales, Universidad Nacional de Colombia.
Fotheringham, A. S., M. Charlton and C. Brunsdon (1996). "The geography of
parameter space: An investigation of spatial non-stationarity." International
Journal of Geographical Information Systems 10(5): 605-627.
Fowler, J., L. Cohen and P. Jarvis (1998). Practical statistics for field biology.
Chichester, England, John Wiley and Sons.
Gale, N. (2000). "The relationship between canopy gaps and topography in a Western
Ecuadorian rain forest." Biotropica 32(4a): 653-661.
Galeano, G., S. Suarez and H. Balslev (1998). "Vascular plant species count in a wet
forest in the Choco area on the Pacific coast of Colombia." Biodiversity and
Conservation 7: 1563-1575.
Gentry, A. (1982). "Patterns of neotropical plant species diversity." Evolutionary
Biology 15: 1-84.
Gentry, A. (1982). Phytogeographic patterns as evidence for a Choco refuge.
Biological Diversification in the Tropics. in G. Prance. New York, Columbia
University Press: 112-136.

353

Gentry, A. H. (1990). Four Neotropical Rainforests. New Haven, Yale University
Press.
Gentry, A. H. (1995). Patterns of diversity and floristic composition in neo-tropical
montane forests. Biodiversity and Conservation of Neotropical Montane
Forests. New York, New York Botanical Garden.
Gessler, P. E., I. D. Moore, N. J. McKenzie and P. J. Ryan (1995). "Soil-landscape
modelling and spatial prediction of soil attributes." International Journal
Geographical Information Systems 4: 421-432.
Givnish, T. J. (1999). "On the causes of gradients in tropical tree diversity." Ecology
87: 193-210.
Gleason, H. A. (1922). "On the relation between species and area." Ecology 3: 158162.
Gobin, A., P. Campling and J. Feyen (2001). "Soil-landscape modelling to quantify
spatial variability of soil texture." Physics and Chemistry of the Earth, Part B:
Hydrology, Oceans and Atmosphere 26(1): 41.
Gonzalez, J. (2000). "Monitoring cloud interception in a tropical montane cloud forest
of the South-Western Colombian Andes." Advances in Environmental
Monitoring and Modelling 1(1): 97-117.
Grainger, A. (1993). "Rates of deforestation in the humid tropics: estimates and
measurements." The Geographical Journal 159(1): 33-44.
Grubb, P. (1977). "Control of forest growth and distribution on wet tropical
mountains: with special reference to mineral nutrition." Annual Review of
Ecology and Systematics 8: 83-107.
Grytnes, J. A. (2003). "Species-richness patterns of vascular plant along seven
altitudinal gradients in Norway." Ecography 26: 291-300.
Hafkenscheid, R. (2000). Hydrology and biogeochemistry of tropical montane rain
forests of contrasting stature in the Blue Mountains, Jamaica., Vrije
University. PhD Thesis.
Hall, J. and M. Swaine (1976). "Classification and ecology of closed-canopy forest in
Ghana." Journal of Ecology 64: 913-951.
Hamilton, L. S., F. N. Scatena and (ed) (1995). Tropical montane cloud forests. New
York, N.Y (USA), Springer Verlag.
Harms, K. E., R. Condit, S. P. Hubbell and R. B. Foster (2001). "Habitat associations
of trees and shrubs in a 50-Ha neotropical forest plot." Journal of Ecology 89:
947-959.
Harner, R. F. and K. T. Harper (1976). "The role of area, heterogeneity, and
favorability in plant species diversity of pinyon-juniper ecosystems." Ecology
57: 1254-1263.

354

Hawkins, B. A. and J. Diniz-Filho (2004). "'Latitude' and geographic patterns in
species richness." Ecography 27(2): 268-272.
Hawkins, B. A., R. Field, H. V. Cornell, D. J. Currie, J.-F. Guegan, D. M. Kaufman,
J. T. Kerr, G. G. Mittelbach, T. Oberdorff, E. M. OBrien, E. E. Porter and J. R.
G. Turner (2003). " Energy, Water, And Broad-Scale Geographic Patterns of
Species Richness." Ecology 84(12): 3105-3117.
He, F., P. Legendre and J. V. LaFrankie (1996). "Spatial pattern of diversity in a
tropical rain forest in Malaysia." Journal of Biogeography 23: 57-74.
Heaney, A. and J. Proctor, Eds. (1989). Chemical elements in littler in forests on
Volcán Barva, Costa Rica. Mineral Nutrients in Tropical Forest and Savanna
Ecosystems, Special Publications of the British Ecological Society No. 9.
Hedrick, P. W. (1986). "Genetic polymorphism in heterogenous environments: A
decade later." Annual Review of Ecology and Systematics 17: 535-566.
Hedrick, P. W., M. E. Ginevan and E. P. Ewing (1976). "Genetic polymorphism in
heterogenous environments." Annual Review of Ecology and Systematics 7: 132.
Heikkinen, R. K. and S. Neuvonen (1997). "Species richness of vascular plants in the
subarctic landscape of northern Finland: modelling relationships to the
environment." Biodiversity and Conservation 6: 1181-1201.
Hill, J. L. and R. A. Hill (2001). "Why are tropical rain forests so species rich?
Classifying, reviewing and evaluating theories." Progress in Physical
Geography 25(3): 326-354.
Howe, H. F. and J. Smallwood (1982). "Ecology of seed dispersal." Annual Review of
Ecology and Systematics 13: 201-228.
Hubbell, S. P. (1979). "Tree dispersion, abundance and diversity in a tropical dry
forest." Science 203: 1299-1309.
Hubbell, S. P. and P. Foster (1986). Biology, chance and history and the structure of
tropical rain forest tree communities. Community Ecology. in J. Diamond and
T. Case. New York, Harper and Row: 314-329.
Hubbell, S. P., R. B. Foster, S. T. O'Brien, K. E. Harms, R. Condit, B. Wechsler, S. J.
Wright and S. Loo de Lao (1999). "Light-Gap Disturbances, Recruitment
limitation and tree diversity in a Neotropical Forest." Nature 283: 554-557.
Huston, M. (1980). "Soil nutrients and tree species richness in Costa Rican forests."
Journal of Biogeography 7: 147-157.
Huston, M. (1982). The effect of soil nutrients and light on tree growth and
interactions during tropical forest succession: Experiments in Costa Rica,
University of Michigan. PhD Thesis.

355

Huston, M. A. (1979). "A general hypothesis of species diversity." American
Naturalist 113: 81-101.
Huston, M. A. (1994). Biological Diversity: The coexistence of species on changing
landscapes. Cambridge, Cambridge University Press.
Hutchinson, G. E. (1959). "Homage to Santa Rosalia or Why are there so many kinds
of animals." American Naturalist 93: 145-159.
Hutchinson, M. (1988). Calculation of hydrologically sound digital elevation models.
Third International Symposium on Spatial Data Handling, Columbus, Ohio,
International Geographical Union.
Hutchinson, M. (1989). "A new procedure for gridding elevation and stream line data
with automatic removal of spurious pits." Journal of Hydrology 106: 211-232.
IaVH (1997). Bio-Diversidad: mapa de prioridades de conservaci›n en Colombia.
Bogota, Colombia, Institude Alexander von Humboldt.
Ingerpuu, N., K. Vellak, T. Kukk and M. Partel (2001). "Bryophyte and vascular plant
species richness in boreo-nemoral moist forests and mires." Biodiversity and
Conservation 10: 2153-2166.
International, C. (1997). Megadiversity. Washington, Conservation International.
Janzen, D. (1970). "Herbivores and the number of tree species in tropical forests." The
American Naturalist 104: 501-528.
Jarvis, A. (2000). "Measuring and Modelling the Impact of Land-use change in
tropical hillsides: The role of cloud interception to epiphytes." Advances in
Evironmental Modelling and Monitoring 1(1): 118-148.
Jarvis, A., J. Rubiano, A. Nelson, A. Farrow and M. Mulligan (2004). Practical use of
SRTM data in the tropics: Comparisons with digital elevation models
generated from cartographic data. Working Document no. 198. Cali,
International Centre for Tropical Agriculture (CIAT): 32.
Johnson, K. H., K. A. Vogt, H. J. Clark, O. J. Schmitz and D. J. Vogt (1996).
"Biodiversity and the productivity and stability of ecosystems." Trends in
Ecology and Evolution 11(9): 372-377.
Johnson, N. C. (1995). Biodiversity in the balance: Approaches to setting geographic
conservation priorities. Washington DC, Biodiversity Support Program.
Junk, W., P. Bayley and R. Sparks (1989). The 'flood pulse' concept in river-system.
Proceedings of the International Large River Symposium, Canadian Special
Publication of Fish and Aquatic Science No. 106.
Kappelle, M. (2004). Tropical Forests: Tropical Montane Cloud Forests.
Encyclopedia of Forest Sciences Vol. 4. in J. Burley, Evams J and J.
Younquist. Oxford, UK, Elsevier: 1782-1793.

356

Kennedy, T. A., NaeemShahid, K. M. Howe, J. M. Knops, D. Tilman and P. Reich
(2002). "Biodiversity as a barrier to ecological invasion." Nature 417: 636638.
Kessler, M. (2002). "The elevational gradient of Andean plant endemism: varying
influences of taxon-specific traits and topography at different taxonomic
levels." Journal of Biogeography 29: 1159-1165.
Kessler, M., S. K. Herzog and J. Fjeldsa (2001). "Species richness and endemism of
plant and bird communities along two gradients of elevation, humidity and
land use in the Bolivian Andes." Diversity and Distributions 7: 61-77.
Kitayama, K. and S. Aiba (2002). "Ecosystem structure and productivity of tropical
rain forests along altitudinal gradients with contrasting soil phosphorus pools
on Mount Kinabalu, Borneo." Journal of Ecology 90(1): 37-51.
Knops, J. M., D. Tilman, N. M. Haddad, S. Naeem, C. E. Mitchell, J. Haarstad, M. E.
Ritchie, K. M. Howe, P. B. Reich, E. Siemann and J. Groth (1999). "Effects of
plant species richness on invasion, disease outbrakes, insect abundances and
diversity." Ecology Letters 2: 286-293.
Kondoh, M. (2000). "Unifying the relationships of species richness to productivity
and disturbance." Proceedings of the Royal Society London B 268: 269-271.
Koponen, P., P. Nygren, D. Sabatier, A. Rousteau and E. Saur (2004). " Tree species
diversity and forest structure in relation to microtopography in a tropical
freshwater swamp forest in French Guiana." Plant Ecology 173: 17-32.
Kremen, C. (1992). "Assessing the indicator properties of species assemblages for
natural areas monitoring." Ecological Applications 2(2): 203-217.
Kress, W. J., W. R. Heyer, P. Acevedo, J. Coddington, D. Cole, T. L. Erwin, B. J.
Meggers, M. Pogue, R. W. Thorington, Vari.RP, M. J. Weitzman and S. H.
Weitzman (1998). "Amazonian biodiversity: assessing conservation priorities
with taxonomic data." Biodiversity and Conservation 7: 1577-1587.
Lambin, E. F. and D. Ehrlich (1997). "The identification of tropical deforestation
fronts at broad spatial scales." International Journal of Remote Sensing
18(17): 3551-3568.
Lawton, J. H., D. E. Bignell, B. Bolton, G. F. Bloemers, P. Eggleton, P. M.
Hammond, M. Hodda, R. D. Holt, T. B. Larsen, N. A. Mawdsley, N. E. Stork,
D. S. Srivastava and A. D. Watt (1998). "Biodiversity inventories, indicator
taxa and effects of habitat modification in tropical forest." Nature 391: 72-76.
Lees, D., C. Kremen and L. Andriamampianina (1999). "A null model for species
richness gradients: bounded range overlap of butterflies and other rainforest
endemics in Madagascar." Biological Journal of the Linnean Society 67: 529584.
Leigh, E. (1999). Tropical Forest Ecology: A View from Barro Colorado Island.
Oxford, UK, Oxford University Press.
357

Leimbeck, R. M., R. Vlaencia and h. Balslev (2004). " Landscape diversity patterns
and endemism of Araceae in Ecuador." Biodiversity and Conservation 13:
1755 -1779.
Letts, M. G. (2003). Carbon Assimilation and Productivity in a North-west Andean
Tropical Montane Cloud Forest. Department of Geography. London, King's
College London. PhD Thesis: 352.
Li, H. and J. F. Reynolds (1994). "A Simulation Experiment to Quantify Spatial
Heterogeneity in Categorical Maps." Ecology 75(8): 2446-2455.
Lillesand, T. M. and R. W. Kiefer (1999). Remote Sensing and Image Interpretation,
4th Edition. London, Wiley.
Lomolino, M. V. (2000). "Ecology's most general, yet protean pattern: the speciesarea relationship." Journal of Biogeography 27: 17-26.
Long, A. (1995). Restricted-range and threatened bird species in tropical montane
cloud forests. Tropical Montane Cloud Forests. in L. Hamilton, J. Juvik and F.
Scatena. New York, Springer-Verlag: 47-65.
Lookingbill, T. and D. Urban (2004). " An empirical approach towards improved
spatial estimates of soil moisture for vegetation analysis." Landscape Ecology
19: 417-433.
Loope, L. L. and T. W. Giambelluca (1998). "Vulnerability of Island Tropical
Montane Cloud Forests to Climate Change, with Special Reference to East
Maui, Hawaii." Climatic Change 39(2-3): 503-517.
Lundholm, J. T. and D. W. Larson (2003). "Relationships between spatial
environmental heterogeneity and plant species diversity on a limestone
pavement." Ecography 26: 715-722.
Macarthur, R. (1965). "Patterns of species diversity." Biological Review 40: 510-533.
Macarthur, R. (1969). "Patterns of communities in the tropics." Biological Journal of
the Linnean Society 1: 19-30.
Mackey, R. L. and D. J. Currie (2001). "The diversity-disturbance relationship: Is it
generally strong and peaked?" Ecology 82(12): 3479-3492.
Magurran, A. (2004). Measuring Biological Diversity. New York, Blackwell.
Maio Mantovani, A. C. and A. W. Setzer (1997). "Deforestation detection in the
Amazon with an AVHRR-based system." International Journal of Remote
Sensing 18(2): 273-286.
Maranon, T. and L. V. Garcia (1997). "The relationship between diversity and
productivity in plant communities: Facts and artifacts." Journal of Ecology 85:
95-96.

358

McNeely, J. A., K. R. Miller, W. V. Reid, R. A. Mittermeier and T. B. Werner (1990).
Conserving the world's biological diversity. Washington, IUCN, WRI, CI,
WWF-US, The World Bank.
Meer, P. v. d. and F. Bongers (1996). "Patterns of tree-fall and branch-fall in a
tropical rain forest in Guiana." Journal of Ecology 84: 19-29.
Meer, P. v. d., F. J. Sterck and F. Bongers (1998). "Tree seedling performance in
canopy gaps in a tropical rain forest at Nouragues, French Guiana." Journal of
Tropical Ecology 14: 119-137.
Melillo, J. M., R. A. Houghton, D. W. Kicklighter and A. D. McGuire (1996).
"Tropical deforestation and the global carbon budget." Annual Review of
Energy and Environment 21: 293-310.
Molino, J.-F. and D. Sabatier (2001). "Tree diversity in tropical rain forests: A
validation of the intermediate disturbance hypothesis." Science 294: 17021704.
Muller, R., C. Nowicki, W. Barthlott and P. Ibisch (2003). "Biodiversity and
endemism mapping as a tool for regional conservation planning - case study of
the Pleurothallidinae (Orchidaceae) of the Andean rain forests in Bolivia."
Biodiversity and Conservation 12: 2005-2024.
Mulligan, M. (1999). "A simple hydrological model for tropical montane cloud
forests." Department of Geography Occasional Paper No.48.
Mulligan, M. (2004). Modelling catchment hydrology. Environmental modelling:
Finding simplicity in Complexity. in J. Wainwright and M. Mulligan.
Chichester, England, John Wiley and Sons: 107-121.
Mulligan, M. and A. Jarvis (submitted). "Laboratory Simulation of Cloud Interception
by Epiphytes and Implications for the Hydrology of the Tambito Experimental
Cloud Forest." Journal of Hydrology.
Myers, N., R. A. Mittermeier, C. G. Mittermeier, G. A. B. Da Fonseca and J. Kent
(2000). "Biodiversity hotspots for conservation priorities." Nature 403: 853858.
Naeem, S., L. J. Thompson, S. P. Lawler, J. H. Lawton and R. M. Woodfin (1994).
"Declining biodiversity can alter the performance of ecosystems." Nature 368:
734-737.
Nelson, A. (2001). "Analysing data across geographic scales in Honduras: detecting
levels of organisation within systems." Agriculture Ecosystems and
Environment 85(1-3): 107-131.
Nelson, A. (2004). The spatial analysis of socio-economic and agricultural data across
geographic scales: Examples and applications in Honduras and elsewhere.
School of Geography. Leeds, University of Leeds. PhD thesis.

359

Nelson, B. W., V. Kapos, J. B. Adams, W. J. Oliveira and O. P. G. Braun (1994).
"Forest Disturbance by Large Blowdowns in the Brazilian Amazon." Ecology
75(3): 853-858.
Newbery, D., J. Gartlan, D. McKey and P. Waterman (1986). "The influence of
drainage and soil phosphorus on the vegetation of Douala-Edea Forest
Reserve, Cameroun." Vegetatio 65: 149-162.
Nieder, J., j. Prosperi and G. Michaloud (2001). "Epiphytes and their contribution to
canopy diversity." Plant Ecology 153: 51-63.
Nilsson, C., G. Grelsson, M. Johansson and U. Sperens (1989). "Patterns of plant
species richness along riverbanks." Ecology 70(1): 77-84.
Nor, S. (2001). "Elevational diversity patterns of small mammals on Mount Kinabalu,
Sabah, Malaysia." Global Ecology and Biogeography 10: 41-62.
Odum, H. (1970). Rain forest structure and mineral cycling homeostasis. A tropical
rainforest: a study of irradiation and ecology at El Verde, Puerto Rico. in H.
Odum and R. Pigeon. Oak Ridge, USA, U.S. Atomic Energy Commission,
Division of Technical Information.
Oliveira, A. d. and S. A. Mori (1999). "A central Amazonian terra firma forest. I High
tree species richness on poor soils." Biodiversity and Conservation 8: 12191244.
Pack, R. T., D. G. Tarboton and C. N. Goodwin (1988). The SINMAP approach to
terrain stability mapping. 8th Congress of the International Association of
Engineering Geology, Vancouver.
Palmiotto, P. A., S. J. Davies, K. A. Vogt, M. S. Ashton, D. J. Vogt and P. S. Ashton
(2004). "Soil-related habitat specialization in dipterocarp rain forest tree
species in Borneo." Journal of Ecology 92(4): 609.
Palmiotto, P. A., K. A. Vogt and P. Ashton (2000). Edaphic specialisation and
nutrient-use efficiency: Mechanisms contributing to high species diversity at
Lambir. Inside CTFS. Summer 2000.
Paradis, E., S. Baillie and J. Sutherland (2002). "Modelling large-scale dispersal
distances." Ecological Modelling 151: 279-292.
Pausas, J. G., J. Carreras, A. Ferre and X. Font (2003). "Coarse-scale plant species
richness in relation to environmental heterogeneity." Journal of Vegetation
Science 14: 661-668.
Peet, R. K. (1974). "The measurement of species diversity." Annual Review of
Ecology and Systematics 5: 285-307.
Pendry, C. A. and J. Proctor (1996). "The causes of altitudinal zonation of rain forests
on Bukit Belong, Brunei." Journal of Ecology 84: 407-418.

