CIAT Research Online - Accepted Manuscript Exploring adaptation strategies of coffee production to climate change using a process-based model The International Center for Tropical Agriculture (CIAT) believes that open access contributes to its mission of reducing hunger and poverty, and improving human nutrition in the tropics through research aimed at increasing the eco-efficiency of agriculture. CIAT is committed to creating and sharing knowledge and information openly and globally. We do this through collaborative research as well as through the open sharing of our data, tools, and publications. Citation: Rahn, Eric; Vaast, Philippe; Läderach, Peter; Van Asten, Piet; Jassogne, Laurence; Ghazoul, Jaboury. 2018. Exploring adaptation strategies of coffee production to climate change using a process-based model. Ecological Modelling 371: 76-89. Publisher’s DOI: https://doi.org/10.1016/j.ecolmodel.2018.01.009 Access through CIAT Research Online: http://hdl.handle.net/10568/90966 Terms: © 2018. CIAT has provided you with this accepted manuscript in line with CIAT’s open access policy and in accordance with the Publisher’s policy on self-archiving. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. You may re-use or share this manuscript as long as you acknowledge the authors by citing the version of the record listed above. You may not change this manuscript in any way or use it commercially. For more information, please contact CIAT Library at CIAT-Library@cgiar.org. 1 Exploring adaptation strategies of coffee production to climate change using 1 a process-based model 2 3 Rahn Eric1,2,3, Vaast Philippe4,5, Läderach Peter2, van Asten Piet3,6, Jassogne Laurence3, Ghazoul Jaboury1 4 5 1Swiss Federal Institute of Technology (ETH) Zurich, Environmental Systems Science, Switzerland 6 2International Center for Tropical Agriculture (CIAT), Cali, Colombia 7 3International Institute of Tropical Agriculture (IITA), Kampala, Uganda 8 4World Agroforestry Centre (ICRAF), Hanoi, Vietnam 9 5Centre de Coopération International en Recherche Agronomique pour le Développement (CIRAD), 10 Eco&Sols, Université de Montpellier, France 11 6 Olam International Ltd, Kampala, Uganda 12 13 Corresponding author: 14 Eric Rahn 15 eric.rahn@usys.ethz.ch 16 ETH Zurich 17 CHN H 71, Universitätsstrasse 16, 8092 Zürich, Switzerland 18 +41 44 633 60 62 19 Field Code Changed 2 1. Abstract 20 The response of coffee (Coffea arabica L.) agronomical performance to changes in climate and atmospheric 21 carbon dioxide concentration ([CO2]) is uncertain. Improving our understanding of potential responses of 22 the coffee plant to these changes while taking into consideration agricultural management is required for 23 identifying best-bet adaptation strategies. A mechanistic crop modeling approach enables the inclusion of a 24 wide range of prior knowledge and an evaluation of assumptions. We adapt a model by connecting it to 25 spatially variable soil and climate data, by which we are able to calculate yield of rain-fed coffee on a daily 26 time-step. The model takes account of variation in microclimate and water use as influenced by shade trees. 27 The approach is exemplified at two East African sites with distinctly different climates (Mt. Elgon, Uganda, 28 and Mt. Kilimanjaro, Tanzania) using a global sensitivity analysis for evaluation of model behavior and 29 prior parameter uncertainty assessment. We use the climate scenario driven by the Hadley Global 30 Environment Model 2 representative for the year 2050 to discuss potential responses of the coffee plant to 31 interactions of elevated [CO2], temperature, and water availability. We subsequently explore the potential 32 for adaptation to this scenario through shade management. The results indicate that under current climatic 33 conditions optimal shade cover at low elevations (1000 m.a.s.l.) is 50%, provided soil water storage capacity 34 is sufficient, enabling a 13.5% increase in coffee yield compared to unshaded systems. Coffee plants are 35 expected to be severely impacted (ranging from 18% to 32% coffee yield reductions) at low elevations by 36 increased temperature (+2.5°C) and drought stress when no elevated [CO2] is assumed. Water competition 37 between coffee and shade trees are projected to be a severe limitation in the future, requiring careful 38 selection of appropriate shade tree species or the adoption of other technologies like conservation measures 39 or irrigation. The [CO2]-fertilization effect could potentially mitigate the negative effect of temperature 40 increase and drought stress up to 13-21% depending on site conditions and will increase yield at higher 41 altitudes. High uncertainty remains regarding impacts of climate change on flowering. The present model 42 allows for estimating the optimal shade level along environmental gradients now and in the future. Overall, 43 it shows that shade proves to be an important adaptation strategy, but this requires improved understanding 44 regarding site-specific management and selection of tree species. Moreover, we do not yet include climate 45 change uncertainty. 46 2. Keyword index 47 Coffea arabica L., agroforestry system, spatial decision support, adaptation to climate change, parameter 48 uncertainty 49 3 3. Introduction 50 Coffee is cultivated in over 70 countries throughout the tropics with approximately 60% of the production 51 being Coffea arabica L. (Arabica coffee) and 40% being Coffea canephora Pierre ex Froehner, syn. Coffea 52 robusta (Robusta coffee) (FAO 2015). Over 70% of the world’s coffee is produced by smallholders 53 managing less than 10 hectares of land (Fridell 2014). Climate change is expected to have substantial 54 impacts on suitable areas for coffee (C. arabica) cultivation (Bunn et al. 2015a; Ovalle-Rivera et al. 2015; 55 Magrach & Ghazoul 2015), pests and diseases pressure (Jaramillo et al. 2011; Magrach & Ghazoul 2015) 56 and genetic resources (Davis et al. 2012), thereby likely changing the agro-ecological zones most suitable 57 for coffee production (Bunn et al. 2015b). Agroforestry systems can both contribute to climate change 58 mitigation while potentially enabling adaptation to climatic changes (Matocha et al. 2012; Mbow et al. 2014; 59 Vaast et al. 2016). 60 Due to the perennial nature of coffee with an economic lifespan typically up to 30 years (Wintgens 2004) 61 and the long time required for agroforestry trees to grow to maturity, decisions regarding adaptation to 62 climate change are challenging. Therefore, there is an urgent need for decision support consisting of accurate 63 estimates of climatic suitability for coffee production and the influence of modified microclimate by shade 64 trees, including competition effects (Luedeling et al. 2014). Where long historical records of coffee 65 performance, weather and soil conditions are available, this is a relatively easy task, but the majority of 66 coffee growing areas lack such data (Luedeling et al. 2014). Statistical species distribution modelling 67 approaches (Schroth et al. 2009, Bunn et al. 2015a; Magrach & Ghazoul 2015) or agro-climatic indices 68 (Lane & Jarvis 2007) have instead been used. These methods are suitable for characterizing broad agro-69 ecological zones (Bunn et al. 2015b) and generating hypotheses on the suitable climatic conditions for 70 coffee, but they lack a mechanistic process representation required to predict crop response outside the 71 current growing domain, including the carbon fertilization effect (CFE) induced by rising atmospheric 72 carbon dioxide concentration ([CO2]). Furthermore, while the above studies have analyzed geographical 73 shifts in coffee suitability, indicating a decrease of available area in the future, they did neither include 74 phenotypic responses of the coffee plant (Nicotra et al. 2010) nor management practices allowing for 75 adaptation to climate change, such as shade management, irrigation, or changes in coffee genotypes (Vaast 76 et al. 2016). 77 Mechanistic crop models are believed to be more appropriate in generating realistic simulations of plant-78 soil-climate interactions. Moreover, they facilitate learning through hypothesis testing and identification of 79 missing knowledge (Sinclair & Seligman 1996; Boote et al. 2010), which enables guidance for management 80 action (Harfoot et al. 2014). However, applying such models without sufficient data for calibration, results 81 4 in large uncertainties of model predictions, next to existing uncertainties in model structure (Beven 2008; 82 Luedeling et al. 2016). The latter is often not identifiable by comparing model outputs with observations 83 alone, as many models can be fitted to the same data leading to the problem of equifinality (Beven and Freer 84 2001). Model comparison and critical reflection of assumptions is considered more appropriate (De Kauwe 85 et al. 2014). 86 Most of the parameters in a crop model are considered “genetic coefficients”, and do not have to be adjusted 87 when applied at different sites (Yin et al. 2004). Yet, some parameters encompass limited process 88 understanding and require calibration when the model is applied to different sites. These include parameters 89 related to the induction of flowering in coffee (Van Oijen et al. 2010b; Rodriguez et al. 2011). Another 90 aspect related to parameter values is phenotypic plasticity, i.e. changes in morphological, chemical, and 91 physical characteristics of a plant in response to the environment. If phenotypic plasticity is explicitly 92 accounted for, there is no need to adapt parameters in different environmental conditions (Yin 2013). 93 Obviously, this is only possible when the required knowledge is available to adequately represent these 94 processes. Considerable understanding is available on phenotypic responses of coffee to water (Poorter & 95 Nagel 2000; Carr 2012; Cavatte et al. 2012; Cannavo et al. 2011) and light availabilities (Matos et al. 2009; 96 Charbonnier et al. 2013; Martins et al. 2014a), but uncertainty is much greater with regard to phenotypic 97 plasticity to atmospheric [CO2] variation (Yin 2013), with only few experimental studies regarding coffee 98 so far (Martins et al. 2014c; Ghini et al. 2015; DaMatta et al. 2016). 99 Other difficulties in exploring possible impacts of climate change on crop production are related to the 100 uncertain responses to climate extremes (Thornton et al. 2014). Depending on the methods used for 101 downscaling Global Climate Model (GCM) output to scales suitable for agriculture, the projected changes 102 in climate only represent mean changes in temperature and precipitation and do not adequately represent 103 changing climate variability, notably temperature and precipitation extremes (Müller et al. 2011; Ramirez-104 Villegas et al. 2013). In addition to uncertainty in boundary conditions, there is also uncertainty in the actual 105 effects of such extremes on the plant (Reyer et al. 2013). 106 The goal of this study is to use a mechanistic coffee model, which integrates current knowledge on coffee 107 ecophysiology, to evaluate potential impacts of climate change in various agro-ecological settings and 108 agricultural managements. By making use of statistical approaches to explore the plausible parameter space, 109 we identify optimal current and future management practices of a wide range of potential genotypes. The 110 objectives of this paper are to 1) present the proposed coffee model, 2) assess model outcome in time and 111 space using mean literature derived parameter values, 3) identify model behavior through global sensitivity 112 analysis and 4) evaluate how robust the predicted change is despite parameter uncertainty conditioned by 113 5 different climate scenarios. We used two contrasting sites of East Africa as case study areas, namely the wet 114 slopes of Mount Elgon, Uganda vs the drier slopes of Mount Kilimanjaro, Tanzania. 