Cocoa plant productivity in West Africa under climate change: a modelling and experimental study

Data were taken from the UK Met Office Unified Model (GA3) ensemble of 25 km resolution global atmosphere-only runs: The U.K. on Partnership for Advanced Computing in Europe (PRACE) Weather-Resolving Simulations of Climate for Global Environmental Risk (UPSCALE) project runs (herein these runs will be referred to as UPSCALE). The UPSCALE simulations use an atmosphere only AMIP (Atmospheric Model Intercomparison Project) type set up, in which an atmospheric model is driven with observed sea surface temperature (SST). Here, OSTIA SSTs represented present climate (Donlan et al 2012); and with OSTIA SSTs modified by HadGEM2 projected SST change (2090–2110 relative to 1990–2010; RCP8.5 scenario). CO₂ is set to constant values of 343 ppmv for the 1990–2012 runs and 935 ppmv for the future climate runs (Mizielinski et al 2014). The ensemble consists of five present day and three future integrations, generated through small perturbations to initial conditions.

The high-resolution model analysis is supplemented with analysis of model output from the 5th Coupled Model Intercomparison Project (CMIP5), which forms the basis of the IPCC 5th assessment. In this study, 29 models were used, comprising a range of modelling centres and including a range of resolutions and parameterisations. A full description of the CMIP5 dataset is given in Taylor et al (2012); and a table of the models used in this study, with information on horizontal resolution, is given in the supplementary material (table S3.1).

Lack of knowledge of future emissions is a key source of uncertainty in climate projections. The CMIP5 approach is to carry out climate model simulations for a range of greenhouse gas concentration trajectories, described by representative concentration pathways (RCPs) (Van Vuuren et al 2011). Although the RCPs are based on CO₂ concentrations, rather than emissions, the two are closely related. To explore the range of uncertainty related to future emissions, the following two RCPs were considered: RCP 4.5, a medium emissions scenario, with CO₂ emissions gradually increasing until ∼2040 and then declining; RCP 8.5, a high emissions scenario, with CO₂ emissions increasing at rates that are close to the present day throughout the 21st Century.

2.2.1. Plant growth conditions

2.2.2. Photosynthetic light response curves

The Joint UK Land Environment Simulator (JULES) is the land surface component of the UK Met Office Hadley Centre climate model. JULES calculates the fluxes of energy and mass between the land surface and atmosphere. It explicitly models the temperature and water content of soil in an arbitrary number of layers, as well as the temporal evolution of leaf area index, when the TRIFFID vegetation model or phenology model is invoked.

In this study, rather than being coupled to a climate model, JULES is run offline — enabling us to explore sensitivities of the land-atmosphere system to individual elements of climate change. The choice of JULES, moreover, ensures consistency with the climate model study using UPSCALE, whose land-surface scheme is equivalent to JULES. In its offline mode, JULES is driven by observed or modelled meteorological variables. Information on soil and land surface properties is provided by ancillary files. The version used here is 5.1, which is described in full at http://jules-lsm.github.io/vn5.1/. A full description of JULES can be found in Clark et al (2011), and Best et al (2011).

Key to our study is model representation of photosynthesis, and the leaf-level exchange of water and carbon in C3 plants. In JULES, the leaf level stomatal conductance (g_s) is linked to the rate of photosynthesis (A) via the CO₂ diffusion equation:

$$\begin{equation}A = {g_{\text{s}}}\left( {{C_{\text{c}}} - {C_{\text{i}}}} \right)/1.6\end{equation} \tag{ 1 }$$

where C_c and C _i are the leaf surface and internal concentrations of CO₂.