360

Pendry, C. A. and J. Proctor (1997). "Altitudinal zonation of rain forest on Bukit
Belalong, Brunei: soils, forest structure and floristics." Journal of Tropical
Ecology 13: 221-241.
Peres, C. A. and C. Baider (1997). "Seed dispersal, spatial distribution and population
structure of Brazilnut trees (Bertholletia excelsa) in southeastern Amazonia."
Journal of Tropical Ecology 13: 595-616.
Phillips, O., A. Gentry, P. Hall, S. Sawyer and R. Vasquez (1994). "Dynamics and
species richness of tropical rain forests." Proceedings of the National Academy
of Science 91: 2805-2809.
Phillips, O. L., P. Nunez Vargas, A. L. Monteagudo, C. A. Pena, Z. M. E. Chuspe, S.
W. Galiano, M. Yli-Halla and S. Rose (2003). " Habitat association among
Amazonian tree species: a landscape-scale approach." Journal of Ecology 91:
757 -775.
Phillips, O. L. and P. H. Raven (1997). A strategy for Sampling Neotropical Forests.
Neotropical Biodiversity and Conservation. in A. C. Gibson. Los Angeles,
University of California.
Phillips, O. L., R. Vasquez Martinez, P. Nunez Vargas, A. Lorenzo Monteagudo, M.E. Chuspe Zans, W. Galiano Sanchez, A. Pena Cruz, M. Timana, M. Yli-Halla
and S. Rose (2003). " Efficient plot-based floristic assessment of tropical
forests." Tropical Ecology 19: 629 -645.
Pickett, S. and P. White (1985). The ecology of natural disturbance and patch
dynamics. New York, Academic Press.
Pimm, S. L., G. J. Russell, J. G. Gittleman and T. M. Brooks (1995). "The future of
Biodiversity." Science 269: 347-350.
Pitman, N. (2000). A large-scale inventory of two Amazonian tree communities.
Department of Botany, Duke University. PhD Thesis: 220.
Pitman, N., J. Terborgh, M. Silman and P. Nunez (1999). "Tree species distributions
in an upper Amazonian forest." Ecology 80: 2651-2661.
Pitman, N. C. A., J. W. Terborgh, M. R. Silman, P. Nunez, D. A. Neill, E. C. Ceron,
W. A. Palacios and M. Aulestia (2001). "Dominance and distribution of tree
species in upper Amazonian Terra Firme forests." Ecology 82(8): 2101-2117.
Poorter, L., Jans L, F. Bongers and R. Van Rampaey (1994). "Spatial distribution of
gaps along three catenas in the moist forest of Tai National Park, Ivory Coast."
Journal of Tropical Ecology 10: 335-398.
Potts, M. D., P. S. Ashton, L. S. Kaufman and J. B. Plotkin (2002). "Habitat patterns
in tropical rain forests: A comparison of 105 plots in Northwest Borneo."
Ecology 83(10): 2782-2797.
Poulsen, A. and H. Balslev (1991). "Abundance and cover of ground herbs in an
Amazonian rain forest." Journal of Vegetative Science 2: 315-322.
361

Pounds, J., M. Fogden and J. Campbell (1999). "Biological response to climate
change on a tropical mountain." Nature 398: 611-615.
Prance, G. (1982). Forest refuges: Evidence from woody angiosperms. Biological
diversification in the tropics. in G. e. Prance. New York, Columbia University
Press: 137-158.
Prendergast, J. R., R. M. Quinn, J. H. Lawton, B. C. Eversham and D. W. Gibbons
(1993). "Rare species, the coincidence of diversity hotspots and conservation
strategies." Nature 365: 335-337.
Proctor, J., J. M. Anderson, P. Chai and H. W. Vallack (1983). "Ecological studies in
four contrasting lowland rain forests in Gunung Mulu National Park, Sarawak:
I. Forest environment, structure and floristics." Journal of Ecology 71(1): 237260.
Rahbek, C. (1997). "The relationship among area, elevation, and regional species
richness in neotropical birds." The American Naturalist 149(5): 875-902.
Rasanen, M., J. Salo and R. Kallioloa (1986). "Fluvial perturbance in the western
Amazon basin - regulation by long-term Sub Andean tectonics." Science 238:
1398-1401.
Raufaste, N. and F. Rousset (2001). "Are partial Mantel tests adequate?" Evolution
55: 1703-1705.
Read, J. M., D. B. Clark, E. M. Venticinque and M. P. Moreiras (2003). "
METHODOLOGICAL INSIGHTS Application of merged 1-m and 4-m
resolution satellite data to research and management in tropical forests."
Applied Ecology 40: 592-600.
Rey Benayas, J. M. and S. M. Scheiner (2002). "Plant diversity, biogeography and
environment in Iberia: Patterns and possible causal factors." Journal of
Vegetation Science 13: 245-258.
Richards, P. (1996). The tropical rain forest, an ecological study, Second edition.
Cambridge, Cambridge University Press.
Ricklefs, R. (1977). "Environmental heterogeneity and plant species diversity: a
hypothesis." The American Naturalist 111: 376-381.
Rincon-Romero, M. (2001). Modelling the hydrological sensitivity of tropical hillside
to land use change. Department of Geography, King's College London. PhD
Thesis.
Rincon-Romero, M. (2001). Modelling the hydrological sensitivity of tropical hillside
to land use change. Department of Geography. London, King's College
London.
Robert, A. and M.-A. s. Moravie (2004). " Topographic variation and stand
heterogeneity in a wet evergreen forest of India." Tropical Ecology 19: 697707.
362

Romoleroux, K., R. Foster, R. Valencia, R. Condit, h. Balslev and E. Losos (1995).
Arboles y arbustos (dap >1) encontrados en dos hectareas de un bosque de la
Amazonia ecuatoriana, Pontificia Universidad Catolica del Ecuador, Quito.
Rosenzweig, M. (1995). Species diversity in space and time. Cambridge, Cambridge
University Press.
Ryan, P., N. McKenzie, D. O'Connell, A. Loughead, P. Leppert, D. Jacquier and L.
Ashton (2000). "Integrating forest soils information across scales: Spatial
prediction of soil properties under Australian forests." Forest Ecology and
Management 138: 139-157.
Salo, J., R. Kallioloa, I. Hakkinen, Y. Makinen, P. Niemela, Puhakka M and P. Coley
(1986). "River dynamics and the diversity of Amazon lowland forest." Nature
322: 254-258.
Salvador-Van Eysenrode, D., J. Bogaert, P. Van Hecke and I. Impens (2000). "Forest
canopy perforation in time and space in Amazonian Ecuador." Acta
Oecologica 21(4-5): 285-291.
Sanders, N. J. (2002). "Elevational gradients in ant species richness: area, geometry,
and Rapoports's rule." Ecography 25: 25-32.
Schaik, v. C. and M. E (1985). "Spatial variation in the structure and litterfall of a
Sumatran rain forest." Biotropica 17: 196-205.
Schmid, B. (2002). "The species richness-productivity controversy." Trends in
Ecology and Evolution 17(3): 113-114.
Schoener, T. (1976). The species-area relation within archipelagos: models and
evidence from island land birds. Proceedings of the XVI International
Ornithological Congress, Australian Academy of Sciences, Canberra.
Shaw, M. (1995). "Simulation of population expansion and spatial pattern when
individual dispersal distributions do not decline exponentially with distance."
Proceedings of the Royal Society of London B 259: 243-248.
Sheil, D. and D. F. Burslem (2003). "Disturbing hypotheses in tropical forests."
Trends in Ecology and Evolution 18(1): 18-26.
Siemann, E., D. Tilman, J. Haarstad and M. E. Ritchie (1998). "Experimental tests of
the dependence of arthropod diversity on plant diversity." The American
Naturalist 152(5): 738-750.
Simonson, S. E., P. A. Opler, T. J. Stohlgren and G. W. Chong (2001). "Rapid
assessment of butterfly diversity in a montane landscape." Biodiversity and
Conservation 10: 1369-1386.
Skole, D. and C. Tucker (1993). "Tropical deforestation and habitat fragmentation in
the Amazon: Satellite data from 1978 to 1988." Science 260(5116): 19051910.

363

Smith, T. and M. Huston (1989). "A theory of the spatial and temporal dynamics of
plant communities." Vegetatio 83: 49-69.
St-Louis, V., M.-J. Fortin and A. Desrochers (2004). "Spatial association between
forest heterogeneity and breeding territory boundaries of two forest
songbirds." Landscape Ecology 19(6): 591-601.
Stattersfield, A. J., M. J. Crosby, A. J. Long and D. C. Wege (1997). Endemic Bird
Areas of the World: Priorities for biodiversity Conservation. Cambridge,
Birdlife International.
Stevens, G. (1992). "The elevational gradient in altitudinal range: an extension of
Rapoport's latitudinal rule to altitude." The American Naturalist(893-911).
Still, C., P. Foster and S. Schneider (1999). "Simulating the effects of climate change
on tropical montane cloud forests." Nature 398: 608-610.
Su, J. C., D. M. Debinski, M. E. Jakubauskas and K. Kindscher (2004). "Beyond
species richness: COmmunity similarity as a measure of cross-taxon
congruence for coarse-filter conservation." Conservation Biology 18(1): 167173.
Sugden, A. M. (1982). "The Ecological, Geographic, and taxonomic relationships of
the flora of an isolated Colombian cloud forest, with some implications for
island biogeography." Journal of the Arnold Arboretum 63: 31-61.
Sullivan, J., A. Teramura and L. Ziska (1992). "Variation in UV-B sensitivity in
plants from a 3000m elevational gradient in Hawaii." American Journal of
Botany 79(7): 737-743.
Takyu, M., S.-I. Aiba and K. Kitayama (2002). "Effects of topography on tropical
lower montane forests under different geological conditions on Mount
Kinabalu, Borneo." Plant Ecology 159(1): 35.
Tarboton, D. (2004). "Terrain Analysis Using Digital Elevation Models (TauDEM)".
available online at http://hydrology.neng.usu.edu/taudem/.
Tateno, R. and H. Takeda (2003). "Forest structure and tree species distribution in
relation to topography-mediated heterogeneity of soil nitrogen and light at the
forest floor." Ecological Research 18: 559-571.
Teramura, A. (1983). "Effects of ultraviolet-B radiation on the growth and yield of
crop plants." Physiology Plantarum 58(3): 415-427.
Terborgh, J. (1977). "Bird species diversity on an Andean elevational gradient."
Ecology 58: 1007-1019.
Terborgh, J., R. B. Foster and P. V. Nunez (1996). "Tropical tree communities: A test
of the nonequilibrum hypothesis." Ecology 77(2): 561-567.

364

Terborgh, J. and J. Mathews (1999). "Partitioning of the understorey light
environment by two Amazonian treelets." Journal of Tropical Ecology 15:
751-763.
Tews, J., U. Brose, V. Grimm, K. Tielborger, M. Wichmann, M. Schwager and F.
Jeltsch (2004). "Animal species diversity driven by habitat
heterogeneity/diversity: the importance of keystone structures." Journal of
Biogeography 31: 79-92.
Tilman, D. (1982). Resource competition and community structure. Princeton,
Princeton University Press.
Tomlin, C. D. (1993). Cartographic Modelling. Geographic Information Systems:
Principles and applications. in D. J. G. Maguire, M.F. Rwind, D.W. New
York, John Wiley and Sons: 361-374.
Tomlinson, P. B. (1987). "Architecture of Tropical Plants." Annual Review of Ecology
and Systematics 18: 1-21.
Tothmeresz, B. (1993). "DivORD 1.50: A program for diversity ordering." Tiscia 27:
33-44.
Toutin, T. (2002). "Three-Dimensional Topographic Mapping With ASTER Stereo
Data in Rugged Topography." Geoscience and remote sensing 40(10): 22412247.
Trichon, V., J. M. Walter and Y. Laumonier (1998). "Identifying spatial patterns in
the tropical rain forest structure using hemispherical photographs." Plant
Ecology 137: 227-244.
Tuomisto, H., K. Ruokolainen, M. Aguilar and A. Sarmiento (2003). " Floristic
patterns along a 43-km long transect in an Amazonian rain forest." Journal of
Ecology 91: 743-756.
Twery, M. J., G. A. Elmes and C. B. Yuill (1991). "Scientific exploration with an
intelligent GIS: Predicting species composition from topography." AI
Applications 5(2): 45-53.
Uhl, C., K. Clark, N. Dezzeo and P. Maquirino (1988). "Vegetation Dynamics in
Amazonian Treefall Gaps." Ecology 69(3): 751-763.
Valencia, R., H. Balslev and C. G. Mino (1994). "High tree alpha-diversity in
Amazonian Ecuador." Biodiversity and Conservation 3: 21-28.
Valencia, R., R. B. Foster, G. Villa, R. Condit, J.-C. Svenning, C. Hernandez and K.
Romoleroux (2004). " Tree species distributions and local habitat variation in
the Amazon: large forest plot in eastern Ecuador." Journal of Ecology 92: 214229.
Vasquez, J. A. and T. J. Givnish (1998). "Altitudinal gradients in tropical forest
composition, structure and diversity in the Sierra de Manantlan." Journal of
Ecology 86: 999-1020.
365

Vetaas, O. R. and J. A. Grytnes (2002). "Distribution of vascular plant species
richness and endemic richness along the Himalayan elevation gradient in
Nepal." Global Ecology and Biogeography 11: 291-301.
Vitousek, P. M. (1984). "Litterfall, nutrient cycling and nutrient limitation in tropical
forests." Ecology 65(1): 285-298.
Vitousek, P. M. and J. Sanford (1986). "Nutrient cycling in moist tropical forests."
Annual Review of Ecology and Systematics 17: 137-167.
Volis, S., Y. Anikster, L. Olsvig-Whittaker and S. Mendlinger (2004). "The influence
of space in genetic-environmental relationships when environmental
heterogeneity and seed dispersal occur at similar scale." The American
Naturalist 163(2): 312-327.
Vormisto, J. (2002). "Palms as rainforest resources: how evenly are they distributed in
Peruvian Amazonia?" Biodiversity and Conservation 11: 1025-1045.
Vormisto, J., T. Hanna and J. Oksanen (2004). "Palm distribution patterns in
Amazonian rainforests: What is the role of topographic variation?" Journal of
Vegetation Science 15: 485-494.
Vormisto, J., O. L. Phillips, K. Ruokolainen, H. Tuomisto and R. Vasquez (2000). "A
comparison of fine-scale distribution patterns of four plant groups in an
Amazonian rainforest." Ecography 23: 349-359.
Waide, R. B., M. R. Willig, C. F. Steiner, G. Mittelbach, L. Gough, S. I. Dodson, G.
P. Juday and R. Parmenter (1999). "The relationship between productivity and
species richness." Annual Review of Ecology and Systematics 30: 257-300.
Wand, M. P. and M. C. Jones (1995). Kernel Smoothing. London, Chapman and Hall.
Wardle, D. A. (1999). "Biodiversity, ecosystems and interactions that transcend the
interface." Trends in Ecology and Evolution 14(4): 125-127.
Wardle, D. A., R. D. Bardgett, Klironomos John N, H. Setala, W. H. van der Putten
and D. H. Wall (2004). "Ecological linkages between aboveground and
belowground biota." Science 304: 1629-1633.
Webb, C. O. and D. R. Peart (2000). "Habitat associations of trees and seedlings in a
Bornean rain forest." Journal of Ecology 88: 464-478.
Webb, E. L., B. J. Stanfield and M. L. Jensen (1999). "Effects of topography on
rainforest tree community structure and diversity in American Samoa, and
implications for frugivore and nectarivore populations." Journal of
Biogeography 26: 887-897.
Wehncke, E., S. Hubbell, R. Foster and J. Dalling (2003). "Seed dispersal patterns
produced by white-faced monkeys: implications for the dispersal limitation of
neotropical tree species." Journal of Ecology 91: 677-685.

366

Weiher, E. (1999). "The combined effects of scale and productivity on species
richness." Journal of Ecology 87: 1005-1011.
Wenny, D. G. (1999). "Two-stage dispersal of Guarea glabra and G. kunthiana
(Meliaceae) in Monteverde, Costa RIca." Journal of Tropical Ecology 15:
481-496.
Whitmore, T. (1984). Tropical rain forests of the Far-East. Oxford, Clarendon Press.
Whitmore, T. (1998). An introduction to tropical rain forests. Oxford, Oxford
University Press.
Whitmore, T. C., R. Peralta and K. Brown (1985). "Total species count in a Costa
Rican tropical forest." Journal of Tropical Ecology 1: 375-378.
Whittaker, R. H. and W. A. Niering (1975). "Vegetation of the Santa Catalina
mountains, Arizona. V biomass, production and diversity along the elevation
gradient." Ecology 56: 771-790.
Whittaker, R. J. (1999). "Scaling, Energetics and Diversity." Nature 401: 865-867.
Wilson, E. O. (1992). The diversity of life. Massachusetts, Belknap Press.
Wolters, V. (2001). "Biodiversity and soil animals and its function." European
Journal of Soil Biology 37: 221-227.
Wong, Y. and T. Whitmore (1970). "On the influence of soil properties on species
distribution in a Malayan lowland dipterocarp rain forest." Malaysian Forester
33: 42-54.
Wood, J. (1996). The Geomorphological Characterisation of Digital Elevation
Models. School of Geography. Leicester, University of Leicester.
Wood, J. (1996). Scale-based Characterisation of Digital Elevation Models.
Innovations in GIS 3: Selected papers from the Third National Conference on
GIS Research UK (GISRUK). in D. Parker. London, Taylor and Francis: 163176.
Wood, J. (1998). Modelling the Continuity of Surface Form Using Digital Elevation
Models. International sumposium on spatial data handling; SDH '98,
Vancouver, International Geographical Union Geographic Information Scince
Study Group.
Wood, J. (2004). Landserf V2.0. London, Department of Information Science, City
University.
Worm, B., K. L. Heike, H. Hillebrand and U. Sommer (2002). "Consumer versus
resource control of species diversity and ecosystem functioning." Nature 417:
848-851.
Wright, S. (1992). "Seasonal drought, soil fertility and the species density of tropical
forest plant communities." Trends in Ecology and Evolution 7: 260-263.

367

Xu, X. and M. Ridout (2001). "Effects of prevailing wind direction on spatial
statistics of plant disease epidemics." Journal of Phytopathology 149: 155166.
Yu, J. Terborgh and Potts (1998). "Can high tree species richness be explained by
Hubbell's null model?" Ecology Letters 1(3): 193-199.
Yumoto, T., K. Kimura and A. Nishimura (1999). " Estimation of the retention times
and distances of seed dispersed by two monkey species, Alouatta seniculus
and Lagothrix lagotricha, in a Colombian forest." Ecological Research 14(2):
179-191.
Zadroga, F. (1981). The hydrological importance of the montane cloud forests area of
Costa Rica. Tropical Agricultural Hydrology. in R. Lal and E. Russell. New
York, John Wiley: 59-73.
Zar, J. (1999). Biostatistical Analysis. New Jersey, Prentice Hall.
Zimmermann, N. (2004). The toposclae and topoclass AML,
http://www.wsl.ch/staff/niklaus.zimmermann/programs/aml4_1.html.
Ziska, L. (1996). "The potential sensitivity of tropical plants to increased ultraviolet-B
radiation." Journal of Plant Physiology 148(1-2): 35-41.
Ziska, L., A. Teramura and J. Sullivan (1992). "Physiological sensitivity of native
plants collected along an elevational gradient in Hawaii to increased UV-B
radiation." American Journal of Botany 79(8): 863-871.