115 6 4. Material and methods 116 A. Coffee Model 117 The original version of the coffee agroforestry model (CAF2007) was described by van Oijen et al. (2010b) 118 and extended by Ovalle-Rivera (2014). This study adapts the CAF2014 model for use as a spatially 119 contextualized decision support tool (SpCAF). This model was chosen as it is specifically designed to deal 120 with coffee agroforestry systems and includes a mechanistic light use efficiency approach that deals with 121 the interaction between temperature and [CO2]. In comparison to CAF2014, we assume no nutrient 122 limitations as we intend to isolate the impact of climate on coffee and therefore focus on yield response to 123 water (i.e. water limited yield according to van Ittersum et al. 2013), temperature and atmospheric [CO2] 124 levels. Consequently, coffee yield is expressed exclusively as a function of climate and soil water 125 availability, excluding nutrient competition, pest and disease alterations, or allelopathic properties of shade 126 trees on understory coffee. Tree shading is simplified to a canopy that provides shade and competes for 127 water through evapotranspiration. Thus, the objective is not to explicitly model a specific shade tree species, 128 but rather allow for exploration of the continuity between no shade and heavy shade and its effects on 129 microclimate and water competition. The model calculates water-limited coffee yield at a daily time-step 130 and is implemented in R statistics (R Core Team 2014). In the following sections, all key processes and 131 model assumptions are presented. An overview of the model is illustrated in Figure 1. 132 [Place Figure 1 here] 133 Coffee growth under optimal water supply 134 Canopy photosynthesis is modelled using a mechanistic light-use efficiency (LUE) approach based on the 135 leaf photosynthesis model of Farquhar et al. (1980) and scaled up to canopy photosynthesis (Charles-136 Edwards 1982), as described in detail by van Oijen et al. (2004). This formula for LUE is calculated on a 137 daily basis and depends on temperature, atmospheric [CO2] concentration, light intensity and the Rubisco 138 content of upper leaves. Instead of modelling photosynthesis and respiration separately, the LUE approach 139 assumes a constant ratio of daily rates of respiration and photosynthesis, which has been explained 140 experimentally (e.g. Gifford 1995, 2003) and theoretically (van Oijen et al. 2010c). The parameters have 141 been adjusted to very low and high light intensity, yielding highest values for LUE at low intensity (van 142 Oijen et al. 2010b). This allows consistency with observations as reported by Franck & Vaast (2009), 143 Cavatte et al. (2012) and Charbonnier et al. (2017) for Arabica coffee. Moreover, by expressing the LUE 144 approach based on the Farquhar et al. (1980) biochemical model, the interaction between temperature and 145 7 elevated atmospheric [CO2] is represented (Yin 2013). Under favorable conditions, growth rate is 146 proportional to the amount of light intercepted by the canopy (Monteith 1977). Therefore, LUE is multiplied 147 with daily intercepted solar radiation, calculated as an exponential function of leaf area index (LAI) and a 148 constant light extinction coefficient according to Beer’s Law for attenuation (Monsi & Saeki 1953). Water 149 stress decreases LUE proportionally via the ratio of actual to potential transpiration rate (Kropff & van Laar 150 1993). The LUE of C. arabica has been shown to range between 0.23 – 0.8 g MJ-1 (Bote et al. 2016a, 151 Charbonnier et al. 2017). It has also been shown that LUE can increase by 50% under shade and thereby 152 largely compensate for a reduction in 60% photosynthetically active radiation (PAR) leading to almost equal 153 net primary productivity (Charbonnier et al. 2017). Therefore, using the LUE approach appears to be 154 particularly appropriate for summarizing differences in phenotypic plasticity to shade (Cavatte et al. 2012; 155 Charbonnier et al. 2017; Bote et al. 2016a). This plasticity is reflected by several physiological and 156 morphological adjustments. Firstly, plants in low light conditions allocate more nitrogen to light harvesting 157 rather than carbon-assimilating enzymes (Cavatte et al. 2012; Poorter et al. 2014), although coffee was 158 found to have a robust photosynthetic machinery even in low light that allows to maximize variable light 159 availability by sunflecks (Martins et al. 2013). Secondly, in high light an increased investment in 160 photoprotective mechanisms was observed (Matos et al. 2009). Thirdly, reduced VPD under shade improves 161 stomatal and mesophyll conductance over the day (Franck & Vaast 2009; Martins et al. 2014b), which is 162 particularly important as coffee photosynthesis is mainly limited by diffusive rather than biochemical 163 limitations (DaMatta et al. 2007; Chaves et al. 2008; Franck & Vaast 2009; Martins et al. 2014b). 164 Fourthly, increases in specific leaf area (Martins et al. 2014b) and diffuse light availability (Charbonnier et 165 al. 2013) under shade might further increase fraction of absorbed photosynthetically active radiation 166 compared to unshaded systems. 167 The fraction of carbon (C) allocated to the different plant organs depends on phenological stage (vegetative 168 growth, bean maturation) and is modulated by water availability. Root C sink strength increases with 169 increasing water stress to the detriment of other plant organs according to the functional equilibrium theory 170 (Brouwer et al. 1983). High fruit load can come at the expense of (i) vegetative growth leading to ‘die back’ 171 (Vaast et al. 2005) in extreme cases, and (ii) the development of fruiting nodes and hence lower sink strength 172 of fruits in the following year (Vaast et al. 2005; Bote & Vos 2016). Phenology is based on thermal time 173 (degree days) and flowering is induced by a threshold amount of rainfall after the dry season (DaMatta et 174 al. 2007; Carr 2012). The intensity of flowering is a function of monthly light intensity around flowering 175 (Cannell 1985; Franck & Vaast 2009). Leaf senescence varies depending on the degree of water stress, while 176 roots were set with a constant lifespan. Leaf growth is calculated as the product of leaf biomass and specific 177 leaf area. The latter is decreased by water stress. 178 8 179 Soil water balance and water stress 180 A simple one-layer soil water balance is used to calculate the amount of water in the root zone, whereby 181 water is added by precipitation and lost through interception, transpiration, evaporation, runoff and drainage 182 (van Keulen 1986; van Oijen et al. 2010b). Runoff is calculated as the amount of water lost through surface 183 flow, which is a function of rainfall interception by vegetation cover (modelled by the system’s leaf area 184 index), soil water holding capacity, and slope. Drainage is calculated as the remaining water not used by 185 evapotranspiration, interception by vegetation or loss through runoff, which is not retained by the soil (van 186 Oijen et al. 2010b). No groundwater influence is assumed. Potential transpiration depends on temperature, 187 solar radiation, vapor pressure and LAI and is calculated with the Penman equation (Penman 1948) 188 according to van Oijen & Leffelaar (2008). Actual transpiration rate decreases below the potential rate when 189 soil water content is below a critical value, which is lower than field capacity by an amount that depends 190 inversely on atmospheric evaporative demand (Driessen 1986; van Oijen et al. 2010b). This critical plant 191 available soil water content additionally depends on the sensitivity of the coffee cultivar to drought. 192 Waterlogging can hamper crop growth when soil water content exceeds a given threshold. Evapo-193 transpiration demand is partly covered by daily rainfall interception through foliage up to a maximum level 194 which is proportional to LAI (van Oijen et al. 2010b; Siles et al. 2010). 195 196 Shade, light transmission, microclimate buffering and water competition 197 Light transmission through shade tree canopy is modelled with Beer’s law using the LAI of the shade tree 198 canopy and an extinction coefficient reflecting different degrees of light transmission and assuming a 199 horizontally homogeneous canopy (Kropff & van Laar 1993, Zuidema et al. 2005). These two variables 200 allow for varying light transmission through shade tree canopy between 0% and 100%. In reality, the shade 201 tree canopy in agroforestry systems is never homogeneous, and includes multiple layers and species, as well 202 as canopy gaps. Charbonnier et al. (2013) showed that the assumption of a homogeneous canopy is 203 nevertheless valid at plot scale, at least under their studied conditions at a humid site in Costa Rica with 204 high planting density dwarf Caturra varieties and low shade tree density with crown projections covering 205 16% of the farm. Although it is difficult to assume their findings are valid under other environmental and 206 shade management conditions, it is an important indication that plot level modelling might be sufficiently 207 approximated with this approach. Therefore, it provides an appropriate reference for comparing the effects 208 9 of climate and shade on coffee. Conceptually, non-shaded and homogeneous shade canopy constitute 209 reference points for the extreme shade canopy types, whereby the LAI and light extinction coefficient 210 represent different shade tree compositions. Changes in coffee leaf functional traits (e.g. SLA and leaf 211 nitrogen concentration) between management systems of different vegetation structures (i.e. various shade 212 tree compositions to no shade) is mainly related to light availability rather than shade tree species 213 composition (Gagliardi et al. 2015). 214 Microclimate modification by shade is reflected as a decrease of daily average temperature proportional to 215 the fraction of radiation intercepted by the shade trees (van Oijen et al. 2010b). Mean daily temperature 216 decreases according to shade level down to a given maximum (e.g. 4 °C; Beer et al. 1998). Water 217 competition starts with rain interception of shade trees which is proportional to LAI up to a specified 218 threshold (Siles et al. 2010). Tree transpiration was modelled as for coffee with the Penman equation 219 (Penman 1948). Shade trees transpire water from the same soil-water layer as used by coffee plants. Hence 220 competition of water occurs when transpiration demands of coffee and shade trees together drive the soil 221 water content below the critical soil water threshold, when actual transpiration rate declines below potential 222 rate. 223 B. Climate and soil data 224 Parameters of the soil-water balance model are retrieved from the gridded functional soil information dataset 225 on root zone plant-available water holding capacity of the Sub-Sahara African Soil (AFSIS; Leenaars et al. 226 2015). Terrain slope is calculated with the terrain function from the “raster” package (Hijmans et al. 2015) 227 based on the SRTM 90-m digital elevation model. Daily weather data is generated from the WorldClim 228 database (Hijmans et al. 2005) using the MarkSim weather generator (Jones & Thornton 2000) on a 30 229 second resolution (approximately 1 km near the equator) for current (interpolation of observed data, 230 representative of 1950-2000) and future climatic conditions (downscaled GCM data using the delta method 231 (Ramirez & Jarvis 2010) and representative of the 2040-2069 time-slice). Since the objective of this study 232 is to present the model and its behavior instead of a climate change impact assessment, we chose only one 233 GCM to exemplify model capability. We focus the study on possible responses of the coffee plant with 234 consideration of its management to a changing climate, rather than assessing uncertainty in climate change 235 predictions. The Hadley Global Environment Model 2 – Earth System (HadGEM2-ES; Jones et al. 2011) 236 from the IPCC 5th assessment report (IPCC 2013) using the representative concentration pathway 6.0 (van 237 Vuuren et al. 2011) was chosen for illustration purposes. HadGEM2-ES has been identified as one of the 238 best performing GCMs (e.g Brands et al. 2013; Perez et al. 2014) and is one of the most commonly used 239 one. Vapor pressure is estimated from temperature, assuming dew point is equal to minimum temperature 240 10 as shown by Wang et al. (2004). Solar radiation is estimated with MarkSim using the model of Donatelli 241 and Campbell (1997, cit. in Jones & Thornton 2000) and adjusted to topography with the ‘slopefactor’ 242 function of the EcoHydRology package of R (Fuka et al. 2014). Because Arabica coffee is mainly cultivated 243 on the slopes of mountain ranges, spatial resolution should not be too low in order to account for the 244 heterogeneity arising through complex topography. To account for the stochasticity of the MarkSim weather 245 generator, 100 replicates of yearly weather data are simulated for each pixel to account for the diversity of 246 possible weather constellations that monthly mean values could be made up of. We calculated four 25-year 247 growth periods per site and then calculated the mean annual yield values of the 100 year simulations. 