IWMI Research Report Evaluating the Flow Regulating Effects of Ecosystems in the 166 Mekong and Volta River Basins Guillaume Lacombe and Matthew McCartney Research Reports The publications in this series cover a wide range of subjects—from computer modeling to experience with water user associations—and vary in content from directly applicable research to more basic studies, on which applied work ultimately depends. Some research reports are narrowly focused, analytical and detailed empirical studies; others are wide-ranging and synthetic overviews of generic problems. Although most of the reports are published by IWMI staff and their collaborators, we welcome contributions from others. Each report is reviewed internally by IWMI staff, and by external reviewers. The reports are published and distributed both in hard copy and electronically (www.iwmi.org) and where possible all data and analyses will be available as separate downloadable files. Reports may be copied freely and cited with due acknowledgment. About IWMI IWMI’s mission is to provide evidence-based solutions to sustainably manage water and land resources for food security, people’s livelihoods and the environment. IWMI works in partnership with governments, civil society and the private sector to develop scalable agricultural water management solutions that have a tangible impact on poverty reduction, food security and ecosystem health. IWMI Research Report 166 Evaluating the Flow Regulating Effects of Ecosystems in the Mekong and Volta River Basins Guillaume Lacombe and Matthew McCartney International Water Management Institute (IWMI) P O Box 2075, Colombo, Sri Lanka i The authors: Guillaume Lacombe is Senior Researcher – Hydrologist and Matthew McCartney is Theme Leader – Ecosystem Services, both based at the Southeast Asia Office of the International Water Management Institute (IWMI), Vientiane, Lao PDR. Lacombe, G.; McCartney, M. 2016. Evaluating the flow regulating effects of ecosystems in the Mekong and Volta river basins. Colombo, Sri Lanka: International Water Management Institute (IWMI). 40p. (IWMI Research Report 166). doi: 10.5337/2016.202 / ecosystems / flow discharge / rain / temperature / floodplains / geomorphology / geography / models / land cover / land use / forest cover / water resources / soils / wetlands / river basins / stream flow / downstream control / catchment areas / paddy fields / policy making / planning / impact assessment / living standards / runoff / Southeast Asia / China / Burma / Lao People's Democratic Republic / Thailand / Cambodia / Vietnam / Ghana / Burkina Faso / ISSN 1026-0862 ISBN 978-92-9090-833-3 Copyright © 2016, by IWMI. All rights reserved. IWMI encourages the use of its material provided that the organization is acknowledged and kept informed in all such instances. Front cover photograph shows rice fields and forest near Vang Vieng, Lao PDR (photo: Matthew McCartney, IWMI). Please send inquiries and comments to IWMI-Publications@cgiar.org A free copy of this publication can be downloaded at www.iwmi.org/Publications/IWMI_Research_Reports/index.aspx Acknowledgements The authors are grateful to the Mekong River Commission (MRC) for the provision of streamflow data, the land cover map and soil data; members of the ‘Asian Precipitation—Highly Resolved Observational Data Integration Towards Evaluation of Water Resources’ (APHRODITE) project for provision of the rainfall database for Southeast Asia; interdisciplinary GLOWA Volta Project for the provision of the flow records in the Volta River Basin; the Climatic Research Unit (CRU) at the University of East Anglia for the provision of rainfall and temperature data for the Volta River Basin; the European Space Agency (ESA) for the provision of the global land cover map GlobCover 2009; and to the United States Geological Survey (USGS) Hydrological data and maps based on SHuttle Elevation Derivatives at multiple Scales (HydroSHEDS) for the provision of the digital elevation model. The authors also thank three anonymous reviewers for their constructive comments on an earlier version of this report. Project This research study was conducted as part of the Optimizing water resource development for poverty alleviation: Combining green (natural) and grey (built) infrastructure project of the CGIAR Research Program on Water, Land and Ecosystems (WLE) under its theme Managing resource variability and competing uses. The study was also a contribution to the United Nations Environment Programme’s (UNEP’s) Approved Programme of Work (PoW) for 2012- 2013 under subprogramme 3 (Ecosystem management - tools to tackle ecosystem degradation). Collaborators This research study is a collaboration of the following organizations: International Water Management Institute (IWMI) United Nations Environment Programme (UNEP) Donors This research study was funded by the following: United Nations Environment Programme (UNEP) This work has been undertaken as part of the CGIAR Research Program on Water, Land and Ecosystems (WLE). IWMI is a member of the CGIAR Consortium and leads this program. Contents Acronyms vi Summary vii Introduction 1 Description of the Basins 3 The Mekong 3 The Volta 6 Method 8 Data 11 Discharge 12 Rainfall 13 Temperature 13 Geomorphological and Geographic Characteristics 13 Soil Characteristics 14 Land Cover 14 Results and Discussion 15 The Power-law Models 15 Interpretation of the Explanatory Variables 18 Limitations of this Study 25 Implications for Water Resources Planning and Policy Making 25 Conclusions and Recommendations 26 References 27 Annex 30 v v Acronyms CRU Climatic Research Unit FDC Flow Duration Curve ITCZ Inter-Tropical Convergence Zone MRC Mekong River Commission vi Summary Natural and agrarian ecosystems regulate scarcity, typical of tropical areas, these power- river f lows by storing and releasing water law models enable the derivation of flow duration between seasons. By smoothing flood peaks curves anywhere along the tributaries of the and enhancing dry-season flows, they protect Mekong and Volta rivers. The models, which human populations against the destruction and perform reasonably well (most R2 pred > 90%), hardship caused by floods and water shortage. allow the hydrological consequences of modified A method to quantify the impact of ecosystems paddy area and forest cover to be determined. on flow regimes has been developed and tested In the Mekong River Basin, extending paddy in the Volta and Mekong river basins: two basins areas results in a decrease in downstream low in which people’s livelihoods are particularly flows. In the Volta River Basin, the conversion dependent on river water. Instead of complex of forests to crops induces greater downstream physical models, the approach utilizes weighted flood flows. A physical interpretation of the least square regressions to derive multivariate model structure was possible for most of power-law models that predict a range of flow the resulting relationships, thus providing an percentiles from catchment characteristics, opportunity to increase our understanding of including geographic, geomorphologic, climatic, the effects of different ecosystems on flow. soil and land cover features. Step-wise and Basin development planners, who tend to best subset regressions were used concurrently neglect f low-related ecosystem services, to select the model variables that best predict should take these relationships into account. flow metrics observed in several gauged sub- Data l imi tat ions meant that the ef fect of catchments. Maximizing the prediction R-squared wetlands on downstream river flow could not (R2 pred), computed by leave-one-out cross- be determined in either basin. There is a need validations, ensured parsimonious, yet accurate, for more data collection and research for this relationships. In a general context of data particular ecosystem. vii Evaluating the Flow Regulating Effects of Ecosystems in the Mekong and Volta River Basins Guillaume Lacombe and Matthew McCartney Introduction Research has been conducted to determine was a first attempt to develop a pragmatic the flow regulating effects of ecosystems in the method for quantifying the flow regulating Mekong and Volta river basins. The research functions of floodplains, forests and headwater conducted built on an earlier study, Factoring the wetlands. Combining elements of hydrological role of ecosystems in the decision-support system regionalization with spatial interpolation of of the Zambezi River Basin, which developed streamflow records, the method developed a method for quantifying the flow regulating utilized observed streamflow records and flow effects of wetlands, floodplains and forests in the duration curves1 (FDCs) to derive a simulated time Zambezi River Basin (McCartney et al. 2013). series of flow in the hypothetical situation that an The research was predicated on the fact ecosystem is absent. This can then be compared that ecosystems, such as wetlands and forests, (using standard hydrological techniques) with an influence the hydrological cycle by affecting rates of observed time series (measured downstream of infiltration and evapotranspiration, and by modifying the ecosystem) to evaluate the impact of that how water is transmitted and stored in a basin particular ecosystem on both high and low flows. (Bruijnzeel 1996; Bullock and Acreman 2003). The method was developed using data obtained Although with little scientific verification, a function from 102 flow gauging stations with more than 25 widely attributed to forests and wetlands is the years of daily flow data, and applied at 16 locations natural regulation of river flow: reducing floods and in the Zambezi River Basin. Results were generally increasing dry-season baseflows. This ‘natural’ flow consistent with other research in southern Africa, regulation is widely perceived as an ‘ecosystem and indicate that the different ecosystems of the service’ that brings significant benefits to people Zambezi affect flows in different and complex ways. and society: reducing flood damage and increasing Broadly: i) floodplains decrease flood flows and water availability during dry periods. In recent increase low flows; ii) headwater wetlands increase years, this has led to the suggestion that natural flood flows and decrease low flows; iii) forests of ecosystems should be considered as ‘natural indigenous ‘miombo’ trees, when covering more than infrastructure’, and more closely incorporated into 70% of the catchment, decrease flood flows and decision-making processes pertaining to water decrease low flows. However, in all the cases, there resources planning and management (Emerton were examples which produced contrary results, and Bos 2004). However, the lack of a trusted and simple relationships between the extent of an evidence base, and uncertainty in quantifying either ecosystem type within a catchment and the impact a reduction in flood risk or increases in baseflow at on the flow regime were not found. a given location are key constraints to incorporating As acknowledged in McCartney et al. (2013), natural solutions into water resources planning and there are several limitations in the method management. developed: The hydrological research conducted in 1. The method attempts to determine the the previous study (McCartney et al. 2013) flow regime in the absence of specific 1 A flow duration curve (FDC) shows the relationship between any given discharge and the percentage of time that flow is equalled or exceeded (Shaw 1984). 