C_c and C_i are related by the Jacobs (1994) formulation:

$$\begin{equation}\frac{{{C_{\text{i}}} - {C_*}}}{{{C_{\text{c}}} - {C_*}}} = {f_0}\left( {1 - \frac{D}{{{D_*}}}} \right)\end{equation} \tag{ 2 }$$

where D is the leaf humidity deficit and ${C_*}$ is the CO₂ compensation point (the point at which no further CO₂ is absorbed from the environment because the CO₂ required for photosynthesis is equal to the amount released by respiration). The variables f₀ and D* are vegetation-specific calibration parameters (for full details see Cox et al 1998).

The rate of photosynthesis if there is no water stress (A_p) is represented as the minimum of three limiting regimes: (i) Rubisco-limited rate (W_C); (ii) light-limited rate (W_L); and (iii) rate of transport of photosynthetic products (W_E):

$$\begin{equation}{A_{\text{p}}} = \min ({W_{\text{C}}},{W_{\text{L}}},{W_{\text{E}}})\end{equation} \tag{ 3 }$$

Each of the limiting rates of photosynthesis is affected by environmental conditions (Collatz et al 1991). In C3 plants, all three limiting regimes are affected by temperature. Photosynthetically active radiation (PAR), however, only affects W_l. Both W_c and W_l are affected by the internal CO₂ concentration (C_i).

Leaf photosynthesis (A) is then derived by scaling A_p with a metric of root-zone soil moisture content ($\beta$):

$$\begin{equation}A = {A_{\text{P}}}\beta \end{equation} \tag{ 4 }$$

The means by which 'A' is scaled to canopy-level gross primary productivity (GPP) is described in Best et al (2011).

As was described above and in Clark et al (2011), the three limiting rates of photosynthesis and the CO₂ diffusion equation (equation (1)) include a number of plant-specific parameters. In order to account for these variations, JULES separates vegetation cover into plant functional types (PFTs). When developing a cocoa PFT, in this study, we begin by considering the response of the tropical broad-leaf PFT (Harper et al 2016) selected for being the closest morphologically to cocoa. To develop a PFT that resembles cocoa, we utilise a data assimilation (DA) method, ensuring that the rate of photosynthesis, and hence the canopy-scale GPP and NPP (Net Primary Productivity) is calculated using a cocoa specific parameterisation.

The method by which the observations described in section 2.2 are utilised to parameterise JULES for cocoa is summarized in the supplementary information section 2.

Both rainfall and temperature are projected to change during the 21st Century in West Africa. Figure 1 shows that, in the high-resolution simulations, precipitation is projected to increase over much of West Africa, albeit with reduced precipitation projected in Western Guinea and in southern coastal regions of Ghana and Côte d'Ivoire. These findings are broadly consistent with the CMIP5 analysis presented in Roehrig et al (2013).

Zoom In Zoom Out Reset image size

Cocoa is a perennial crop and is thus sensitive to the seasonal cycle in rainfall. A metric of rainfall seasonality (used in Laderach et al (2013) and Schroth et al (2016)) is the number of consecutive months of the climatological season cycle during which rainfall is less than 100 mm. It is suggested in these studies and in previously published work on cocoa (Wood and Lass 2008) that regions that climatologically experience more than three consecutive months of < 100 mm of rainfall are less suitable for cocoa cultivation, and that this determines the northern boundary of the cocoa belt in Ghana and Côte d'Ivoire. Figure 2 shows that the projected changes in rainfall do not result in large changes in this metric within the present-day cocoa growing regions delineated by Schroth et al (2016). There is some indication, however, that the projected increases in rainfall have a favorable effect on the cocoa growing capacity for some regions of Nigeria. These results are consistent with Schroth et al (2016), but may be considered more robust because of the improved representation of the seasonal cycle in the high-resolution climate simulations used in this study.

