Assessing climate change impacts on sorghum and millet yields in the Sudanian and Sahelian savannas of West Africa

Climate change is likely to adversely affect food security in many regions of the world, especially in developing countries where a large fraction of the population is already facing chronic hunger and malnutrition (Lobell et al 2008, Schmidhuber and Tubiello 2007). In such countries, the survival of the population will depend on the effective adaptation of agriculture to climate change. Better knowledge of climate impacts on crop yields is a pre-requisite for breeding more resilient crop varieties, or for adopting existing varieties more resistant to climate-induced stress—two different strategies to counteract the adverse effects of climate change on crop yields (Barnabás et al 2008). However any varietal adaptation will be challenging because of uncertainties both in regional climate changes scenarios (Christensen et al 2007) and in the response of crops to changing climate (Barnabás et al 2008). For Africa in particular, the meta-analysis of the results of 16 studies over West Africa (Müller 2011, Roudier et al 2011) shows that although projected impacts on yield are most frequently slightly negative (−10%), there is a large spread between results. This variation indicates a low confidence in future yield projections, with projected impacts on crop yields ranging from −50% to +90%. More recently, Knox et al (2012) estimate that mean yield change for all crops is −8% by the 2050s with strong variations among crops and regions. This diversity of impacts reflects the differences in these impact studies, which focus on different locations, and rely on different climate projections, crops and crop models, and downscaling techniques; the result is a large spread of crop yield projections (Sultan 2012). This low degree of confidence makes it difficult to aggregate and upscale such projections in order to provide a consistent assessment of future yield changes at regional or continental scale (Berg et al 2013).

There is thus a need to perform more consistent assessments of climate change impacts on crop yields at the regional scale in West Africa. In particular, it is necessary to account, as far as possible, for uncertainties in regional climate change scenarios. Indeed, the considerable spread in current climate model projections of regional climate change over West Africa, especially with respect to rainfall (Müller 2011, Roudier et al 2011), results in uncertainty in estimates of future crop yields; with the existing studies generally being based on too narrow a selection of climate projections. In addition, impact assessments need to quantify the uncertainty associated with the response of each crop type to a changing climate. Several different approaches exist to derive large-scale climate change impacts on crop yields, such as the use of empirical relationships between crop growth and climate (Lobell et al 2008, Schlenker and Lobell 2010) or the recent development of large-scale vegetation models including an explicit representation of croplands (Arneth et al 2010, Müller et al 2009, Berg et al 2013). Among these approaches the up-scaling of simulations from plot-scale process-based crop models over a larger regional domain (Jones and Thornton 2003, Liu et al 2008) allows the diversity of responses to be assessed accurately, including the effects of different cultivars and/or management practices. Especially in West Africa, farmers are able to select adapted varieties (e.g. late or early millet) or to adapt their practices (e.g. delayed or early sowing) to a changed environment (Dingkuhn et al 2006). The ability to identify the most suitable crop varieties, with the most robust characteristics for withstanding climate change, is crucial for formulating adaptation strategies in the region.

In this context, we studied an extensive set of yield simulations using a process-based crop model enabling us to assess the impacts of climate change on the yield of sorghum and millet, two staple crops in the Sudanian and Sahelian savannas of West Africa. The study took into account the diversity of local cultivars and the full range of climate scenarios. We first simulated future sorghum and millet crop yields at the scale of West Africa, and then identified the areas and crop cultivars most vulnerable to climate change. Finally, differences among cultivars in the response of millet and sorghum yields were discussed, identifying the characteristics of the robust cultivars needed to minimize yield loss in the face of climate change.

2.1. Climate data

Three climate data sets are used in this study. First a set for 1961–90, compiled from 35 meteorological stations with time series at daily resolution located in 9 countries across West Africa (figure 1(a)). These data sets have been compiled by the AGRHYMET Regional Centre and National Meteorological Agencies. These stations record rainfall and several meteorological parameters at 2 m above ground level, such as solar radiation, surface wind speed, humidity and temperature. The data were used to perform crop growth simulations.