1 ecosystems as if this was the only Volta river basins. However, soon after initiating difference between the catchments of the study it became clear that, in comparison to the interest. This ignores the fact that, in all Zambezi, there are relatively few gauging stations cases, the presence of the ecosystem is with long daily flow records in either the Mekong or dependent on the wider geological and the Volta river basin. Consequently, it was decided climatic setting: they are a function of the to modify the approach. landscape in which they are located. One option considered was the application of 2. Lack of data meant that an arbitrary, a rainfall-runoff model. Such models can be used rather ad hoc, approach had to be used in different ways. One possibility is to calibrate to determine the reference conditions on the model using data prior to land-use change a case-by-case basis. As a result, the and then use the model as a ‘virtual control’ in method is ultimately subjective. conjunction with rainfall observed after the land- use change, in order to reconstitute runoff as if 3. The method makes no allowance for no change in the catchment had occurred (e.g., changes in mean annual discharge. The Lacombe et al. 2010). Underlying assumptions mean discharge of the simulated ‘without of this approach are that the model adequately ecosystem’ approach was assumed to be simulates the full range of available rainfall-runoff the same as that of the ‘with ecosystem’ responses, and that the catchment is stationary approach. Given that the presence of the before and after the land-use change. Another ecosystem causes changes in flood flows approach involves the application of a spatially- as well as low flows, both of which affect distributed, physically-based model, in which mean flow, this is unlikely to be the case. land use is explicitly incorporated in the model However, without knowledge of how the parameterization. It is then possible to evaluate mean flow is affected by the presence the effect of any change in land cover by modifying of the ecosystem, it was not possible to the model parameters. In this way, different land- modify the mean flow. use scenarios can be simulated directly (e.g., Homdee et al. 2011). However, this requires a very Overall, the previous study concluded: good understanding of the hydrological processes occurring under different land covers, including the “…effects on flow are a function not effect of land conversion on soil surface properties just of the presence/absence of different (Lacombe et al. 2015). ecosystem types, but also of a range Both approaches are constrained, if there of other biophysical factors, including are l imited data for model calibration and topography, climate, soil, vegetation and validation. In our case studies, land-use data geology. Not surprisingly, the hydrological were scarce and understanding of the biophysical functions of natural ecosystems depend processes affecting flow generation is very limited. to a large extent on location-specific Consequently, it was not possible to undertake characteristics that make it difficult to such detailed hydrological modelling. The most generalize. To identify distinctive functions reliable data are flow and rainfall. For this reason, much more detailed research that takes the approach developed relies predominantly on into account the full range of biophysical these data, and attempts to detect relationships factors af fect ing f low is required. ” with other environmental characteristics without (McCartney et al. 2013). pre-assumptions. As with the previous Zambezi study, the Against this background, the original intention of method is based on the derivation of FDCs. the research conducted for the current study was to However, in the current study, rather than basing verify the applicability of the method derived in the FDCs on ‘reference catchments’ (see McCartney Zambezi, through application in the Mekong and et al. 2013), the FDCs were derived from 2 statistical relationships that linked different flow the shape of the FDC in some circumstances. percentiles to specific catchment characteristics. Past studies developed in other parts of the In this way, it was possible to derive FDCs at world (Blöschl et al. 2013; Salinas et al. 2013) ungauged sites, and the equations developed have demonstrated that such features can be were then used to determine the FDCs in incorporated into statistical tools, linking points on situations ‘without’ the ecosystem of interest. the FDC to catchment characteristics. Such techniques rely on the fact that the shape Compared to the method developed in the of any FDC is largely a function of the catchment Zambezi River Basin, the current study has characteristics that govern the partitioning of two advantages (i.e., points 2 and 3 above): (i) precipitation into interception, infiltration and rapid reference catchments are not required; and (ii) runoff, as well as subsurface storage, delayed the impact of an ecosystem on mean annual flow drainage and evaporation. can be evaluated. For these reasons, the new Such approaches have been used elsewhere approach is considered to be less subjective. (Homa et al. 2013; Castellarin et al. 2013), and This report summarizes the research empirical evidence from around the world indicates conducted. To provide context, both the Mekong that the catchment characteristics that determine and the Volta river basins are described briefly the shape of the FDC typically relate to climate, in the next section. The report then provides a geology, soil depth and permeability, vegetation detailed description of the method developed cover dynamics and catchment geomorphology. to estimate the impact of catchment features/ Of particular interest to the current study is that ecosystems on river flow. The results obtained the presence and absence of specific landscape for both basins are presented. The strengths features (e.g., wetlands) have been shown to affect and weaknesses of the method are discussed. Description of the Basins The Mekong season, which usually lasts from May until late September or early October. Tropical cyclones The Mekong is the world’s 12th longest river, occur over much of the area during August and flowing approximately 4,350 km through three September, and even October (in the Mekong provinces of China, and continuing into Myanmar, Delta). The northeast monsoon brings lower Lao People’s Democratic Republic (Lao PDR), temperatures from China and causes dry weather Thai land, Cambodia and Vietnam before in the Lower Mekong River Basin (i.e., the portion discharging into the South China Sea. The basin of the Mekong River Basin south of China) from of the Mekong River drains a total land area late October until April. Mean annual rainfall is of 795,000 km2 from the eastern watershed of significantly different between the east and west the Tibetan Plateau to the Mekong Delta. Mean banks of the Lower Mekong River. Rainfall of annual flow is 457 km3, which is equivalent to a more than 2,500 mm y-1 occurs in the western runoff depth of 600 mm y-1 (MRC 2005). mountain regions of Lao PDR. In contrast, The climate of the Mekong River Basin is typically, less than 1,000 mm y-1 of rainfall occurs dominated by the monsoon, which generates wet in the Khorat Plateau in Northeast Thailand. and dry seasons of more or less equal lengths. The Tibetan Plateau, where the Mekong The southwest monsoon generates the wet River originates, is predominantly covered by 3 alpine grass and rangeland. Below the tree formations and shrublands in the highlands of line, at approximately 4,000 m, needle-leaved Lao PDR, which play an important role in securing forest format ions and shrubland become the livelihoods of local people (Heinimann 2006), dominant, which increasingly give way to have been converted to mono-species rubber broad-leaved, evergreen forests and woodlands plantations which compete with food crops and from elevations of 2,000-2,500 m downwards only provide a fraction of the ecosystem services (Figure 1). However, in the more densely associated with natural forests. populated southern part of Yunnan Province Water resources of the Mekong region are and northern Lao PDR, in particular, large undergoing significant development, particularly extents of these evergreen forests have been for hydropower, and also for water diversion, degraded by practices of shifting cultivation and water supply and irrigation. Besides hydropower other large-scale human disturbance, resulting dams, many of the weirs and dams that already in very heterogenic patterns of forests and exist on the rivers in the Mekong region are forest regrowth interspersed with patches of for irrigation. Countries in the Mekong region shrubland, grassland and cropland. The land continue to see irrigated agriculture as a central cover distribution in the lowlands of Lao PDR pillar of rural development, and withdrawals for and Cambodia shows lower proportions of this purpose dominate water use in the region. In evergreen vegetation, but higher distributions the Lower Mekong River Basin, the irrigated area of cropland, cropland/vegetation mosaics, (ca. 1.2 Mha) is currently less than 10% of the and dry-deciduous, broad-leaved wood and total agricultural area (15 Mha). There are plans scrublands. Intensively cultivated areas, mainly to increase dry-season irrigation by 50% (i.e., rice, are concentrated in the Mekong Delta, to 1.8 Mha) over the next 20 years. Lao PDR lowlands surrounding the Tonle Sap Lake and plans to expand irrigation from about 100,000 in the extensive plains of the Khorat Plateau ha (i.e., 4,000 small- to medium-scale schemes, in Thailand. In the Mekong Delta, double- and mostly pumping water directly from rivers) to triple-season rice paddies, orchard cultivation, over 300,000 ha. Major irrigation expansion is aquaculture, coastal mangrove forests and being studied in Cambodia, linked to investments a mixture of the latter, known as integrated in flood control in the undeveloped Cambodian shrimp-mangrove farming systems, dominate delta and to hydropower development elsewhere. (Leinenkugel et al. 2013). Mainstream water transfers have long been Across much o f the Mekong reg ion , considered by Thailand to complement national investment in land, including foreign direct approaches to alleviate drought in the northeast investment, has been promoted as an effective of the country. development tool by several governments. For The Mekong River Basin supports the example, in Lao PDR, the establishment of livelihoods of a population of over 70 million policies and regulatory frameworks that are people. Climate variability and, in particular, favorable to land- and resource-intensive floods and droughts have a major effect on the investment has resulted in a rapid increase in the livelihoods of many of these people. The severe area of land granted for development. Excluding economic, social and environmental impacts mining, land deals totalling 1.1 million hectares of droughts are conf i rmed by a growing (Mha) (ca. 5% of the national territory of Lao PDR level of vulnerability among people living in and an area greater than the total area under rice the affected areas. For example, mil l ions cultivation) have been recorded (Schönweger et of farmers and low-income earners were al. 2012). A range of products are cultivated or affected by the drought in 2004, which caused extracted from lands under investment: rubber, considerable agricultural losses in Northeast teak, eucalyptus, cassava and sugarcane. Hence, Thailand and Cambodia, a significant reduction large-scale land cover changes are occurring in rice crop in Lao PDR and critical levels throughout the region. Many secondary forest of sa l ine in t rus ion in the Mekong Del ta. 4 5 FIGURE 1. (a) Land use, and (b) location of the 65 gauging stations in the Lower Mekong River Basin. (a) (b) Source: (a) GlobCover 2009 (Bontemps et al. 2011), and (b) Lacombe et al. 2014b. Note: GLWD – from the Global Lakes and Wetlands Database (GLWD). Unlike floods, which reach a high level of The Volta severity within relatively short time periods, droughts develop more slowly over periods of Although just one-third of the Mekong River in terms several months and within certain regional areas. of length, the Volta River (1,600 km) drains an area However, once established, the economic, social of 403,000 km2, which represents half of the surface and environmental consequences of droughts area of the Mekong River Basin. The Volta River pose a serious threat to those who rely on secure Basin is shared by six countries in West Africa. water supplies, including farmers, fisherfolk and It lies mainly in Ghana (42%) and Burkina Faso domestic households. (43%) with the remainder in Benin, Cote d’Ivoire, The role of floods is much more complex Mali and Togo (Figure 2). The river, one of Africa’s than that of droughts, because, in addition to most important in terms of length and discharge, costs, floods also bring significant benefits for comprises three major tributaries corresponding both ecology and agriculture. For example, in the to large sub-catchments: the Black Volta (147,000 Lower Mekong River Basin, the annual costs of km2), White Volta (106,000 km2) and the Oti (72,000 flooding (mainly related to damage to infrastructure km2), which come together to form the Lower and crops) equates to an annual average of USD Volta (325,000 km2). The total annual flow varies 60-70 million. In contrast, the benefits of floods considerably, but is approximately 40,400 Mm3, on (associated primarily with fisheries and agriculture) average (Andah et al. 2004), which is equivalent to are estimated to average approximately USD a runoff depth of 100 mm y-1. 