368

8

Appendix 1

Acquisition and processing of digital high resolution aerial imagery over tropical
forests
This Appendix shows some of the preliminary results achieved from a pilot study at
using high resolution images to “monitor” tree canopy diversity.

The regular presence of cloud and steep topographic characteristics of the montane
tropical forest found in Reserva Tambito make aerial photography an unrealistic
means of assessing biodiversity.

Logistically it is very complicated to take the

necessary data in Reserva Tambito, so aerial photography is only to be applied to
Tiputini Biodiversity Station.

Data Collection

Images were acquired using two complementary approaches, at both high and low
altitude, designed to provide images with different spatial coverages and resultant
differences in resolution. The first approach uses aircraft based aerial photography to
acquire high altitude/low resolution images for the whole of the TBS field site.
Within the Tiputini Biodiversity Station the second approach concentrated on the
acquisition of high resolution (but low spatial coverage) images of the forest canopy,
especially over ground-based diversity plots through the use of a tethered helium
balloon.

Instruments and technical specifications

369

The instruments used for the photo acquisition vary depending on the aerial or
tethered techniques, but all take full colour photos (red, green, blue) from digital
cameras.

The aircraft based acquisition of aerial photos is not limited by weight, and so two
cameras were used to ensure varying image resolutions and qualities. These were a
Canon Digital Video camera, and a Kodak DCS420 digital camera. The Canon
digital video takes 30 frames per second in DV format to a 1-hour DV tape.
Individual frames of size 768 x 567 pixels were extracted from the resultant video.
The recorded video was viewed externally by a small LCD TV monitor from within
the aircraft to ensure that the imagery was properly acquired.
The digital camera used in the aerial photography was a Kodak DCS420
professionally calibrated digital aerial photography camera, which takes images of
size 1524 x 1012 pixels. This camera was attached to a frame rate generator to
automatically takes photos at a given time interval. The imagery is written to a
PCMCIA memory card, and requires a minimum of 4 seconds between photos to
completely save the imagery to memory. The camera will hold a maximum of 201
images in memory, before it must be downloaded to computer. These limitations
were important when planning the flight path for image acquisition. Figure 1 shows
the instrumental setup of the apparatus.

370

Figure 111 Instrumental setup for aerial photography acquisition

Camera calibration

Each camera was subjected to a calibration experiment to determine the relationships
between the distance of the sensor from the target and the spatial resolution and
spatial coverage. With this information it was also possible to calculate the time
required between frames (frame-rate) to ensure a 33% overlap in images, and also the
elevation at which blurring was likely to occur. This information is important for
flight planning, in order to ensure that the imagery covers the entire TBS study region.
The results of these calibrations are shown in Figures 112-115.

371

1200

900
CANON Sw ath length
(m)
KODAK Sw ath length
(m)
NIKON Sw ath length
(m)

800

800
700
600
500

600
400
400

300

Image width (m)

Image length (m)

1000

200
200
100
0
0

200

400

600

800

1000

0
1200

Flight Altitude (m)

Figure 112 Ground spatial coverage of images from the three different cameras depending on
elevation of the sensor above the ground surface.

200
180

CANON resolution
(cm)

Resolution (cm)

160
140

KODAK Resolution
(cm)

120

NIKON resolution
(cm)

100
80
60
40
20
0
0

200

400

600

800

1000

1200

Flight Altitude (m)

Figure 113 Resolution of images from the three different cameras depending on elevation of
the sensor above the ground surface.

372

12
CANON time bet frames
w ith 1/3 overlap (s)

Time between frames (secs)

10

KODAK time betw frames
w ith 1/3 overlap (s)
NIKON time betw frames
w ith 1/3 overlap (s)

8

6

4

2

0
0

200

400

600

800

1000

1200

Flight Altitude (m)

Figure 114 Required time between frames to ensure 1/3 overlap in images depending on
elevation of the sensor above the ground surface.

1200
KODAK blurring alt
NIKON blurring alt
CANON blurring alt

Blurring altitude (cm)

1000

800

600

400

200

0
300 600 900 1200 1500 1800 2100 2400 2700 3000 3300 3600

Flight Altitude (m)

Figure 115 Elevation at which blurring is expected to occur for each of the three cameras.
This was calculated based on the spatial resolution of the pixels, the exposure of the camera

373

and the speed of the aeroplane (fixed at 200km/h). If during the exposure time the aeroplane
movement exceeds the equivalent of 1 ground pixel, it was assumed that blurring would be
present in the image.

Aerial photo acquisition

Based on these calibrations, a flight path was constructed which involved image
acquisition at three elevations, producing images with spatial resolution of
approximately 30cm (1200m elevation), 15cm (600m elevation) and 10cm (300m
elevation) with the Kodak DCS420. The 1200m and 600m swaths were designed to
produce continuous stereo images with 33% overlap. The flight path was loaded as a
background file in the Trimble ProXL GPS, and used in flight to navigate. The
planned and the actual flight path are shown in Figure 116.

374

Reserve Region

wnlo

ad

ms
wa t
hs

a do
Dat

600

Flight path
loaded into
Trimble Asset
Surveyor

Lose altitude to 600m

12

600m
swaths

ch
roa
app
m
00

1200m
swaths

Deviation from
Planned flight path

300m
swath

To airport

a do
Dat

oad
wnl

Planned flight path
Actual flight path

Figure 116 Planned and actual flight path for aerial image acquisition over TBS, March 2002.

The images were taken during one single flight on the 11th March, 2003. The sky was
around 80% clear, with just a few isolated clouds at approximately 1100m elevation
above the ground surface. As can be seen in Figure 116, turbulence meant that the
planned flight path could not be exactly maintained, and fuel limitations mean that
only seven out of eleven of the 600m swaths were completed. Due to the cloud
height, the planned 1200m imagery was actually taken at approximately 1100m
elevation above the ground surface. All images were of reasonable quality, and a
large percentage of the reserve was covered with at least one resolution of aerial
imagery.

Helium balloon photo acquisition
375

The helium balloon based photo acquisition is limited by weight. The helium balloon
carries some 7 cubic metres of helium, and is capable of a maximum payload of
approximately 3kg. This limitation means that only the one sensor can be safely
attached, accompanied by servos and remote control receivers and transmitters. In
this case a high-resolution Nikon Coolpix 990 digital camera is used, with a 28mm
lens and resolution of 3.3 mega-pixels with an image size of 2048 x 1536 pixels. The
camera is attached to the balloon, and the balloon held on a cable of up to 200m
length.

There is no form of propulsion so the path of the balloon is at the whim of the wind.
The balloon requires a large clearing to avoid snagging on tree branches, so the
balloon can only be flown from the river, or from some especially large gaps in the
forest floor (Figure 117).

This restricted the potential coverage of the balloon

imagery. The photos were taken by means of a remote control servo attached to the
camera platform. Figure 118 and Figure 119 show the strategy used for the balloon
within the forest, and an image of the camera apparatus.

376

Figure 117 Method for helium balloon image acquisition from forest gaps

Figure 118 The helium balloon in flight

377

Figure 119 Instrumental setup of tethered balloon photo acquisition

At full balloon height (200m) the camera provides images with a pixel size of 5.6cm
and a spatial coverage of 115m x 86m. Images were taken along the river covering
almost continuously some 3 km of river, and extending as far as 100m into the
surrounding forest. The balloon was also flown from a canopy tower, capturing high
resolution images of plot 6, and from a gap in the center of the 1Ha Pitman plot,
capturing approximately 80% of the plot with high resolution imagery.

Georeferencing of images

The method for georeferencing the images vary between the balloon and the aircraft.
The aerial photos were georeferenced using GPS data through an iterative approach,
starting with the 1200 m imagery.

First of all in-flight GPS data and camera

characteristics derived from the calibration were used to approximately georeference
the images. The orientation of the image is calculated based on the change in position
from the GPS for the second prior to and after image acquisition. The GPS height is
378

used to calculate the spatial coverage, and geographic positions of the corner points
are calculated, and used as control points in the georeferencing. Starting with the
image closest to the accommodation cabins, identifiable features were selected in the
crudely georeferenced image. The easiest feature to identify were buildings, but these
were only present in this one 1200m images. Logs and outcroppings in the riverbank
were also easily identifiable in the image and in the field. Inland and away from
buildings it was more difficult. Some features that were used included large gaps,
flowering trees, large leafless ceibas, and some especially prominent palms. The
approximate coordinates of each identified feature was entered into a Garmin12 GPS
unit, and this location then visited in the field. Once the precise point was found, the
Trimble ProXL GPS unit was left taking a geographic position for at least 10 minutes.
In each image at least 5 GPS points were taken, whenever possible covering the
corners and a central point. ERDAS Imagine was used to georeference the images,
using bilinear interpolation.
Once the first image was georeferenced accurately, adjoining images were
georeferenced firstly based on a crude stitching to the 33% overlap with the first
image. This crudely georeferenced image was then used to locate features in the field,
and the final georeferenciation of the image performed using bilinear interpolation of
the GPS points.

This “dispersive” method was used to georeference all 1200m

images. The root mean square error was approximately 3-5 pixels (equivalent to 12m).
Once the 1200m images had been georeferenced, they were mosaiced together to
form a single image of the study site. This mosaic was then used as a base map with
which to georeference the 600m and 300m imagery, also using bi-linear interpolation.

379

In this case common features were identified for control points, using at least 10 per
image.

The helium balloon images were more easily georeferenced due to the clarity in the
photos.

Features were located on the ground, and ERDAS Imagine used to

georeference each image.

Only high quality images were georeferenced.

The

approximate accuracy of the georeferencing using this method is 1m.

Table 66 summarises the number of images acquired, how many of which were
georeferenced and their approximate ground coverage, and shows the georeferenced
imagery available for the study site.

Altitude of image
acquisition (metres
above ground level)
1200
600
300
100-200 (balloon)

Total Number of
Images Acquired

Total Number of images
Georeferenced

Average Resolution
of Images (cm)

Area coverd by
images (Ha)

60
154
18
679

33
46
14
30

33.6
21.4
8.4
7.6

1006
300
24
51

Table 66 Summary of image acquisition and georeferencing results.

380

Figure 120 Georeferenced imagery for TBS at all resolutions

Image Analysis

Firstly, object-oriented remote sensing techniques were used on the 1500m imagery to
separate out individual tree forms, in order to provide a vector coverage of each
canopy tree. This analysis was performed using eCognition Version 2. eCognition
makes a segmentation of an image into shapes based on the colour and texture of the
image. These shapes attempt to maintain internal homogeneity, but represent the
external heterogeneity of the image. The scale of these objects can be adjusted using
the arbitrary scale parameter, and the segmentation can be manipulated by weighting
the importance of each spectral band, and the relative weight between colour and
shape.

381

The second phase of analysis in eCognition is a hierarchical classification. A number
of levels are produced using the multi-resolution segmentation, of increasing levels of
detail using the scale parameter. Within each level, rules are produced based on the
spectral and textural properties of each shape, including neighbourhood functions, to
create a classification of key objects. At each level, the multiresolution segmentation
is tailored to the specific classification needs. In the case of this work, classification
hierarchies were constructed to classify the image into shade / canopy hole, tree
canopy, and river (the classification hierarchy is shown in Figure 121).

Figure 121 Classification hierarchy constructed in eCognition.

Once individual shapes classified as “tree crown” were identified, the image data for
each shape was extracted, including mean, max, min, mode and standard deviation of
pixels for each of the three bands, along with shape-based characteristics such as
“roundness”, perimeter, and area. The attribute values for each tree crown were then
subjected to a clustering analysis (using Ward’s method), and 100 arbitrary “canopy
forms” extracted from this analysis.

382

Finally, a grid-based analysis of canopy form richness was performed using DIVAGIS (http://diva-gis.org), to provide a map of possible canopy diversity.

Results and Discussion

The classification process was applied to a 1500m image, producing a realistic
classification or image regions representing tree crowns (Figure 122). When this
classification was combined with the image segmentation, and multi-variate statistical
analysis a map of diversity in tree crown canopy forms was produced (Figure 123).

Figure 122 Results of the hierarchical classification of eCognition, identifying river (blue),

gaps / shade (black) and tree crowns (green).

383

Figure 123 Modelled canopy diversity, derived from object-oriented analysis of canopy

forms.

The analysis has not been rigorously performed, rather has been performed as a pilot
test to examine the potential of this kind of analysis in measuring tree canopy
diversity over large areas. The results are promising, but the methods require detailed
study which is considered beyond the possibilities of this PhD thesis.

Several

problems were identified:

1. The scale-dependent image segmentation failed to account for different sized
tree crowns, resulting in a segmentation that, depending on the scale parameter
selected, clumped or separated individuals trees into single or multiple objects
2. The statistical methods behind the classification of tree forms requires greater
rigor, and detailed development

384

3. Validation of the image analysis is required, and the ten diversity plots
established for this study were not suitable for validation of these images (see
Figure 124).

Fieldwork in 2003 attempted to gather ground-based measurements of canopy tree
diversity using the matching of individual crowns visible in the imagery with crowns
identified from the ground, but this proved highly difficult.

Figure 124 Plot 1 (25m x 25m) from the air, in an image taken using the helium balloon. As

can be seen, few trees penetrate the canopy, with just 5-10 of the 80 trees identified on the
ground actually visible in this high-resolution image.

385

For the reasons briefly outlined above, and others not-mentioned the image-based
measurement of tree diversity was dropped from the research for this thesis, and is not
included in main text. However, it still merits further study.

386

9

Appendix 2

Heterogeneity model for quantitative variables
/* Integrated model for studying the spatial heterogeneity of the environment from a
biological standpoint
/*
/* The model requires :
/*
a kernel file, in ASCII format (or several if you wish to perform a multi-scale
analysis - file format = kernel_50x50_"scale".asc)
/*
input environmental variable (can be integer, floating)
/*
/* The model makes the following steps:
/*
* Reclassifies the environmental variable to between 25 and 30 classes
/*
* For each reclassified map, creates seperate binary grids for each class
/*
* Calculates the probability of receiving seeds from each environmental class
in the surrounding area (defined by the kernel)
/*
* Combines the probabilities of receiving each class through calculation of the
Shannon Index to
/*
produce a grid of environmental diversity
/*
* Normalises the final diversity map (mean_div_rel) and adjusts for edge
effect (mean_div_adj)
/*
/* The model also carries out validation on your data, writing a text file with the rsq
correlation.
/*
/*
/*
/* by Andy Jarvis
/*
2003
/* a.jarvis@cgiar.org
/* Model variables
&terminal 9999
&s inputvariable = d:\datosproyecto\ecuador\dem_comparison\srtm\mean_curv_dd
&s thresholdlow = 25
&s thresholdhigh = 30
&s kernel = kernel_50x50
&s killer = no
/* Validation variables
&s samplesites = d:\datosproyecto\ecuador\dem_comparison\srtm\plot_richness
&s kernelfiles := [listfile kernel*.asc -file]
&if [null %kernelfiles%] &then &type There are no kernel ASCII files
&else

387

&do
&s num := [token %kernelfiles% -count]
&do k := 1 &to %num%
&sv name = [extract %k% %kernelfiles%]
&sv scale = [substr %name% 14 3]
/* This section reclassifies the maps into %thresholdhigh% classes, so that the
focalsum calculation can work
describe %inputvariable%
&sv mapmin = %grd$zmin%
&type %mapmin%
&sv mapmax = %grd$zmax%
&do j := %thresholdlow% &to %thresholdhigh%
&type Classifying map into %j% classes
&do i := 1 &to %j%
&sv interval1 = [calc %mapmax% - %mapmin%]
&sv interval = [calc %interval1% / %j%]
&sv countless = [calc %i% - 1]
&if [EXISTS tempclass%i% -grid] &then kill tempclass%i% all
tempclass%i% = con (%inputvariable% <= [calc %i% * %interval% +
%mapmin%], con (%inputvariable% > [calc %interval% * %countless% +
%mapmin%], %i%, 0), 0)
&end
&do i := 1 &to %j%
&if %i% = 1 &then &sv list = tempclass%i%
&else &sv list = %list%, tempclass%i%
&end
&if [EXISTS intermed -grid] &then kill intermed all
intermed = sum (%list%)
&do i := 1 &to %j%
kill tempclass%i% ALL
&end
&if [EXISTS classes_%j% -grid] &then kill classes_%j% all
classes_%j% = con (isnull(intermed), 0, intermed)
&type Separating into component class maps
&do i := 1 &to %j%
&if [EXISTS class_%j%_%i% -grid] &then kill class_%j%_%i% all
class_%j%_%i% = con(classes_%j% == [calc %i%], 1, 0)
&end
kill classes_%j% all
/* This section of the model makes the spatial heterogeneity calculation on each class
&type Running focalsum model
&do i := 1 &to %j%
&if [EXISTS het_%j%_%i% -grid] &then kill het_%j%_%i% all

388

het_%j%_%i% = focalsum (class_%j%_%i%, weight, %kernel%_%scale%.asc,
DATA)
&end
&do i := 1 &to %j%
&if %i% = 1 &then &sv thelist = het_%j%_%i%
&else &sv thelist = %thelist%, het_%j%_%i%
&end
&type Combining class layers
&if [EXISTS het_sum_%j% -grid] &then kill het_sum_%j% all
het_sum_%j% = sum (%thelist%)
/* This section calculates the Shannon Index for diversity, normalises and adjusts for
density
&type Creating temporary frequencies
&do i := 1 &to %j%
&if [EXISTS tem%i% -grid] &then kill tem%i% all
tem%i% = (het_%j%_%i% / het_sum_%j%) * (het_%j%_%i% / het_sum_%j%)
&end
&do i := 1 &to %j%
&if %i% = 1 &then &sv thelist2 = tem%i%
&else &sv thelist2 = %thelist2%, tem%i%
&end
&type Creating sum-squared layer
&if [EXISTS het_sum2_%j% -grid] &then kill het_sum2_%j% all
het_sum2_%j% = sum (%thelist2%)
&do i := 1 &to %j%
kill tem%i%
&end
&type Calculating relative diversity layer
&if [EXISTS rel_div_%j% -grid] &then kill rel_div_%j% all
rel_div_%j% = (1 - het_sum2_%j%)
&type Calculating adjusted diversity layer
&if [EXISTS adj_div_%j% -grid] &then kill adj_div_%j% all
adj_div_%j% = (rel_div_%j% * het_sum_%j%)
&type normalizing final maps
describe adj_div_%j%
&sv divmin = %grd$zmin%
&sv divmax = %grd$zmax%