248 C. Study sites 249 Two study sites of East Africa were chosen with distinctly different climates (Fig 2 and Table S1). An 250 altitudinal transect (1000 – 2200 m.a.s.l.) was selected along the slopes of Mt. Elgon, Uganda, with hot and 251 dry climate at low altitudes, and cool and wet climate at high altitudes. On the other hand, Mount 252 Kilimanjaro, Tanzania, was chosen as a prominent coffee growing region with severe drought issues and a 253 high predicted climate change impact (Craparo et al. 2015). The precipitation values of the WorldClim data 254 for Kilimanjaro above 1500 m.a.s.l. were corrected with the insights gained from weather monitoring along 255 the altitudinal transect as recently published by Appelhans et al. (2016, Figure 4). At the two study sites, the 256 HadGEM2-ES estimates a slightly higher temperature increase and a lower change in annual rainfall 257 compared to four other commonly used GCMs (Table S2). Therefore, the HadGEM2-ES represents a 258 warmer and drier climate with respect to the other evaluated GCMs. The soils on the slopes of Mt. Elgon 259 and Mt. Kilimanjaro are predominantly Nitisols (WRB for soil classification (IUSS Working Group 2015), 260 developed on basaltic outflows. Water holding capacity is highly heterogeneous due to the complex 261 topography and is highest at mid altitude (Figure S1). The selected areas have an extension of 1739 km2 in 262 Mt. Elgon and 1430 km2 in Mt. Kilimanjaro and are important coffee production areas with the majority of 263 households growing coffee. The coffee is often intercropped with banana (Musa spp) and a diversity of 264 shade trees with Cordia Africana, Albizia schimperiana and Ficus mucosa being the most frequent ones 265 (Hemp 2006; van der Wolf et al. 2016). 266 [Place Figure 2 here] 267 D. Experimental design 268 The full continuum of light and water competition was modelled. To calculate optimal shade level per pixel 269 for maximum coffee yield, we varied the shade tree leaf area index between 0 and 3 and the shade tree light 270 extinction coefficient between 0.4 and 0.8, yielding shade levels between 0 and 90%. The continuum of 271 11 water competition was represented by the two extremes of full potential and no water competition; full 272 potential water competition between coffee and shade trees is assumed due to presence of coffee and tree 273 roots in the same single and homogeneous soil layer and with no preferential access of coffee roots to soil 274 water. For the sensitivity and uncertainty analyses on the other hand, we did not use the full continuum of 275 light and water competition due to high computation demand. We compared non shaded systems with 276 systems covered by 50% shade (shade tree leaf area index = 1 and shade tree light extinction coefficient = 277 0.7) assuming full water competition. 278 Climate scenarios consisted of current climatic conditions referring to 1950-2000 according to WorldClim 279 and denoted as year 2000, and future climate representative of the 2040-2069 time-slice (in the following 280 referred to 2050) without elevated atmospheric [CO2] (future A) and with elevated atmospheric [CO2] 281 (future B). Atmospheric [CO2] concentrations for current and future A scenarios were 380 ppm and 478 282 ppm for future B scenario, according to the IPCC representative concentration pathways (RCP) 6.0 (van 283 Vuuren et al. 2011). For the future B scenario, we assumed neither photosynthetic downregulation nor 284 changes in carbon allocation to represent the maximum possible CO2 fertilization effect (CFE). The future 285 A scenario represents the “temperature” effect of climate change, and the difference between future B and 286 future A, represents the uncertainty in CFE. Maximum CFE reflects increased photosynthesis and water use 287 efficiency assuming no phenotypic responses to elevated [CO2]. The uncertainty range, therefore, reflects 288 possible physiological and/or morphological changes as a response to elevated [CO2] not accounted for by 289 the model. For the sensitivity analysis and prior (i.e. uncalibrated) uncertainty assessment, we chose two 290 locations each on Mt. Elgon and Mt. Kilimanjaro, one at low altitude (1097 and 1133 m.a.s.l., respectively) 291 and a second one at high altitude (1863 and 1629 m.a.s.l, respectively). 292 293 E. Evaluation 294 Parameter ranges and assumptions 295 Parameter ranges were selected as marginally uniform distributions based on published studies as presented 296 by van Oijen et al. (2010a) and from more recent literature (see Table S3 and S5). It is important to 297 differentiate between parameter values that can be considered as generally valid for all Arabica coffee 298 species and parameter values that are genotype specific (Boote et al. 2003). Genotype specific parameters 299 include carbon allocation parameters, transpiration coefficient, light extinction coefficient, leaf Rubisco 300 content, time between seedling planting and reproductive phase, specific leaf area, and parameters of plant 301 12 organ senescence. Differences between traditional low yielding and improved high yielding genotypes can 302 be largely explained by differences in carbon allocation patterns, whereby the efficiency of the 303 photosynthetic apparatus remained almost the same (Génard et al. 2008; Ericsson et al. 1996). For example, 304 the difference at juvenile stage between the coffee cultivars Coffea arabica cv. Caturra, Catuai, and Catimor 305 was mainly related to morphological variables (i.e. height, leaf area, internode number, root to shoot ratio), 306 while physiological variables (i.e. leaf net photosynthetic rate, stomatal conductance, intercellular [CO2], 307 and intrinsic water-use efficiency) were not significantly different (Zhang et al. 2017). 308 Sensitivity analysis 309 A global sensitivity analysis was performed using the extended Fourier amplitude sensitivity test (eFAST) 310 of Saltelli et al. 1999. The eFAST estimates the contribution of individual input factors to output variance 311 and evaluates the variability of the output for the entire uncertainty domains of the parameters. Additionally, 312 it addresses higher-order interaction between input factors. The method makes use of a sampling approach 313 following a systematic search trajectory designed to explore the parameter space via a transformation 314 function. A first-order sensitivity and a total-effect sensitivity index are computed from the generated 315 trajectories using a spectral decomposition of the model output variability. The total output variance is 316 decomposed to estimate the influence of individual parameters by its main effects and the interactions with 317 other parameters. This method was shown to be more efficient than Sobol’s method (Sobol’, 1993) which 318 uses a Monte Carlo approach to estimate first and total sensitivity indices (Makowski et al. 2006) and has 319 been shown to be more robust than several other global sensitivity methods (Chan et al. 1997; Saltelli & 320 Bolado 1998). Furthermore, computational demand is significantly lower with eFAST and interaction 321 effects among parameters are computed, which is not the case with Sobol’s algorithm. We used the 322 sensitivity R package by Pujol et al. (2015). Conversion of the sensitivity indices was found for a sample 323 size of factor 1000 resulting in 26 000 model runs with 26 parameters, i.e. 26 000 parameter vectors were 324 sampled and run with the model per climate and management scenario. 325 326 Prior parameter uncertainty analysis 327 As the exact parameter values are unknown, we evaluate how the different shade management and climate 328 scenarios impact on the different parameter sets within the range of parameter uncertainty. This approach is 329 called forward uncertainty analysis (Beven 2008), as the outputs depend entirely on the prior assumptions 330 of the parameters, which are propagated forward through the model predictions. To assess forward 331 13 parameter uncertainty, we use the parameter vectors as sampled with the eFAST method. The uncertainty 332 is exemplified for a hypothetical but realistic (i.e. derived from existing literature) coffee Arabica genotype. 333 Therefore, we set the light extinction coefficient and Rubisco content in leaves to 0.65 (m2 m-2) and 0.54 (g 334 m-2), respectively. The uncertain parameters are given in table S4. The soil parameters representing the 335 fraction of water content at field capacity, air dryness, wilting point, and water saturation have been set to 336 the values of the respective pixels of the AFSIS database (Leenaars et al. 2015). 337 338 5. Results 339 A. Model runs with mean parameter values 340 Mount Elgon: 341 On Mt. Elgon, the high altitude is currently the most suitable area for coffee production with a peak at 1780 342 m.a.s.l. (Figure 3). Due to the optimal mean annual temperatures (19.3 °C), annual precipitation (1654 mm) 343 and high cloud cover and hence low daily mean solar radiation (19.7 MJ m-2 d-1), coffee is not dependent 344 on shade for buffering microclimate. Shade becomes beneficial from 1500 m.a.s.l. on downwards, whereby 345 the benefit increases as altitude decreases. Maximum yield gain with optimal shade management assuming 346 full potential water competition is 11% for current and 13% and 10% for future A and B, respectively. At 347 low altitude, shade up to 50% increases coffee yield, despite the conservative assumption of non-preferential 348 access of coffee roots to soil water and hence full potential water competition between coffee and shade 349 trees. Furthermore, some areas at low altitudes have soils with low water holding capacity resulting in 350 increased competition for water between coffee and shade trees. In the scenario of future climate (2050) as 351 predicted by the HadGEM2-ES without considering elevated atmospheric [CO2], coffee yield is negatively 352 affected up to 2000 m.a.s.l., with a maximum decrease of 44% on poor soils at low altitude (mean = -25%), 353 while at altitudes higher than 2000 m.a.s.l. yield is increased due to predicted higher temperatures. Optimal 354 coffee suitability increases up to 2200 m.a.s.l. equaling an altitudinal shift of 400 m; this drastically reduces 355 the suitable area for coffee cultivation as exemplified by a comparison of maps on current and future A 356 scenarios (Figure 3). The increased yield in the future, limited to the highest altitudes, is additionally 357 attributed to a foreseen precipitation increase [100-300 mm yr-1]. However, this area coincides with the 358 protected Afromontane forest starting at around 2200 m.a.s.l., hence agricultural expansion to these higher 359 altitudes is not permitted. Shade as an adaptation strategy allows to adapt currently non-shaded systems in 360 the future but only if coffee is associated with tree species not competitive but rather complementary in 361 14 terms of water use. Unless water competition is adequately dealt with, the benefit of modified microclimate 362 by shade trees will not be able to sustain the current yield level. Assuming no phenotypic plasticity as a 363 response to elevated [CO2], the CFE cancels the negative effect of high temperature at low and mid altitudes 364 on Mt. Elgon, while at high altitude yield is increased [10-15%]. 