8-10 billion annually (MRC 2009). Of course, the Climatically, the basin is dominated by the spatial distribution of costs and benefits is very rain-bearing, southwesterly tropical maritime uneven, with benefits, to a large extent, being air mass and the dry, northeasterly tropical realized in the lower part of the basin (i.e., the continental air mass (Dickson and Benneh delta and the Cambodian floodplain). The costs are 1988). The two air masses meet at the Inter- generally associated with flooding further upstream, Tropical Convergence Zone (ITCZ). At any particularly as a consequence of flash floods on location, the rainy season begins when ITCZ has tributaries (MRC 2009). passed overhead moving north and ends with The impac t o f f l oods and d rough t s its southwards retreat. Consequently, there is a (posit ive and negative) are inf luenced by general tendency for rainfall to decrease from the natural ecosystems in the basin. The significant south to the north, though this general effect is seasonal variation in water levels in the river disrupted in a few places as a consequence of (typically wet-season levels are 8-10 m higher local relief. Between May and August (i.e., the than that in the dry season along the main West African monsoon), the ITCZ moves to the stem and main tributaries) creates a rich and north and the entire basin lies under the influence extensive series of floodplains, backwaters, of the tropical maritime. These months yield swamps and other wetlands in the basin. approximately 75% of the total annual rainfall. In Many rural l ivelihoods are founded on the the vicinity of the coast, rainfall is approximately integrated use of a wide range of natural 1,500 mm y-1 and bimodal, falling between May resources, adapting to the seasonal changes and October, with a short dry season in July/ of flooding and recession. Consequently, the August, separating two peaks (Dickson and hydrological functioning of natural ecosystems Benneh 1988). The two rainfall peaks tend to is intimately tied to people’s livelihoods. As built disappear northward, and in the northern part of infrastructure in the basin increases, particularly Ghana, the rainfall distribution is uni-modal and as more hydropower dams are constructed, averages approximately 500 mm y-1. In addition to and flow regimes are altered, many of these seasonality, inter-annual variability is considerable important ecosystems are under threat. (Nicholson 2005). 6 7 Figure 2. (a) Land use, and (b) location of the 20 gauging stations in the Volta River Basin. (a) (b) Source: (a) GlobCover 2009 (Bontemps et al. 2011), and (b) created as part of this study. Note: GLWD – from the Global Lakes and Wetlands Database (GLWD). Grassland is the dominant land cover rainfall variability, floods and droughts. River flows throughout the basin, ranging from 76% of the are extremely sensitive to precipitation (Andreini delta catchment in the south to 98% of the Arly et al. 2000). catchment in the north. As in the Mekong, natural Floods in northern Ghana in 2007, particularly ecosystems, wetlands, floodplains and forests along the White Volta, are reported to have (though much degraded) influence the basin affected more than 275,000 people with many hydrology. Within Burkina Faso, wetland areas are of them being displaced from their homes. The not extensive, but there are floodplains along the flooding also damaged farmland and resulted in major tributaries of the Volta River. It is estimated the loss of crops. It is estimated that more than that, within Ghana, there are at least 238,600 ha 12,200 ha of farmland were ‘washed away’, and of swamps and floodplains on tributaries flowing some 160,000 metric tonnes of food was lost as into Lake Volta (Hughes and Hughes 1992). a consequence of the flooding and the drought Water resources in the basin have come that immediately preceded it (IRIN 2007). The under increasing pressure in recent years. vagaries of rainfall influence not only livelihoods Population growth in Ghana and Burkina Faso has and food security but also economic development. resulted in larger abstractions to meet increasing It is anticipated that climate change, in conjunction water demand (van de Giesen et al. 2001). with increasing population, may aggravate the Population is projected to reach 34 million in situation (Lacombe et al. 2012). Against this 2025, up from 18.6 million in 2000 (Biney 2010). background, there are plans to build more dams As in the Mekong, much of the agriculture in the to increase electricity production and expand basin is rainfed, and thus highly susceptible to irrigation in the basin. Method Detecting and quantifying the flow regulating be attributed to the differences in the catchment effects of ecosystems is challenging in data- characteristics. scarce areas such as the Mekong and Volta Multiple regression analyses determine linear river basins, where accurate and sufficiently long or log-linear relationships between a dependent records of land use and land cover are limited. variable (in this case, flow percentiles) and In most cases, only one remote sensing product independent variables (in this case, catchment is available for a particular date, making time characteristics). Flow variables can be correlated comparison of land use and flow for correlation to more than one explanatory variable at the analyses impossible. An alternative option is same time. For example, mean annual flow may to use a statistical approach to assess the be positively correlated to rainfall, catchment area relationship between human basin modifications and the mean catchment slope. Consequently, it and the resulting impact on flow. Instead of is in correct to derive simple correlations between exploring basin modifications over time, it can be the variable and a single catchment characteristic. easier to make the comparison between gauged Rather, multiple regression techniques, which basins – with different land cover – over the enable variables with statistically significant same period. Then, if the statistical relationships explanatory power to be identified, must be are valid and sufficiently strong, the spatial applied. Multiple linear regressions enable the variability of flow (i.e., inter-basin variability) can statistical characterization of these correlations, 8 and provide the opportunity to identify the causal ln(q) = β0 + β1 ⋅ ln(X 1 ) + β 2 ⋅ ln(X 2 ) + links between catchment characteristics and catchment hydrology. Knowing that catchment ... + βm ⋅ ln(X m ) + ε (2) alterations, such as land cover change, will affect low flow and high flow differently, it is important to apply the multiple regression analyses to β0 is a constant. v (equation [1]) and ε a wide range of flow percentiles. This way, it (equation [2]) are the log-normal and normally is possible to observe how the environmental distributed errors (or residual) of the models, characteristics of a catchment determine the respectively. The natural logarithm (ln) being shape of the corresponding FDC. Following the defined for strictly positive values only, catchment method proposed by Homa et al. (2013), we characteristics Xi and flow q with possible zero developed a two-step approach: values are incremented by one prior to being used - in the regression analysis (Homa et al. 2013). In Derivation of a set of multiple linear these cases, Xi and/or q should be replaced by regression models to estimate a range X of daily flow percentiles from a number i+1 and/or q+1, respectively, in equations (1) and (2). For each predicted flow percentile q of the of catchment characteristics, including FDC, selection of the variables Xi (i.e., catchment the area of natural ecosystems and land characteristics) with the highest explanatory power cover types (i.e., forests, crops). and calculation of their respective coefficients - Use of these relationships to determine βi was performed by weighted least squares what the FDC would have been if the regressions applied to n observations qj (j = 1, catchment areas covered by each ..., n) of q and their respective m catchment land cover type were modi f ied by characteristics Xij. A description of the approaches setting the relevant area to different used to calculate the dependent variables qj and va lues in the regress ion models . the independent variables Xij is presented in the section Data. Relatively simple regression approaches have Unlike ordinary least square regressions shown value for determining flow percentiles in treating the n observations of qj equally, weighted data-scarce regions of the world (Hrachowitz et least square regression (Tasker 1980) enables al. 2013). Since a power-law is more suitable for the varying number kj of hydrological years (April environmental processes (hydrological variables 1 - March 31 for both basins) used to calculate are not normally distributed and highly skewed) each flow statistic qj and its associated climate than a linear relationship, we used equation (1) characteristics to be taken into account. Values of to estimate a flow metric q from m catchments qj derived from a greater number of hydrological with characteristics Xi (i=1,…, m). A logarithmic years are more precise (have lower variance) transformation of equation (1) results in a log- and thus should have a greater weight in the linear model (equation [2]), where coefficients βi regression. However, this reliability decreases as (i=1,…, m) can be determined by multiple linear the variance of qj increases. To account for these regressions. It should be noted that the original two counteracting factors, the multiple regression linear form of the equation (without logarithmic analysis was performed by weighted least square transformation) was first tested and proved to regressions (Tasker 1980) using two different underperform, compared to the multiplicative formulae for the Mekong and the Volta regions. version requiring logarithmic transformation. In the Mekong, the following equation was used: (1) k w = j q = expβ0 ⋅ X β1 ⋅ X β2 ⋅ ⋅ ⋅ X βm 1 2 m ⋅ν j Stdev(q j ) (3) 9 Where: Stdev(qj) is the standard deviation of qj. For each gauged sub-catchment, the In the Volta, the computation of the weight predicted FDC was refined using a scaling wj using equation (3) would have resulted in high factor α (qpred = a×qobs). The value of α was bias caused by the high disparity in the number optimized to minimize the deviation between kj of observations between different stations. kj qobs and qpred for each flow percentile predicted varies between 1 and 45 with a median value with a power-law equation, where qobs and qpred of 4. In addition, wj is not computable when kj are the observed and predicted flow variables, = 1, because Stdev(qj) becomes null. For these respectively. The value of α was optimized reasons, an alternative equation was set up as by maximizing the Nash-Sutcliffe efficiency follows: wj = kj. This way, the increased reliability criteria. This scaling factor was not intended to of qj, as the number of kj observations increases, ‘artificially’ improve the predictive power of the is taken into account. power-law models, but rather to reduce possible For each estimated flow metric qj, selection bias while assessing the effect of land-use of the best set of explanatory variables Xi was changes on FDC by altering values taken by guided by the combined use of the selection the explanatory variables Xi of the power-law algorithms known as ‘best subsets regression’ equations. and ‘step-wise regression’, both of which are The FDCs produced using the methodology widely available in statistical packages. This described above provide a useful overview of selection was intended to maximize the prediction the impact of ecosystems/land cover on the flow R-squared (R2 pred) calculated by leave-one-out regime at any given location. However, analysis cross-validations. Unlike the classical R-squared, of specific flow events, and determination of the the maximization of which can lead to model cumulative impact on flow volumes, requires over-fitting and loss of robustness, R2 pred reflects comparison of actual streamflow time