Zoom In Zoom Out Reset image size

Second, Mitchell and Jones (2005) have produced a global gridded climate-observation database at the monthly timescale for 1901–2009 at 0.5° resolution (CRU TS 3.1 database; www.cru.uea.ac.uk/). We computed annual values across West Africa by averaging their rainfall and temperature data over the grid points from longitude 19°W to 17°E and from latitude 2.5°N to 25°N and computing the summer mean (May–September) rainfall amount and mean temperature. These data were used to define decadal climate anomalies relative to the 1961–90 reference period. For example, for the first decade of the twentieth century, we computed the summer temperature anomalies as the temperature differences between the 1901–10 average and the 1961–90 baseline average. Similarly, past summer rainfall anomalies were defined as the ratio of decadal rainfall to 1961–90 rainfall.

Third, monthly outputs from simulations over the 20th (20C3M, 1961–2000) and 21st (2001–2100) centuries were retrieved from the Program For Climate Model Diagnosis and Intercomparison (PCMDI) archive for several climate models and scenarios from CMIP3 (the World Climate Research Programme's Coupled Model Intercomparison Project phase 3 multi-model dataset); (table S3 available at stacks.iop.org/ERL/8/014040/mmedia) and CMIP5 (table S4 available at stacks.iop.org/ERL/8/014040/mmedia). Note that not all simulations from the CMIP5 exercise were available at the time of submitting this paper. For each Global Climate Model (GCM), we considered rainfall and mean daily temperature projections during summer (May–September) for two time periods (2031–50 and 2071–90), three emission scenarios based on the Special Report on Emissions Scenarios (A2, A1B, B1) for CMIP3 runs and three Representative Concentration Pathways (RCP 4.5, 6.0 and 8.5, the new emission scenarios used for the 5th IPCC report, Moss et al (2010)) for CMIP5 runs. We computed the summer temperature anomalies as the temperature differences between the future projection and the 1961–90 reference period from the GCM. Similarly, future summer rainfall anomalies were defined as the ratio of future rainfall to 1961–90 rainfall.

2.2. Crop yield simulations

The simulated yield indices over the 35 weather station locations agreed well with the yield indices derived from statistics for each country available from FAO (figure 1(b)). Note that the term yield index refers to the standardized time series of crop yield. Averaging sorghum and millet yields, the inter-annual correlation coefficient between simulated and observed detrended yield indices is R = 0.70 (details per crop and per country are given in figure S1, trend equations are given in SI appendix available at stacks.iop.org/ERL/8/014040/mmedia). Compared to other crop impact studies at the continental scale (e.g. Berg et al 2010), SARRA-H shows improved simulation of crop yield variability in West Africa. Moreover, its skill is close to that shown by GLAM for groundnut in India, in our opinion, one of the best crop growth models for the tropics (Challinor et al 2004). However although our crop model could capture quite well the variability of crop yields, the mean yield (table S2 available at stacks.iop.org/ERL/8/014040/mmedia) was overestimated, by a factor two, which is a common shortfall in many crop model studies in Sub-Saharan Africa, because crop models are usually calibrated against data collected in controlled environments, and thus do not account for non-climatic factors like pests, weeds and soil-related constraints (Challinor et al 2004, 2005, Bondeau et al 2007). Despite this overestimation of the mean yield, the model accurately simulated the effect of changes in rainfall on crop yield anomalies (figure S3 available at stacks.iop.org/ERL/8/014040/mmedia).