365 Mount Kilimanjaro: 366 On the slopes of Mt. Kilimanjaro, the highest yield of rain-fed coffee is between 1600 and 1800 m.a.s.l. 367 (Figure 4). Shade improves coffee yield at lower altitudes (<1600 m.a.s.l.) only if non-water-competitive 368 shade tree species are used. Assuming full potential water competition, shade is beneficial solely up to 1370 369 m.a.s.l., due to the dry climate reducing the net benefit of shade trees. Furthermore, the temperature effect 370 of predicted climate change, according to the HadGEM2-ES without considering elevated [CO2], decreases 371 yield up to 1730 m.a.s.l.. Higher evaporative demand and hence water stress will be a serious problem and 372 the issue of water competition between coffee and shade trees even more critical. Adapting coffee systems 373 to this foreseen climate will require substantial knowledge on shade tree selection and management. 374 Elevated [CO2] might mitigate the negative effects of predicted climate change at low altitude if no 375 physiological and/or morphological acclimation processes take place. At mid altitude, future elevated [CO2] 376 conditions will not be able to maintain the same yield level as under current conditions. Maximum yield 377 gain with optimal shade management assuming full potential water competition is 12% for current and 18% 378 and 17% for future A and B, respectively. 379 [Place Figure 3 here] 380 [Place Figure 4 here] 381 B. Prior parameter uncertainty 382 Prior parameter uncertainty analysis provides insights on the qualitative changes to be expected using 383 different parameter sets when evaluating different shade management and climate scenarios (Figure 5). The 384 mean responses of the impact of different management practices (no shade vs. shaded systems) at current 385 and future (with CFE vs no CFE) climatic conditions over a longer time period, give a clear picture of the 386 climate signal. Incorporating shade trees with 50% canopy cover at low altitude (1092 m.a.s.l) on Mt. Elgon 387 benefits coffee yield on average by 13.5% over a 25 year growth period, despite high parameter uncertainty 388 (Figure 5). This shade level also constitutes the optimal shade level for highest coffee yield (Figure 6). A 389 few parameter sets reveal no change or even a yield decrease under shaded conditions. This occurs under 390 conditions of combined low leaf sink strength, low maximum leaf life-span, and short thermal time 391 15 requirement for bean maturation. Such a parameter constellation is unlikely present in a real genotype, 392 therefore making this scenario rather unrealistic. Yet, to improve our model understanding it is still helpful 393 to identify such conditions. In these parameter constellations, LAI is too small to adapt to the limited light 394 availability under shade, which outweighs the benefits of improved microclimate. Furthermore, the short 395 thermal time requirement for bean maturation results in an increased photo-assimilate demand by berries 396 that cannot be provided by the low leaf area and low light availability. At high altitudes, on the other hand, 397 shade (with 50% canopy cover) is not beneficial for coffee yield compared to unshaded systems, due to the 398 already low temperatures, high humidity associated to high annual rainfall, and low solar radiation due to 399 high cloud cover. None of the evaluated parameter constellations allow for improved yield under shade at 400 high altitude. Unshaded coffee outperforms shaded coffee in terms of coffee yield at current climatic 401 conditions at high altitude (Figure 6). 402 In a future climate as projected by the HadGEM2-ES, coffee yield is hampered by increased temperature 403 and drought stress at low altitude (-18%) if no elevated atmospheric [CO2] concentration is assumed (Figure 404 5). For most parameter constellations, this negative impact on yield can be mitigated if shade is included 405 allowing to maintain current yield in the future. A shade level of 50% continues to provide the highest net 406 benefit resulting in highest yield (Figure 6). The shade effect for different parameter constellations has been 407 described above. In a future scenario with elevated atmospheric [CO2] reaching 478 ppm, the negative 408 impact of increased temperature and drought stress is mitigated (Figure 5), allowing to maintain current 409 yield in the future even without shade. This is assuming neither a downregulation in photosynthesis nor 410 changes in carbon allocation as a response to elevated [CO2]. As shown in Figure 5 and 6, optimizing shade 411 management under these conditions results in increased yield compared to current unshaded systems (18%). 412 At high altitude, by contrast, increased temperatures will benefit coffee yield in both scenarios (no/with 413 elevated [CO2]). Shade with 50% canopy cover at these altitudes will not be able to outperform unshaded 414 systems either today or in the future, if the sole objective is to maximize coffee yield (Figure 5). But shade 415 up to 18% does not lead to a yield reduction (Figure 6). 416 On Mt. Kilimanjaro, Tanzania, shade also improves coffee yield at low altitude (24%) but not at high 417 altitude. Compared to Mt. Elgon, future predicted climate change has a more negative effect in both 418 scenarios, without (-32%) and with elevated [CO2] (-18%) in unshaded conditions. But yield can be 419 improved by shading to (-2%) and (16%) in scenarios without and with elevated [CO2], respectively. 420 Although shading coffee at high altitude reduces potential yield at current climatic conditions, it buffers 421 coffee systems against future climate scenarios and helps maintain yield (no elevated [CO2]) or even 422 increases yield (with elevated [CO2]). 423 16 [Place Figure 5 here] 424 [Place Figure 6 here] 425 C. Parameter sensitivity 426 The most sensitive parameters are those related to soil water availability (i.e. amount of water available for 427 coffee and trees between field capacity and wilting point), light interception (light extinction coefficient), 428 photosynthesis (Rubisco content in leaves), and carbon allocation (i.e. sink strength for leaves and storage 429 organs) (Figure 7). These parameters need to be well known or appropriately estimated to reduce 430 uncertainty. The remaining parameters do not significantly influence output variance and hence can be set 431 as constant values allowing for model simplification. The difference in parameter sensitivity to changes in 432 environment (Uganda vs Tanzania / low vs high altitude) is less than to shade level. 433 Shade reduces the energy input reaching the soil and the coffee plants, and therefore reduces water use by 434 coffee transpiration and to a lesser extent soil evaporation. Yet, adding transpiration of the shade trees, hence 435 assuming full potential water competition all year around, leads to high sensitivity of soil parameters. The 436 soil water budget of coffee plants is less important under shaded conditions when there is no water 437 competition. The strong influence of soil hydraulic parameters (parameters related to field capacity and 438 wilting point) highlights the need for using high quality soil data as inputs. As water becomes less limiting, 439 parameters related to light interception, canopy photosynthesis, and carbon allocation become more 440 important. The C sink strength of roots gains importance in water limited conditions when soil hydraulic 441 parameters are excluded of the sensitivity analysis (data not shown), while sink strengths of leaves and 442 berries are relatively more important in light limited conditions. 443 As shown in Figure 5, the general pattern of model parameter sensitivity between Mt. Elgon and Mt. 444 Kilimanjaro is consistent, despite the differences in climatic conditions. The main difference is the higher 445 sensitivity of parameters related to thermal time, namely base temperature for maturation (TMATB) and 446 thermal time to maturation (TMATT) particularly in unshaded systems located at low altitude. 447 The soil water balance parameters are provided as input by the AFSIS database, whereby the uncertainty of 448 this database is not considered in this study. Yet, as the model is very sensitive to soil hydraulic parameters, 449 accurate estimates are required to reduce uncertainty. For calibration purposes, it is important to know which 450 parameters are critical when removing them from the analysis. When excluding the soil water balance 451 parameters, the sensitive parameters (>0.1) are (in descending order): carbon sink strength of leaves 452 17 (SINKL), Rubisco content of upper leaves (RUBISC), sink strength for storage organs (SINKPMAX), light 453 extinction coefficient (KEXT), and carbon sink strength of roots (SINKR). 454 455 [Place figure 7 here] 456 18 6. Discussion 457 We presented a model allowing to taking into account spatially-explicit soil and climate characteristics to 458 explore management options for coffee systems via shade tree association. One GCM was used, as the main 459 purpose of this study is the presentation of the model features rather than a full climate change impact 460 assessment. Nonetheless, consistent with earlier studies, we found negative impacts of increased 461 temperature at low altitude, while high altitude areas should gain in terms of yield potential (Bunn et al. 462 2015a, Magrach & Ghazoul 2015). Our model suggests that the CFE potentially could mitigate the negative 463 effects of increased temperature to a large extent, hence suggesting that previous studies might have 464 overestimated the negative impact of climate change. Additionally, one of the most salient features of the 465 present model is to calculate the optimal shade level along environmental gradients now and in the future 466 according to different climate scenarios. Below we discuss differences among coffee genotypes, the effect 467 of shade and [CO2] fertilization, the importance of soil water availability, the interaction between 468 temperature and [CO2] effect, as well as implications for smallholders in Mt. Elgon and Mt. Kilimanjaro, 469 and ultimately model improvements. 