series the ability of the model to predict observations ‘with’ and ‘without’ specific ecosystems. To which were not used in the model calibration. generate these time series, we took the daily Maximizing R2 pred generally leads to greater flow at locations where it had been measured parsimony in the number of explanatory variables. and converted all flows to a flow percentile. An explanatory variable was considered to be Each flow percentile was then converted back statistically significantly different from zero if its to a daily f low value using the regression p-value, derived from Student’s t-test, was lower models: once with the ecosystem set to the than 0.05. coverage actually present and once with it set The required homoscedasticity (homogeneity to zero (e.g., rice area) or alternatively to a of variance) of the model residuals ε was verified maximal value (e.g., forest or crop area). This by visual inspection of the residual plots. Possible approach assumes that there is no change multi-collinearity among the explanatory variables in the frequency of specific flow events as a was controlled with the variance inflation factor consequence of land-use change. (VIF), which should never exceed 8. VIFs for all Depending on the FDCs, two types of explanatory variables of our models were found interpolations were used to assess flow values to never and rarely exceed 3 and 2, respectively. corresponding to flow percentiles positioned The influence statistic Cooks D (Cook and between two model led values (e.g. , f low Weisberg 1982) was used to identify and remove value associated with the 0.85 flow percentile outlier catchments exhibiting high influence on the was calculated by interpolating flow values estimation of the model coefficients. Removal of corresponding to the 0.80 and 0.90 f low these outliers (between 2 and 5, depending on the percentiles): fifth-order polynomial interpolation flow metrics) was found to systematically increase or exponential interpolation. Comparison of the the performance of the models. For further two time series enabled quantification of the background on R-squared, VIF and influence impact of land-use change on the catchment statistics, refer to Helsel and Hirsch (2002). hydrology. 10 Data It should be noted that the set of candidate explanatory the basins. The discrepancies do not affect the variables were not exactly the same in the two river results because the main objective was to observe basins (cf. data on geomorphology, geography, how different classes of land cover have an impact climate, soil and land cover characteristics) – Mekong on the flow metrics. As indicated in the section Land (Table 1) and Volta (Table 2). The reason for this is Cover, particular care was devoted to ensuring that that data for each basin were collected over different representative land cover classes were selected as periods of time and data availability varied between explanatory variables in each basin. TABLE 1. List of the candidate explanatory variables (catchment characteristics) used in the multiple regression analyses for the Mekong River Basin. Variable Definition Unit Minimum Median Maximum name value value value Climatic characteristics Rain Median annual rainfall mm/year 880 1,416 2,093 Geomorphological characteristics Area Drainage area km2 207 3,278 106,748 Peri Perimeter km 76 401 2,090 Slop Mean slope % 2 15 32 Elev Mean elevation m 84 562 1,168 Drai Drainage density = total stream km-1 0.09 0.13 0.17 lengths/drainage area S Ratio of area draining south % 21 37 45 SW Ratio of area draining southwest % 26 44 62 W Ratio of area draining west % 16 35 52 NW Ratio of area draining northwest % 28 41 51 Geographic characteristics (coordinates of the flow gauging stations) Lati Latitude Decimal 12.33 16.70 20.70 Long Longitude Degree 99.35 104.03 108.00 Soil characteristics (averaged over catchment area) Sdep Mean soil depth 4-unit scale 0.00 3.07 4.00 Sste Mean top soil texture 0.00 2.08 2.91 Land cover characteristics (ratio of catchment area coverage) Fore Forest % 3 75 98 Padd Paddy % 0 4 77 Wetl Wetlands (marsh and swamp) % 0 0 1.23 Source: Lacombe et al. 2014b. 11 TABLE 2. List of the candidate explanatory variables (catchment characteristics) used in the multiple regression analyses for the Volta River Basin. Variable Definition Unit Minimum Median Maximum name value value value Climatic characteristics Temp Median annual temperature Degrees Celcius 27.0 28.1 29.0 Rain Median annual rainfall mm/year 601 917 1,075 Geomorphological characteristics Area Drainage area km2 4,419 17,863 138,086 Peri Perimeter km 338 869 3,261 Slop Mean slope % 1.2 1.6 2.8 Elev Mean elevation m 184 299 416 Reli Relief m 16 52 99 Drai Drainage density = total stream lengths/drainage area km-1 0.10 0.11 0.14 DDir Drainage direction % 0.70 0.86 1.27 Land cover characteristics (ratio of catchment area coverage) Fore Forest 1.4 12.4 45.6 Shru Shrubland 0.1 22.0 45.6 Herb Herbaceous 0.0 0.1 10.5 NoVe No vegetation % 0.0 0.0 1.0 Flood Flooded vegetation 0.0 0.0 0.2 Wate Water bodies 0.0 0.1 0.6 Crop Cultivated and managed lands 8.9 62.9 93.5 Discharge Council for Scientific and Industrial Research - Water Research Institute (CSIR-WRI), Ghana. In The power-law models were parameterized the Mekong River Basin, out of 71 gauged sites, using flow records provided by the Mekong River 65 stations were selected. The streamflow time Commission (MRC) for the Lower Mekong River series comprised between 1 and 41 years of Basin, and downloaded from the GLOWA Volta records with a median value of 17 years. In the databases for the Volta River Basin. In the two two basins, records are available between 1951 regions, the study sub-catchments were selected and 2007. The full list of the stations with their based on data availability and reliability, and on characteristics is available in Annex A. the basis that they provide records which were not In order to best represent and estimate FDC, subject to dam regulation. 11 flow percentiles (i.e., exceedance probabilities) In the Volta River Basin, out of 23 gauged were selected for prediction with the power-law sites, 20 stations were selected, based on data- models in the Lower Mekong River Basin: 0.05, quality control and data gaps. These stations 0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70, 0.80, 0.90 had between 1 and 45 years of records, and and 0.95. Additionally, we computed the annual have been maintained by either the Office de minimum, maximum and mean flow - henceforth la Recherche Scientifique et Technique d’Outre- referred to as Min, Max and Mean, respectively. Mer (ORSTOM) (former name of the Institut de Since daily flow values below 1 m3.s−1 are not Recherche pour le Développement [IRD]) or the provided in the MRC database, regression models 12 had to be computed using sub-catchments with covering the period 1901-2009 and provided free median values of flow percentiles greater than by the Consortium for Spatial Information (CGIAR- 1 m3.s−1. This resulted in the removal of 15, 11, CSI) (available at http://www.cgiar-csi.org/data/ 10, 7 and 5 sub-catchments from the datasets uea-cru-ts-v3-10-01-historic-climate-database - used to compute the Min, 0.95, 0.90, 0.80 and accessed on March 15, 2014). 0.70 flow percentiles, respectively. For the Volta Several rainfall variables were tested for River Basin, 13 flow percentiles were selected for correlation with each of the 11 selected flow prediction with the power-law models: 0.01, 0.05, percentiles: annual and monthly rainfall depths, 0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70, 0.80, 0.90, rainfall depth cumulated over the n-day rainiest 0.95 and 0.99. Additionally, we computed the periods of the hydrological year (n=5, 10 and 15). annual mean flow. Annual rainfall was found to exhibit the greatest For each sub-catchment j in the Volta and correlation coefficients with all the flow metrics q Mekong river basins, the value of each of these in the Volta and Mekong river basins, and was flow percentiles, and the Min, Max and Mean included as the only candidate explanatory rainfall values were obtained by selecting the median of variable in the power-law models for both the the kj annual values as suggested by Vogel and Mekong and Volta river basins. Median rainfall Fennessey (1994). Compared to the period-of- and median flow values used in the regression record FDC, which indicates the percentage of analyses were derived from the same hydrological time (duration) a particular value of streamflow years (Tables A1 and A2 in Annex A). is exceeded over a historical period, the median FDC reflects the percentage of time a particular value of streamflow is exceeded in a typical year. Temperature This second approach is particularly relevant when the start and end dates of the flow time Areal temperature variables were generated series used to compute the flow percentiles vary following the same procedure as for rainfall a lot between the study sub-catchments. All these using the CRU database (Harris et al. 2014). flow metrics (flow percentiles and Min, Max and Through several correlation analyses, only annual Mean flow values) are the dependent variables q temperature was finally selected as a candidate in equations (1) and (2). explanatory variable in the Volta River Basin. Temperature was not studied in the Mekong, because the new CRU database (used for Rainfall the Volta) was not available at the time of the analysis. Two different gridded precipitation databases were used in the Mekong and Volta river basins to derive areal daily rainfall time series for each Geomorphological and Geographic study sub-catchment. In the Mekong, we used Characteristics the high-resolution (0.25° × 0.25°), daily gridded precipitation database ‘Asian Precipitation - Highly S e v e r a l g e o m o r p h o l o g i c a l c a t c h m e n t Resolved Observational Data Integration Towards characteristics, likely to influence hydrology, were Evaluation of Water Resources’ (APHRODITE) derived from the United States Geological Survey (Yatagai et al. 2012) (freely available at http:// (USGS) Hydrological data and maps based www.chikyu.ac.jp/precip/ - accessed February on SHuttle Elevation Derivatives at multiple 03, 2014), covering the period 1951-2007. In the Scales (HydroSHEDS), a quality-controlled Volta, we used the medium resolution (0.50° × 90-meter digital elevation model (Lehner et al. 0.50°), monthly gridded precipitation database 2006) (freely available at http://hydrosheds. from the Climatic Research Unit (CRU) at the cr.usgs.gov/index.php (accessed on March 15, University of East Anglia (Harris et al. 2014), 2014). These characteristics include drainage 13 area, perimeter, mean slope, mean elevation, Land Cover drainage density, drainage direction and relief. The drainage density is the cumulative length of Land cover percentages for land cover types all streams within the catchment, normalized by in each of the study basins were derived from the drainage area of the catchment. The stream two different remote sensing products. In the network consists of all outlet points draining an Mekong River Basin, we used a digitized 2003 area greater than 40 km2. This threshold value land cover map prepared by MRC (2011). Forest was selected to best capture the variability cover was produced by merging four forest types of drainage densities among the study sub- available as separate land cover classes in the catchments. For the Mekong River Basin, four published map: coniferous forest, deciduous forest, variables representing mean drainage directions evergreen forest and forest plantation. The other were calculated - South, Southwest, West and land cover classes (wetlands and paddy) were Northwest. A value of 1 (or zero) meant that the directly available, because they correspond to sub-catchment was draining toward the named distinct land cover classes on the published map. direction (or opposite to the named direction). In the Volta River Basin, we used the global For the Volta River Basin, drainage direction land cover map GlobCover 2009 (Bontemps was defined slightly differently: only one variable et al. 2011). This is a 300-m resolution global was defined - the ratio of the number of pixels land cover map produced from an automated draining to the northeast divided by the number classification of the MEdium Resolution Imaging of pixels draining to the southwest. Relief is Spectrometer (MERIS) Full resolution (FR) time the standard