389

&if [EXISTS adj_div_%j%_n -grid] &then kill adj_div_%j%_n all
adj_div_%j%_n = (adj_div_%j% - %divmin%) / (%divmax% - %divmin%)
describe rel_div_%j%
&sv divmin2 = %grd$zmin%
&sv divmax2 = %grd$zmax%
&if [EXISTS rel_div_%j%_n -grid] &then kill rel_div_%j%_n all
rel_div_%j%_n = (rel_div_%j% - %divmin2%) / (%divmax2% - %divmin2%)
&do i = 1 &to %j%
kill het_%j%_%i%
kill class_%j%_%i%
&end
&end
&do i := %thresholdlow% &to %thresholdhigh%
&if %i% = %thresholdlow% &then &sv thelist3 = rel_div_%i%_n
&else &sv thelist3 = %thelist3%, rel_div_%i%_n
&if %i% = %thresholdlow% &then &sv thelist4 = adj_div_%i%_n
&else &sv thelist4 = %thelist4%, adj_div_%i%_n
&end
&type Calculating means
/* THis section puts the different maps for each class number together
&if [EXISTS div_rel -grid] &then kill div_rel all
div_rel_%scale% = sum(%thelist3%) / (%thresholdhigh% - %thresholdlow% + 1)
&if [EXISTS div_adj_%scale% -grid] &then kill div_adj_%scale% all
div_adj_%scale% = sum(%thelist4%) / (%thresholdhigh% - %thresholdlow% + 1)
/* Cleans up your directory of temporary files
&do i := %thresholdlow% &to %thresholdhigh%
kill het_sum_%i% all
kill het_sum2_%i% all
kill adj_div_%i% all
kill adj_div_%i%_n all
kill rel_div_%i% all
kill rel_div_%i%_n all
&if [EXISTS intermed -grid] &then kill intermed all
&end
/* Performs the validation using your sample sites map
model_comparison_%scale%.txt = sample (%samplesites%, div_adj_%scale%,
div_rel_%scale%)
correlation %samplesites% div_rel_%scale%
&sv rsqtable = [open rsq_results.txt openstat -append]

390

&if %openstat% ne 0 &then
&do
&type could not open rsqtable (Close arc/info and start again)
&return
&end
&else
&type %openstat%
&do
&if %k% = 1 &then
&do
&sv writeres = [write %rsqtable% 'Scale
Rsquared']
&sv writeres = [write %rsqtable% %scale%' '%.correlation_out%]
&end
&else
&sv writeres = [write %rsqtable% %scale%' '%.correlation_out%]
&sv closestat = [close %rsqtable%]
&end
&end
&end
&return

391

10 Appendix 3

Heterogeneity model for categorical variables
/* Integrated model for studying the spatial heterogeneity of the environment from a
biological standpoint
/*
/* The model requires :
/*
a kernel file, in ASCII format (or several if you wish to perform a multi-scale
analysis - file format = kernel_50x50_"scale".asc)
/*
input environmental variable (can be integer, floating)
/*
/* The model makes the following steps:
/*
* Reclassifies the environmental variable to between 25 and 30 classes
/*
* For each reclassified map, creates seperate binary grids for each class
/*
* Calculates the probability of receiving seeds from each environmental class
in the surrounding area (defined by the kernel)
/*
* Combines the probabilities of receiving each class through calculation of the
Shannon Index to
/*
produce a grid of environmental diversity
/*
* Normalises the final diversity map (mean_div_rel) and adjusts for edge
effect (mean_div_adj)
/*
/* The model also carries out validation on your data, writing a text file with the rsq
correlation.
/*
/*
/*
/* by Andy Jarvis
/*
2003
/* a.jarvis@cgiar.org
/* Model variables
&terminal 9999
&s inputvariable = d:\datosproyecto\ecuador\dem_comparison\srtm\featu_dd
/* Validation variables
&s samplesites = d:\datosproyecto\ecuador\dem_comparison\srtm\plot_richness
&s kernelfiles := [listfile kernel*.asc -file]
&if [null %kernelfiles%] &then &type There are no kernel ASCII files
&else
&do
&s num := [token %kernelfiles% -count]
&do k := 1 &to %num%
&sv name = [extract %k% %kernelfiles%]

392

&sv scale = [substr %name% 14 3]
/* This section reclassifies the maps into %thresholdhigh% classes, so that the
focalsum calculation can work
describe %inputvariable%
&sv mapmin = %grd$zmin%
&type %mapmin%
&sv mapmax = %grd$zmax%

&type Separating into component class maps
&do i := 1 &to 6
&if [EXISTS class_%i% -grid] &then kill class_%i% all
class_%i% = con(%inputvariable% == [calc %i%], 1, 0)
&end
/* This section of the model makes the spatial heterogeneity calculation on each class
&type Running focalsum model
&do i := 1 &to 6
&if [EXISTS het_%i% -grid] &then kill het_%i% all
het_%i% = focalsum (class_%i%, weight, %name%, DATA)
&end
&do i := 1 &to 6
&if %i% = 1 &then &sv thelist = het_%i%
&else &sv thelist = %thelist%, het_%i%
&end
&type Combining class layers
&if [EXISTS het_sum -grid] &then kill het_sum all
het_sum = sum (%thelist%)
/* This section calculates the Shannon Index for diversity, normalises and adjusts for
density
&type Creating temporary frequencies
&do i := 1 &to 6
&if [EXISTS tem%i% -grid] &then kill tem%i% all
tem%i% = (het_%i% / het_sum) * (het_%i% / het_sum)
&end
&do i := 1 &to 6
&if %i% = 1 &then &sv thelist2 = tem%i%
&else &sv thelist2 = %thelist2%, tem%i%
&end
&type Creating sum-squared layer

393

&if [EXISTS het_sum2 -grid] &then kill het_sum2 all
het_sum2 = sum (%thelist2%)
&do i := 1 &to 6
kill tem%i%
&end
&type Calculating relative diversity layer
&if [EXISTS rel_div -grid] &then kill rel_div all
rel_div = (1 - het_sum2)
&type Calculating adjusted diversity layer
&if [EXISTS adj_div -grid] &then kill adj_div all
adj_div = (rel_div * het_sum)
&type normalizing final maps
describe adj_div
&sv divmin = %grd$zmin%
&sv divmax = %grd$zmax%
&if [EXISTS adj_div_n -grid] &then kill adj_div_n all
adj_div_n = (adj_div - %divmin%) / (%divmax% - %divmin%)
describe rel_div
&sv divmin2 = %grd$zmin%
&sv divmax2 = %grd$zmax%
&if [EXISTS rel_div_n -grid] &then kill rel_div_n all
rel_div_n = (rel_div - %divmin2%) / (%divmax2% - %divmin2%)
&do i = 1 &to 6
kill het_%i%
kill class_%i%
&end

&type Calculating means
/* THis section puts the different maps for each class number together
&if [EXISTS div_rel_%scale% -grid] &then kill div_rel_%scale% all
div_rel_%scale% = rel_div_n
&if [EXISTS div_adj_%scale% -grid] &then kill div_adj_%scale% all
div_adj_%scale% = adj_div_n
/* Cleans up your directory of temporary files
&if [EXISTS adj_div -grid] &then kill adj_div all
&if [EXISTS het_sum -grid] &then kill het_sum all

394

&if [EXISTS het_sum2 -grid] &then kill het_sum2 all
&if [EXISTS rel_div -grid] &then kill rel_div all
&if [EXISTS adj_div_n -grid] &then kill adj_div_n all
&if [EXISTS rel_div_n -grid] &then kill rel_div_n all
&if [EXISTS intermed -grid] &then kill intermed all
/* Performs the validation using your sample sites map
model_comparison_%scale%.txt = sample (%samplesites%, div_adj_%scale%,
div_rel_%scale%)
correlation %samplesites% div_rel_%scale%
&sv rsqtable = [open rsq_results.txt openstat -append]
&if %openstat% ne 0 &then
&do
&type could not open rsqtable (Close arc/info and start again)
&return
&end
&else
&type %openstat%
&do
&if %k% = 1 &then
&do
&sv writeres = [write %rsqtable% 'Scale
Rsquared']
&sv writeres = [write %rsqtable% %scale%' '%.correlation_out%]
&end
&else
&sv writeres = [write %rsqtable% %scale%' '%.correlation_out%]
&sv closestat = [close %rsqtable%]
&end
&end
&return

11 Appendix 4
Submitted to Novedades Colombiana
PLANTS OF TAMBITO I. DICOTILEDONOUS. A PRELIMINARY LIST

Carlos E. González and Andrew Jarvis

Resumen
Se presenta una lista preliminar de las plantas dicotiledóneas registradas desde 1998 hasta el 2002 en
la reserva Tambito, Cauca, Colombia (2º 30’ N- 76º 59’ W). Se reportaron 585 especies distribuidas en
86 familias. Se organizaron según su hábito de crecimiento: hierbas (52 especies), arbustos (132
especies), árboles (301 especies), trepadoras (59 especies) y epífitas (5 especies). Las familias de
mayor número de especies fueron Rubiaceae, Gesneriaceae, Melastomataceae y Piperaceae. Esta es
una lista preliminar centrada en ofrecer una estimación inicial de las especies en Tambito. El patrón
numérico demuestra que se deben hacer esfuerzos de colecta en lianas, epífitas y trepadoras además
de nuevas identificaciones a los grupos menos conocidos taxonomicamente para completar el

395

inventario de dicotiledóneas. Iguales esfuerzos de deben hacerse en Helechos, monocotiledones y
grupos de plantas criptógamas.

Palabras claves: Flora, Bosque de Niebla, Tambito, Cordillera Occidental, Cauca,
Colombia

Abstract
This study presents a preliminary list of dicotiledonous plants that have been registered 1998 – 2002 in
Reserva Tambito, Cauca, Colombia (2º 30’ N- 76º 59’ W). Some 585 species are reported, in 86
families. The species included in this list were classified into herbs (52 species), shrubs (132 species),
trees (301 species), climbers (59 species) and epiphytes (5 species). The most speciose familes are
Rubiaceae, Gesneriaceae, Melastomataceae and Piperaceae. This is a preliminary list aimed to
provide an initial estimate of the plant species in Tambito. It is suggested that more collections are
made in lianas, epiphytes and climbers to further complete the inventory of dicotiledonous plants, and
that similar studies should be made for monocotiledonous and cryptogams (including mocees and ferns)
species.

Key words: Flora, Cloud Forest, Tambito, Western Andes, Cauca, Colombia

Introduction
This is the first compilation of Tambito´s plant species, which presents a preliminary list of
dicotiledonous plants. A series of future lists are in preparation under a Florula project (II:
Monocotiledonous III: Ferns and similars IV: Mosses and affines). Previous studies in
Tambito suggest that there might be as many as 640 species collected (including
monocotiledonous), in 269 genera and 110 families including monocotiledonous (González
C.E, 2000). This paper produces a rigorous review of collections in Tambito between 1998
and 2002, providing a more conclusive figure for the number and taxonomy of the species
now registered in Tambito. Additionally we make suggestions for the focus of future
botanical collection in Tambito.

Materials and Methods
Study Site

Reserva Tambito, Cauca, Colombia (2º 30’ N- 76º 59’ W) lies on the western slopes of the
western cordillera of the Andes. The altitudinal range rises from 1053 to 2860 m.a.s.l.. The
total area of the reserve is 2150 hectares, with areas of secondary and primary forest, and
some pasture area around the Tambito cabin. Annual rainfall ranges from 3900mm around the
cabin, to as much as 6560mm at 2000m.a.s.l. (Jarvis, 2000). Relative humidity averages at
over 90%, with daily variation between 60 and 100% (Jarvis, 2000). It is a private nature
reserve which forms the Pacific Enviromental Studies Centre to the University of Cauca,
Popayan.
Field data collection

Botanical collections were made from 1998 – 2002, many as part of the fieldwork for HERB
Project and Project Negret (King’s College London). The areas covered were web distributed
around the twin catchments, between 1300 and 2500 m.a.s.l.. Lack of access to the higher
elevations prevented collections above 2500m.a.s.l..
Samples were collected and preserved temporarily in alcohol according to the traditional
method of botanical sample preservation. The samples were then dried for 48 hours at 90oF,
and stored in the CAUP herbarium in the Natural History Museum of Popayan. Identification
of the samples were made to species level whenever possible, though many identifications

396

stop at genus. Some samples remain unidentified to family, and so have been omitted from
this list.

Results and Discusion
Table 1 shows the species which have been registered, and are seperated into families. For
each species additional information is provided with regard to the Holdridge life zone in
which the species was found, in addition to the plant form and information on the collection
number for the herbarium.
In total, 585 species have been identified, in 86 families. Some 301 of these are tree species,
whilst 132 are shrubs, 59 are climbers, 52 are herbs and 5 are epiphytes. Only 60-65%
(around 326 species) of identifications are confirmed; 4% (around 22 especies) remain
unidentified and 21 especies are being investigated as new species for science (4 have already
been described). Some of the species which are unidentified might also be undescribed
species. Melastomataceae and Rubiaceae are generally poorly identified families. Tree
families like Lauraceae and Myrtaceae were found to be less taxonomically documented.
Genera like Nectandra, Ocotea, Miconia, Piper, Besleria, Solanum, Ficus, Palicourea and
Psychotria have the highest probabilities for finding new undescribed or non identified
species.
These figures show the weaknesses in collections, and highlight the need for further study in
lianas, epiphytes and climbers. For both these groups the species richness is likely to be
significantly higher. Some data about rare or endemic species are marked, but must be
considered more thoroughly in subsequent analyses.
Figure 1 shows the species composition with respect to families. The most speciose familes
are Rubiaceae (67 species), Gesneriaceae (64 species), Melastomataceae (59 species) and
Piperaceae (36 species). A total of 38% of all registered species are in these four families.

Conclusions
This preliminary list provides a basic documentation of the dicotiledonous flora found in
Tambito. It demonstrates the exceptional level of plant diversity in Tambito, and throws
down the challenge to further study and document the biological resources within the Reserve
and the surrounding area. This kind of information is crucial to comply with the Convention
on Biological Diversity, where basic inventory of biological resources is essential to construct
relevent and effective national conservation policies.
It is hoped that this paper will contribute to the establishment of a Florula project for Tambito
(González, 2002), which integrates the results of biological research in Tambito over the past
years. Amongst these findings are new species to science (Gonzalez, 2002), and significant
advances in knowledge of the ferns (Casañas, 2002) and palms (Cortes, 1996). This
preliminary list could be lengthened and finalised under such a project, and extended into the
surrounding Munchique National Park and Choco Biogeographic region.

Acknowledgements
The authors would like to thank anyone who has worked in Tambito for making this research
possible. We are grateful to the Project HERB and Project Negret team for their support and
aid in fieldwork. Thanks to the national COL and CAUP Herbarium of the Natural History
Museum at the University of Cauca; the Popayán National Parks officers; to Fundación

397

Proselva; to King´s College London and local communities around Tambito. We would like
to acknowledge everyone who has been involved with plant collections or field work
assistance:
Sandra L. Díaz, Andrew Jarvis, Mike Salazar, Juliana González, Matthew Letts,
Alexandre Cortéz, Juan C. Garcia, Tatiana Arias, Luis A. Ortega, Carolina Alcazar,
Beatriz Salgado, Alvaro J. Negret, Mark Mulligan, Natalia Gómez, Martha L. Burbano.

References
CASAÑAS, O.L. Análisis vertical de los helechos epifíticos en Tambito. Popayán, 2002,
100 p. Trabajo de grado (Biólogo). Universidad del Cauca. Facultad de Ciencias
Naturales, Exactas y de la Educación. Departamento de Biología.
CORTES, A. Estructura y diversidad de una comunidad de palmas (Wlffia regia) en la
reserva Tambito. Popayán, 1996, 100 p. Trabajo de grado (Licenciatura en Biólogía).
Universidad del Cauca. Facultad de Ciencias Naturales, Exactas y de la Educación.
Departamento de Biología
GONZÁLEZ, C.E. Diversidad y composición florística del bosque de niebla en el
Departamento del Cauca, Colombia. Cespedesia. Vol. 24, No 75 (2000)
_______. Nuevas especies de Panopsis (Proteaceae) para Sur América. En preparación.
Mus Hist Nat. 2002. Popayán
_______. La Florula de Tambito: Una contribución al conocimiento de la flora del
parque nacional Munchique. Popayán. 2002. Propuesta de Proyecto en realización.
HOLDRIDGE, L.R. Life zone ecology. San José de Costa Rica: Tropical Science Centre,
1967. 206 p.
JARVIS, A. Measuring and Modelling the Impact of Land-use change in tropical
hillsides: The role of cloud interception to epiphytes. Advances in Evironmental
Modelling and Monitoring Vol. 1, (2000); p 118-148.

Tables and Figures
Figure 1 Number of species per family for Reserva Tambito.

398

Table 1 Taxonomic list
Life Zone (Lf): Refers to the ecological forest type. bh-pm (1300-1600 m) =
Premontane Wet Forest; bh-mb ( 1600-2000 m) = Lower Montane Wet Forest; bll-mb
( 2000-2500 m) = Lower Montane Rain forest; bll-m (> 2500 m) = Upper Montane
Rain forest. According to classification (Holdridge, 1967)
* = Refering to endemic or rare species
Plant form: hb = Herb; tr = Tree; sh = Shrub; cl = Climber; ep= Epiphytic
Elevation range: Refers to the records of altitudinal range
Reference of collection (Only main collector): CG Carlos González, RS Ramón
Serna, SD Sandra Díaz, RR Rosalba Ruiz, OC Olga Casañas, RSA Roberto
Sánchez, LA Liza y Aleja, BR Bernardo Ramírez, FG Fernando González; MB
Martha Burbano; HB HERB

Taxon
Acanthaceae
Aphelandra acanthus Nees

Aphelandra sp nov
Justicia chlorostachya Leonard
Justicia
Lepidagathis lanceolata (Nees.) Wasshausen

Odontonema cf
Pseuderanthemum
Trichantera gigantea (Bonpl.) Nees

Actinidaceae
Saurauia brachybotrys Turcz
Saurauia micayensis Killip & Soejarto

Saurauia aff omnichlophila R.E.
Schult.
Amaranthaceae
Alternanthera cf elongata (W. ex R.) Schinz
Iresine diffusa Humb. & Bonpl ex Willd

Life zone Lf Elevation Collectio
n
bh-mb
1800
sh
bh-mb/pm
1655-1850
CG 3359
hb
bh-pm
1600
CG 3383
bh-mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-mb
bll-mb
bh-pm
bh-pm
bh-pm
bh-pm