The results presented in figure 2 show the effect of rainfall and temperature changes on mean crop yield (averaged across all varieties of sorghum and millet). We found that mean crop yields decreased in 31 of the 35 climate scenarios (figure 2) mainly as a response to temperature warming. When warming exceeded 2 °C, mean crop yields were always found to decrease significantly; rainfall changes being only able to modulate the magnitude of this negative impact. The worst case occurred when both heat and water stresses were combined. Based on the analysis of crop simulations we found that three key processes, and their interactions, accounted for the detrimental effect of temperature increase on crop yield. First the water stress was amplified with an increase of potential evapotranspiration (+8% for a 4 °C warming) in water-limited soils, where the crop roots could not take up more water. Second, the maintenance respiration per biomass unit increased dramatically (+30% for a 4 °C warming) and the biomass production consequently decreased. Third, the crop-cycle length was reduced (by up to 10 d for a 4 °C warming), which also resulted in reduced biomass production. As temperature is expected to increase in all IPCC climate change scenarios (table S3 available at stacks.iop.org/ERL/8/014040/mmedia), the negative impact on crop yields is a very robust result, especially for long-term projections (2071–90) in which warming over sub-Saharan West Africa is projected to exceed +2 °C in about 70% of the IPCC climate scenarios. Despite uncertainty in future rainfall changes, which are inconsistent across climate models and scenarios in CMIP3 simulations, and the projected increase of rainfall in CMIP5 simulations, the warming signal led to a yield reduction greater than 10% from the present-day value in nearly 50% of the climate medium-term projections (2031–50), and in more than 80% of long-term projections (figure 3). Future projections lie outside the domain of observed climate fluctuations over the past century (figure 2). Historical climate data from Climatic Research Unit (CRU) show important rainfall decadal fluctuations relative to the 1961–90 baseline: from +19% in the 1950s to −9% in the 1980s. These fluctuations led to major variations in the observed yield, which are well reproduced by the crop model (figure S3 available at stacks.iop.org/ERL/8/014040/mmedia). However, historical temperature variations, around a warming trend of about +1 °C since 1901, have remained moderate compared to the magnitude of the future warming signal projected by IPCC models. Given the dominant control by heat stress in all scenarios, future climate impacts on crop yields have no equivalents in historical conditions, for which the main controlling variable was rainfall variation. In all climate change scenarios, the projected yield decrease is modelled to be as severe as, or more severe than, during the record droughts of the 1970s and 1980s, which had dramatic consequences (Mellor and Gavian 1987).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The response of crop yields to climate change may vary strongly with variety and be region-dependent. First, when considering a warming of +4 °C (Thornton et al 2011), we found that the probability of a yield reduction higher than 10% was greater in the Sudanian region (southern Senegal, Mali, Burkina Faso, northern Togo and Benin), because of an exacerbated sensitivity to temperature changes compared to the Sahelian region (Niger, Mali, northern parts of Senegal and Burkina Faso), where crop yields are more sensitive to rainfall changes (figure 4(a)). In the Sahelian countries, future yield projections are more uncertain because of the inconsistency of rainfall changes among scenarios. Second, when looking at the specific response of each variety of sorghum and millet in our ensemble of simulations (table S1 available at stacks.iop.org/ERL/8/014040/mmedia), we found that traditional photosensitive varieties tend to be more resistant to the adverse effects of climate change (figure 4(b)). Although these traditional varieties have a lower present-day simulated yield compared to modern cultivars, their future yields are less affected by the projected temperature increase (table S1, figure S4 available at stacks.iop.org/ERL/8/014040/mmedia). Because traditional varieties of millet and sorghum calibrated in our study are photoperiod-sensitive, the duration of phenological phases depends not only on temperature and sums of GDD, but also on photoperiod (astronomic day length). Photoperiod of course remains constant while temperature is increasing (figure 5). As a result the crop-cycle length of traditional varieties is less reduced by the temperature increase than it is for the modern photoperiod-insensitive varieties. Biomass production is therefore less affected by the temperature increase (figure S5 available at stacks.iop.org/ERL/8/014040/mmedia). However, the impact of climate change remains negative for a +4 °C warming scenario whatever the variety. Furthermore, given the large difference in mean yields of the modern versus traditional varieties, the modern varieties still yield more under optimal fertility conditions in a warmer world (+4 °C) with 20% less rainfall based on the relative yield changes estimated in figure 4.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Finally we simulated the effects of an 'adaptation scenario', whereby in the future farmers systematically select photoperiod-insensitive varieties of millet and sorghum with a higher thermal time requirement under higher temperatures. To do so, we increased the modelled thermal time requirement of the basic vegetative phases of the two current varieties of millet and sorghum so that the length of their growth cycle remained the same under higher temperatures scenarios. This adaptation scenario clearly mitigates the effect of temperature increase and reduces crop yield losses due to a reduction of the crop-cycle length (figure 4(b)). However the traditional photoperiod-sensitive cultivars still minimize crop yield losses in the worst climate change scenario (+4 °C and 20% decrease of rainfall). Furthermore the crop yield of photoperiod-sensitive cultivars is less variable than that of modern varieties, irrespective of whether rainfall increases or decreases.