deviation of all pixel elevations series, with 22 land cover classes. We defined within each sub-catchment. A value of zero seven land cover types as candidate explanatory corresponds to perfectly flat, horizontal land. variables: ‘Crop’ was produced by merging The geographic coordinates of the flow three land cover classes: rainfed crops, mosaic gauging stat ions ( lat i tude and longi tude) croplands/vegetation and mosaic vegetation/ were selected as two additional candidate croplands. ‘Forest’ was the result of merging four explanatory variables for the Mekong River land cover classes: closed to open broad-leaved Basin (Table 1). These geospatial variables evergreen or semi-deciduous forest, open broad- may control solar radiation that has an effect leaved deciduous forest, mosaic forest-shrubland/ on evapotranspiration and local climate patterns grassland and mosaic grassland/forest-shrubland. influenced by the regional topography, not ‘Herbaceous’ resulted from merging two land only in the sub-catchment itself, but also in its cover classes: ‘closed to open grassland’ and neighborhood. ‘sparse vegetation’. The land cover type ‘No vegetation’ was produced by merging two land cover classes: ‘artificial areas’ and ‘bare areas’. The three other land cover types – ‘flooded Soil Characteristics vegetation’, ‘water bodies’ and ‘shrubland’ – correspond to single land cover classes already Due to the lack of available data for the Volta, available in the original database: ‘closed to open we included soil characteristics only in the vegetation regularly flooded’, ‘water bodies’ and Mekong. Two soil characteristics, likely to control ‘closed to open shrubland’, respectively. hydrological processes, were selected from the The multiple regression analyses enabled MRC soil database (MRC 2011): soil depth and the flow regulating effect of the land cover top soil texture. A four-unit scale suggested by types that have a significant explanatory power MRC was used for quantification. Average values on the flow metrics (i.e., p-value < 5%) to be for each soil characteristic and each sub-catchment selected and their associated coefficient (i.e., were averaged by weighting each scale unit by the exponent) to be calculated. If a selected land- respective area covered in the sub-catchment. use type (variable Xi in equations [1] and [2]) 14 has a positive exponent (coefficient βi in equation in the model signifies that this (land cover type) [1]), it means that the flow metric q will increase has a negligible hydrological effect compared to as the spatial extent of that land cover type other candidate explanatory variables listed in increases. A negative coefficient means that the Tables 1 and 2. It is also possible to compare flow metric q will decrease as the spatial extent the respective effects of various land cover types of the land cover type increases. The absence of by comparing the values of their associated any land cover type as an explanatory variable coefficients in the power-law models. Results and Discussion The Power-law Models flows, indicating that the explanatory variables tested in this study do not correspond to the Tables 3 and 4 present the results of the catchment characteristics that predominantly multiple regression analyses for the 14 flow control low flows. This suggests that more effort metr ics ( l isted in column 1) in the Lower is needed to generate catchment characteristics Mekong River Basin and the Volta River Basin, suitable for multivariate low flow predictions. respectively. Column 2 provides the value For the Lower Mekong River Basin, values of of the intercept term βo. Columns 3 to 11 (in the explanatory variable ‘Padd’, and of the flow Table 3) and columns 3 to 12 (in Table 4) metrics 0.50, 0.60, 0.70, 0.80, 0.90, 0.95 and provide the coefficients βi associated with each Min (Table 3) should be incremented by 1 for explanatory variable Xi included in the power- inclusion in equation (1). For the Volta River law models (cf. equations [1] and [2]). The last Basin, only the values of the flow metrics 0.30, column of the tables indicates the performance 0.40, 0.50, 0.60, 0.70, 0.80, 0.90, 0.95 and of the models. The model performance is 0.99 (Table 4) should be incremented by 1 for good in the Mekong River Basin for any flow inclusion in equation (1) (cf. section Method). metric: R2 pred>89%. In the Volta River Basin, the As examples for the Lower Mekong River Basin, model performs well for flow percentiles lower equations (4) and (5) show how to predict the than 80%: R2 pred>80%. In the two basins, the 0.95 flow percentile (q0.95) and mean annual flow predictive power of the models declines for low (qmean) using the coefficients provided in Table 3. q0.95 = exp−27.857 × Rain2.698 × Peri1.436 × Elev0. 9 6 6 × Lati−1.291 × (Padd+1)−0.285 − 1 (4) qmean = exp−18.989 × Rain2.543 × Area0.883 × Drai 1.089 (5) Figure 3 illustrates the performance of the 3[c], respectively); the 0.95 and 0.05 flow power-law models by comparing observed and percentiles in the Lower Mekong River Basin predicted flows in each gauged sub-catchment (Figure 3[b]) and the Volta River Basin (Figure j (one plot = one sub-catchment) for some 3[d]), respectively. Refer to Lacombe et al. flow metrics: mean annual flow in the Mekong (2014b) for further interpretations of Figures and the Volta river basins (Figures 3[a] and 3(a) and 3(b). 15 TABLE 3. Coefficients and performance of the power-law models predicting the flow metrics q in the Lower Mekong River Basin (cf. Table 1 for the full names of the variables). No coefficient value means that the corresponding variable was found not to have statistically significant explanatory power to predict the flow metric and therefore should not be included in the equation predicting q. q β0 E xplanatory variables (βi, i>0) R2 pred (m3/s) (%) Rain Peri Elev Area Drai Slop Lati Padd Fore Max 1.870 -0.796 0.668 2.694 0.798 -1.423 89.1 0.05 -14.434 2.376 0.862 2.016 94.1 0.10 -21.435 2.608 0.970 93.5 0.20 -23.087 2.742 0.988 94.3 0.30 -24.135 2.519 0.335 0.992 91.8 0.40 -29.234 2.603 1.789 0.566 92.5 0.50 -31.247 2.529 1.798 0.714 0.262 92.1 0.60 -24.521 2.289 1.600 0.963 -1.526 -0.155 92.4 0.70 -24.023 2.307 1.469 1.074 -1.820 -0.155 90.7 0.80 -25.761 2.582 1.411 1.080 -1.852 -0.189 92.2 0.90 -28.562 2.613 1.467 0.844 -1.706 0.587 -2.503 89.5 0.95 -27.857 2.698 1.436 0.966 -1.291 -0.285 90.5 Min -32.951 3.027 1.416 0.803 -2.684 0.535 -2.598 89.1 Mean -18.989 2.543 0.883 1.089 94.7 Source: Adapted from Lacombe et al. 2014b. TABLE 4. Coefficients and performance of the power-law models predicting the flow metrics q in the Volta River Basin (cf. Table 2 for the full names of the variables). No coefficient value means that the corresponding variable was found not to have statistically significant explanatory power to predict the flow metric and therefore should not be included in the equation predicting q. q β0 E xplanatory variables (βi, i>0) R2 pred (m3/s) (%) Fore Crop Area Peri Reli Slop Drai Elev Rain Temp Mean -23.09 -0.354 1.223 -2.487 4.074 94.22 0.01 172 -0.694 1.132 -7.552 -40.97 91.77 0.05 186.1 -0.583 1.243 -7.371 -45.87 94.39 0.10 202.3 -0.584 1.313 -7.611 -50.67 93.74 0.20 -28.62 -0.342 1.302 -2.738 5.047 92.13 0.30 -8.622 -0.387 1.989 1.370 -1.467 96.04 0.40 88.49 -0.416 1.303 0.814 -2.512 -26.73 92.36 0.50 0.516 1.047 2.452 4.445 95.61 0.60 -5.218 0.867 2.421 0.791 3.815 91.21 0.70 -3.567 0.888 1.522 0.912 2.681 80.52 0.80 1.226 3.067 0.388 3.056 73.07 0.90 5.590 1.732 4.550 3.541 54.84 0.95 5.325 1.608 4.145 2.818 45.6 0.99 4.504 1.340 3.440 2.233 33.75 16 17 FIGURE 3. Comparison of observed and predicted flows. (a) and (b): Lower Mekong River Basin. (c) and (d): Volta River Basin. (a) (b) 1,000 1,000 R2 pred)= 94.7% R2 pred)= 90.5% 100 100 10 1 10 0.1 1 0.01 1 10 100 1,000 0.01 0.1 1 10 100 1,000 Observed mean annual flow (m3.S-1) Observed 95% flow percentile (m3.S-1) (c) (d) 1,000 1,000 R2 pred)= 94.2% R2 pred)= 94.4% 100 100 10 1 10 1 10 100 1,000 10 100 1,000 Observed mean annual flow (m3.S-1) Observed 5% flow percentile (m3.S-1) Predicted mean annual flow (m3.S-1) Predicted mean annual flow (m3.S-1) Predicted 5% flow percentile (m3.S-1) Predicted 95% flow percentile (m3.S-1) Interpretation of the Explanatory Variables sub-catchments of the Volta River Basin used in Physical Variables this study. The coefficient of other explanatory variables, not related to land cover change (i.e., In the Lower Mekong River Basin, annual rainfall drainage area, perimeter, relief, drainage density is an explanatory variable in all the models (Table and temperature), are also difficult to explain 3). Coefficient values are much greater than unity and suggest that further research is required (average = 2.59), indicating that an increase to better understand how the climate and the of x% in annual rainfall would induce a greater geomorphology controls flow production in the than x% increase in any of the studied flow sub-catchments of the Volta River Basin. metrics. Drainage area is an explanatory variable In both the Lower Mekong and the Volta for mean annual flow and high-flow variables. river basins, the high values of R2 pred (Tables The coefficients for this variable are slightly 3 and 4) could largely be attributed to mass lower than 1, depicting a slight tendency for a balance considerations, i.e., they indicate that reduction in runoff depth (i.e., specific runoff) as a larger catchment produces more flow. To catchment size increases, reflecting the tendency verify if this scaling behavior actually magnifies for increased seepage in larger catchments the performance of the models, we carried out (Pilgrim et al. 1982). In contrast, low-flow variables the multiple regression analyses for the Lower are better explained by the catchment perimeter Mekong River Basin again, using specific runoff rather than the catchment area. The perimeter (in mm y-1) as a dependent variable and computed provides information related to the shape of the Nash-Sutcliffe efficiency coefficient based the catchment. For a given catchment area, a on volumetric runoff (m3 s-1) for the two sets of greater perimeter implies a longer time for water power-law models predicting either specific or to reach the catchment outlet, thus explaining the volumetric runoff. According to this efficiency positive correlation with low-flow variables. The coefficient, the models predicting specific runoff drainage density quantifies the level of catchment were found not to outperform those described in drainage by stream channels. Lower drainage this report. In addition, homoscedasticity of the density corresponds to flatter land with less residual of the models predicting specific runoff differentiated drainage paths. High values imply was more difficult to obtain compared to the steep-sided valleys, shorter flow transfer time models predicting volumetric runoff. The values and a flashy (i.e., steeper) hydrograph. As would of the variance inflation factor (VIF) associated be anticipated, the coefficients of the drainage with each explanatory variable were higher for density are consistently positive and negative for the models predicting specific runoff compared to high and low flows, respectively. Flow percentiles the models predicting volumetric runoff, revealing of intermediate magnitude are not influenced by greater dependency between the explanatory the drainage density (Lacombe et al. 2014b). variables. Finally, the explanatory variables of In the Volta River Basin, elevation and slope the model predicting specific runoff were found are the predominant controls on high and low to vary a lot depending on the flow percentiles flows, respectively. According to the signs of considered, thus hampering the physical these coefficients, high flows tend to decrease interpretation of the set of equations. For all as elevation increases, while low flows tend to these reasons, the models predicting volumetric increase in steeper catchments. It is difficult runoff (reported in Tables 3 and 4) were found to provide a physical interpretation for these to be more reliable than those predicting specific behaviors. While rainfall is always an explanatory runoff (not included in this report). Furthermore, variable in the Mekong River Basin, it is only multiple regression, including drainage area as used to predict mean flow and the 20% flow an independent variable, has been shown to percentile in the Volta River Basin, suggesting that better account for the