470 7.1 Genotypic variation 471 Data availability for calibration purposes is a major constraint. Explicitly analyzing the uncertainty range in 472 parameters is therefore of great importance (Beven 2008). Due to the lack of certainty concerning which 473 parameter sets best simulate the actual genotypes and planting densities currently cultivated at the sites of 474 interest, we did our simulations based on the entire range of parameter values by sampling the plausible 475 parameter space. Despite the uncertainty in the absolute change induced by the assessed scenarios compared 476 to current non-shaded systems (Figure 5 and 6), the model allows for identifying the direction of change in 477 coffee yield (Reyer et al. 2016). It further exemplifies the potential of improved genotypes in the context of 478 climate change adaptation (Bertrand et al. 2016). The maps in Figures 3 & 4 represent only one general 479 Arabica genotype at a fixed planting density resulting in relatively low observed water-limited yield up to 480 2.5 t/ha. The use of modern high yielding dwarf varieties (e.g. Catimor derived varieties) planted at high 481 density (4000-5000 plants/ha) as commonly found on coffee plantations in Brazil and Colombia, for 482 example, can reach yield levels over 3 t/ha (van der Vossen et al. 2005; DaMatta et al. 2007). Different 483 coffee varieties exhibit different yield potentials and responses to environmental stresses. Differences in 484 morphological traits have been identified to be the main determinants of yield differences and drought 485 tolerance among Arabica cultivars (Tausend et al. 2000; Zhang et al. 2017). Open questions remain 486 regarding genotypic differences in temperature tolerance and response to elevated [CO2]. Furthermore, 487 hybrid varieties between Arabica, Robusta and hybrids of Timor have been developed mainly for pest and/or 488 19 disease resistance (Bertrand et al. 2016; Herrera & Lambot 2017) that could potentially also benefit of 489 Robusta’s tolerance to higher temperature. To analyze this, data of different varieties on the physiological 490 (particularly stomatal conductance) and morphological characteristics are required. The sensitivity analysis 491 (Figure 7) highlights that plant architecture (light extinction coefficient) plays an important role in plant 492 performance under shaded or non-shaded conditions. Traditional varieties which have a less dense canopy 493 (Unigarro et al. 2016) are usually better adapted to shade, while modern dwarf varieties with high self-494 shading often grow well without shade (Montagnon et al. 2012). Genotypic differences regarding responses 495 to light availability are of great interest for using agroforestry systems as adaptation strategy (Bertrand et 496 al. 2016). 497 7.2 Effect of shade 498 The model allows for estimating the optimal shade level along environmental gradients now and in changing 499 future climates. In areas with optimal water supply, elevated temperatures and high solar radiation, a range 500 of shade tree species can be used with beneficial effects for coffee yield. In these areas, trees with high 501 timber value or fast-growing ones can substantially improve farmers’ livelihoods (Schaller et al. 2003; Vaast 502 et al. 2008; Vaast et al. 2015). Where water is limiting, shade trees need to be selected with high root niche 503 differentiation (Cannavo et al 2011; Padovan et al. 2015) and non-competing phenology. The latter also 504 provides options to optimize light availability (Beer et al. 1998). Knowledge of appropriate selection of 505 shade tree species (van der Wolf et al. 2016) and their management related to timing and amount of pruning 506 is thus required (Beer et al. 1998). It is important to further consider the economic impact of agroforestry 507 systems. At locations where shade does not decrease coffee yield, or when such decreases are relatively 508 small, increased farm profitability and revenue diversification can be achieved by farmers due to the co-509 production of other agroforestry products, such as fruits, firewood, and timber (Rice 2011; Vaast et al. 510 2015). 511 We showed that the model responds well to decreasing light availability, by increasing the sensitivity of 512 parameters relevant to light interception (Figure 7). Soil hydraulic parameters had a strong effect on yield, 513 which emphasizes the importance of sound soil input data. Different soil types will therefore have an 514 important influence on yield output, which need to be accounted for in addition to changes in climate 515 variables. We suggest the use of recently established spatial databases on root zone plant-available water 516 holding capacity (Leenaars et al. 2015), but uncertainties in these data should be taken into account in future 517 studies. These soil types in combination with precipitation regime and root distribution of shade tree species 518 will largely determine the risk for water competition or potential for complementarity between coffee and 519 intercrops/shade trees (van Kanten & Vaast 2006; Cannavo et al. 2011; Padovan et al. 2015). 520 20 In this study, light and water competitions were estimated conservatively, assuming homogeneous shade 521 cover and full potential water competition within a single soil layer. Nevertheless, the selected approach 522 allows the calculation of the biophysical determinants of successful agroforestry systems (Sanchez 1995) at 523 the critical endpoints of the continuum of different degrees of light and water competition. Shade tree species 524 differ substantially in the way they intercept light and use water based on their canopy, root architecture and 525 phenology (e.g. deciduous vs. evergreen trees) (Luedeling et al. 2016). On the other hand, the model does 526 not account for the effect of shade and climate change on pests and diseases. Shade can either have beneficial 527 or negative effects on pests and diseases depending on environmental context (Jonsson et al. 2015; Liebig 528 et al. 2016) which additionally requires careful consideration for shade tree management (Staver et al. 2001). 529 7.3 Effect of [CO2] fertilization 530 We identified that an increase in [CO2] might substantially mitigate some of the negative impacts caused by 531 climate change, although the magnitude of that mitigation effect is not clear due to possible responses in 532 phenotypic plasticity (Aspinwall et al. 2015). To account for this uncertainty in CFE, we modelled zero 533 effect (modelled by omitting an increase in [CO2] in the future) and the effect resulting when no phenotypic 534 plasticity is assumed as a response to elevated [CO2] from 380 ppm to 478 ppm. The model indicates an 535 averaged maximum effect on coffee yield between 13.5 – 21 % (Figure 5), with higher effect at the more 536 humid Mt. Elgon site, compared to the drier Kilimanjaro site. Although water-use efficiency generally 537 increases due to elevated [CO2] at an instantaneous time-scale, the duration of stomatal opening at longer 538 time scales is increased at more humid sites which can lead to higher overall benefit of [CO2]. Our modelled 539 values agree well with the average response of 13% found in free-air [CO2] enrichment experiments (FACE) 540 across a range of C3 crops (Long et al. 2006) and a recent one conducted on coffee in Brazil, resulting in a 541 yield increase of 12.0 – 14.6%, depending on genotype, due to increasing atmospheric [CO2] concentrations 542 from 390 ppm to 550 ppm over a two-year period without increasing ambient temperature (Ghini et al. 2015; 543 DaMatta et al. 2016). This yield increase was even larger (approx. 40%) in the third year leading to a mean 544 increase of 28% (Raquel Ghini, personal communication). This high CFE found in this FACE experiment 545 is due to the absence of downregulation in photosynthetic capacity or acclimation of stomatal conductance. 546 There was no decrease in leaf nitrogen content and no alteration in leaf Rubisco content. Differences 547 between the two evaluated genotypes in their responses to elevated [CO2] were found, suggesting some 548 uncertainties on how other common Arabica varieties might respond. Anyway, this general lack of 549 phenotypic plasticity in response to elevated [CO2] is exceptional when compared to many other C3 plants 550 (Ainsworth & Long 2005) and suggests that Arabica coffee benefits comparatively well from CFE. This 551 appears to be due to the fact that coffee photosynthesis is highly limited by diffusional rather than 552 21 biochemical limitations (Martins et al. 2014b). Research on other C3 plants with similar diffusional 553 limitations have also indicated above-average benefit of CFE (Niinemets et al. 2011; Flexas et al. 2014). 554 Additionally, as a perennial woody species it has a larger carbon sink capacity compared to annual crops 555 (Ainsworth & Long 2005). Higher water use efficiency is thereby achieved by increasing net photosynthesis 556 at maintained stomatal and mesophyll conductance (DaMatta et al. 2016). Maintaining stomatal 557 conductance is beneficial as a reduction might result in an increased leaf temperature via its effects on leaf 558 energy balance (Bernacchi et al. 2007), which could exacerbate the warming induced by climate change. 559 7.4 Interaction between temperature and [CO2] effect 560 To what extent elevated [CO2] might mitigate increased temperature is unclear and might be different for 561 different genotypes and dependent on the accompanying change in air humidity and soil water availability. 562 Growth chamber studies have found increased tolerance of coffee to supra-optimal temperature at elevated 563 [CO2], mainly due to increased investment into protective measures (Martins et al. 2016; Rodrigues et al. 564 2016). The degree of this increased tolerance will depend on the temperature and precipitation induced 565 changes in soil moisture and VPD and the vigor of the coffee plant. Therefore, hot and humid areas are 566 expected to benefit more from the interaction between elevated [CO2] and supra-optimal temperature, 567 compared to areas that will become hot and dry. This explains why our model implies that elevated [CO2] 568 does not allow to maintain current yield in Kilimanjaro, while at Mt. Elgon it might cancel out the negative 569 effect of increased mean temperature. Irrigation practices are widespread in the Kilimanjaro area (Hemp 570 2006) and the last two decades (1990 - 2010) have been identified as quite dry compared to average climate, 571 with reductions in precipitations during the main growing season (Otte et al. 2016). Further drying is 572 projected by the HadGEM2-ES due to decreased rainfall in combination with higher evaporative demand 573 as a result of higher temperatures, particularly during bean filling. This likely results in increased C demand 574 for maintenance and respiration as well as for bean filling, with simultaneous decrease in photosynthesis 575 due to increased VPD and drought stress. Importantly, we have to be aware that there remain high 576 uncertainties on how the interactions between warming and elevated [CO2] will impact flowering, bean 577 maturation and cup quality. For example, there are indications that flowering might be severely reduced 578 under increased temperature (Drinnan & Menzel 1995; Craparo et al. 2015), which is currently not included 579 in the model due to limited process knowledge. Pest and disease pressure will additionally change and likely 580 continue to constitute a primary production constraint to coffee cultivation, for which we have only limited 581 information at present (Jaramillo et al. 2011; Ghini et al. 2015; Jonsson et al. 2015). 582 7.5 What does the [CO2] effect mean for smallholder farmers in Mt. Elgon and Mt. Kilimanjaro? 583 22 The vast majority of farmers on Mt. Elgon and Mt. Kilimanjaro are smallholders, who Smallholder farmers 584 in Mt. Elgon and Kilimanjaro constitute the vast majority and are faced withface several constraints to 585 access adequate nutrient and pest and disease management options, leading to coffee systems with very low 586 yield (Wang et al. 2015; Craparo et al. 2015; Liebig et al. 2016). Therefore, it is unlikely that they will 587 benefit of the CFE, unless appropriate actions to promote climate-smart agricultural management are taken 588 (Vaast et al. 2016). The challenge of the CFE is that it favors highly intensified systems, thereby further 589 increasing their competitive advantage compared to extensive ones of smallholders, and hence likely making 590 the formers more vulnerable (Morton 2007). The opportunity of CFE, however, is that improving 591 smallholders’ farming conditions might allow more of them to continue to cultivate cultivating coffee in the 592 future than previously estimated. 