399

hb
hb
hb
sh
hb
tr
tr
tr
sh

hb
hb
tr
tr

1800
1350-1600
1300
1600
1600

1450-1600
1600-2000
2300
1600
1600
1300
1200

Herbariu
m

OC 135
CG 3359
BR 7724
CG 1482
RS 844
RS 1065

CAUP
CAUP
CAUP
CAUP
CAUP PSO
CAUP
CAUP
CAUP

LG 036
CG 3187
CG 1254

AFP CAUP
CAUP
CAUP

RS 996
RS 997

CAUP
CAUP

CG 3154

CAUP

Anacardiaceae
Toxicodendron striatum (Ruiz & Pavon) Kuntze
Ochoterenaea
Annonaceae
Guatteria
Guatteria
Malmea
Apiaceae
Hydrocotyle
Apocynaceae
Mandevilla
Aquifoliaceae
Ilex
Ilex
Ilex cf pernervata Cuatrec
Ilex cf karstenii Loes
Ilex cf laurina HBK
Araliaceae
Dendropanax cf caucanus (Harms.) Harms
Dendropanax macrophyllum Cuatrec
Oreopanax cf arboreus (L.) Dec ex Plank
Oreopanax peltatus Linden
Schefflera
Schefflera
Schefflera trifolium
Asteraceae
Ambrosia cumanensis HBK
Austroeupatorium inulaefolium (Kunth.) K & H
Critoniopsis cf occidentalis (Cuatrec.) H. Rob
Lepidaploa lehmanii (Hieron.) H. Robinson
Mikania banisteriae DC
Munnozia hastifolia (Poepp.) H. R & Brettel
Pollalesta aff macrophylla * (Sch. Bip) Aris
Balanophoraceae
Langsdorffia hypogea Matinus
Balsaminaceae
Impatiens cf balsamina L.
Begoniaceae
Begonia tiliifolia C. DC
Begonia cf maynensis A. DC.
Begonia geminiflora L.
Begonia secunda L.
Begonia killipiana * Smith & Schubert
Begonia urticae L. var. urticae
Begonia rossmanniae A. DC.
Begonia aff guaduensis Kunth
Bignoniaceae
Schlegelia cf roseiflora Ducke
Bombacaceae
Matisia bolivarii Cuatrec
Matisia aff dolychopoda sp nov
Matisia
Matisia
Matisia cf cordata Bonpl.
Quararibea cf
Spirotheca rhodostyla Cuatrec
Boraginaceae
Cordia cf cylindrostachia (Ruiz & Pavón) R & S
Tournefortia gigantifolia Killip
Brunelliaceae
Brunellia cf glabra Cuatrec.
Burseraceae
Protium
Protium
Protium
Protium
Trattinikia

bh-pm
bh-mb
bh-pm

tr
tr
tr

bh-pm

RS 1182

CAUP

1450

CG 1383
CG 3132
SD 367

CAUP
CAUP
CAUP

-

CG 0000

CAUP

1300-2000
1500-1966
1856-2200
1650
1450-1600

-

CAUP

CG 3197
CG 3143
CG 3248
CG 3008
CG 1145

CAUP
CAUP
CAUP
CAUP
CAUP

CG 1908
CG 1921
CG 1886
CG 3271
CG 3280
CG 3283
CG 2933

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

2200

RS 1056
RS 1078
CG 1150
RS 847
RR 1371
RS 1068
CG 1400

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

1400-2200

CG 1262

CAUP

1500-1700
1300
2300
1800
2300-2500
2100
1550
1300

-

CAUP

CG 1328
CG 1358
CG 1292
CG 3107
CG 1225
CG 1423
CG 1154
CG 1513

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3257

CAUP

CG 1104
CG 1149
CG 1750
CG 1323
CG 1344
CG 1181
CG 3268

CAUP
CAUP COL
CAUP
CAUP
CAUP
CAUP
CAUP

CG 2884
CG 2863

CAUP
CAUP

1480
1686-1966
1500

hb
cl
bh-pm/mb
bh-mb
bh/ll-mb
bh-pm
bh-pm
bh-pm
bh-mb
bh-pm
bh-mb
bh-mb
bh-mb
bh-pm
bh-pm
bh-pm
bh-mb
bh-pm
bh-mb
bh-pm
bh/ll-mb

tr
tr
tr
tr
tr
tr
tr
tr
tr
cl
cl
cl
sh
sh
tr
sh
cl
cl
tr

bll-mb

1500
1650
1500
1856
1950
1856
1200-1300
1450
1450
1600
1580
1600-2000
1450
1580-2400

hb
bh/ll-mb
hb
bh-pm/mb
bh-pm
bll-mb
bh-mb
bll-mb
bll-mb
bh-pm
bh-pm

cl
cl
hb
cl
cl
cl
cl
cl

-

cl

bh-mb
bh-mb
bh-pm
bll-mb
bh-pm/mb
bll-mb
bh-mb
bh-pm
bh-mb

tr
tr
tr
tr
tr
tr
tr
sh
sh

bll-mb

1950
1600
1300
2300
1350-1400
2300
1856
1400
1700
2500

tr
bh-pm
bh-pm
bh-pm/mb
bh-pm
bh-mb
bh-pm

tr
tr
tr
tr
tr
tr

bh-pm

1300
tr

400

1450-1600
1300
1650-1686
1450
1700
1500

CAUP
CG 3544
CG 3155
CG 3057
CG 1902
CG 1897
CG 3094

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

Indet
Caesalpinaceae
Brownea
Campanulaceae
Burmeistera cf oblongifolia E. Wimm.
Burmeistera cf cylindrocarpa Zahlbr.
Burmeistera cf ceratocarpa Zahlbr.
Centropogon cf papillosus E. Wimm.
Centropogon congestus Gleason
Centropogon grandis (l.f.) Presl
Siphocampylus scandens (Kunth.) G. Don
Capparaceae
Cleome
Podandrogyne cf. polychroma Woodson (Colc)
Caprifoliaceae
Viburnum lehmanii Killip & Smith
Viburnum pichinchense Benth.
Caricaceae
Carica microcarpa Jacq
Jacaratia aff chocoensis Gentry & Forero
Cecropiaceae
Cecropia garciae Standl.
Cecropia cf telealba Cuatrec.
Cecropia bullata C. Berg & P. Franco
Cecropia
Cecropia
Celastraceae
Celastrus colombianus Cuatrec
Gymnosporia gentryi Lundell
Perrottetia maxima Cuatrec
Perrottetia * sp nov
Chloranthaceae
Hedyosmum bonplandianum Kunth
Hedyosmum cuatrecazanum Occhioni
Hedyosmum racemosum (Ruiz & Pavón) G Dn
Chrysobalanaceae
Hirtella cf americana Cuatrec
Licania aff cuatrecasasii Prance
Chletraceae
Chletra fagifolia Kunth
Clusiaceae
Calophyllum brasiliense Camb.
Clusia alata Pl. & Tr.
Clusia bernardoi * Pipoly & Cogollo
Clusia bracteosa Cuatrec.
Clusia cf. crenata Cuatrec.
Clusia discolor Cuatrec or T. Parviflora cf1279
Cluisa aff dixonii Little
Clusia cf ellipticifolia Cuatrec.
Clusia aff inesiana or longistyla Cuatrec.
Clusia loranthacea Planchon & Triana
Clusia pentandra Cuatrec.
Clusia hirsuta * Hammel sp nov
Clusia
Clusia
Clusia
Clusia
Chrysoclamis cf bracteolata Cuatrec.
Chrysoclamis colombiana (Cuatrec.) Cuatrec
Chrysoclamis cf tenuifolia
Chrysoclamis
Chrysoclamis
Tovomita weddelliana Planchon ex Triana
Vismia cavanillesiana Cuatrec.
Vismia baccifera (L.) Tr. ex Pl.
Vismia lauriformis (Lamb.) Choisy
Vismia mandur Hieron.

bh-mb
bh-mb
bh-pm
bll-mb
bll-mb
bh/ll-mb
bh-pm
bh-pm
bh-mb
bh/ll-mb
bh-pm
bh-pm
bh-pm/mb
bh-pm
bll-mb
bll-mb
bh-mb
bll-mb
bh/pm/mb
bh/ll-mb
bll-mb
bh-pm/mb
bh-mb
bh-pm
bh/ll-mb
bh-pm

sh
sh
sh
sh
sh
sh
sh
hb
sh
tr
tr
sh
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr

bll-mb

1900
1710
1450
2400
2000
1600-2400
-

CG 3160

CAUP

CG 1425
OC 324
RS 1063
CG 1301
CG 1269
CG 1424
-

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
-

CG 1363
CG 1452

CAUP
CAUP

CG 3265
CG 1183

CAUP
CAUP

CG 2887
CG 3166

CAUP
CAUP

CG 3249
RS 1001
CG 1277
CG 3302
CG 3387

CAUP
CAUP
CAUP
CAUP
CAUP

RSA 2437
CG 3192
CG 3145
CG 1312

AFP
CAUP
CAUP
CAUP COL

CG 3067
CG 3133
CG 1394

CAUP
CAUP
CAUP

CG 3084
CG 3171

CAUP
CAUP

CG 3232

CAUP

2300

CG 3389
CG 3101
RS 831
RS 889
CG 3030
CG 3218
CG 3037
CG 1754
CG 3200
CG 2877
CG 1255
CG 1353
CG 1911
LA 017
CG 1768
CG 1508
CG 1172
CG 3016
CG 3105
CG 1496
CG 1101
CG 3100
CG 1432
CG 1129
LA 031
CG 3382

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

-

CG 1227

CAUP COL

1400-1500
1400-1600
1600-1800
1500-2200
1300
1400
1300-1856
1580
1950
2200
1686
2100
1600-1800
1950-2000
2100-2300
1580-1700
1900-2000
1400
1500-1850
1300-1600
2250

tr
bh-mb
bh-pm
bh/ll-mb
bh-mb
bh-mb
bh-pm
bh-mb
bh-pm
bh-mb
bh-mb
bll-mb
bh-pm
bh-mb
bh-pm
bll-mb
bh/ll-pm/mb
bh-mb
bh-pm
bh-pm
bh-pm/mb
bll-mb
bh-pm
bh-pm
bll-mb

tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr

bll-mb

1600-1686
1500-1600
1600-2000
1600
1650
1300
1650
1400-1600
1750
1856
2300
1200
1600-1800
1490
2300
1200-2200
1600-1966
1300
1600
1400-1800
2100
1600
1450
1900-2000

cl
-

401

Cornaceae
Cornus periviana J.F. Macbr.
Cucurbitaceae
Gurania macrophylla (Collect)
Melothria pendula L.
Cunnoniaceae
Weinmannia
Elaeocarpaceae
Sloanea cf brevispina Smith
Sloanea cf echinocarpa Uittien
Ericaceae
Antropterus cf verticillatus
Bejaria mathewsii Fielding & Gardner
Cavendishia bracteata (Ruiz & Pavon. ex JH)
Cavendishia
Gaultheria cf
Psammisia cf
Macleania cf recumbens A.C. Sm
Macleania
Satyria
Psammisia
Psammisia
Psammisia sodiroi Hoerold
Themistoclesia cf cuatrecasasii A. C. Smith
Pernettya
Sphyrospermum buxifolium Poepp & Endl.
Themistoclesia
Themistoclesia
Erythroxilaceae
Erythroxilum lucidum Kunth
Erythroxilum cf macrophyllum Cav
Euphorbiaceae
Acalypha macrostachya Jacq.
Acalypha platyphylla Müll. Arg.
Alchornea coelophylla Pax & K. Hoffmann
Alchornea aff glandulosa Endl. & Poepp.
Alchornea triplinervia Müell
Alchornea sp nov
Euphorbia goudotii Boiss
Hyeronima oblonga (Tul.) Müll. Arg.
Hyeronima
Mabea cf. occidentalis Benth.
Richeria grandis Vahl.
Sapium cf marmieri Huber.
Sapium
Tetrorchidium macrophyllum Müll. Arg
Fabaceae
Andira
Dussia cf lehmannii Harms.
Dussia cf
Desmodium
Crotalaria
Flacourtiaceae
Banara guianensis Aublet.
Casearia arborea (L.C. Rich.) Urban
Casearia cajambensis Cuatrec.
Casearia silvestris Sw.
Casearia
Pleuranthodendron lindenii (Turcz.) Sleumer
Gentianaceae
Irlbachia alata (Aublet.) P. Maas
Gesneriaceae
Alloplectus bolivianus (Britton.) Wiehler
Alloplectus icthyoderma Hanst
Alloplectus medusaeus L.E. Skog
Alloplectus aff panamensis Morton
Alloplectus schultzei Mansfeld

bll-mb

cl
cl

2300
-

-

CG 1287

CAUP

-

-

CG 3120
CG 3013

CAUP
CAUP

CG 1505
FG 3184
CG 2937
CG 2956
CG 1119ª
CG 1119
RS 808
CG 3202
CG 1233ª
RS 1146
CG 1495
CG 1420
CG 1216
CG 1202
BR 7755
CG 1240
CG 1245

CAUP
CAUP COL
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 1387
RS 846

CAUP
CAUP

CG 3162
RS 794
CG 1221
CG 1128
CG 3214
CG 3264
RS s.n
CG 3210
CG 3180
CG 3156
CG 3535
CG 3168
CG 3313
CG 3250

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
AFP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3262
CG 3102
CG 3072
CG 3548
-

CAUP
CAUP
CAUP
CAUP
-

LA 034
CG 3029
CG 3278
168
CG 1940
CG 3018

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

-

CAUP

CG 1940
CG 2834
CG 1313
RS 853
CG 2832
CG 1751
CG 1369

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

tr
bll-mb
bh-pm/mb
bll-mb
bll-mb/m
bh-mb
bh-pm
bll/h-mb
bh-mb
bh-mb
bh-pm
bll-mb
bh-mb
bh-pm
bll-mb
bll-mb
bll-mb
bll-mb
bll-mb
bll-mb
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bh-pm
bh-mb
bll-mb
bh/bll-mb
bh/ll-mb
bh-pm
bh-pm
bh-pm
bh-mb
bh-mb
bh-mb
bh-pm/m
bh-mb
bh-pm
bh-pm
bh-mb
bh/ll-mb
bh/ll-mb
bh-mb

tr
tr
sh
hb
sh
sh
sh
sh
sh
sh
sh
hb
hb
cl
cl
hb
hb
tr
tr
tr
tr
tr
tr
tr
sh
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
hb

tr
tr
tr
tr
tr
hb
tr

1966
1650
2000
2580-2640
1800-2000
1400
1900-2000
1600
1600
1700
2200
1600
1400
2000
2400
2400
‘2000
2300
2400
1450-1500
1580
1300
1600
2200
1600
1750
2200-2500
1600-2200
1300-2200
1300
1600-1850
1300
1856
1856
1856
1350-1966
1856
1400
1450
1650-1856
1950
2200
1650
1200

bh-pm
sh
bh-pm
bh/ll-mb
bll-mb
bh-mb
bll-mb
bh/ll-mb
bh/ll-mb
bh/ll-mb
bh-mb

402

sh
sh
sh
sh
sh
sh
cl
cl

1350-1600
1800-2300
2300
1580-1600
2200-2500
1200-2200
1400-2000
1650-2100
1400-1600

Alloplectus tetragonoides Mansfeld
Alloplectus tetragonus (Oerst.) Hanst
Alloplectus teuscheri (Raymond.) Wiehler
Alloplectus weirii Wiehler
Alloplectus sp nov
Alloplectus sp nov
Alloplectus sp nov
Besleria quadrangulata L.E. Skog
Besleria reticulata Fritsch
Besleria miniata C.V. Morton
Besleria aff riparia C.V. Morton
Besleria sp nov
Besleria sp.nov
Besleria sp nov
Besleria villosa Fritsch
Besleria cf tambensis C.V. Morton
Besleria
Besleria
Capanea affinis Fritsch
Capanea grandiflora (Kunth) Decne & Planch.
Columnea angustata (Wiehler) Skog
Columnea anisophylla DC.
Columnea dimidiiata (Benth.) Kuntze
Columnea fuscihirta Kvist & Skog
Columnea gigantifolia Kvist & Skog
Columnea lehmannii Mansfeld
Columnea medicinalis (Wiehler.) Kvist & Skog
Columnea minor (Hook.) Hanst var nov 1
Columnea minor (Hook.) Hanst var nov 2
Columnea nicaraguensis Oerst var nov 1
Columnea nicaraguensis Oerst var nov 2
Columnea cf picta H. Karst.
Columnea strigosa Bentham
Columnea aff villosissima Kvist & Skog
Columnea sp nov
Cremosperma hirsutissimum var hirsutissimum
Cremosperma hirsitissimum var album K & S
Cremosperma humidium Kvist & Skog
Cremosperma cf nobile C.V. Morton
Diastema affine Fritsch
Drymonia cf aciculata Wiehler
Drymonia serrulata (Jacq.) Mart
Drymonia cf sulphurea Wiehler
Drymonia turrialvae Hanst.
Drymonia warscewicziana Hanst.
Gasteranthus corrallinus (Fritsch) Wiehler
Gasteranthus delphinioides (Scem.) Wiehler
Gasteranthus leopardus M. Freiberg
Hepiella ulmifolia (Kunth.) Hanst
Kohleria inaequalis Benth var inaequalis
Kohleria inaequalis Benth. Var ocellata
Kohleria spicata Oerst
Monopyle inaequalis C.V. Morton
Monopyle cf macrocarpa Benth
Paradrymonia
Paradrymonia
Paradrymonia sp nov
Hippocastanaceae
Billia columbiana Planch. & Linden ex Tr & Pl.
Hippocrateaceae
Salacia cf gigantea Loes.
Hydrangeaceae
Hydrangea
Icacinaceae
Calatola colombiana Slearm
Citronella silvatica Cuatrec.
Lamiaceae

bh-mb
bh-mb
bll-mb
bh/ll-mb
bll-mb
bh-mb
bh/ll-mb
bll-mb
bll-mb
bh-mb
bh/ll-mb
bll-mb
bh-mb
bh/ll-mb
bh-pm
bll-mb
bh-pm
bh-mb
bh/ll-mb
bh-pm/mb
bh-pm/mb
bh-pm
bh-mb
bh-pm/mb
bh-mb
bh/ll-mb
bh-mb
bh-mb
bll-mb
bh-mb
bll-mb
bh-mb
bh/ll-pm/mb
bll-mb
bh-mb
bh-pm
bll-mb
bh-pm
bh-pm
bh-mb
bh/ll-mb
bh/ll-mb
bh-mb
bh-pm
bll-mb
bh-pm/mb
bh-pm
bh-pm
bh-pm
bh-pm/mb
bh-pm/mb
bh-mb
bh-mb

sh
sh
sh
sh
hb
sh
sh
sh
sh
sh
sh
sh
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
cl
hb
hb
hb
hb
hb
sh
sh
sh
sh
cl
sh
sh
sh
hb
hb
hb
hb
sh
sh
cl
cl
cl
cl
tr
cl

bh/ll-mb

1500-2350

CG 3356
CG 2867
CG 3276
CG 2846
CG 3240
CG 2838
CG 1257
CG 1444
CG 2843
CG 1215
CG 1207
CG 3391
CG 2966
CG 2871
CG 2848
CG 2844
CG 1461
CG 1179
CG 2870
CG 1451
CG 1336
CG 3547
CG 2836
CG 1893
CG 1765
CG 2866
CG 3000
CG 2941
CG 2942
RS 1013
CG 1166
CG 2996
CG 1412
CG 2849
CG 2847
CG 2960
CG 2990
CG 157
CG 1293
CG 2934
CG 2928
CG 1354
CG 2850
CG 1447
CG 1752
CG 1762
CG 1361
CG 1289
RS 1002
CG 1753
CG 1473
CG 2998
CG 2997

CAUP
CAUP
CAUP
CAUP
CAUP US
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP COL
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP COL
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP COL
CAUP COL
CAUP COL