heterogeneity of those basin rainfall is not the main discriminatory variable that attributes which are correlated to drainage area differentiates catchment hydrology in the gauged and thus reduces omitted variable bias (Farmer et 18 al. 2015). This scaling behavior is reflected in the FDC. Figures 4(b), (d) and (f) show the effect exponent values of the variable ‘drainage area’, of paddy on low flows. When the paddy area is which are different from 1 (cf. Tables 3 and 4), set to zero (this case is equivalent to replacing and by the fact that this variable is replaced by paddy by a land cover equal to a weighted the variable ‘perimeter’ in the models predicting average of all other land covers included in the low-flow percentiles. studied sample of sub-catchments), low flows increase. In these examples, the difference in Ecosystem-related Variables low flow is large because the three selected sub- In both the Volta and Mekong river basins, catchments in Figure 4, i.e., Nam Mun, Nam Chi the unavoidable time discrepancy between and Nam Mae Ing rivers, were covered by 60%, the dates of the flow record and the land 50% and 39% of the paddy area, respectively. cover maps probably weakens the relationship It should be noted that the 0.90 flow percentile between hydrology and land cover. However, is an outlier in the FDC corresponding to actual a f e w n o t a b l e c o r r e l a t i o n s a r e w o r t h paddy coverage in figures 4(b), (d) and (f). This highlighting and are discussed in the following is likely to be attributed to the lower performance subsections. of the power-law model predicting this flow percentile, which has the second lowest value of i. Paddy fields in the Mekong River Basin R2 pred in Table 3. The surface ratio of paddy rice is negatively Figure 5 illustrates the effect of an expansion correlated to four low-flow variables (f low in the rice area on low flows at Tha Ngon in Lao percentiles 0.60, 0.70, 0.80 and 0.95). One PDR. Figure 5(b) compares estimated FDCs with possible explanation is that the rice is irrigated current paddy areas to a hypothetical FDC that and dry-season abstractions are reducing low would correspond to an increase in the paddy flows. However, only a small proportion of the area from 6% (current paddy coverage in the lowland rice in the Lower Mekong River Basin is Nam Ngum River Basin at Tha Ngon) to 25% irrigated. In this case, an alternative explanation (projection for the next 25 years). The Nam Ngum is that the traditional practice of puddling (i.e., River Basin is particularly relevant for projecting churning of the soil prior to transplanting) creates expansion in the paddy area: it is one of the most a hardpan layer of rice below the rice root populated catchments in Lao PDR with easily zone, which reduces groundwater recharge, and accessible flat and fertile lands, including potential contributes to the maintenance of ponded water for paddy expansion (Lacombe et al. 2014a). and hence increased evapotranspiration from the Figure 5(b) shows that, even in the case of an bunded rice fields (Bouman et al. 2007). extreme scenario of expansion in the paddy area Figures 4(a), (c) and (e) i l lustrate the in the Nam Ngum catchment, the effect on low accuracy of the power-law models to predict the flow would remain moderate. 19 FIGURE 4. FDC at three stations along the main tributaries of the Mekong River in Thailand. (a), (c) and (e): comparison of observed and predicted FDC in actual conditions of paddy coverage (as of 2003, date of the land cover map used in this analysis). (b), (d) and (f): comparison of FDC predicted under actual paddy coverage and no paddy coverage. (a) (b) 10,000 Nam Mun River at Ubon Ratchathani 10,000 Nam Mun River at Ubon Ratchathani Observed FDC Predicted FDC (no paddy coverage) Predicted FDC (actual paddy coverage) 1,000 1,000 Predicted FDC (actual paddy coverage) 100 100 10 10 1 1 0 20 40 60 80 100 0 20 40 60 80 100 Flow percentile (%) Flow percentile (%) (c) (d) 10,000 10,000 Nam Chi River at Yasothon Nam Chi River at Yasothon 1,000 1,000 Predicted FDC (no paddy coverage) Observed FDC Predicted FDC (actual paddy coverage) Predicted FDC (actual paddy coverage) 100 100 10 10 1 1 0 20 40 60 80 100 0 20 40 60 80 100 Flow percentile (%) Flow percentile (%) (e) (f) 1,000 Nam Mae Ing River at Thoeng 1,000 Nam Mae Ing River at Thoeng Observed FDC Predicted FDC (no paddy coverage) 100 Predicted FDC (actual paddy coverage) 100 Predicted FDC (actual paddy coverage) 10 10 1 1 0 20 40 60 80 100 0 20 40 60 80 100 Flow percentile (%) Flow percentile (%) 20 Flow (m3/s) Flow (m3/s) Flow (m3/s) Flow (m3/s) Flow (m3/s) Flow (m3/s) FIGURE 5. FDC of the Nam Ngum River at Tha Ngon, Lao PDR. (a) comparison of observed and predicted FDC under actual paddy coverage; and (b) comparison of FDC predicted under actual paddy coverage (6% of the sub-catchment area) and extended paddy coverage (25% of the sub-catchment area). (a) 5,000 Nam Ngum River at Tha Ngon Observed FDC Predicted FDC (actual paddy coverage) 500 50 0 20 40 60 80 100 Flow percentile (%) (b) 5,000 Nam Ngum River at Tha Ngon Predicted FDC (extended paddy coverage) Predicted FDC (actual paddy coverage) 500 50 0 20 40 60 80 100 Flow percentile (%) In Figure 6, the impact of lowland rainfed portion of the sub-catchment covered with rice rice on seasonal flows is estimated in the sub- (64,049 km2). This difference is equivalent to the catchment of the Nam Mun River at Ubon difference between evapotranspiration of rice Ratchathani in Thailand. Currently, 60% of the and the combined evapotranspiration of other sub-catchment is covered with rice. Without any land covers in the sub-catchment. This low value paddy fields in this sub-catchment, low flows remains negligible compared to average annual would be about 200 × 106 m3/month higher evapotranspiration rates in the region (about from January to April. Over a full year, rainfed 1,300 mmy-1) (Tanaka et al. 2008). However, the paddy causes a flow reduction of about 800 × hydrological effect is significant (Figure 6), due to 106 m3, which is equivalent to 12 mm over the the large size of the sub-catchment. 21 Flow (m3/s) Flow (m3/s) FIGURE 6. Average hydrograph of the Nam Mun River at Ubon Ratchathani under current conditions (with paddy) and calculated assuming the absence of paddy fields in the sub-catchment. Nam Mun River at Ubon Ratchathani 6,400 3,200 1,600 800 No paddy coverage Actual paddy coverage 400 200 100 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec ii. Forest and crops in the Volta River Basin because many causes are possible through Table 4 shows that the only two explanatory changes in evapotranspiration, infiltration, soil variables selected from the land cover types water storage capacity, velocity of water transfers, investigated to predict the flow metrics q are etc. It is also possible that crops are grown in ‘crop’ and ‘forest’. According to the signs of the areas which are generally wetter, and therefore associated coefficients, increasing forest cover the results reflect a relationship but not a reduces high flow and increases low flow. It causative one. Further analyses and research are is possible that, by increasing infiltration and required to understand the observed relationship. evapotranspiration, surface runoff is reduced, Figure 7 illustrates the effect of land cover thereby reducing high flows. Furthermore, by changes on FDC at Saboba along the Oti River favoring infiltration and protecting soil against in Ghana. Two cases are compared to the erosion, forest increases the water storage original situation: a substitution of forest by crops capacity of the soil, thus increasing water table - changing the areal proportions of these two drainage into the stream during the dry season, land-use types from 19% and 38% to 1.4% and which increases low flows. However, these 55.4%, respectively, and a substitution of crops by suppositions need to be confirmed by process forests - changing the areal proportions of these studies and plot-level experiments. two land-use types from 38% and 19% to 8.9% The percen tage o f a rea covered by and 47.9%, respectively. These extreme cases crops alters the low flow only: according to were bounded by the extreme values of crop and the relationships resulting from the regression forest coverage reported in Table 2, because the analysis, this land cover type has no effect on predictive power of all models may reduce if they high flows. The positive coefficients indicate that are applied to catchments with characteristics low flows tend to increase as crop coverage outside the range of values reported in Table increases. It is difficult to interpret these results 2. The relative positions of the original, forest 22 Flow (10 6m3/month) and crop FDC in Figure 7(b) indicate that the is managed. This was not taken into account substitution of crop by forest reduces high flows here. It is possible that two land cover changes, slightly, while the opposite change (replacement apparently opposite in direction, both result in of forest by crop) increases high flow. Due to a drastic reduction of low flows. It depends on the much lower predictive power of the low flow how the new vegetation cover is managed, and models (Table 4), low flow changes in Figure 7(b) the impacts of land conversion on soil surface should be interpreted with caution. As mentioned properties and infiltration (e.g., Lacombe et al. above, low flow modelling, especially in the 2015). Less ambiguity is observed for high flows, Volta River Basin, would require further effort and our results should be used to anticipate the to improve the accuracy of predictions. Another possible risk of enhanced floods in response to confounding factor is the way the vegetation cover deforestation in the Volta River Basin. FIGURE 7. FDC of the Oti River at Saboba. (a) comparison of observed and predicted FDC under actual condition of land cover; and (b) predicted FDC under actual condition, and two scenarios either maximizing forest cover or crop cover. (a) 10,000 Oti River at Saboba Observed FDC 1,000 Predicted FDC (actual land cover) 100 10 1 0 20 40 60 80 100 Flow percentile (%) (b) 10,000.00 Oti River at Saboba Predicted FDC (extended forest) 1,000.00 Predicted FDC (actual land cover) Predicted FDC (extended crop) 100.00 10.00 1.00 0.10 0.01 0 20 40 60 80 100 Flow percentile (%) 23 Flow (m3/s) Flow (m3/s) It should be noted that investigation of the in the annual crop area from 38% to 8.9%) hydrological effect of such substitution of land results in an overall flow reduction of 2,100 cover (e.g., replacing forest by crops) could not × 106 m3/year, which is equivalent to 39 mm/ be carried out in the Mekong, because only year over the sub-catchment area (54,320 km2). one land cover category (paddy) was found to Flow losses in the dry season are relatively significantly alter river flow in this basin. greater (75% reduction in January) than that F igure 8 d isp lays the mean month ly in the wet season (29% reduction in August). hydrographs associated with each of the three This overall reduction in basin water yield could FDCs depicted in Figure 7(b). An increase in be attributed to higher evapotranspiration from the forest area from 19% to 48% of the sub- deep-rooted forest trees, compared to that of catchment area (at the expense of a reduction annual crops. FIGURE 8. Average hydrograph of the Oti River at Saboba for three land cover types: ‘Original’: actual conditions of land cover (38% and 19% of the sub-catchment covered by crop and forest, respectively). ‘Forest’ assumes an increase in forest cover (8.9% and 48% of the sub-catchment covered by crop and forest, respectively). ‘Crop’ assumes an increase in crop cover (55.4% and 1.4% of the sub-catchment covered by crop and forest, respectively). 