593 7.6 Model improvements 594 Several aspects need to be highlighted for future improvements regarding the present coffee model. Firstly, 595 further insights are required on how coffee light use efficiency is affected by different light availabilities at 596 different time scales. This allows for clarifying to what extent the mechanistic light use efficiency approach 597 (van Oijen et al. 2004) is an appropriate summary model of changes in microclimate due to shade (i.e. 598 changes in minimum and maximum temperatures, decreased VPD and increased diffuse light) and changes 599 in phenotypic plasticity of the coffee plant. Secondly, we still lack a sufficient mechanistic understanding 600 of flowering induction and associated asynchronous flowering events. A first attempt has been done by 601 Rodriguez et al. (2011) by using demographic population models which should be further evaluated. To 602 improve our understanding, it will require long term data sets of flowering events combined with weather 603 and soil data of different environmental contexts. The same is required to better understand factors leading 604 to flower abortion. Thirdly, a more detailed modelling routine might be required for branch dynamics that 605 adequately represent competition between fruit load and increment in branch and new branch development. 606 Depending on the new branches formed and extension of existing branches, the nodes and potential 607 flowering should be calculated. Finally, improvement is also required in representing the impact of extreme 608 events, such as temperature and rainfall, on coffee productivity. As more data become available, more 609 complex process representation will be possible, either through direct parameterization or inverse 610 modelling. Particularly, this is needed for extending the soil water balance to multilayered soil 611 compartments enabling root niche differentiation between the coffee plant and shade trees as recently 612 exemplified in Nicaragua (Padovan et al. 2015). In order to better account for differences in canopy structure 613 of the shade trees, summary models as presented by Duursma & Mäkelä (2007) and Forrester (2014) could 614 23 be implemented, with intermediary complexity between our present approach and the MAESTRA approach 615 as used by Charbonnier et al. (2013, 2017). 616 These improvements strongly depend on data availability. The collection of functional trait data for different 617 genotypes in different environmental contexts will allow to analyze genotypic and phenotypic plasticity and 618 significantly contributes to the required data for modelling endeavors. This requires the consolidation of 619 available functional trait data for coffee (particularly flowering phenology) and shade trees within the plant 620 trait database (Kattge et al. 2011; Martin & Isaac 2015) but also on coffee yield, management practices and 621 local knowledge (Ordonez et al. 2014; van der Wolf et al. 2016). Consolidation of such a database would 622 improve either direct parameterization or calibration through inverse modelling, for example by using a 623 Bayesian approach, thereby allowing to estimate parameter values in different climates where weather, soil 624 and production data are available (van Oijen et al. 2005). 625 626 24 7. Conclusions 627 We presented a mechanistic model that processes spatially-explicit soil and climate data to guide coffee 628 cultivation and shade tree management for sustainable coffee production systems. We applied the model to 629 two case studies, namely the slopes of Mt. Elgon, Uganda, and Mt. Kilimanjaro, Tanzania. According to the 630 GCM used (HadGEM2-ES), Mt. Kilimanjaro will be severely affected by a drying climate in the future, 631 while Mt. Elgon is impacted to a lesser extent. The present modelling results confirm earlier studies on 632 coffee suitability changes, namely that low-altitude areas will be more strongly affected by climate change. 633 However, previous studies did not include neither CFE nor shade management and consequently seem to 634 have overestimated the negative effects of mean changes in temperature and shifts in precipitation regimes; 635 i.e. yield improvements due to elevated [CO2] seem to largely mitigate the negative impacts of elevated 636 temperature. Shade trees can play a strong beneficial role at the low-altitude areas that have more marginal 637 climates and will suffer most from climate change. Higher altitude areas (>1500m) may see little to no direct 638 benefit from shade trees on coffee yield. The model outputs can guide initiatives related to trees in 639 agricultural landscapes, such as biodiversity conservation and/or carbon sequestration, and in minimizing 640 potential trade-offs with farm productivity. This will aid in decision making towards improvement of current 641 coffee landscapes and their adaptation to climatic change. In integrating the current understanding of the 642 coffee system, we also identify knowledge gaps related to flowering and bean maturation and hence cup 643 quality in response to climate change with particular focus on climate extremes which require further study. 644 The positive effect of CFE will particularly benefit intensified systems, hence significant improvements in 645 agronomic management of smallholder systems will be required. Therefore, sustainable intensification 646 remains an important aspect of climate change adaptation. Although we showed that shade management 647 consists of an important adaptation strategy, increased knowledge on adequate selection and management 648 of shade tree species will be required to maximize complementarity in the use of resources, particularly 649 water. This will also require more insights into the impact of shade trees on nutrient complementarity and 650 pest and disease pressure for coffee under various climate change scenarios. 651 652 8. Acknowledgements 653 This research was funded by the Federal Ministry for Economic Cooperation and Development (BMZ) the 654 Professorship of Ecosystem Management, ETH Zurich and the Research Program on Forest, Trees and 655 Agriculture (FTA). It was implemented as part of the CGIAR Research Program on Climate Change, 656 Agriculture and Food Security (CCAFS), which is carried out with support from CGIAR Fund Donors and 657 25 through bilateral funding agreements. For details please visit https://ccafs.cgiar.org/donors. The views 658 expressed in this document cannot be taken to reflect the official opinions of these organizations. We would 659 like to thank Bruno Rapidel and Oriana Ovalle-Rivera for sharing their insights on the CAF2014 model and 660 Beatriz Rodriguez for helping in preparing the climate data with MarkSim. Finally, we thank two 661 anonymous reviewers for their valuable comments. 662 26 9. References 663 Ainsworth E. & Long S. (2005) What have we learned from 15 years of free-air CO2 enrichment (FACE)? 664 A meta-analytic review of the responses of photosynthesis, canopy properties and plant production to 665 rising CO2. New Phytologist 165:351-372. 666 Appelhans T., Mwangomo E., Otte I., Detsch F., Nauss T. & Hemp A. (2016) Eco-meteorological 667 characteristics of the sourthern slopes of Kilimanjaro, Tanzania. International Journal of Climatology 668 36:3245-3258. 669 Aspinwall M., Loik M., de Dios V., Tjoelker M., Payton P. & Tissue D. (2015) Utilizing intraspecific 670 variation in phenotypic plasticity to bolster agricultural and forest productivity under climate change. 671 Plant, Cell and Environment 38:1752-1764. 672 Beer J., Muschler R., Kass D. & Somarriba E. (1998) Shade management in coffee and cacao plantations. 673 Agroforestry Systems 38:139-164. 674 Bernacchi C., Kimball B., Quarles D., Long D. & Ort D. (2007) Decreases in stomatal conductance of 675 soybean under open-air elevation of [CO2] are closely coupled with decreases in ecosystem 676 evapotranspiration. Plant Physiology 134-144. 677 Bertrand B., Marraccini P., Villain L., Breitler J.C. & Etienne H. (2016) Healthy tropical plants to mitigate 678 the impact of climate change – as exemplified in coffee. In: Torquebiau E. (ed.) Climate Change and 679 Agriculture Worldwide, chapter 7. Springer Netherlands. P. 83-95. 680 Beven K. (2008) Environmental Modelling: An uncertain future? Routledge: London. 681 Beven K. & Freer J. (2001) Equifinality, data assimilation, and uncertainty estimation in mechanistic 682 modelling of complex environmental systems using the GLUE methodology. Journal of Hydrology 683 249:11-29. 684 Boote K., Jones J., Batchelor W., Nafziger E. & Myers O. (2003) Genetic coefficients in the CROPGRO-685 soybean model: Links to field performance and genomics. Agronomy Journal 95:32-51. 686 Boote K., Jones J., Hoogenboom G. & White J. (2010) The role of crop systems simulation in agriculture 687 and environment. Int. J. Agric. Environ. Inf. Syst. 1:41-54. 688 27 Bote A. & Jan V. (2016) Branch growth dynamics, photosynthesis, yield and bean size distribution in 689 response to fruit load manipulation in coffee trees. Trees 30:1275-1285. 690 Bote A., Ayalew B., Anten N. & Vos J. (2016a) Analysis of coffee (Coffea arabica L.) performance in 691 relation to radiation levels and rates of nitrogen supply I. Vegetative growth, production and distribution 692 of biomass and radiation use efficiency. PhD thesis, chapter 2, Wageningen University, The Netherlands. 693 Brouwer R. (1983) Functional equilibrium: sense or nonsense? Netherlands Journal of Agricultural Sciences 694 31:335-348. 695 Bunn C., Läderach P., Ovalle Rivera O. & Kirschke D. (2015a) A bitter cup: climate change profile of 696 global production of Arabica and Robusta coffee. Climate Change 129:89-101. 697 Bunn C., Läderach P., Pérez Jimenez J.G., Montagnon C. & Schilling T. (2015b) Multiclass classification 698 of agro-ecological zones for Arabica coffee: An improved understanding of the impacts of climate 699 change. PLoS ONE 10(10):e0140490. Doi:10.1371/journal.pone.0140490. 700 Cannavo P., Sansoulet J., Harmand J.M., Siles Gutierrez P., Dreyer E. & Vaast P. (2011) Agroforestry 701 associating coffee and Inga densiflora results in complementarity for water uptake and decreases deep 702 drainage in Costa Rica. Agriculture, Ecosystems and Environment, 140 (1-2): p. 1-13. 703 http://dx.doi.org/10.1016/j.agee.2010.11.005 704 Cannell M. (1985) Physiology of the coffee crop. In: Clifford M. & Willson K. (eds.) Coffee: Botany, 705 biochemistry and production of beans and beverage. Springer, Boston, USA, p. 108-134. 706 Carr M. (2012) Advances in irrigation agronomy. Cambridge University Press, UK. Chapter 5, coffee; p. 707 105-144. 708 Cavatte P., Oliveira A., Morais L., Martins S., Sanglard L. & DaMatta F. (2012) Could shading reduce the 709 negative impacts of drought on coffee? A morphophysiological analysis. Physiologia Plantarum 710 144:111-122. 711 Chan K., Saltelli A. & Tarantola S. (1997) Sensitivity analysis of model output: variance-based methods 712 make the difference. In: Proceedings of the 1997 Winter Simulation Conference, Atlanta, Georgia. 713 28 Charbonnier F., le Maire G., Dreyer E., Casanoves F., Christina M., Dauzat J., Eitel J., Vaast P., Vierling 714 L. & Roupsard O. (2013) Competition for light in heterogeneous canopies : Application of MAESTRA 715 to a coffee (Coffea arabica L.) agroforestry system. Agricultural and Forest Meteorology 181 :152-169. 716 Charbonnier F., Roupsard O., le Maire G. et al. (2017) Increased light-use efficiency sustains net primary 717 productivity of shaded coffee plants in agroforestry system. Plant, Cell and Environment, 718 doi:10.1111/pce.12964. 