1600-2000

CG 3226

CAUP

2500

CG 3006

CAUP

1650-1800
1750-1950

178

CAUP

CG 3022
CG 3191

CAUP
CAUP

CG 3564
CG 1522

CAUP
CAUP

1800
1700
2100
1600-2300
2150
1900
1800-2200
2300
2300-2500
1450-1800
1400-2100
2100
1500-1700
1500-2100
1600
2200
1200-1400
1600-1800
1800-2400
1300-2100
1600
1400-1600
1400-1900
1400-1700
1800
1800-2250
2180
1600
2300-2500
1580-1800
2000
1700-1900
1500-2100
2000-2100
1700
1300-1450
2300
1300
1200
1300-1600
1400-2000
1300-2100
1700
1400
2500
1300-2000
1300-1600
1300-1600
1600
1400-1700
1500-1800
1600-1800
1800

tr
bh-mb
tr
bll-mb
tr
bh-mb
bh-mb
bh-pm
bh-pm
bh-mb

403

hb
hb
hb
hb

1400
1200
1684

Hyptis
Salvia
Lauraceae
Beilschmiedia?
Endlicheria
Nectandra aff longifolia (Ruiz & Pavon) Nees
Nectandra aff membranacea (Sw.) Griseb.
Nectandra aff mollis (Kunth.) Nees
Nectandra reticulata (Ruiz & Pavón) Mez
Nectandra aff purpurea (Ruis & Pavón) Mez
Nectandra
Nectandra
Nectandra
Nectandra
Nectandra
Nectandra
Nectandra
Nectandra?
Nectandra?
Ocotea aff babosa C.K. Allen
Ocotea aff duqei Korstom.
Ocotea oblonga (Meissner) Mez
Ocotea aff simulans C.K. Allen
Ocotea
Ocotea
Ocotea
Ocotea
Persea aff caerulea (Ruis & Pavón) Mez
Persea aff hexanthera Kupp
Persea americana Mill.
Persea
Pleurothyrium cf
Rhodostemonadaphne cf
Indet
Indet
Indet
Indet
Lecythidaceae
Eschweilera
Eschweilera
Eschweilera aff coriaceae
Eschweilera aff caudiculata Kunth
Lentibulariaceae
Utricularia jamesoniana Oliver
Loranthaceae
Gaiadendron punctatum (Ruiz & Pavón) Don
Ixocanthus hutchinsonii Kunth
Orycthanthus asplundii Kunth
Phorandendron cf piperoides (Kunth) Trell.
Struthanthus
Lythraceae
Cuphea racemosa (L.f.) Sprengel
Malpighiaceae
Bunchosia cf
Stigmaphyllon bogotense Triana & Planchon
Malvaceae
Sida rhombifolia L.
Wercklea feroz (Hook. F.) Fryxell
Marcgraviaceae
Marcgravia cf browneii (Tr. & Pl.) Krug & Urb
Melastomataceae
Aciotis cf
Axinaea cf lehhmanii Cogn ex Chair
Axinaea cf.
Arthrostema aff macrodesmum Gleason
Arthrostema cf
Blakea alternifolia (Gleason) Gleason

bh-pm
bh-mb
bh-mb
bh-mb
bh-pm/mb
bh-mb
bh-pm/mb
bh-mb
bh-mb
bh-mb
bh-mb
bh-pm
bll-mb
bh-pm
bh-pm
bh-mb
bh-mb
bh-mb
bh-mb
bh-pm/mb
bh-mb
bh-mb/pm
bh-mb
bll-mb
bh-mb
bh-pm
bh-mb
bh-pm
bll-mb
bh-mb
bll-mb
bh-pm
bll-mb
bh/ll-mb
bll-mb

tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
sh
tr
tr
tr

bh-pm

1300
1600
1600
1600
1300-1600
1400-2000
1400-1600
1500-1966
1684
1600
1600
1500
2200
1750
1400-1856
1400-1600
1650
1600
1966
1350-1966
1684
1600-1966
1600
1966
1500
1650
1600
1300
2200
1856
2200
1500-1686
1900-2000
1856-2200
2200
1550

CG 3064
CG 3181
CG 1098b
CG 1391
NG 018
CG 3399
CG 3123
CG
CG 1920
CG 3098
CG 3063a
RS 1046
CG 3542
CG 3092
CG 3301
CG 3188
CG 3255
CG 3526
CG 3186
RS 874
CG 3134
CG 3395
CG 3065
CG 3274
RS 872
CG 3144
CG 3012
CG 1095
CG 1759
CG 3165
CG 3303
CG 3272
CG 3233

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3074
CG 3294
CG 3270
CG 3235

CAUP
CAUP
CAUP
CAUP

OC 393

CAUP

CG 2861
BR 7732
BR
CG 1104
LA 27

CAUP
PSO
PSO
CAUP
CAUP

RS 1072

CAUP

CG 3525
RS

CAUP
AFP

RS 1070
CG

CAUP
CAUP

CG 3076

CAUP

MB 039
CG 3225
CG 2946
RS 792
CG 1349
CG 3028
CG 3220
CG 3040

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

ep
bll-mb
bh-pm
bh-pm
bh-pm

ep
ep
ep
ep

bh-pm

2500
1500
1600
1450
1450

hb
bh-pm
bh-pm
bh-pm
bh-pm

sh
cl
hb
tr

bh-mb

1600
1500
1600
1400
1400-1800

cl
bh-pm
bh/ll-mb
bh-mb
bh-pm
bll-mb
bh-pm
bll-mb
bh-pm
bh-pm

404

hb
tr
tr
sh
sh
tr
tr
cl
tr

1600
1856-2200
1800
1600
2300
1650
2200-2400
1450-1680
1700

Blakea punctulata (Triana) Wurdack
Blakea cf squamiger L. Uribe
Blakea
Bellucia calliptrata Gleason
Centronia phlomoides Tr.
Centronia
Clidemia
Conostegia cf. cuatrecasasii Gleason
Conostegia aff. icosandra (Sw.) Urban
Graffenrieda colombiana
Henrrietella L.
Huilaea mutissina
Killipia quadrangularis Gleason
Killipia
Leandra
Leandra
Miconia aff aegrotans Naud.
Miconia agregata Gleason
Miconia aeruginosa Naudin.
Miconai aff. brachygyna
Miconia aff caudata (Bonpl.) DC
Miconia cf. dolichopoda Naud.
Miconia floribunda (Bonpl.) C.CD.
Miconai aff. haughtii
Miconia aff. laetivirens L. Uribe
Miconia cf lehmannii Cognniaux
Miconia aff longifolia 8ªubl.) C.DC.
Miconia aff. triplinervis Ruiz & Pavon
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia
Miconia?
Miconia
Miconia
Miconia
Pterolepis trichomata (Rottb.) Cogn
Tibouchina
Tibouchina
Topobea
Topobea
Indet
Indet
Indet
Indet
Indet
Indet
Meliaceae
Carapa guianensis Aubl.
Guarea cf glabra Vahl.
Guarea cf guidonia L. (Sleum)
Guarea cf kunthiana A. Juss.
Guarea
Guarea
Guarea
Ruagea cf pubescens H. Karst.
Trichilia
Mimosaceae
Inga chartaceae
Inga densiflora Benth
Inga exalata T.S. Elias

bh-mb/pm
bh-mb
bh-mb
bh-pm
bh-pm
bh-mb
bh-pm
bh-pm
bll-mb
bll-mb
bh-mb
bh-mb
bh-pm
bll-mb
bh-pm
bll-mb
bh-pm
bh-pm
bh-pm/mb
bh-pm
bh-pm
bll-MB
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb

bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh/ll-pm/mb
bh/ll-pm/mb
bh-pm
bll-mb
bh/ll-pm
bh-pm
bh-pm
bh-pm
bh/ll-pm
bh-pm
bh-pm
bh-pm
bh-pm

405

tr
tr
tr
sh
tr
tr
tr
tr
tr
tr
sh
hb
hb
tr
tr
tr
tr
sh
sh
tr
tr
tr
tr
tr
tr
tr
tr
sh
sh
sh
sh
sh
tr
tr
sh
tr
tr
tr
sh
tr
tr
tr
tr
sh
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr

1400-2000
1750-1966
1400-1500
1684
1500
1686-1966
1600
1600
2150
2200
1100
1200
1650-1684
2400
1800
2200
1966
1856
1650-1966
1750
1950
2450-2550
1684-1950
1686-1700
1856
1500
1800
1800
1800
1800
2300
1450
1800
1250
1600
1500
1600
2100
1700
1856
1856-1950
1300
1686-1950
1350-1686
1600-1650
1600-2200
1600-2500
1686
2200
1900-1950
1600-1750
1500
1300
1856-1966
1300
1600
1450-1500
1500

CG 2876
OC 675
CG 3198
CG 3116
CG 1454
CG 3082
CG 3103
CG 3266
CG 3024
CG 3531
CG 1197
CG 1167
CG 1351
CG 1356
CG 3043
CG 406
CG 1878
CG 3365
CG 3111
CG 3111ª
CG 3119
CG 3199
CG 3291
CG 1760
CG 3073
CG 3209
CG 3109
CG 3258
CG 3104
CG 1896
CG 1434
CG 1180
CG 1281
CG 1202
LA 032
CG 1777
CG 1342
CG 3362
CG 1889
CG 1884
CG 3555
CG 3122
CG 1935
CG 3260
CG 3247
CG 1915
CG 3518
CG 3311
CG 3140
CG 3073
CG 1132

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
AFP,CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3070
CG 3080
CG 3011
CG 3039
CG 3516
CG 3079
CG 3364
CG 3281
CG 3189

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3096
CG 3400
CG 3245
CG 3158
CG 2886

CAUP
CAUP
CAUP
CAUP
CAUP

Inga
Inga
Inga
Inga
Monimiaceae
Mollinediia tomentosa (Benth.) Tull
Siparuna cf gesnerioIdes (Kunth) A..DC.
Siparuna laurifolia (Kunth) A. DC.
Siparuna cf lepidota A. DC.
Siparuna
Siparuna
Moraceae
Brosimum cf.
Clarisia cf.
Ficus cf gigantocyse Dugand
Ficus cuatrecassana
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Helicostylis tovarensis (Klotzsch & Karst.) Berg
Naucleopsis cf
Perebea xantochyma Dugand
Pseudolmedia laevigata Trecul.
Sorocea pubivena (Akker & Berg) C.C. Berg
Indet
Myristicaceae
Compsoneura aff. rigidifolia W. Rodriguez
Otoba gordonifolia A.C (Gentry)
Otoba lehmannii (A.C. Smith) Gentry
Otoba novogranatensis Moldenke
Myrsinaceae
Ardisia
Cybianthus aff montanus Lundell (Agost)
Cybianthus occigranatensis (Cuatrec.) Agosti
Cybianthus
Geissanthus cf mezianus Agost
Geissanthus cf
Myrsine coriaceae (Sw) R. Br. Ex Roem. & Sh
Myrsine
Myrtaceae
Myrcianthes rhopaloides (Kunth) Mc Vaugh
Psidium guianensis Sw.
Eugenia cf
Eugenia cf
Eugenia cf
Eugenia cf
Myrcianthes
Myrcia
Indet
Indet
Indet
Indet
Ochnaceae
Ouratea
Onagraceae
Fuchsia macrostigma Bentham
Fuchsia cf sessilifolia
Oxalidaceae
Oxalis
Papilionaceae

tr
bh-pm
bh-pm
bh/ll-pm
bll-mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bh-pm
bh/ll-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh/ll-pm
bh-mb
bh-pm
bh-mb
bh-pm
bh-pm
bh-pm
bh/ll-pm/mb
bh-pm
bh-pm/mb
bll-mb
bh-mb
bh-pm/mb
bh-pm
bh/ll-mb
bh-pm/mb
bh-pm
bh-pm
bh-mb
bh-pm
bh-pm
bh-mb
bh-pm/mb
bh-pm
bll-mb
bh-pm
bh-pm-mb
bh-pm
bh/bll-mb
bh-pm

tr
sh
sh
sh
sh
sh
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr

bh-pm

1300-1600
1400-1800
1800-1900
2300
1450-1856
1400-1500
1750
1300
1500
2200
1686
1950-2000
1300
1400
1500-1600
1400-1500
1600
1600
1300
1400-1600
1750-2000
1950
1300
1950
1300
1500-1600
1500
1500-2000
1700
1600
2200
1600-1856
1600-2000
1600
1684-2200
1500-1686
1500
1300
1856
1400-1500
1300-1600
1950
1650
1450
2300
1300
1300-1686
1300-1350
1950-2000
1650
1300

CG 1392
CG 3108

CAUP
CAUP

CG 3158
CG 3195
CG 3115
CG 1300
CG 3259
CG 1492

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3215
CG 1917
CG 3087
CG 3237
CG 3058
CG 3295
CG 3176
CG 1924
HB 169
LA 40
RS 882
CG 1517ª
CG 1517
CG 2987
CG 3124
CG 3293
CG 3152
CG 3292
CG 3153
HB 147

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3090
CG 3023
CG 1435
RS 773

CAUP
CAUP
CAUP
CAUP

CG 3229
CG 3044
CG 3538
NG 05
CG 3056
CG 3066
CG 3099
CG 3164

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3242
CG
CG 3536
CG 3282
CG 3046
CG 1907
CG 3300
CG 3169
CG 3174
CG 3163
CG 3118
CG 3045

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 3178

CAUP

CG 1428
CG 1220

CAUP
CAUP

BR 7727

CAUP

CG 3548
FG 2944

CAUP
CAUP

tr
bh-mb
bll-mb

sh
sh

bh-pm

1200-1800
2300
1300-1600

hb
bh-pm
bh-pm
bh-pm

406

hb
hb
sh

1400
1500
1350

Desmodium
Zornia curvata Mohl
Indet
Passifloraceae
Passiflora cumbalensis Karsten
Passiflora antioquensis H. Karst.
Passiflora
Passiflora
Piperaceae
Peperomia alberthssmithii Trell & Yunker
Peperomia alwynii Callejas & Betancur
Peperomia angularis C. DC.
Peperomia caliginigaudens Trell & Yunker
Peperomia hernandifolia (Vahl) A. Dietr.
Peperomia lancifolia Hooker
Peperomia obtusifolia (L. ) A. Dietr. Cf
Peperomia peltoidea Kunth
Peperomia striata Ruiz & Pavón
Pperomia swartziana Miquel
Peperomia
Peperomia
Peperomia
Peperomia
Peperomia
Peperomiia
Peperomia
Peperomia
Peperomia
Peperomia
Piper carpunya Ruiz & Pavón
Piper crassinervium H.B.K.
Piper curvinervium Callejas y Betancur
Piper dryadum C.DC.
Piper echinocaule Yunck
Piper lanciifolium C.DC.
Piper pilibracteum Trell & Yunck
Piper trianae C.DC.
Piper turnidinodum Yunck
Piper
Piper
Piper
Piper
Piper
Piper
Polygalaceae
Monnina pulcra Chodat
Monnina
Proteaceae
Panopsis sp nov
Panopsis sp nov
Rosaceae
Prunus integrifolia (Presl.) Walp.
Rubus robustus C. Presl.
Rubiaceae
Amphidasya
Borreria
Cinchona cf pubescens Vahl.
Coccocypselum
Dioicodendron dioicum cf (Sch. & Kr.) Steye
Elaeagia utilis (Goud) cf
Elaeagia cf pastoensis
Elaeagia
Elaeagia
Elaeagia
Elaeagia
Elaeagia
Faramea calyptrata C.M. Taylor

bll-mb
bh-pm
bh-mb
bh-pm
bh-mb
bh-mb/pm
bh-pm/mb
bh-pm
bll-mb
bh/ll-mb
bh-pm
bh-pm/mb
bll-mb
bll/h-mb/pm
bll-mb
bh-mb
bh-pm
bh-mb
bh-pm
bh-pm
bll-mb
bh-pm
bh-pm
bh-mb
bh-pm
bh-pm
bh/ll-pm/mb
bh-mb
bh/ll-mb
bh-pm
bh/ll-mb
bh-pm
bh-mb
bh/ll-mb
bh-pm
bh-mb
bh-mb
bll-mb
bh-pm
bll-mb
bh-mb
bll-mb
bh-mb
bh-pm
bll-mb
bh-pm
bh-pm
bh/ll-pm/mb
bh-mb
bh-pm/mb
bh/ll-pm/mb
bh-pm
bh-pm
bh/ll-pm/mb
bh-pm
bll-mb
bh-pm
bll-mb
bh/ll-pm/mb
bh-pm
bh-pm
bh-pm

407

cl
cl
cl
cl
cl
hb
cl
sh
sh
hb
sh
sh
sh
cl
hb
hb
hb
cl
hb
hb
cl
cl
hb
cl
sh
hb
sh
sh
cl
cl
hb
sh
sh
sh
sh
sh
sh
sh
sh
tr
tr
tr
tr
tr
cl
sh
hb
tr
hb
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr
tr

2500
1450-1550
1700
1200-1300
1800
1600-2000
1600
1300
2350
1655-2100
1100
1800
2100
1600-2300
1900
1800
1620
1800
1100
1000
2100
1300-1600
1300-1400
1900
1200
1450
1700-2100
1900
1800-2000
1450
1900-2200
1700
1900
1800-2200
1400
1800
1800
2100
1400
2250
2000
2000-2100
1500-1900
1300
2500
1600
1200-1300
1400-2400
1400-1600
1300-1700
1750-2000
1550
1400
2200
1684
2200
1600
1900-2000
1400-2500
1450-1650
1650
1600

CG 1485

CAUP

CG 1763
CG 1458
CG 1414
CG 3546

CAUP
CAUP
CAUP
CAUP

CG 1439
CG
CG 1163
CG 2915
ML 02
CG 2952
CG 2924
CG 2868
CG 2977
CG 3150
CG 2944
CG 2991
CG 2982
CG 2978
CG 2925
CG 2927
CG 1271
CG 2270
CG 1343
CG 1433
CG 1932
LA 010
CG 2862
CG 3127
CG 3557
LA 09
CG 2858ª
CG 2852
CG 3217
CG 3552
CG 3565
CG 2972
JG 017
CG 2949
CG 2674

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
AFP,CAUP
CAUP
CAUP
CAUP
AFP,CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 1321
CG 1430

CAUP
CAUP

CG 1774
CG 3097

CAUP,COL
CAUP,COL

CG 3167
CG

CAUP
CAUP

MB 037
CG 2931
CG 1487
CG
CG 3173
CG 3212
RS 1148
CG 1490
CG 3222
CG 3071
CG 3299
CG 1380
CG 3121
CG 3252
CG 3041

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

Faramea aff chlorophylla Muell-Arg
Faramea oblongifolia Standl.
Faramea
Galium cf hypocarpium (L.) Endl. Ex Griseb.
Gonzalagunia dependens Ruiz & Pavón
Guettarda crispiflora Vahl.
Guettarda turnefortiopsis Standl.
Hamelia?
Hoffmannia sprucei Standl.
Hoffmannia
Hoffmannia
Hoffmannia
Isertia pittieri (Standl.) Standl
Ladenbergia macrocarpa (Vahl.) Klotzsch
Ladenbergia magnifolia (R. & P.) Kl.
Ladenbergia
Palicourea crocea (sw.) R. & S.
Palicourea cutrecasasii Standl. & Steyerm
Palicourea gibbosa Dwter
Palicourea lasiantha K. Krause.
Palicourea killipii Standl.
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Palicourea
Posoqueria
Psychotria allenii Standl.
Psychotria aff beteriana D.C.
Psychotria gentryi (Dwyer) C.M. Taylor
Psychotria cf officinalis (Aubl.) Raeusch.
Psychotria
Psychotria
Psychotria
Psychotria
Psychotria
Psychotria
Psychotria
Rondeletia aff colombiana Rusby
Rondeletia cf glabrata Klotzsch
Rondeletia
Spermacoce
Indet
Indet
Indet
Indet
Indet
Rutaceae
Zanthoxylum
Sabiaceae
Meliosma
Sapindaceae
Allophylus mollis (H.B.K) Radlk
Allophylus
Allophylus
Cupania cinerea Poepp
Talaisia
Serjania
Tapirira cf guianensis Aubl.
Indet

bh-pm
bll-mb
bll-mb
bh-pm
bh-mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bh/ll-pm/mb
bh/ll-mb
bll-mb
bh-pm
bh-pm
bll-mb
bll-mb
bh/ll-mb
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bh-pm
bll-mb
bh-pm
bh-pm
bh-mb
bh-pm
bh-pm
bll-mb
bll-mb
bll-mb
bh-pm
bh-pm
bh-mb
bh-mb
bh-pm
bh-pm/mb
bh-pm
bh-pm
bll-mb
bh-pm
bll-mb
bh-pm/mb

hb
sh
tr
tr
tr
sh
sh
sh
sh
tr
tr
tr
tr
sh
tr
sh
tr
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
tr
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
sh
tr
tr
tr
hb
tr
tr
tr
tr
tr

bll-mb

2200-2500

CG 3014
MB 011
LA 014
CG 1278
CG 1246
CG 1341
CG 3358
CG 3203
CG 1106
LA 020
CG 1413
CG 3062
CG 3170
CG 1488
CG 1148
CG 3219
CG 3112
CG 2973
CG 1171
JG 034
CG 1401
CG 3310
CG 1415
CG 1427
CG 1397
RS 1111
RS 1164
RS 850
CG 1306
CG 1453
CG 1416
CG 1900
CG 3086
CG 3286
CG 3052
CG 1151
CG 1417
CG 1189
CG 1498
CG 1107
CG 3175
CG 3296
CG 1876
CG 3151
CG 3522
CG 2936
NG 012
CG 3246
CG 3308
CG 3309
CG 3231
CG 3052