10,000 Oti River at Saboba 1,000 100 Extended forest cover Actual land cover Extended crop cover 10 1 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 24 Flow (106m3/month) Limitations of this Study decreased [cf. Figure 6]), this land-use change does not explicitly account for the nature of any The variables listed in Tables 1 and 2 but not replacement in land cover. Rather, the assumption appearing in Tables 3 and 4, respectively (e.g., made implicitly is that the area is effectively drainage directions, soil characteristics, longitude replaced by an ‘average’ land-use equivalent to and wetland) were found not to have any a weighted average of all other land cover types explanatory power for any of the predicted flow included in the studied sample of sub-catchments metrics at the 5% significance level. However, (i.e., 65 sub-catchments for the Mekong). The these exclusions do not indicate that these impact could be considerably different depending variables have no hydrological impacts on on the actual replacement land use (e.g., if rice is downstream flow. For instance, the effects of replaced by another annual crop or forest). soils and wetlands on catchment hydrology are The method is an improvement to that complex, and depend on various context-specific developed for the Zambezi River Basin, because situations (Ribolzi et al. 2011). Acreman and it enables mean annual discharge to change as Holden (2013) show that the effects of wetlands a consequence of land cover change. However, on river regimes depend on their location within the approach remains limited by the assumption the catchment and on the geological nature that there is no change in the frequency of of the substratum. In our analysis, the lack of specific flow events as a consequence of land- explanatory power of these variables may be use change or the presence/absence of particular due to the selected metrics used to characterize ecosystems. In reality, this is unlikely to be the soils and wetlands (surface area, guided by data case. However, without knowledge of how the availability), which do not necessarily capture frequency of specific flows change (which is the main properties controlling catchment-scale likely to be location-specific), it is not possible to hydrological behavior. In addition, it should be modify the percentiles. More research is required noted that the surface area of wetlands never to quantify the effect of different land cover types exceeds 1.23% of the surface area of the studied on the frequency of flows. sub-catchments in the Lower Mekong River Basin A final constraint relates to the evaluation and 0.6% (water bodies) in the Volta River Basin, of land cover in each basin. Due to limitations which likely explains their apparent negligible role in available data, we were only able to utilize in the hydrological responses of the study sub- land cover assessments at a single point in time. catchments. There was no possibility of determining land cover The multivariate relationships reported in changes over time. Thus, the assumption was Tables 3 and 4 were used to predict what the made that catchment conditions were stationary. FDC would have been if the catchment areas This is, of course, not the case in reality, but covered by each land-use type were modified by without additional information on land use/land setting the relevant area to different values in the cover changes over time, it was necessary to regression models. This procedure enables an make this assumption. approximate assessment of the impact on flows of either an increase or a decrease in the area of a specific land use. When two different land-use Implications for Water Resources types are selected as explanatory variables in the Planning and Policy Making same regression model, the conversion of one land-use type to the other is easily determined Recent modelling efforts undertaken by river (e.g., in the Volta Basin, where crop is replaced basin commissions to include land management by forest or forest is replaced by crop [cf. Figure in basin development plans have focused on 8]). However, when the model includes only one irrigation and their impact on downstream flow land-use type as an explanatory variable (e.g., in (e.g., basin development plan of the Mekong River the Mekong River Basin, where the rice area is Basin: MRC [2011]). This study has demonstrated 25 that rainfed lowland paddy fields (the majority of implications of converting significant areas of land rice production in the Mekong River Basin is not to plantations, particularly of teak and rubber. irrigated) tend to significantly reduce downstream For the Volta River Basin, this study has low flow in this basin. This is particularly important highlighted the potential of increased high flows because the difference between water supply and likely additional flood risk, associated with and demand in the dry season is narrowing, as the conversion of forest to crops. This result a result of population growth and increasing dry- confirms that forest management is of tremendous season water demand. Therefore, in the Mekong importance in West Africa, not only for local River Basin, hydrological modelling efforts for communities, but also for downstream populations basin development planning should include paddy living in the vicinity of the major rivers. Given areas as an important component when modelling the relatively low runoff in the basin and high catchment-scale hydrological processes. This sensitivity to annual rainfall totals (in conjunction recommendation is of particular importance in with rapidly increasing population), there is an Lao PDR where paddy is seen as a major crop in urgent need for further research to identify the agricultural development plans. Similarly, further impacts of land-use change (deforestation, in research is essential to understand the possible particular) on low flows. Conclusions and Recommendations The p r imary goa l o f t h i s s tudy was to entities, guided by the data available. In addition, investigate how different types of ecosystems the surface area of wetlands never exceeds alter downstream river flows in two major river 1.23% of the gauged sub-catchment areas in the basins, which have very different hydro-ecological Mekong and 0.6% (water bodies) in the Volta, conditions: the Volta and the Mekong. Although likely explaining their apparent negligible role the method developed inevitably has limitations, in the hydrological responses of the study sub- overall, it is more rigorous and more objective catchments. We surmise that, in catchments with than that used previously in the Zambezi River a greater proportion of wetlands, they would (as Basin study. However, in part, because of data was the case in the Zambezi) have a significant limitations, the results were seemingly less clear impact on river flows. than those derived for the Zambezi. In addition to flow prediction under various Using multivariate power-law models to conditions of paddy, forest and crop cover, the predict streamflow, we found that land cover/land multivariate power-law models derived in this use does indeed control downstream flows. The study can be used for a range of applications, primary land cover types controlling downstream including prediction of the impact of change in flows are paddy areas in the Mekong and crop/ rainfall amounts on mean, low and high basin forest in the Volta. In contrast, wetlands and water yields (especially in the Mekong, where soil types were found not to have any significant annual rainfall is an explanatory variable in all effect on downstream flows in the sub-catchments flow percentile models), and regional impact investigated. This lack of statistical explanatory assessment of local hydrological alterations power was most likely a consequence of the through the comparison of water yields from selected metrics used to characterize these nested basins. 26 References Acreman, M.; Holden, J. 2013. How wetlands affect floods. Wetlands 33: 773-786. Andah, W.; van de Giesen, N.; Huber-Lee, A.; Biney, C.A. 2004. Can we maintain food production without losing hydropower? The Volta Basin (West Africa). In: Climate change in contrasting river basins: Adaptation for water, food and environment, eds., Aerts, J.C.J.H.; Droogers, P. Wallingford, UK: CABI Publishing. Pp. 181-194. Andreini, M.; van de Geisen, N.; van Edig, A.; Fosu, M.; Andah, W. 2000. Volta Basin water balance. ZEF Discussion Papers on Development and Policy No. 21. Bonn, Germany: Zentrum für Entwicklungsforschung (ZEF) (Center for Development Research). Biney, C.A. 2010. Connectivities and linkages within the Volta Basin. Available at www.gwsp.org/fileadmin/GCI_ conference/Products/Pres_-_Biney_-_Connectivities.pdf (accessed on October 7, 2015). Blöschl, G.; Sivapalan, M.; Wagener, T.; Viglione, A.; Savenije, H. (Eds.). 2013. Runoff prediction in ungauged basins: Synthesis across processes, places and scales. New York, NY, USA: Cambridge University Press. 462p. Bontemps, S.; Defourny, P.; Van Bogaert, E.; Arino, O.; Kalogirou, V.; Perez, J.R. 2011. GLOBCOVER 2009: Products description and validation report. Université catholique de Louvain; European Space Agency (ESA). 53p. Available at http://due.esrin.esa.int/files/GLOBCOVER2009_Validation_Report_2.2.pdf (accessed on February 1, 2016). Bouman, B.A.M.; Humphreys, E.; Tuong, T.P.; Barker, R. 2007. Rice and water. Advances in Agronomy 92: 187- 237. Bruijnzeel, L.A. 1996. Predicting the hydrological impacts of land cover transformation in the humid tropics: The need for integrated research. In: Amazonian deforestation and climate, eds., Gash, J.H.C.; Nobre, C.A.; Roberts, J.M.; Victoria, R.L. Chichester, UK: John Wiley. Pp. 15-55. Bullock, A.; Acreman, M. 2003. The role of wetlands in the hydrological cycle. Hydrology and Earth System Sciences 7: 358-389. Castellarin, A.; Botter, G.; Hughes, D.A.; Liu, S.; Ouarda, T.B.M.J.; Parajka, J.; Post, D.; Sivapalan, M.; Spence, C.; Viglione, A.; Vogel, R. 2013. Prediction of flow duration curves in ungauged basins. In: Runoff prediction in ungauged basins: Synthesis across processes, places and scales, eds., Blöschl, G.; Sivapalan, M.; Wagener, T.; Viglione, A.; Savenije, H. Cambridge, UK: Cambridge University Press. Pp. 135-162. Cook, R.D.; Weisberg, S. 1982. Residuals and influence in regression. New York: Chapman and Hall. Dickson, K.B.; Benneh, G. 1988. A new geography of Ghana. London: Longman. Emerton, L.; Bos, E. 2004. Value: Counting ecosystems as an economic part of water infrastructure. Gland, Switzerland; Cambridge, UK: International Union for Conservation of Nature (IUCN). 88p. Farmer, W.H.; Over, T.M.; Vogel, R.M. 2015. Multiple regression and inverse moments improve the characterization of the spatial scaling behavior of daily streamflows in the Southeast United States. Water Resources Research 51(3): 1775-1796. Harris I.; Jones, P.D.; Osborn, T.J.; Lister, D.H. 2014. Updated high-resolution grids of monthly climatic observations - the CRU TS3.10 Dataset. International Journal of Climatology 34: 623-642. Heinimann, A. 2006. Patterns of land cover change in the Lower Mekong Basin: The relevance of mesoscale approaches. PhD thesis. Faculty of Natural Sciences, University of Bern, Switzerland. National Centre of Competence in Research (NCCR) North-South/Centre for Development and Environment (CDE), Bern, Switzerland. Helsel, D.R.; Hirsch, R.M. 2002. Statistical methods in water resources. Reston, VA: United States Geological Survey (USGS). Homa, E.S.; Brown, C.; McGarigal, K.; Compton, B.W.; Jackson, S.D. 2013. Estimating hydrologic alteration from basin characteristics in Massachusetts. Journal of Hydrology 503: 196-208. 27 Homdee, T.; Pongput, K.; Kanae, S. 2011. Impacts of land cover changes on hydrologic responses: A case study of Chi River Basin, Thailand. Annual Journal of Hydraulic Engineering, JSCE 55: 31-36. Hrachowitz, M.; Savenije, H.H.G.; Blöschl, G.; McDonnell, J.; Sivapalan, M.; Pomeroy, J.W.; Arheimer, B.; Blume, T.; Clark, M.P.; Ehret, U.; Fenicia, F.; Freer, J.E.; Gelfan, A.; Gupta, H.V.; Hughes, D.A.; Hut, R.W.; Montanari, A.; Pande, S.; Tetzlaff, D.; Troch, P.A.; Uhlenbrook, S.; Wagener, T.; Winsemius, H.C.; Woods, R.A.; Zehe, E.; Cudennec, C. 2013. A decade of Predictions in Ungauged Basins (PUB)—a review. Hydrological Sciences Journal 58(6): 1198-1255. Hughes, R.H.; Hughes, J.S. 1992. A directory of African wetlands. Gland, Switzerland and Cambridge, UK: IUCN; Nairobi: United Nations Environment Programme (UNEP); Kenya: World Conservation Monitoring Centre (WCMC). 820p. IRIN (Integrated Regional Information Networks). 