Zoom In Zoom Out Reset image size

The projected changes in rainfall are small in comparison to variability between models (see supplementary information). This is not the case, however, for temperature. Figure 3 shows that minimum, maximum and mean daily temperature are all projected to increase in this region; by approximately 5 °C by the end of the 21st Century under an RCP8.5 scenario. Moreover, figure 4 indicates that the projected shifts in temperature are fairly uniform across the full range of values. In other words, a 30 °C event in the present occurs as often as a 35 °C event is projected to occur in the future; and the same is true for a 20 °C present/25 °C future event.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The high-resolution climate simulations provide detailed spatial information and a good representation of seasonality, but they are only available for the end of the 21st Century and for a high emissions scenario. Although there are deficiencies in the CMIP5 simulations, the multi-model mean broadly captures the spatial variability of present day mean annual rainfall amount and temperature (Roehrig et al 2013, Dunning et al 2017). There is value, therefore, in using the CMIP5 simulations to explore projected time series of regional annual temperature and rainfall, and to explore sensitivity to emissions scenarios. Figure 4 shows time series of anomalies in temperature, total rainfall, and amount of rain during intense events for 7⁰ W–0, 5⁰N –10⁰ N—comparing RCP4.5 and RCP8.5 emissions scenarios for 29 CMIP5 models. The dots show where the multi-model anomaly distribution for a given year is significantly different to the multi-model anomaly distribution for the climatological period (1986–2005)—in other words, when the signal of climate change emerges from the noise of internal model variability. The anomalies are calculated relative to each model's own climatology. As would be expected, there is a large warming trend. Scenario differences become apparent in the 2030s, but significant warming relative to 1986–2005 is evident for both scenarios throughout the future model integration.

In contrast to temperature, when averaged over a large region, a signal of increase or decrease in annual precipitation barely emerges from the noise, even by the end of the 21st Century. Comparing RCP4.5 and RCP8.5, there is some indication that differences between models are slightly greater at the end of the 21st Century for the RCP8.5 scenario, but beyond that, there is little sensitivity to scenario. Previous studies have shown, however, that even when the projected change in mean precipitation is small, there may be significant projected changes in the daily extremes (Field et al 2012). The bottom panel of figure 4 shows that the amount of seasonal precipitation during high rainfall days is projected to increase by the end of the 21st Century in RCP8.5.

As was described in section 2, the first step of the modelling is using a DA technique to parameterize the land-surface model for cocoa (Pinnington et al 2020). The results of the DA are shown in figure 5. Figure 5 (top left panel) shows the observations and the prior JULES prediction before DA. The observed photosynthesis initially increases with irradiance and then saturates at around 250 µmol m⁻² s⁻¹ (consistent with cocoa being a shade-adapted plant). In comparison, the photosynthetic rate for JULES has only just begun to saturate at 450 µmol m⁻² s⁻¹, as is typical for standard C3 vegetation. For both model and observations, the level of photosynthesis for the plant is always higher under increased CO₂ levels. Figure 5 (top right panel) is the same as the top left panel but now using the new parameterisation after applying the DA technique to the JULES model and observational data. It is clear that after DA, the JULES model predictions fit the cocoa observations well and are always within the standard deviation of the observations. This is further illustrated by the scatter plots in two bottom panels of figure 5, where, after DA, the value of r² increases whereas root-mean squared error (RMSE) decreases considerably, for the JULES predictions compared to the observations. These findings provide evidence that we now have a version of the JULES model capable of representing key aspects of photosynthetic response of cocoa to changes in CO₂ and other key atmospheric variables.

Zoom In Zoom Out Reset image size

In order to explore the effect of climate change on cocoa productivity, net primary productivity was simulated using JULES driven with the present and future CO₂ levels and climates. NPP is the accumulation of biomass by plants before any litter or mortality. Although it is not a direct indicator of yield it does represent the resource available to the plant for growth, including allocation into fruiting bodies. In this study, we focus on NPP rather than GPP because NPP is more closely related to crop yield. The driving data were derived from the same high-resolution climate simulations described in section 2.1, and all results are presented for the JULES cocoa PFT (see section 2.3).