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
AFP,CAUP
CAUP
CAUP,MO
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

1600

CG 1168

CAUP

1600
1300
1600
1600
1500-1600
1500-1650
1500-1600
1500-1600

CG 3527

CAUP

CG 1130
CG 3177
RS 878
RS 1059
HB 361
CG 1152
CG 1143
CG 1377

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

CG 1119

CAUP

1490
2300
2400-2500
1350
1800
1550
1600
1490-1500
1550-1700
1400-2000
1300
1400
1600
2200
1950-2000
2000-2400
2300
1600
1600
2200
2300
2000
1480
1600
1600
1600
2400
1500
2300
1500
1500
1950
1350-1650
1500
2400
2350
2100
1600
1300
1950
1900
1300
1300-1750
1100
1600
2200
1600
2200
1600-1850

tr
bh-pm
tr
bh-pm/mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm/mb
bh-pm
bh-pm
bh-pm
bh/ll-pm/mb
bll-mb

408

tr
tr
tr
tr
tr
cl
tr
tr
tr
tr

1300-1600
1600-2200
1950

Sapotaceae
Pouteria cf maguirei (Aubre) Penn.
Pouteria parcki (Ducke)
Pouteria cf wurdackii Aublet
Simaroubaceae
Picramnia cf magnifolia J.F. Macbr.
Solanaceae
Brugmansia cf alba L.
Cestrum
Cestrum
Cestrum
Cestrum
Cuatresia
Markea pilosa S. Knapp.
Physalis
Physalis
Solandra
Solanum aturense Dunal
Solanum nudum Dunal
Solanum
Solanum
Solanum
Solanum
Solanum
Solanum
Solanum
Solanum
Trianae
Staphyleaceae
Huertea cf glandulosa R. & P.
Turpinia occidentalis (Sw.) G. Don.
Styracaceae
Styrax tesmannii Perkins
Theaceae
Freziera sessiliflora A. Gentry
Tropaeolaceae
Tropaeolum deckerianum Maritz & Karsten
Ulmaceae
Trema micrantha 8L.) Blume
Urticaceae
Pilea daguensis Killip
Urera
Urtica
Verbenaceae
Aegiphila novogranatense Moldenke
Citharexylum aff kunthianum Moldenke
Stachitarpheta cayenensis Vahl.
Verbena litoralis Kunth
Violaceae
Viola
Viola
Vochysiaceae
Vochysia aff duquei Pilg

tr
bh-pm/mb

CG 3519
CG 3131

CAUP
CAUP

CG 3035

CAUP

CG 3390
CG 3138
CG 3521
CG 2940
RS 1026
LA 037
CG 3545
CG 1250
CG 1123
CG 1471
RS 881
CG 3113
CG 1370
LA 023
RS 1000
CG 3314
CG 2860
CG 1418
CG 1259
CG 1291
CG 3205

CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
AFP,CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP
CAUP

1600-1850

CG 3539
CG 1489

CAUP
CAUP

1850

CG 3376

CAUP

1500-2300

CG 3251

CAUP

-

CG 1286

CAUP

1600
1450
1450

-

-

RS 999
RS 156
LA 022

CAUP
CAUP
AFP,CAUP

CG 3137
RS 1080
RS 1081

CAUP
CAUP
CAUP

CG 3559
CG 1429

CAUP
CAUP

RS 966

CAUP

1500-1650
tr

bh-pm
bh-mb
bh-pm
bh-pm
bh-pm
bh-pm
bh-pm/mb
bll-mb
bh-pm
bh-mb
bh-pm
bh-pm-mb
bh-pm
bh-pm
bh-pm
bh-pm
bll-mb
bll-mb
bll-mb
bll-mb
bh-pm/mb
bh-pm
bh/ll-mb

sh
tr
tr
sh
sh
sh
sh
sh
sh
cl
sh
sh
tr
cl
sh
cl
sh
sh
sh
sh
cl
tr
tr

bh-mb

1400
1650-1966
1600
1400
1650
1450-1580
1500-1600
2300
1600
1900
1450-1580
1950-2000
1400
1450
1600
1400-1500
2000
2100
2300
2400
1950
1600
1600-2300

tr
bh-mb
tr
bh/ll-pm/mb
cl
bh-pm
tr
bh-pm
bh-pm
bh-pm

bh-mb
bh-pm
bh-pm
bh/ll-mb
bll-mb

cl
hb
hb
tr
tr
hb
hb
hb
hb

bh-pm

1600-1966
1600
1450
1900-2100
2000
1600

tr

409

12 Appendix 5

PLANTS OF TIPUTINI BIODIVERSIY STATION I. DICOTILEDONOUS. A
PRELIMINARY LIST
Carlos Gonzalez and Andy Jarvis

This is a preliminary list of tree plot collections at Tiputini Biodiversity Station. The
collection was made in ten 25m x 25m distributed evenly around TBS, and one 1-Ha
plot which was adapted to the HERB tree plot methodology after Pitman (2002)
established the plot.
The collections can be found in the University San Francisco of Quito herbarium, and
copies in the National Herbarium, Quito and the herbarium of the Universidad
Catolica. Photos have been takn f most individuals, and are available from the
authors upon request.
The tree database includes 2375 individuals which have been marked, measured and
identified. Of these, some 603 species have been separated, and this list reports these
findings. The database is in a continual state of improvement, and it is stressed that
the identifications made here are preliminary.

Taxon

Anacardiaceae
Astronium cf
Spondias cf mombin L.
Spondias cf venulosa (Endl.) Endl
Spondias cf
Tapiria cf guianensis
Anacardium cf (sp novnigelperu)
Indet
Annonaceae
Crematosperma cauliflorum aff
Duguettia spixiana Mart.
Duguettia
Duguettia
Guatteria cargadero cf
Guatteria
Guatteria cf
Guatteria cf
Guatteria
Annona cf
Rollinia
Xylopia
Indet
Pseudomolmea
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet (Guatteria?)
Indet
Indet
Apocynaceae
Aspidosperma
Aspidosperma

No
individuals

Elevation
(m)

Collection

Photo
Gallery

199-200
261
227
227

CG 3672
CG 4413
CG 4550
CG 3604
CG 4423
CG 4252
CG 4112

1516A/B
55A
1502A/B
1172A/B
697A/B

1
6
3

227
200-210

CG 4442
CG 4204
CG 4197

3
1
1
1
1
1
1
1
1
1
1
1
1
2
1
3
1
1
1
1

220
227
219
261
238
223
227
261
196
217
200
227
227
227
227
227
227
217
227

CG 4225
CG 4093
CG 3679
CG 4436
CG 4493
CG 4302
CG 4172
CG 4424
CG 3753
CG 3759
CG 3805
CG 4372
CG 4070
CG 4158
CG 4142
CG 4178
CG 4323
CG 3777
CG 4220
CG 3596

1648A/B
916A/B
938A/B
1294A/B
988A/B
641A/B
1652A/B
1754A/B
850AA/BB
1503A/B
240A/B
268A/B
342A
672A
744A/B
778A/B
857A/B
1319A/B
235A
970A/B
22A/B

1

219

CG 4362
CG 3644

1463A/B
40A/B

1
1
1
1
1
1
1

410

219
261

Couma
Couma
Couma
Himathantus
Lachmellea
Lachmellea
Aquifoliaceae
Ilex
Ilex cf inundata
Ilex
Araliaceae
Dendropanax arboreus (L.) Dec & Pl
Dendropanax caucanus (Harm.) Har
Dendropanax
Arecaceae
Astrocarium chambira Burret
Astrocarium murumuru Mart.
Euterpe precatorae
Geonoma maxima cf
Iriarthea deltoidea Ruiz & Pavon
Oenocarpus bataua Mart.
Socratea exorrhiza (Mart.) H. Wendl.
Wettinia maynensis Spruce
Indet sp1
Bignoniaceae
Jacaranda copaia (Aublet) D. Don
Memora cladothica Sandwith
Tabebuia serratifolia (Vahl.) G. Nich
Tabebuia cf
Indet
Bombacaceae
Ceiba
Matisia cordata Bonpl
Matisia cf longiflora Gleason
Matisia cf bracteolosa Ducke
Matisia sp2
Ochroma pyramidale (Cav. ex Lam.) U.
Pachira acuatica Aubl.
Quararibea
Quararibea
Quararibea
Indet
Boraginaceae
Cordia
Cordia
Indet
Burseraceae
Crepidospermum rhoifolium Bent (T& Pl)
Crepidospermum
Protium ecuadorense Benoist
Protium cf fimbriatum Swart
Protium glabrescens Swart
Protium polybotrium (Turcz.) Engl
Protium cf robustum (Swart.) DM Porter
Protium trifoliolatum Engl.
Tetragastris panamensis (Engl.) Kuntz
Trattinikia
Protium
Protium
Protium
Protium
Protium
Protium
Protium
Protium
Indet
Indet

1
1
1
1
1
1

261
227
227
199
227
227

CG 4419
CG 4046
CG 4177
CG 3606
CG 4199
CG 4245

605A/B
864A
918A/B
1037A/B

1
1
1

217
220
220

CG 3795
CG 3891
CG 3926

518A/B
532A/B

4
3
1

227
219-261
220

CG 4067
CG 3676
CG 3882

673A/B
2098/99
3409/11

6
1
4
17
25
1
2
1
10

217-220
200
219-261
227
199-227

CG 3796
CG 3849
CG 4348
CG 4057
CG 3625
CG 4552
CG 3847
CG 3904
CG 3699

290A/B
394A
49A/B
360A/B
256A/B
151A/B

4
11

219-261
227

-

219
261

CG 4278
CG 4208
CG 4468
CG 4337
CG 4367

1482A/B
655A/B
1593A/B
1434A/B
1524A/B

1
3
13
2
19
1
1
1
2
1

219
220
217-227
227
199-261
100
227
261
238
223

CG 3651
CG 3714
CG 4087
CG 4247
CG 3608
CG 4161
CG 4395
CG 4454
CG 4515
CG

187A/B
102A/B
1198A/B
644A/B
1547A/B
1646B
1695B
1690A/B

1
1
1

199
227
228

CG 3621
CG 4163
CG 4482

42A
898A/B
1628A/B

CG 4075
CG 4457
CG 3839
CG 3916
CG 3818
CG 3731
CG 4053
CG 4418
CG 4481
CG 4368
CG 4157
CG 4386
CG 4531
CG 4040
CG 4159
CG 4356
CG 4157
CG 4439
CG 4036
CG 4184

665A/B
1636A/B
370A/B
570A/B
399A/B
212A/B
595B
1497A/B
1583A/B
1513A/B
751B
1533A/B
743A/B
1459A
887A/B
1665A/B
619A
835A/B

1
2
1
1
1
1
1
1
5
2
1
1
1
-

411

200-227
217
199-21

199
238
199-200
220
220
217
227
261
238
261
219-227
238-261
223
227
227
219
227
238
227
227

Protium
Capparaceae
Capparis
Caricaceae
Jacaratia cf digitata (Poepp & Endl.) Solm
Cecropiaceae
Cecropia tomentosa
Cecropia sciadophylla Mart.
Cecropia ficifolia
Cecropia
Cecropia
Cecropia
Cecropia
Coussapoa cf villosa Poepp & Endl.
Pourouma bicolor Mart.
Pourouma ferruginea cf
Pourouma guianensis Aubl.
Pourouma napoensis C. Berg.
Pourouma cf ovata
Pourouma cf villosa
Pourouma sp1
Pourouma sp?
Pourouma
Pourouma
Celastraceae
Maytenus
Maytenus
Maytenus
Chrysobalanaceae
Couepia
Couepia
Couepia
Couepia
Licania cf elliptica Standl
Licania cf caudate Prance
Licania glablanca
Licania
Licania
Licania
Licania
Licania
Licania
Licania
Licania
Licania
Licania
Parinari cf
Indet
Indet
Clusiaceae
Chrysoclamis cf membranacea Pl & Tr
Chrysoclamis
Tovomita cf amazonica
Vismia cf sprucei Sprage
Vismia
Indet
Combretaceae
Buchenavia
Dichapetalaceae
Dichapetalum cf rugosum
Tapura
Elaeocarpaceae
Sloanea
Sloanea
Sloanea
Sloanea
Sloanea

1704A/B
1

261

CG 4405

1558A/B

5

219-227

CG 4553

625A/B

4
4
3
3
1
1
2
2
1
?
4
1
?
2
3
3
-

217-227
219-261
219-227
220-227
227
227
200
199
219
?
223-238
223
?
199-227
199-261
219-227
227

CG 3718
CG 4345
CG 3721
CG 4276
CG 4301
CG 3885A
CG 4121
CG 3833
CG 3600
CG 3681
CG 3710
CG 3781
CG 4526
CG 4474
CG 3609
CG 3599
CG 4377
CG 4271

293A/B
1422A/B
158A/B
690A/B
369A/B
50A/B
200B
243A/B
886A/B
1291A/02

1
1
1

210
238
227

CG 3867
CG 4444
CG 4265

447A/B
1642A/B
1122B

1
2
1
1
1
1
1
1
2
1
1
1
1
1
1
1
1
2

217
261
227
227
227
219

199
217
200
210
238
227
227
227
227
227
219
227

CG 3732
CG 4378
CG 4320
CG 4305
CG 4197
CG 3695
CG 4341
CG 4524
CG 3581
CG 3769A
CG 3801
CG 3861
CG 4451
CG 4211
CG 4244
CG 4269
CG 4081
CG 4216
CG 3887
CG 4102

258A/B
1523A/B
1320A/B
1388A/B
938AA/B
166A
1418A/B
1727A/B
277A/B
344A/B
423A/B
1008A/B
1036A/B
1104A/B

1
1
2

223
199
217
217
227-261
219

CG 3589
CG 4510
CG 3750
CG 3756?
CG 4127
CG 3712?

1683A/B
265A/B
812A/B
105A/B

1

227

CG 4090

632A/B

-

217
261

CG 3730
CG 4387

222A
1537A

1
1
4
-

199
261
217-219
217
200

CG 3628
CG 4404
CG 3762
CG 3780
CG 3825

72A/B
1571A/B
275A/B
283A/B
331A/B

2

412

1000A/B
-

Sloanea
Sloanea
Sloanea
Sloanea
Euphorbiaceae
Pausandra trianae (Muell. Arg.) Baill.
Acidoton nicaraguensis
Hasseltia floribunda
Alchornea triplinervia (Sprengel) Muell. Arg
Conceveiba
Alchornea latifolia Sw
Hyeronima alchorneiodes Allemao
Mabea “comun”
Mabea superbrondu
Mabea occidentalis Benth.
Mabea
Phyllanthus cf urinaria L.
Richeria racemosa
Alchornea triplinervia
Acalypha cf
Acalypha cf cuneata Poepp & Endl.
Sapium cf ovobatum K ex Müll. Arg
Croton
Nealchornea cf
Pera cf
Pera duguet
Hyeronima
Hyeronima cf
Acalypha cf
Indet
Richeria racemosa
Indet
Indet
Indet
Indet
Indet
Fabaceae
Inga auristellae Harms
Inga
Inga
Inga velutina Willd
Inga rusbyi Pittier
Inga brachyrachis/capitata
Inga aff umbratica
Inga
Inga
Inga cordatoalata Ducke
Inga umbratica Poepp
Inga
Inga acreana Harms
Inga
Inga
Inga multijuga o ruiziana
Inga cf microcoma Harms
Inga sarayacuensis T.D.Penn
Inga
Inga alata
Inga
Inga
Inga tenuistepula Ducke
Inga cf splendens Willd.
Inga
Inga
Inga
Inga umbellifera (Vahl) Steud ex DC.
Inga cf stenoptera Benth.
Inga spectabilis (vahl) Willd.