In our simulations, the major effect of climate change on the yields of millet and sorghum in West Africa was yield losses induced by higher temperature leading to increased potential evapotranspiration, crop maintenance respiration and a reduction of the crop-cycle length. Millet and sorghum yields are likely to decrease by some 0–41% in the 21st century over West Africa because of the expected warming, irrespective of whether rainfall increases or decreases. Although the different climate scenarios, time slices, crops and methodologies, hamper exact comparison to the results from other studies, this negative impact on crop yield is broadly consistent with previous findings for maize by Jones and Thornton (2003) and Schlenker and Lobell (2010). Based on a similar approach, but using the CERES crop model driven by climate projections from the HadCM2 climate model, Jones and Thornton (2003) projected an overall decrease of maize yield of −10% for Africa by 2050, with impacts varying between −30 and +2% over Sub-Saharan countries (−14% on average). This result suggests the primary role of temperature increase as the driver of maize yield change (Thornton et al 2009). With a completely different methodology based on empirical large-scale relationships between climate and yields, Schlenker and Lobell (2010) projected a yield decrease for mid-century (2046–65) of about −20% for C4 cereals, driven by the increase in temperature.

Although the size of this impact on yield is uncertain because of uncertainty in rainfall change projections, future climate and potential impacts will be very different from those observed in recent history because of the adverse role of temperature. Defining adaptation strategies for this unknown situation in West Africa is thus an unprecedented challenge. We must be able to identify the most vulnerable areas and to specify the crop varieties with the most robust characteristics for withstanding climate change. Here, our simulations have shown that the effect of climate change is not identical for all major grown cultivars of millet and sorghum. Crop yield losses are expected to be greater for modern cultivars with a high biological yield potential but a short, nearly fixed, growth cycle. Until recently, these varieties have been recommended as a way of intensifying cropping systems in West Africa (Dingkuhn et al 2006, De Rouw 2004). Traditional photoperiod-sensitive cultivars are less affected by temperature increase since the photoperiod limits the reduction of the crop duration. The effects of the shortening of crop-cycle duration could be partly counteracted by breeding varieties with higher thermal requirements but traditional photoperiod-sensitive cultivars will still present the advantage of lower yield variability and higher resistance to drought—factors reducing the relative advantage of modern cultivars. Furthermore, the high potential yield of modern varieties is only achieved under optimal fertility conditions which are rarely met in West Africa. In addition, photoperiod sensitivity would likely present some advantages in the event of future change in the timing of the rainy season. Indeed, it allows for flowering at the end of the rainy season for a wide range of planting dates and avoids incomplete grain filling, a problem for late maturing varieties faced with water shortage at the end of the rainy season (Dingkuhn et al 2006). Identifying such specific local crop characteristics is a key to developing a breeding strategy that addresses the trade-off between intensification and resilience to climate change (Dingkuhn et al 2006, De Rouw 2004).

Every modelling study has its limitations, and we recognize a few caveats to the design of our experiments. First, we only considered the effect of changes in mean temperature and rainfall, yet changes in variability such as heat-waves occurring during critical stages of crop development can also affect crop yield (Wheeler et al 2000) and floods from heavy rainfall may amplify the adverse effect of climate change. However, it is unlikely that the simulations from even the state-of-the-art GCMs we have used can give any robust prediction of such extreme events at local scale. Second, the crop simulations did not consider the effect of increased atmospheric CO₂ concentration on crop yields. However, we should not expect the increased atmospheric CO₂ concentration to offset the adverse effect of temperature increase on millet and sorghum yields, because they are both C4 plants. Photosynthesis rates for C4 plants do not respond strongly to higher ambient CO₂, although water stress resistance can be slightly improved as a result of lower stomatal conductance and greater intercellular CO₂ concentration (Leakey 2009, Long et al 2006a, 2006b). Indeed, comparison between crop simulations where CO₂ is increasing, or is kept constant in the future, shows very little difference amongst millet yield changes in Africa under several climate change scenarios (Berg et al 2013). Finally, our simulations considered only one representative soil for the entire Sudanian and Sahelian savanna, while in reality soil properties might vary across the region. Indeed, rainfed millet and sorghum are often grown along toposequences with differences in elevation of a few metres. This natural rainfall harvesting can lead to differentiation in soil properties and hydrological conditions, which in turn may affect crop performance and yield.

In summary, we have attempted to perform a consistent assessment of climate change impacts on crop yields in West Africa by accounting for uncertainties in future climate scenarios. The adverse role of higher temperature on crop yield is a robust result, irrespective of rainfall changes, and seems to be modulated across millet and sorghum varieties and across regions in West Africa. This points to the need for further work on identifying varieties that are more resilient to elevated mean temperature during the growing period.

This work has been funded by REGYNA (GIS-CES), the French National Research Agency (ANR) through the ESCAPE project (ANR-10-CEPL-005) and the R2DS network.