719 Charles-Edwards D. (1982) Physiological determinants of crop growth. Academic Press, Sydney. 720 Chaves A., Ten-Caten A., Pinheiro H., Ribeiro A. & DaMatta F. (2008) Seasonal changes in photoprotective 721 mechanisms of leaves from shaded and unshaded field-grown coffee (Coffea arabica L.) trees. Trees 722 22:351-361. 723 Craparo A., van Asten P., Läderach P., Jassogne L. & Grab S. (2015) Coffea arabica yields decline in 724 Tanzania due to climate change: Global implications. Agricultural and Forest Meteorology 207:1-10. 725 DaMatta F., Ronchi C., Maestri M. & Barros R. (2007) Ecophysiology of coffee growth and production. 726 Brazilian Journal of Plant Physiology 19:485-510. 727 DaMatta F., Godoy A., Menezes-Silva P., Martins S. Sanglard L., Morais L., Torre-Neto A. & Ghini R. 728 (2016) Sustained enhancement of photosynthesis in coffee trees grown under free-air CO2 enrichment 729 conditions: disentangling the contributions of stomatal, mesophyll, and biochemical limitations. Journal 730 of Experimental Botany 67(1):341-352. 731 Davis A., Gole T., Baena S. & Moat J. (2012) The impact of climate change on indigenous Arabica coffee 732 (Coffea arabica): Predicting future trends and identifying priorities. PLoS ONE 7(11): e47981. 733 Doi:10.1371/journal.pone.0047981. 734 De Kauwe M., Medlyn B., Zaehle S. et al. (2014) Where does the carbon go? A model-data intercomparison 735 of vegetation carbon allocation and turnover processes at two temperate forest free-air CO2 enrichment 736 sites. New Phytologist 203:883-899. 737 Driessen P. (1986) The water balance of the soil. In: van Keulen H., & Wolf J. (eds) Modelling of 738 agricultural production: weather, soils and crops. PUDOC, Wagentingen, pp. 76-116. 739 29 Duursma R. & Mäkelä A. (2007) Summary models for light interception and light-use efficiency of non-740 homogeneous canopies. Tree Physiology 27:859-870. 741 Ericsson T., Rytter L. & Vapaavuori E. (1996) Physiology of carbon allocation in trees. Biomass Bioenergy 742 11:115-127. 743 FAO (2015) FAO statistical pocketbook: Coffee. Food and Agricultural Organization of the United Nations, 744 Rome. 745 Farquhar G., von Caemmerer S. & Berry J. (1980) A biochemical model of photosynthetic CO2 assimilation 746 in leaves of C3 species. Planta 149:78-90. 747 Flexas J., Carriqui M., Coopman R., Gago J., Galmés J., Martorell S., Morales F. & Diaz-Espejo A. (2014) 748 Stomatal and mesophyll conductances to CO2 in different plant groups : underrated factors for predicting 749 leaf photosynthesis responses to climate change ? Plant Science 226:41-48. 750 Forrester D. (2014) A stand-level light interception model for horizontally and vertically heterogeneous 751 canopies. Ecological Modelling 276:14-22. 752 Franck N. & Vaast P. (2009) Limitation of coffee leaf photosynthesis by stomatal conductance and light 753 availability under different shade levels. Trees 23:761-769. 754 Fridell G. (2014) Coffee. Polity Press, Cambridge, UK. 755 Fuka D., Walter M., Archibald J., Steenhuis T. & Easton Z. (2014) EcoHydRology: A community modeling 756 foundation for Eco-Hydrology. R package version 0.4.12. http://CRAN.R-757 project.org/package=EcoHydRology. 758 Génard M., Dauzat J., Franck N., Lescourret F., Moitrier N., Vaast P. & Vercambre G. (2008) Carbon 759 allocation in fruit trees : from theory to modelling. Trees 22:269-282. 760 Ghini R., Torre-Neto A., Dentzien A., Guerreiro-Filho O., Iost R., Patricio F., Prado J., Thomaziello R., 761 Bettiol W. & DaMatta F. (2015) Coffee growth, pest and yield responses to free-air CO2 enrichment. 762 Climatic Change 132:307-320. 763 30 Gifford R. (1995) Whole plant respiration and photosynthesis of wheat under increased CO2 concentration 764 and temperature: long-term vs. short-term distinctions for modelling. Global Change Biology 1:385-396. 765 Gifford R. (2003) Plant respiration in productivity models: conceptualisation, representation and issues for 766 global terrestrial carbon-cycle research. Functional Plant Biology 30: 171-186. 767 Harfoot M., Newbold T., Tittensor D., Emmott S., Hutton J., Lyutsarev V., Smith M., Scharlemann J. & 768 Purves D. (2014) Emergent global patterns of ecosystem structure and function from a mechanistic 769 general ecosystem model. PLoS Biol 12(4): e1001841. Doi:10.1371/journal.pbio.1001841. 770 Hemp A. (2006) The banana forests of Kilimanjaro: biodiversity and conservation of the Chagga 771 homegardens. Biodiversity and Conservation 15:1193-1217. 772 Herrera J.C. & Lambot C. (2017) The coffee tree – genetic diversity and origin. In: Folmer B. (ed.) The 773 craft and science of coffee. Elsevier. Chapter 1, pp. 1-16. 774 Hijmans R., Cameron S., Parra J., Jones P. & Jarvis A. (2005) Very high resolution interpolated climate 775 surfaces for global land areas. International Jounal of Climatology. 25:1965-1978. 776 Hijmans R. (2015) raster: Geographic Data Analysis and Modeling. R package version 2.3-40. 777 http://CRAN.R-project.org/package=raster. 778 IPCC (2013) Climate change 2013: The physical science basis. Contribution of working group I to the fifth 779 assessment report of the Intergovernmental Panel on Climate Change [Stocker, T.F., D. Qin, G.-K. 780 Plattner, M. Tignor, S.K. Allen, J. Boshung, A. Nauels, Y. Xia, V. Bex and P.M. Midgley (eds.)]. 781 Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1535 pp. 782 IUSS Working Group (2015) World reference base for soil resources 2014. World Soil Resources Reports 783 no. 106. FAO, Rome, Italy. 784 Jaramillo J., Muchugu E., Vega F., Davis A., Borgemeister C. & Chabi-Olaye A. (2011) Some like it hot: 785 the influence and implications of climate change on coffee berry borer (Hypothenemus hampei) and 786 coffee production in East Africa. PLoS ONE 6(9): e24528. Doi:10.1371/journal.pone.0024528. 787 Jarvis A., Reuter H., Nelson A., Guevara E. (2008) Hole-filled SRTM for the globe Version 4, available 788 from the CGIAR-CSI SRTM 90m Database (http://srtm.csi.cgiar.org). 789 31 Jones P. & Thornton P. (2000) MarkSim: Software to generate daily weather data for Latin America and 790 Africa. Agronomy Journal 92:445-453. 791 Jones C.D., Hughes K., Bellouin N. et al. (2011) The HadGEM2-ES implementation of CMIP5 centennial 792 simulations. Geosci. Model. Dev. 4:543-570. 793 Jonsson M., Raphael I., Ekbom B., Kyamanywa S. & Karungi J. (2015) Contrasting effects of shade level 794 and altitude on two important coffee pests. J Pest Sci 88:281-287. 795 Kattge J., Díaz S., Lavorel S. et al. (2011) TRY – a global database of plant traits. Global Change Biology 796 17:2905-2935. 797 Kropff M.J. & van Laar H.H. (1993) Modelling crop-weed interactions. CAB International, Wallingford. 798 Lane A. & Jarvis A. (2007) Changes in climate will modify the geography of crop suitability: agricultural 799 biodiversity can help with adaptation. Journal of the Semi-Arid Tropics 4(1):1-12. 800 Leenaars J.G.B., T. Hengl, M. Ruiperez Gonzalez, J. Mendes de Jesus, G.M.B. Heuvelink, J. Wolf, L. van 801 Bussel, L. Claessens, K.G., H. Yang & Cassman, K. (2015) Root Zone Plant-Available Water Holding 802 Capacity of the Sub-Saharan Africa soil, version 1.0. Gridded functional soil information (dataset RZ-803 PAWHC SSA v. 1.0). ISRIC Report 2015/02. Collaboration project of Africa Soil Information Service 804 (AfSIS) and Global Yield Gap and Water Productivity Atlas (GYGA). ISRIC - World Soil Information, 805 Wageningen, the Netherlands. 806 Liebig T., Jassogne L., Rahn E., Läderach P., Poehling H.M., Kucel P., van Asten P. & Avelino J. (2016) 807 Towards a collaborative research: A case study on linking science to farmers’ perceptions and knowledge 808 on Arabica coffee pests and diseases and its management. PLoS ONE 11(8): e0159392. 809 Doi:10.1371/journal.pone.0159392. 810 Long S., Ainsworth E., Leaky A., Nösberger J. & Ort D. (2006) Food for thought: lower-than-expected crop 811 yield stimulation with rising CO2 concentrations. Science 312:1918-1921. 812 Luedeling E., Kindt R., Huth N. & Koenig K. (2014) Agroforestry systems in a changing climate – 813 challenges in projecting future performance. Current Opinion in Environmental Sustainability 6:1-7. 814 32 Luedeling E., Smethurst P., Baudron F., Bayala J., Huth N., van Noordwijk M., Ong C., Mulia R., Lusiana 815 B., Muthuri C. & Sinclair F. (2016) Field-scale modeling of tree-crop interactions: Challenges and 816 development needs. Agricultural Systems 142:51-69. 817 Magrach A. & Ghazoul J. (2015) Climate and pest-driven geographic shifts in global coffee production: 818 implications for forest cover, biodiversity and carbon storage. PLoS ONE 10(7): e0133071. 819 Doi:10.1371/journal.pone.0133071. 820 Makowski D., Naud C., Jeuffroy M., Barbottin A. & Monod H. (2006) Global sensitivity analysis for 821 calculating the contribution of genetic parameters to the variance of crop model prediction. Reliability 822 Engineering & System Safety 91:1142-1147. 823 Martin A. & Isaac M. (2015) Plant functional traits in agroecosystems: a blueprint for research. Journal of 824 Applied Ecology 52:1425-1435. 825 Martins S., Detmann K., dos Reis J., Pereira L., Sanglard L., Rogalski M. & DaMatta (2013) Photosynthetic 826 induction and activity of enzymes related to carbón metabolism: insights into the varying net 827 photosyntheisis rates of coffee sun and shade leaves. Theoretical and Experimental Plant Physiology 828 25:62-69. 829 Martins S., Galmés J., Cavatte P., Pereira L., Ventrella M. & DaMatta F. (2014a) Understanding the low 830 photosynthetic rates of sun and shade coffee leaves: bridging the gap on the relative roles of hydraulic, 831 diffusive and biochemical constraints to photosynthesis. PLoS ONE 9(4): e9557. 832 Doi:10.1371/journal.pone.0095571. 833 Martins S., Araújo W., Tohge T., Fernie A. & DaMatta F. (2014b) In high-light-acclimated coffee plants 834 the metabolic machinery is adjusted to avoid oxidative stress rather than to benefit from extra light 835 enhancement in photosynthetic yield. PLoS ONE 9(4): e94862. Doi:10.1371/journal.pone.0094862. 836 Martins L., Tomaz M., Lidon F., DaMatta F. & Ramalho J. (2014c) Combined effects of elevated [CO2] 837 and high temperature on leaf mineral balance in Coffea spp. plants. Climatic Change 126:365-379. 838 Martins M., Rodrigues W., Fortunato A., Leitao A., Rodrigues A., Pais I., Martins L., Silva M., Reboredo 839 F., Partelli F., Campostrini E., Tomaz M., Scotti-Campos P., Ribeiro-Barros A., Lidon F., DaMatta F. & 840 33 Ramalho J. (2016) Protective response mechanisms to heat stress in interaction with high [CO2] 841 conditions in Coffea spp. Frontiers in Plant Science. Doi: 10.3389/fpls.2016.00947. 842 Matocha J., Schroth G., Hills T. & Hole D. (2012) Integrating climate change adaptation and mitigation 843 through agroforestry and ecosystem conservation. In: Nair P. & Garrity D. (eds) Agroforestry – The 844 future of global land use. Springer, Netherlands, p. 105-125. 845 Matos F., Wolfgramm R., Goncalves F., Cavatte P., Ventrella M. & DaMatta F. (2009) Phenotypic plasticity 846 in response to light in the coffee tree. Environmental and Experimental Botany 67:421-427. 847 Mbow C., Smith P., Skole D., Duguma L. & Bustamante M. (2014) Achieveing mitigation and adaptation 848 to climate change through sustainable agroforestry practices in Africa. Current Opinion in Environmental 849 Sustainability 6:8-14. 850 Monsi M. & Saeki T. (1953) Über den Lichtfaktor in den Pflanzengesellschaften und seine Bedeutung für 851 die Stoffproduktion. Jpn J Botany 14:22-52. 852 Montagnon C., Marraccini P. & Bertrand B. (2012) Breeding for coffee quality. In: Oberthür T., Läderach 853 P., Pohlan H., Cook J. (eds.) Specialty coffee: Managing quality. International Plant Nutrition Institute, 854 Southeast Asia Program. 855 Monteith J. (1977) Climate and the efficiency of crop production in Britain. Philos. Trans. R. Soc. Lond. 856 281:277-294. 