Over much of West Africa, NPP is projected to increase (figure 6), suggesting that cocoa trees will become more productive. Over the key cocoa growing regions, highlighted in the box plots, the changes are significant at the 95% level. Small decreases (<10%) are seen in the coastal and northwest regions, in which precipitation is projected to decline. In the light of previous studies (Laderach et al 2013, Schroth et al 2016), the increases in NPP might seem unexpected, given that temperatures are higher and the projected precipitation changes are small.

Zoom In Zoom Out Reset image size

To investigate further, a series of idealised experiments was conducted, in which temperature, precipitation and specific humidity were altered in turn for both present day and elevated CO₂ (figures 7 and S2.1). In these experiments, temperature was increased by a uniform value, and precipitation and humidity were altered by a uniform proportion. Two sets of scenarios were considered, representing CMIP5 RCP8.5 average temperature, precipitation and humidity changes for the middle and end of the 21st Century, under an RCP8.5 scenario. Additional sets of simulations, in which the vegetation suffers no water stress at all—i.e. $\beta$ is kept at 1 throughout the simulation period (equation (4)), were carried out for both scenarios. Areal mean NPP for all the experiments for the region highlighted in figure 6 is shown in figure 7. Comparison between the idealized runs suggests that the projected increase in NPP is driven primarily by elevated CO₂, with the effect modulated by change in temperature, precipitation and humidity. As would be expected, NPP increases with higher precipitation and atmospheric humidity—both of which act to reduce water stress (which affects photosynthesis, see equation (4)). The sensitivity to water stress is confirmed by the high NPP in the simulations in which water stress is set to zero, i.e. $\beta = 1$. These effects are highlighted in figures S2.1(b) and S2.1(c). Here, the projected increase in specific humidity has a greater effect on NPP than the projected increase in precipitation, especially for the end of century scenarios. It should be noted that because of the large warming, in the full future climate scenarios, the projected increase in specific humidity is still generally associated with a reduction in relative humidity, and hence an increase in vapour pressure deficit (D). In the model, the effect of increased D is to reduce the internal concentration of CO₂ and hence to reduce the rate of photosynthesis (equation (2)).

Zoom In Zoom Out Reset image size

The response to regional temperature change is more difficult to anticipate. GPP and hence NPP will be reduced if the temperature is either well below or well above the optimum for photosynthesis. As temperature increases towards the optimum, the rate of photosynthesis increases and then decreases sharply once the optimal temperature is exceeded (Lambers and Oliveira 2019). In West Africa, field studies have shown a weak positive correlation between temperature and cocoa production (Ojo and Sadiq 2010, Lawal and Emaku 2007), suggesting that in the present day, the optimum temperature for cocoa is not frequently exceeded in hot years, and also that the temperature is not so far below the optimum that it becomes the primary limitation on yield. This is consistent with the relationship between temperature and GPP in the model experiments (figure 8), which shows that for the West African climate, GPP is not heavily dependent on daily temperature. Note that we compare GPP, rather than NPP against temperature in order to make inferences about changes in the optimal temperature for photosynthesis. At the highest temperatures, however, there is a sharp drop off in GPP, suggesting that the optimal temperature for photosynthesis is exceeded on the hottest days in both the present and future climate.

Zoom In Zoom Out Reset image size

Based on these results, we would expect that if temperature were to increase by the amount projected in RCP8.5 simulations by the end of the century, GPP would be low because temperatures are frequently well above the optimum. And indeed, this behaviour is evident in the factorial experiments (figure 7). However, the factorial experiments also show that when CO₂ is elevated, GPP at high temperatures is similar to that simulated for present day conditions. The reason for this can be inferred from figure 8, which plots GPP against temperature for present day and future conditions (with no water stress). This comparison suggests that the optimal temperature increases as CO₂ increases and that the temperatures found in the future are not as far beyond the optimum as would be expected if the optimal temperature was the same as for the present day. This shift in optimal temperature under elevated CO₂ is an emergent property of the land surface model, related to changes to the limiting conditions on photosynthesis (Sage and Kubien 2007).