1
1
1
1

238
217
227
199

CG 4499
CG 4097
CG 4203
CG 4499

735A/B
1603A/B

7

219-227

1
1
3
2
6
1
1
1
1
5
1
1
1
1
1
1
1
1
5
1
1
1
2
1

227
199
227
227
227
217-219
220-223
223
227
227
227
261
227
220
261
227
220
227
227
227
227
227
217
217
220
238
227

CG 3698
CG 3782
CG 4171
CG 3624
CG 4288
CG 4171
CG 4084
CG 3768
CG 3807
CG 4546
CG 4232
CG 4105`
CG 4240
CG 4433
CG 4218
CG 3906
CG 4390
CG 4230
CG 3927
CG 4318
CG 4250
CG 3601
CG 4496
CG 4218
CG 4404
CG 3774
CG 3787
CG 3923
CG 4475
758
CG 4262

1003A/B
232A/B
1071A/B
51A/B
1223A/B
877A/B
648A/B
280A/B
339A/B
1678A/B
1081A/B
710A/B
1068A/B
1509A/B
993A/B
585A/B
1554A/B
1094A/B
547A/B
1367A/B
1185A/B
993
274A/B
-

1
1
1
1

200
199
238
199
219
219

1
1
1
1
1
2
1
7
2
1
1
1
2
1
2

219
219
219
219
217
217
217
217
217-227
200
200
210
210
210
210
220
219
219
261
261
261
261
261-223

CG 4035
CG 3583
CG 4484
CG 3588
CG 3662
CG 3665
CG 4455
CG 3667
CG 3682
CG 3653
CG 3655
CG 3778
CG 3741
CG 3740
CG 3913
CG 3745
CG 3798
CG 3763
CG 3873
CG 3856
CG 3855
CG 3872
CG 4141
CG 4350
CG 4338
CG 4311
CG 4523
CG 4432
CG 4434
CG 4426

10A/B
1618AA/B
16A/B
1668A/B
186A/B
193A/B
196A/B
220A/B
252A/B
343A/B
285A/B
452A/B
461A/B
1430A/B
1437A/B
1501A/B
1724A/B

413

Inga capitata Desvaux
Inga
Inga
Inga cf umbelliferae (Vahl) Steud. ex DC.
Inga velutina Willd.
Inga
Inga
Inga
Inga
Inga cf suaveolens Ducke
Inga spectabilis Vahl. Willd
Inga aff setulifera T.D. Penn
Inga
Inga
Inga punctata
Inga tessmannii Harms
Inga
Inga
Inga marginata
Inga aff heterophylla Willd
Inga
Abarema jupunba
Lecointea peruviana
Bauhinia brachycalyx
Dalbergia frutescens (Vell.) Britton
Macrolobium angustifolium (Benth.) Cowman
Macrolobium gracile Spruce & benth
Macrolobium archeri R.S.Cown
Macrolobium colombianum aff
Zygia heteroneura Barneby & Crimes
Hymenae oblongifolia
Zygia aff latifolia (L.) Fawcett & Rendle
Zygia schultzeana
Zygia?
Indet
Brownea macrophylla Hort ex Mast
Calliandra trinervia
Marmaroxillon
Indet
Dalbergia
Dussia
Zygia
Browneopsis ucayalina Huber
Macrolobium
Swartzia multujuga
Brownwea grandiceps
Flacourtiaceae
Casearia arborea (Rich.)Urb.
Casearia
Casearia cf javitensis Kunth
Casearia nigricans Sleumer
Casearia
Carpotroche cf
Lozania
Ryania cf speciosa Vahl
Tetratylacium macrophyllum Poepp
Casearia cf prunifolia Kunth( Ryania en list)
Casearia
Hasseltia floribunda Sw
Casearia
Neosprucea cf
Lacistema cf
Casearia
Casearia
Casearia
Hippocrateaceae
Salacia cf spectabilis Ac. Smith

1
2
2
1
1
1
9
1

261
227-261
261
261
261
238
238
223
223-238
227

1
1
5
1

227
227
227
227

1

227

1

227

1
1
1

210
217
227-238

6
3
1
1
1
1
3
1

227
227
227
277

1
1
1
2
78
5

277
277
277
261
223-238
217

1
7

1
1

219
277
277

277
277
277
277
199
227
277
217-277
277
277
277
277
277
261
238
277

414

CG 4473
CG 4394
CG 4408
CG 4556
CG 4428
CG 4403
CG 4473
CG 4513
CG 4540
CG 4513
CG 4111
CG 4523
CG 4099
CG 4215
CG 4213
CG 4289
CG 3789
CG 4285
CG 4374
CG 4272
CG 3794
CG 3871
CG 3769
CG 4452
CG 3822
CG 3636
CG 3709
CG 3744
CG 3686
CG 4086
CG 4088
CG 4476
CG 3589
CG 4505
CG 4338
CG 4189
CG 4293
CG 4410
CG 4100
CG 4188
CG 4479
CG 4374
CG 4328
CG 3728

1602A/B
1539A/B
1541A/B
1760A/B
1550A/B
1556A/B
1602A/B
1691A/B
703A/B
1763A/B
716A/B
1001A/B
1005A/B
1221A/B
250A/B
1235A/B
1481A/B
1282A/B
296A/B
492A/B
272A/B
1669A/B
381A/B
78A/B
128A/B
251A/B
132A/B
645A/B
643A/B
1602A/B
1555A
1732A/B
1437A/B
937A/B
1204A/B
1532A/B
734A/B
831A/B
1592A/B
1481A/B
1450A/B
940A/B

CG 3622

32A/B

CG 4296
CG 4257
CG
CG 4257
CG 4414
CG 3642
CG 4060
CG 4166
CG 3754
CG 4096
CG 4450
CG 4115
CG 4257
CG 4535
CG 4132
CG 4431
CG 4450
CG 4221

432A/B
813A/B
1511A/B
69A/B
593A/B
866A/B
301A/B
738A/B
843A/B
701A/B
1148A/B
1714A/B
803A/B
1505A/B
1660A/B
980A/B

CG 4226

921A/B

Salacia
Humiriaceae
Vantanea
Ventanea
Lauraceae
Cinnamomun triplinervia
Endlicheria
Edlicheria
Nectandra
Nectandra
Nectandra
Nectandra
Ocotea?
Ocotea?
Ocotea?
Ocotea
Ocotea
Persea
Mezilaurus
Rhodostemonodaphne
Indet
Indet.
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Lecythidaceae
Eschweilera andina
Eschweilera cf rufifolia O tessma S.A. Mori
Eschweilera
Eschweilera
Eschweilera
Grias neuberthii J.F. Macbr
Gustavia longifolia Poepp ex berg
Lecythis

CG 4077

661A/B

CG 3770
CG 4343

277A/B
1412A/B

277
277
277
277
277
277

CG 4174
CG 4168
CG 4463
CG
CG 4421
CG 4398
CG 4449
CG 3585
CG 3617
CG 3602
CG 4420
CG 4101
CG 4180
CG 3712
CG 3790
CG 3817
CG 3611
CG 3634
CG 3717
CG 3671
CG 3652
CG 3775
CG 3776
CG 3760
CG 3770
CG 3799
CG 3869
CG 3875
CG 3874
CG 3859
CG 3920
CG 4393
CG 4527
CG 4514
CG 4528
CG 4561
CG 4164
CG 4198
CG 4207
CG 4236
CG 4228
CG 4227
CG 4268
CG 4260
CG 4290
cf
CG 4273
CG 4279
CG 4085
CG 4125
CG 4185
CG 4280

873A/B
881A/B
1587A/B
8
1530A/B
1564A/B
1657A/B
6
38
57A/B
1488A/B
730A/B
854A/B
105A/B
297A/B
307A/B
24A/B
76A/B
107A/B
177A/B
194A/B
233A/B
234A/B
267A/B
297A/B
377A/B
413A/B
432A/B
442A/B
489A/B
500A/B
1542A/B
1686A/B
1715A/B
1723A/B
1738A/B
888A/B
949A/B
1026A/B
1067A/B
1095A/B
1098A/B
1107A/B
1142A/B
1218A/B
1266
1276A/B
1278A/B
647A/B
819A/B
840A/B
1240A/B
1706A/B

222-277
223-261
199
261
238
222
220-277
200

CG 3702
CG 4416
CG 3590
CG 4370
CG 4469
CG 3700
CG 3632
CG 3865

154A/B
1577A/B
7A/B
1477A/B
1594A/B
115A/B
68A/B
478A/B

219

3
1
1
2
1
1
2
1
1
2
1
1
3
1
1
2
3
1
3
2
1
1
2
1
1
2
3
1
1
1
1
3
1
1
1
1
1
3
1
1
1
1
2
1
1
1
1
1

8
2
1
4
1
2
8
1

415

277
277
238
199
261
261
238-277
199
199
199-217
261
219
227
222-219
217
199-227
199
222-227
199-200
222
217-277
217-238
217
217
200
200
200
200-261
200-222
220
261
223
223
223-277
223
277
277
277
277
200-277
277
277
277
277

Lecythis
Indet
Malpighiaceae
Byrsonima cf
Melastomataceae
Blakea
Miconia
Miconia
Miconia
Miconia
Miconia
Indet
Indet
Indet
Indet
Indet
Indet
Mouriri
Mouriri
Indet
Meliaceae
Cedrela cf
Guarea cf goma Pulle
Guarea pterorachis Harms
Guarea purusana cf
Trichillia septentrionalis
Guarea
Guarea
Guarea “gomma”
Guarea
Cabralea cangeriana
Guarea silvatica
Guarea guentheriana
Guarea
Trichilia
Guarea
Guarea
Guarea
Trichilia
Trichilia
Indet
Indet
Menispermaceae
Abuta cf
Monimiaceae
Mollinedia
Siparuna
Siparuna
Moraceae
Brosimum
Brosimum
Castilla cf
Castilla
Clarisia
Clarisia
Ficus
Ficus
Ficus
Ficus
Ficus
Ficus
Helicostylis
Maquira
Naucleopsis
Naucleopsis cf
Naucleopsis
Perebea

2
2

223-238
277

CG 4462
CG 4154

1581A/B
?

1

238

CG 4472

4A/B

2
1
1
5
1
1
1
1
1
1
1
1
1
1

223-261
217
261
223-261
277
277
217
200
200
200
223
277
223
277

CG 4567
CG 3784
CG 4425
CG 4506
CG 4044
CG 4150
CG 3771
CG 3803
CG 3826
CG 3868
CG 4539
CG 4104
CG 4533
CG 4267

1747A/B
?
1492A/B
1701A/B
610?
765A/B
279?
318A/B
403A/B
494A/B
1725?
712A/B
1716A/B
1111?
1707A/B

1
1
7
5

261
219
200-220
277

1
1
6
5
1
2
12
3
1
7
1
1
16
7
2
1

222
217
220-277
220-238
261
261-277
220-277
238-220
223
277
277
277
222-277
238-261
277
277

CG 4417
CG 4357
CG 3736
CG 4063
CG 3650
CG 3715
CG 3772
CG 3928
CG 3912
CG 4435
CG 4415
CG 4466
CG 4448
CG 4545
CG 4176
CG 4217
CG 4306
CG 3684
CG 4368
CG 4210
CG 4316

1476A/B
1448A/B
242A/B
589A/B
195A/B
122?
236?
527A/B
542A/B
1508A/B
1517A/B
1579A/B
1650A/B
1687A/B
847A/B
992A/B
1392?
150?
1479A/B
1009?
1325A/B

1

219

CG 4363

1464

1
1
1

277
277
277

CG 4147
CG 4258
CG 4324

774A/B
1147A/B
1317A/B

2
1
3
4
4
2
2
1
2
1
1
2
2
1
1
2

200
277
217-261
238-277
199-277
261-277
222
238
223
277
277
199
277
261
277
222-217

6

220-222

CG 3832
CG 4275
CG 3751
CG 4478
CG 3603
CG 4382
CG 3720
CG 4497
CG 4518
CG 4287
CG 4283
CG 3703
CG 4095
CG 4411
CG 4195
CG 3737
CG 3841
CG 3673

357A/B
1269A/B
246?
1627?
54A/B
1572A/B
97
1624A/B
1731A/B
1229A/B
1258?
93?
737A/B
1535A/B
924A/B
214A/B
383A/B
180?

416

Poulcenia armata
Pseudolmedia
Sorocea
Sorocea
Batocarpus “ramaroja”
Sorocea
Sorocea
Sorocea?
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Myristicaceae
Otoba parvifolia cf
Compsoneura
Compsoneura
Virola duckei A.C. Smith
Virola mollis (AC. DC) Warb
Virola multinervia
Virola obovata
Virola flexuosa
Virola dixonii
Virola
Virola
Virola
Virola
Virola
Virola
Virola pavoni
Iryanthera ulei
Iryanthera
Otoba
Iryanthera
Otoba glycicarpa
Otoba
Myrsinaceae
Ardisia
Ardisia “semiovada”
Cybianthus
Stylogyne
Myrtaceae
Eugenia
Eugenia
Eugenia
Eugenia
Eugenia aff stipitata
Eugenia
Eugenia feijoi
Calophyllum
Plinia
Plinia
Indet
Indet
Indet
Indet
Indet
Eugenia
Indet
Indet
Indet

1
1
2
56
1
2
1
3
2
2
1
1
2
1
1
5
9
2
1

417

219
238
199
200-222
220
219
277
217
199-238
222-217
217
222
222
217-22
200
200-277
238
277
277

CG 4344
CG 4438
CG 3689
CG 3800
CG 3884
CG 4349
CG 4291
CG 3691
CG 3626
CG 3725
CG 3669
CG 3677
CG 3739
CG 3758
CG 3863
CG 3911
CG 4487
CG 4069
CG 4325

1424A/B
1670A/B
143A/B
337A/B
516A/B
1432A/B
1215A/B
112?
18?
125A/B
147A/B
189?
244A/B
261A/B
486A/B
582?
1597?
675A/B
1306A/B

CG 4456
CG 4128
CG 4446
CG 4340
CG 3619
CG 4234
CG 3866
CG 4536
CG 3791
CG 4461
CG 4091
CG 4264
CG 4091
CG 3845
CG 3680
CG
CG
CG 4308
CG 4559
CG 3917
CG 3685

1647A/B
816A/B
1640A/B
1411A/B
44?
1072A/B
417?
1712A/B
205A/B
1634?
634A/B
1128A/B
634A/B
367A/B
174A/B
1009A/B
718?
1384A/B
1749?
550A/B
135A/B
1702A/B

CG 3795
CG 4455
CG 4130
CG 4335

303A/B
1641A/B
808A/B
1436A/B

CG 3645
CG 3899
CG 4052
CG 4183
CG 4292
CG 4358
CG 4360
CG 4134
CG 4229
CG 3910
CG 3706
CG 3761
CG 3804
CG 3838
CG 4360
CG 4422
CG 4483
CG 4465
CG 4555

43A/B
502A/B
598?
836A/B
1213A/B
1475A/B
1460A/B
796A/B
1024A/B
538A/B
98?
288?
341A/B
396A/B
1460A/B
1504A/B
1584A/B
1588A
1762A/B

Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Plinia
Indet
Indet
Nyctaginaceae
Neea “supercrasa”
Neea
Neea “altomina”
Neea “popular”
Ochnaceae
Ouratea “flaquita”
Olacaceae
Heisteria
Heisteria
Heisteria
Indet
Indet
Piperaceae
Piper
Polygonaceae
Triplaris Americana
Coccoloba fallax
Coccoloba densifrons
Indet
Proteaceae
Roupala montana
Rhamnaceae
Colubrina arborescens (Mill.) Sarq.
Rubiaceae
Borojoa
Duroia hirsuta cf
Faramea glandulosa Poepp
Psychotria
Pentagonia spathicalyx
Pentagonia
Posoqueria latifolia
Posoqueria?
Posoqueria?
Cousarea brevicaulis Krause
Cousarea cephaloides Taylor
Randia
Randia
Simira cordifolia
Simira wurdackii Steyerm.
Wittmackanthus stanleyanus cf
Psychotria stenostachya
Randia armata (Sw.) DC.
Indet
Indet
Cousarea
Simira
Alseis
Cousarea cf
Indet
Ladenbergia
Indet
Indet
Cousarea macrophylla aff

CG 4549
CG 4551
CG 4037
CG 4119
CG 4152
CG
CG 4175
CG 3901
CG 4352
CG 4392
CG 4512
CG 4542
CG 4520

1767?
1768?
618A/B
691A/B
762A/B
86A/B
861A/B
575A/B
1454A/B
1543A/B
1680A/B
1693A/B
1722A/B

CG 3877
CG 4488
CG 4502
CG 4117

485A/B
1672A/B
1728A/B
695A/B

CG 4467

1622A/B

CG 4315
CG 4266
CG 3769
CG 4076
CG 4544

1333A/B
1109A/B
269A/B
658?
1692?

CG

1459A/B

CG 3815
CG 4110
CG 3919
CG 4133

1429A/B
704A/B
543A
800?

CG

59A/B
1607

CG 4193
CG 4353
CG 4565
CG 4445
CG 3713
CG 4329
CG 4108
CG 4169
CG 3816
CG 3812
CG 4277
CG 3659
CG 4253
CG 4107
CG 4492
CG 4071
CG 3616
CG 4355
CG 3584
CG 3593
CG 3711
CG 3663
CG 3729
CG 3813
CG 3876
CG 3925
CG 4492
CG 4534
CG 4530

418

925A/B
1449A/B
1748A/B
1644A/B
104A/B
1439A/B
688A/B
880A/B
321A/B
335A/B
1262A/B
208A/B
1178A/B
700A/B
1643A/B
671A/B
788A/B
1458A/B
12?
19?
94A/B
209A/B
228A/B
317A/B
483A/B
526?
1643A/B
1711A/B
1719A/B

Indet
Indet
Indet
Alibertia
Indet
Indet
Sabiaceae
Discophora guianensis
Sapindaceae
Allophylus
Allophylus
Allophylus
Talisia
Indet
Sapotaceae
Chromolucuma cf
Chrysophyllum
Sarcaulus
Micropholis venulosum
Micropholis “molihoniana”
Pouteria bilocularis
Pouteria
Pouteria
Pouteria
Pouteria
Pouteria
Pouteria
Pouteria
Chrysophyllum venezuelanense (Pierre)Penn
Pouteria vernicosa T.D. Penn
Pouteria quianensis Aubl.
Pouteria
Pouteria multiflora (A.DC) Eyme
Pouteria platiphylla cf
Pouteria
Pouteria
Pouteria
Pouteria
Micropholis
Simaroubaceae
Picramnia
Simaba “smed”
Simaruba
Solanaceae
Cestrum
Solanum
Staphyleaceae
Turpinia occidentalis
Tapura
Sterculiaceae
Herrania
Theobroma subincanum cf
Theobroma speciosum cf
Styracaceae
Styrax argenteus
Theophrastaceae
Clavija
Tiliaceae
Apeiba membranacea
Apeiba aspera
Apeiba
Mollia gracilis
Ulmaceae
Celtis schippii Standl.
Celtis
Trema micrantha
Urticaceae

419

CG 4558
CG 4522
CG 4554
CG 4223
CG 4322
CG 4298

1742?
1750A/B
1765?
983A/B
1321A/B
1403A/B

CG 4079

657A/B

CG 4347
CG 4409
CG 4146
CG 4330
CG 4326

1425A/B
1544A/B
770A/B
1446A/B
1305?

CG 3705
CG 4508
CG 4205
CG 3828A
CG 3757
CG 4034
CG 3765
CG 3620
CG 3704
CG 3708
CG 3724
CG 3690
CG 3701
CG 3693
CG
CG 4224
CG 3657
CG 3746
CG
CG 3654
CG 3860
CG 4489
CG 4504

134A/B
1684A/B
915A/B
374A/B
1614A/B
623A/B
80A/B
48A/B
88A/B
100A/B
108A/B
118A/B
124A/B
157A/B
215A/B
986A/B
188A/B
238A/B
1068?
183A/B
443A/B
1655?
1733A/B
1619A/B

CG 4373
CG 4156
CG 4196

1522A/B
752A/B
939A/B

CG 4300
CG 4192

1398A/B
957A/B

CG 4051
CG 4162

599A/B
894A/B

CG 4179
CG 3914
CG 4443

856A/B
551A/B
1651A/B

CG 3648

60A/B

CG 3905

565A/B

CG 4047
CG 4041
CG 3900
CG 4209

606A/B
612A/B
524A/B
1014A/B

CG
CG 4491
CG

512A/B
1617A/B
121A/B

Urera
Urera cf
Violaceae
Gloesospermum ecuadorensis
Leonia crassa
Leonia
Rinorea apiculata Hekking
Rinorea lindeniana (Tul.) Kuntze
Rinorea viridifolia Rusby
Vochysiaceae
Qualea paraensis
Vochysia
Collected PLOT 1-9 and problems PITMAN
Indet
Indet
Annonaceae-Guatteria
Indet
Indet
Indet
Fabaceae
Fabaceae
Indet
Moraceae
Indet
Lauraceae
Indet
Meliaceae?
Fabaceae?
Fabaceae/Inga
Indet
Annonaceae
Indet
Sterculiaceae-Theobroma
Indet
Rubiaceae?
Annonaceae cf
Theaceae?
Indet
Indet
Indet
Indet
Anacardiaceae/Burseraceae
Euphorbiaceae- Mabea
Fabaceae?
Indet
Indet
INdet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet
Indet

CG 4098
CG 4256

714?
1156

CG 3892
CG 4136
CG
CG 4045
CG 4282
CG

506A/B
791A/B
1438A/B
607A/B
1252A/B
1435A/B

CG 3842
CG 4511

348A/B
1689A/B
H1708
1710A/B
1713A/B
1726A/B
1735A/B
1743A/B
1752B
26A/B
161A/B
00A/B
218A/B
231A/B
255A/B
262A/B
531A/B
325A/B
332A/B
372A/B
000A/B
429A/B
531A/B
541A/B
564A/B
1413A/B
1423A/B
1468A/B
1607A/B
1632A/B
1667A/B
273A/B
1759A/B
273A/B
8A/B
1769A/B
1551A/B
1755A/B
298A/B
438A/B
1736A/B
450A/B
90A/B
528A/B
H86A/B
PPP1083
1273A/B
1295A/B
1301A/B
1329A/B
1336A/B
1369A/B
882A/B

420