857 Morton J. (2007) The impact of climate change on smallholder and subsistence agriculture. Proc. Natl. Acad. 858 Sci USA 104:19680-5. 859 Müller C., Cramer W., Hare W. & Lotze-Campen H. (2011) Climate change risks for African agriculture. 860 PNAS 108(11):4313-4315. 861 Nicotra A., Atkin O., Bonser S., Davidson A., Finnegan E., Mathesius U., Poot P., Purugganan M., Richards 862 C., Valaladares F. & van Kleunen M. (2010) Plant phenotypic plasticity in a changing climate. Trends 863 in Plant Science 15(12):684-692. 864 Niinemets Ü, Flexas J. & Peñuelas J. (2011) Evergreens favored by higher responsiveness to increased CO2. 865 Trends in Ecology and Evolution 26(3):136-142. 866 34 Ordonez J., Luedeling E., Kindt R., Lestara Tata H., Harja D., Jamnadass R. & van Noordwijk M. (2014) 867 Constraints and opportunities for tree diversity management along the forest transition curve to achieve 868 multifunctional agriculture. Current Opinion in Environmental Sustainability 6:54-60. 869 Otte I., Detsch F., Mwangomo E., Hemp A., Appelhans T. & Nauss T. (2016) Multidecadal trends and 870 interannual variability of rainfall as observed from five lowland stations at Mt. Kilimanjaro, Tanzania. 871 Journal of Hydrometeorology 18(2):349-361. 872 Ovalle-Rivera O. (2014) Impacto del cambio climático en al productividad de café (Coffea arabica) bajo 873 sistemas agroforestales en Costa Rica y Nicaragua. Master thesis, Centro Agronómico Tropical de 874 Investigación y Enseñenza (CATIE), Turrialba, Costa Rica. URL: 875 http://repositorio.bibliotecaorton.catie.ac.cr/bitstream/handle/11554/7204/Impacto_del_cambio_climati876 co.pdf;jsessionid=426080B22C242F2CB326613530DD199C?sequence=1 877 Ovalle-Rivera O., Läderach P., Bunn C., Obersteiner M. & Schroth G. (2015) Projected shifts in Coffea 878 arabica suitability among major global producing regions due to climate change. PLoS ONE 10(4): 879 e0124155. Doi:10.1371/journal.pone.0124155. 880 Padovan M., Cortez V., Navarrete L., Navarrete E., Deffner A., Centeno L., Munguía R., Barrios M., 881 Vílchez-Mendoza J., Vega-Jarquín C., Costa A., Brook R. & Rapidel B. (2015) Root distribution and 882 water use in coffee shaded with Tabebuia rosea Bertol. And Simarouba glauca DC. compared to full 883 sun coffee in sub-optimal environmental conditions. Agroforestry Systems 89:857-868. 884 Penman HL. (1948) Natural evaporation for open water, bare soil and grass. Proc R Soc Lond, Ser A 885 193:120-146. 886 Poorter H. & Nagel O. (2000) The role of biomass allocation in the growth response of plants to different 887 levels of light, CO2, nutrients and water: a quantitative review. Australian Journal of Plant Physiology 888 27:595-607. 889 Poorter H., Lambers H. & Evans J. (2014) Trait correlation networks: a whole-plant perspective on the 890 recently criticized leaf economic spectrum. New Phytologist 201(2):378-382. 891 Pujol G., Iooss B., Janon A. et al. (2015) sensitivity : Sensitivity Analysis. R package version 1.11.1. 892 http://CRAN.R-project.org/package=sensitivity. 893 35 R Core Team (2014) R: A language and environment for statistical computing. R Foundation for Statistical 894 Computing, Vienna, Austria. URL: http://www.R-project.org/. 895 Ramirez-Villegas J. & Jarvis A. (2010) Disaggregation of Global Circulation Model Outputs. International 896 Center for Tropical Agriculture (CIAT), Cali, Colombia 897 Ramirez-Villegas J., Challinor A., Thornton P. & Jarvis A. (2013) Implications of regional improvement in 898 global climate models for agricultural impact research. Environmental Research Letters 8(2): 024018 899 Reyer C., Leuzinger S., Rammig A., et al. (2013) A plant’s perspective of extremes : terrestrial plant 900 responses to changing climatic variability. Global Change Biology 19:75-89. 901 Reyer C., Flechsig M., Lasch-Born P. & van Oijen M. (2016) Integrating parameter uncertainty of a process-902 based model in assessments of climate change effects on forest productivity. Climate Change 903 137(3):395-409. 904 Rice R. (2011) Fruits from shade trees in coffee: how important are they? Agroforestry Systems 83:41-49. 905 Rodrigues W., Machado J., da Silva J., de Assis F., Ferraz L., da Silva L., de Abreu D., de Paula W., Cespom 906 L., de Sousa E., Glenn D., Ramalho J. & Campostrini E. (2016) Whole-canopy gas exchanges in Coffea 907 sp. is affected by supra-optimal temperature and light distribution within the canopy: The insights from 908 an improved multi-chamber system. Scientia Horticulturae 211:194-202. 909 Rodriguez D. Cure J., Cotes J., Gutierrez A. & Cantor F. (2011) A coffee agroecosystem model: I. Growth 910 and development of the coffee plant. Ecological Modelling 222:3626-2639. 911 Saltelli A. & Bolado R. (1998) An alternative way to compute Fourier amplitude sensitivity test (FAST). 912 Computational Statistics & Data Analysis 26(4):445-460. 913 Saltelli A., Tarantola S. & Chan K. (1999) A quantitative model-independent method for global senstivity 914 analysis of model output. Technometrics 41(1):39-56. 915 Sanchez P. (1995) Science in agroforestry. Agroforestry Systems 30:5-55. 916 36 Schaller M., Schroth G., Beer J. & Jiménez F. (2003) Species and site characteristics that permit the 917 association of fast-growing trees with crops: the case of Eucalyptus deglupta as coffee shade in Costa 918 Rica. Forest Ecology and Management 175:205-215. 919 Schroth G., Läderach P., Dempewolf J., Philpott S., Haggar J., Eakin H., Castillejos T., Garcia J., Soto L., 920 Hernandez R., Eitzinger A. & Ramirez-Villegas J. (2009) Towards a climate change adaptation strategy 921 for coffee communities and ecosystems in the Sierra Madre de Chiapas, Mexico. Mitigation and 922 Adaptation Strategies for Global Change 14:605-625. 923 Siles P., Vaast P., Dreyer E. & Harmand J.M. (2010) Rainfall partitioning into throughfall, stemflow and 924 interception loss in a coffee (Coffea arabica L.) monoculture compared to an agroforestry system with 925 Inga densiflora. Journal of Hydrology 395:39-48. 926 Sinclair T. & Seligman N. (1996) Crop modeling: From infancy to maturity. Agronomy Journal 88(5):698-927 704. 928 Sobol’ I.M. (1993) Sensitivity estimates for nonlinear mathematical models. Math. Model. Comput. Exp. 929 14:407-414. 930 Staver C., Guharay G., Monterroso D. & Muschler G. (2001) Designing pest-suppressive multistrata 931 perennial crop systems: shade-grown coffee in Central America. Agroforestry Systems 53:151-170. 932 Tausend P., Meinzer F. & Goldstein G. (2000) Control of transpiration in three coffee cultivars: the role of 933 hydraulic and crown architecture. Trees 14:181-190. 934 Thornton P., Ericksen P., Herrero M. & Challinor A. (2014) Climate variability and vulnerability to climate 935 change: a review. Global Change Biology 20:3313-3328. 936 Unigarro C., Jaramillo A., Ibarra L. & Flórez C. (2016) Canopy structure and theoretical coefficient of 937 extinction in genotypes of arabic coffee in Colombia. Acta Agronómica 65:383-389. 938 Vaast P., Bertrand B., Perriot J.J., Guyot B. & Génard M. (2005). Fruit thinning and shade improve bean 939 characteristics and beverage quality of coffee (Coffea arabica L.) under optimal conditions. Journal of 940 the Science of Food and Agriculture, 86 (2):197-204. 941 37 Vaast P., van Kanten R., Siles P., Angrand J. & Aguilar A. (2008) Biophysical interactions between timber 942 trees and Arabica coffee in suboptimal conditions of Central America. In: Jose S. & Gordon A. (eds.) 943 Toward agroforestry design. Springer, Netherlands, pp. 133-146. 944 Vaast P., Martínez M., Boulay A., Castillo B. & Harmand J.M. (2015) Diversifying Central American coffee 945 agroforestry systems via revenue of shade trees. In: Ruf F. & Schroth G. (eds.) Economics and Ecology 946 of Diversification. Springer, Netherlands, pp. 271-281. 947 Vaast P., Harmand J.-M., Rapidel B., Jagoret P. & Deheuvels O. (2016) Coffee and cocoa production in 948 agroforestry – A climate-smart agriculture model. In: Torquebiau E. (ed.) Climate Change and 949 Agriculture Worldwide, Chapter 16, pp. 209-224. Springer, Netherlands. 950 Van der Vossen H. (2005) A critical analysis of the agronomic and economic sustainability of organic coffee 951 production. Experimental Agriculture 41(4):449-473. 952 Van der Wolf, J., Jassogne, L., Gram, G. & Vaast, P. (2016) Turning local knowledge on agroforestry into 953 an online decision-support tool for the tree selection in smallholders’ farms. Experimental Agriculture, 954 1-17. doi:10.1017/S001447971600017X 955 Van Ittersum M., Cassman K., Grassini P., Wolf J., Tittonell P. & Hochman Z. (2013) Yield gap analysis 956 with local to global relevance – A review. Field Crops Research 143:4-17. 957 Van Kanten R. & Vaast P. (2006) Transpiration of arabica coffee and associated shade tree species in sub-958 optimal, low altitude conditions of Costa Rica. Agroforestry Systems 67:187-202. 959 Van Keulen H. (1986) A simple model of water-limited production. In: Van Keulen H., Wolf J. (Eds.) 960 Modelling of Agricultural Production: Weather, Soils and Crops. Simulation Monogr. PUDOC, 961 Wageningen, pp. 130-152. 962 Van Oijen M., Dreccer M., Firsching K. & Schnieders B. (2004) Simple equations for dynamic models of 963 the effects of CO2 and O3 on light-use efficiency and growth of crops. Ecological Modelling 179:39-60. 964 Van Oijen M., Rougier J. & Smith R. (2005) Bayesian calibration of process-based forest models: bridging 965 the gap between models and data. Tree Physiology 25:915-927. 966 38 Van Oijen M. & Leffelaar P. (2008) Lintul-2: wáter limited crop growth. Wageningen University Plant 967 Sciences. URL: http://models.pps.wur.nl/node/944. Accessed 16.02.2016. 968 Van Oijen M., Dauzat J., Harmand J-M., Lawson G. & Vaast P. (2010a) Coffee agroforestry systems in 969 Central America: I. A review of quantitative information on physiological and ecological processes. 970 Agroforestry Systems 80:341-359. 971 Van Oijen M., Dauzat J., Harmand J-M., Lawson G. & Vaast P. (2010b) Coffee agroforestry systems in 972 Central America: II. Development of a simple process-based model and preliminary results. Agroforestry 973 Systems 80:361-378. 974 Van Oijen M., Schapendonk A. & Höglind M. (2010c) On the relative magnitudes of photosynthesis, 975 respiration, growth and carbon storage in vegetation. Annals of Botany 105:793-797. 976 Van Vuuren D., Edmonds J., Kainuma M. et al. (2011) The representative concentration pathways: an 977 overview. Climate Change 109:5-31. 978 Wang E., Smith C., Bond W. & Verburg K. (2004) Estimation of vapour pressure deficit and crop water 979 demand in APSIM and their implications for prediction of crop yield, water use, and deep drainage. 980 Australian Journal of Agricultural Research 55:1227-1240. 981 Wang N., Jassogne L., van Asten P., Mukasa D., Wanyama I., Kagezi G. & Giller K. (2015) Evaluating 982 coffee yield gaps and important biotic, abiotic, and management factors limiting coffee production in 983 Uganda. European Journal of Agronomy 63:1-11. 984 Wintgens (2004) The coffee plant. In: Wintgens J. (ed.) Coffee: Growing, processing, sustainable 985 production. Wiley-VCH Verlag, Germany. 986 Yin X., Struik P. & Kropff M. (2004) Role of crop physiology in predicting gene-to-phenotype relationships. 987 Trends in Plant Science 9(9):426-432. 988 Yin X. (2013) Improving ecophysiological simulation models to predict the impact of elevated atmospheric 989 CO2 concentration on crop productivity. Annals of Botany 112:465-475. 990 39 Zhang Z., Cai Z., Liu G., Wang H., Huang L. & Cai C. (2017) Effects of fertilization on the growth, 991 photosynthesis, and biomass accumulation in juvenile plants of three coffee (Coffea arabica L.) 992 cultivars. Photosynthetica 55(1):134-143. 993 Zuidema P., Leffelaar P., Gerritsma W., Mommer L. & Anten N. (2005) A physiological production model 994 for cocoa (Theobroma cacao): model presentation, validation and application. Agricultural Systems 995 84:195-225. 996