2007. Ghana: Nearly 275,000 affected by floods in little-known disaster. Available at http://www.irinnews.org/report/74278/ghana-nearly-275-000-affected-by-floods-in-little-known- disaster (accessed on March 3, 2014). Lacombe, G.; Pierret, A.; Hoanh, C.T.; Sengtaheuanghoung, O.; Noble, A. 2010. Conflict, migration and land-cover changes in Indochina: A hydrological assessment. Ecohydrology 3(4): 382-391. Lacombe, G.; McCartney, M.; Forkuor, G. 2012. Drying climate in Ghana over the period 1960-2005: Evidence from the resampling-based Mann-Kendall test at local and regional levels. Hydrological Sciences Journal 57(8): 1594-1609. Lacombe, G.; Douangsavanh, S.; Baker, J.; Hoanh C.T.; Bartlett, R.; Jeuland, M.; Phongpachith, C. 2014a. Are hydropower and irrigation development complements or substitutes? The example of the Nam Ngum River in the Mekong Basin. Water International 39(5): 649-670. Lacombe, G.; Douangsavanh, S.; Vogel, R.; McCartney, M.; Chemin, Y.; Rebelo, L.; Sotoukee, T. 2014b. Multivariate power-law models for streamflow prediction in the Mekong Basin. Journal of Hydrology: Regional Studies 2: 35-48. Lacombe, G.; Ribolzi, O.; de Rouw, A.; Pierret, A.; Latsachak, K.; Silvera, N.; Pham Dinh, R.; Orange, D.; Janeau, J.-L.; Soulileuth, B.; Robain, H.; Taccoen, A.; Sengphaathith, P.; Mouche, E.; Sengtaheuanghoung, O.; Tran Duc, T.; Valentin, C. 2015. Afforestation by natural regeneration or by tree planting: Examples of opposite hydrological impacts evidenced by long-term field monitoring in the humid tropics. Hydrology and Earth System Sciences Discussions 12: 12615-12648. Lehner, B.; Verdin, K.; Jarvis, A. 2006. HydroSHEDS technical documentation. Washington, DC, USA: World Wildlife Fund (WWF). Leinenkugel, P.; Oppelt, N.; Kuenzer, C. 2013. A new land cover map for the Mekong: Southeast Asia’s largest transboundary river basin. Pacific Geographies 41: 10-14. McCartney, M.P.; Cai, X.; Smakhtin, V. 2013. Evaluating the flow regulating functions of natural ecosystems in the Zambezi River Basin. Colombo, Sri Lanka: International Water Management Institute (IWMI). 59p. (IWMI Research Report 148). MRC (Mekong River Commission). 2005. Overview of the hydrology of the Mekong Basin. Vientiane: Mekong River Commission. 82p. MRC. 2009. Annual Mekong flood report 2008. Vientiane: Mekong River Commission. 84p. MRC. 2011. Basin development plan programme - phase 2. Assessment of basin-wide development scenarios. Main report. Vientiane: Mekong River Commission. Nicholson, S. 2005. On the question of the “recovery” of the rains in the West African Sahel. Journal of Arid Environments 63: 615-641. Pilgrim, D.H.; Cordery, I.; Baron, B.C. 1982. Effects of catchment size on runoff relationships. Journal of Hydrology 58(3-4): 205-221. 28 Ribolzi, O.; Patin, J.; Bresson, L.M.; Latsachack, K.O.; Mouche, E.; Sengtaheuanghoung, O.; Silvera, N.; Thiébaux, J.P.; Valentin, C. 2011. Impact of slope gradient on soil surface features and infiltration on steep slopes in northern Laos. Geomorphology 127: 53-63. Salinas, J.L.; Laaha, G.; Rogger, M.; Parajka, J.; Viglione, A.; Sivapalan, M.; Blöschl, G. 2013. Comparative assessment of predictions in ungauged basins – Part 2: Flood and low flow studies. Hydrology and Earth System Sciences 17: 2637-2652. Schönweger, O.; Heinimann, A.; Epprecht, M.; Lu, J.; Thalongsengchanh, P. 2012. Concessions and leases in the Lao PDR: Taking stock of land investments. Centre for Development and Environment (CDE), University of Bern. Bern and Vientiane: Geographica Bernensia. Shaw, E.M. 1984. Hydrology in practice. Wokingham, UK: Van Nostrand Reinhold (UK) Ltd. Tanaka, N.; Kume, T.; Yoshifuji, N. ; Tanaka, K.; Takizawa, H.; Shiraki, K.; Tantasirin, C.; Tangtham, N.; Suzuki, M. 2008. A review of evapotranspiration estimates from tropical forests in Thailand and adjacent regions. Agricultural and Forest Meteorology 148(5): 807-819. Tasker, G.D. 1980. Hydrologic regression with weighted least squares. Water Resources Research 16(6): 1107- 1113. van de Giesen, N.; Andreini, M.; van Edig, A.; Vlek, P. 2001. Competition for water resources of the Volta Basin. In: Regional Management of Water Resources: Proceedings of a Symposium held during the Sixth IAHS Scientific Assembly at Maastricht, the Netherlands, July 2001. IAHS Publication no. 268. Vogel R.M.; Fennessey, N.M. 1994. Flow-duration curves. I: New interpretation and confidence intervals. Journal of Water Resources Planning and Management 120(4): 485-504. Yatagai, A.; Kamiguchi, K.; Arakawa, O.; Hamada, A.; Yasutomi, N.; Kitoh, A. 2012. APHRODITE: Constructing a long-term daily gridded precipitation dataset for Asia based on a dense network of rain gauges. Bulletin of the American Meteorology Society 93(9): 1401-1415. 29 Annex TABLE A1. Characteristics of the river gauging stations that were included in the analysis carried out in the Mekong River Basin. Country Station name River name Latitude Longitude Selected hydrological years Cambodia Ban Kamphun Se San 13.55 106.05 1960-1963, 1965-1968, 1994- 1996, 1999-2001 Cambodia Ban Khmoun Se Kong 13.74 106.19 1961-1968 Cambodia Battambang Stung Sangker 13.05 103.20 1999-2001 Cambodia Kompong Putrea Stung Sen 13.22 105.26 1965-1968 Cambodia Kompong Thmar Stung Chinit 12.50 105.13 1997-2001 Cambodia Kompong Thom Stung Sen 12.71 104.87 1961-1969, 1982-2001 Cambodia Lumphat Sre Pok 13.47 106.98 1965, 1967-1969 Cambodia Treng Stung Sangker 12.87 103.14 1963-1971 Lao PDR Attapeu Se Kong 14.81 106.84 1989-1990, 1993-1995, 1998- 2004 Lao PDR Ban Hin Heup Nam Lik 18.63 102.36 1967-1974, 1993-1995 Lao PDR Ban Keng Done Se Bang Hieng 16.19 105.32 1965-1970, 1973-1977, 1979, 1992-2003 Lao PDR Ban Kok Van Nam Pa 19.96 102.30 1988-1991, 1996-2003 Lao PDR Ban Mixay Nam Khan 19.79 102.18 1960-1983, 1985-1991, 1995- 2004 Lao PDR Ban Muong Chan Se Pon 16.66 106.29 1999-2002 Lao PDR Ban Na Luang Nam Ngum 18.91 102.78 1987-2008 Lao PDR Ban Phalane Se Xangxoy 16.66 105.57 1997-1999 Lao PDR Ban Sibounhom Nam Suong 19.97 102.27 1968-1971, 1987-1991, 1995- 2003 Lao PDR Ban Signo Nam Theun 17.85 105.05 1989-2004 Lao PDR Dong Hen Se Champhone 16.70 105.29 1995-2003 Lao PDR Highway Bridge Se Thamouak 16.58 105.91 1995-2002 Lao PDR Keng Kok Se Champhone 16.45 105.21 1988, 1990, 1992, 1995-1996, 1998-2001 Lao PDR Kham Keut Nam Theun 18.24 104.66 1986, 1990 Lao PDR Khong Sedone Se Done 15.58 105.81 1989 Lao PDR Mahaxai Se Bang Fai 17.41 105.20 1989-2000, 2002-2004 Lao PDR Muong Borikhan Nam Sane 18.56 103.74 1985-1990, 1993-2004 Lao PDR Muong Kasi Nam Lik 19.24 102.26 1987-2003 Lao PDR Muong Mai Nam Nhiap 18.50 103.66 1987-1994, 1997-2004 Lao PDR Muong Ngoy Nam Ou 20.70 102.67 1989-1991, 1995-2001 Lao PDR Muong Nong Se La Nong 16.37 106.51 1995-2003 Lao PDR Road Nb 13 Bridge Se Bang Fai 17.08 104.98 1961, 1963-1971, 1973-1977, 1979-1984, 1994-2003 Lao PDR Saravanne Se Done 15.69 106.45 1988-1991, 1994-1996, 2000, 2002-2004 (Continued) 30 TABLE A1. Characteristics of the river gauging stations that were included in the analysis carried out in the Mekong River Basin. (Continued) Country Station name River name Latitude Longitude Selected hydrological years Lao PDR Souvannakhili Se Done 15.40 105.83 1986-1990 Lao PDR Tchepon bridge Se Bang Hieng 16.68 106.21 1999-2003 Lao PDR Tha Ngon Nam Ngum 18.13 102.62 1962-1970 Lao PDR Vang Vieng Nam Song 18.92 102.44 1988-1994, 2002-2004 Thailand Ban Chot Nam Chi 16.10 102.58 1976-1986, 1988-1992, 1994- 2002 Thailand Ban Fang Phe Lam Dom Yai 14.69 105.15 1969-1998 Thailand Ban Huai Khayuong Huai Khayuong 15.01 104.64 1980-2002 Thailand Ban Huai Yano Mai Nam Mae Chan 20.11 99.79 1976-2002 Thailand Ban Kae Nam Pong 16.86 102.18 1980-1996 Thailand Ban Na Kham Noi Huai Bang Sai 16.72 104.63 1985-2002 Thailand Ban Na Thom Nam Yang 16.05 104.03 1980-1998 Thailand Ban Nong Kiang Huai Rai 16.13 101.66 1976-1977, 1979-2002 Thailand Ban Pa Yang Nam Mae Kham 20.23 99.80 1981-1997, 1999-2002 Thailand Ban Pak Huai Nam Heung 17.71 101.41 1967-1993, 1996-2002, 2006 Thailand Ban Tad Ton Huai Pa Thao 15.94 102.03 1977-1986, 1988-1989, 1991- 2002 Thailand Ban Tha Kok Daeng Nam Songkhram 17.86 103.78 1965-1974, 1992-2000 Thailand Ban Tha Mai Liam Nam Mae Fang 20.02 99.35 1970-2002 Thailand Ban Tha Sai Nam Mae Lao 19.86 99.84 1972-1998 Thailand Ban Tha Ton Nam Mae Kok 20.06 99.36 1970-1993, 1996-2004 Thailand Ban Wang Sai Nam Loei 17.05 101.52 1976-1979, 1981-2002 Thailand Chiang Rai Nam Mae Kok 19.92 99.85 1977-1992 Thailand Dam site Nam Mae Suai 19.70 99.52 1976-1993, 1996-1997 Thailand Dam site Nam Mae Pun Luang 19.43 99.46 1976-2002 Thailand Dam site Nam San 17.43 101.27 1966-2002 Thailand Dan Sai Nam Man 17.28 101.15 1968-2002 Thailand Rasi Salai Nam Mun 15.33 104.16 1979-1993, 1996-2000 Thailand Thoeng Nam Mae Ing 19.68 100.19 1969-2002 Thailand Ubon Ratchathani Nam Mun 15.22 104.86 1951-1965 Thailand Wang Saphung Nam Loei 17.30 101.78 1967-1988, 1991-2002, 2006 Thailand Yasothon Nam Chi 15.78 104.14 1952-1965 Vietnam Cau 14 Ea Krong 12.61 107.93 1984-2003 Vietnam Duc Xuyen Krong Kno 12.33 107.86 1985-2006 Vietnam Kontun Dak Bla 14.34 108.00 1984-1988, 1990-2005 Vietnam Trung Nghai Krong Po Co 14.37 107.87 1992, 1995-1997 31 TABLE A2. Characteristics of the river gauging stations that were included in the analysis carried out in the Volta River Basin. Country Station name River name Latitude Longitude Selected hydrological years Benin Porga Pendjari 11.05 0.97 1954-1955, 1959, 1973 Burkina Faso Dakaye Red Volta 11.78 -1.6 1977, 1979-1980, 1982 Burkina Faso Diebougou Bougouriba 10.93 -3.17 1978-1979 Burkina Faso Samandeni Black Volta 11.46 -4.46 1977-1978, 1981-1982 Burkina Faso Samboali Singou 11.28 -1.02 1978, 1981 Burkina Faso Yilou Nakanbe 13 -1.55 1977-1980 Ghana Bamboi Black Volta 8.15 -2.03 1951, 1955, 1957, 1960, 1964, 1968-1971, 1973, 1975, 2003 Ghana Bui Black Volta 8.28 -2.23 1971-1976, 1981, 1983-2007 Ghana Chache Black Volta 9.17 -2.72 1998, 2002 Ghana Kpasenkpe White Volta 10.43 -1.05 2005 Ghana Nakong Sissili 10.8 -1.5 1972, 1974, 2003 Ghana Nangodi Red Volta 10.87 -0.62 1972 Ghana Nasia Black Volta 10.15 -0.8 1969-1972, 1990, 1992-1995, 2000, 2002, 2004-2005 Ghana Nawuni White Volta 9.7 -1.08 1954-1979, 1984, 1986-1989, 1991, 1993-2000, 2002-2003, 2005-2007 Ghana Pwalagu White Volta 10.58 -0.85 1952, 1963, 1966, 1968, 1971- 1972, 2003, 2005 Ghana Sabari Oti 9.28 0.23 1976-1977, 1982, 1984, 1997, 2004-2005 Ghana Saboba Oti 9.6 0.31 1954, 1956, 1961, 1963-1966, 1971, 1976, 1985-1987, 1989, 1991-1993, 1995-1998, 2004- 2007 Ghana Wiasi Sissili 10.33 -1.35 1966, 1976, 1990, 1992 Ghana Yarugu White Volta 10.98 -0.4 1962-1969, 1971-1972, 1975, 2005 Ghana Yagaba Kulpawn 10.23 -1.28 1965, 1971, 1992, 2005 32 IWMI Research Reports 166 Evaluating the Flow Regulating Effects of Ecosystems in the Mekong and Volta River Basins. Guillaume Lacombe and Matthew McCartney. 2016. 165 Controlling Floods and Droughts through Underground Storage: From Concept to Pilot Implementation in the Ganges River Basin. Paul Pavelic, Brindha Karthikeyan, Giriraj Amarnath, Nishadi Eriyagama, Lal Muthuwatta, Vladimir Smakhtin, Prasun K. Gangopadhyay, Ravinder P. S. Malik, Atmaram Mishra, Bharat R. Sharma, Munir A. Hanjra, Ratna V. Reddy, Vinay Kumar Mishra, Chhedi Lal Verma and Laxmi Kant. 2015. 164 Integrated Assessment of Groundwater Use for Improving Livelihoods in the Dry Zone of Myanmar. Paul Pavelic, Sonali Senaratna Sellamuttu, Robyn Johnston, Matthew McCartney, Touleelor Sotoukee, Soumya Balasubramanya, Diana Suhardiman, Guillaume Lacombe, Somphasith Douangsavanh, Olivier Joffre, Khin Latt, Aung Kyaw Zan, Kyaw Thein, Aye Myint, Cho Cho and Ye Thaung Htut. 2015. 163 Demonstrating Complexity with a Role-playing Simulation: Investing in Water in the Indrawati Subbasin, Nepal. John Janmaat, Suzan Lapp, Ted Wannop, Luna Bharati and Fraser Sugden. 2015. 162 Landlordism, Tenants and the Groundwater Sector: Lessons from Tarai- Madhesh, Nepal. Fraser Sugden. 2014. 161 Is ‘Social Cooperation’ for Traditional Irrigation, while ‘Technology’ is for Motor Pump Irrigation? Mengistu Dessalegn and Douglas J. Merrey. 2014. 160 Understanding Farmers’ Adaptation to Water Scarcity: A Case Study from the Western Nile Delta, Egypt. Wafa Ghazouani, François Molle, Atef Swelam, Edwin Rap and Ahmad Abdo. 2014. 159 Climate Change, Out-migration and Agrarian Stress: The Potential for Upscaling Small-scale Water Storage in Nepal. Fraser Sugden, Lata Shrestha, Luna Bharati, Pabitra Gurung, Laxmi Maharjan, John Janmaat, James I. Price, Tashi Yang Chung Sherpa, Utsav Bhattarai, Shishir Koirala and Basu Timilsina. 2014. 158 Water for Food in Bangladesh: Outlook to 2030. Upali A. Amarasinghe, Bharat R. Sharma, Lal Muthuwatta and Zahirul Haque Khan. 2014. Electronic copies of IWMI's publications are available for free. Visit www.iwmi.org/publications/index.aspx Postal Address P O Box 2075 Colombo Sri Lanka Location 127 Sunil Mawatha Pelawatta Battaramulla Sri Lanka Telephone +94-11-2880000 Fax +94-11-2786854 E-mail iwmi@cgiar.org Website www.iwmi.org ISSN: 1026-0862 ISBN: 978-92-9090-833-3