South African maize production scenarios for 2055 using a combined empirical and process-based model approach

In South Africa, a semi-arid country with a diverse agricultural sector, climate change is projected to negatively impact staple crop production. Our study examines future impacts to maize, South Africa's most widely grown staple crop. Working at finer spatial resolution than previous studies, we combine the process-based DSSAT4.5 and the empirical MAXENT models to study future maize suitability. Climate scenarios were based on 9 GCMs run under SRES A2 and B1 emissions scenarios down-scaled (using self-organizing maps) to 5838 locations. Soil properties were derived from textural and compositional data linked to 26422 landforms. DSSAT was run with typical dryland planting parameters and mean projected CO2 values. MAXENT was trained using aircraft-observed distributions and monthly climatologies data derived from downscaled daily records, with future rainfall increased by 10% to simulate CO2 related water-use efficiency gains. We assessed model accuracy based on correlations between model output and a satellite-derived yield proxy (integrated NDVI), and the overlap of modeled and observed maize field distributions. DSSAT yields were linearly correlated to mean integrated NDVI (R2 = 0.38), while MAXENT's relationship was logistic. Binary suitability maps based on thresholding model outputs were slightly more accurate for MAXENT (88%) than for DSSAT (87%) when compared to current maize field distribution. We created 18 suitability maps for each model (9 GCMs X 2 SRES) using projected changes relative to historical suitability thresholds. Future maps largely agreed in eastern South Africa, but disagreed strongly in the semi-arid west. Using a 95% confidence criterion (17 models agree), MAXENT showed a 241305 km2 suitability loss relative to its modeled historical suitability, while DSSAT showed a potential loss of only 112446 km2. Even the smaller potential loss highlighted by DSSAT is uncertain, given that DSSAT's mean (across all 18 climate scenarios) projected yield change within this area is +7.7% (range: -28 to 54%), and only 66182 km2 show negative yield changes. In terms of new suitability, both models show small increases relative to their modeled historical suitability, mostly in the southeast (DSSAT = 5197 km2, MAXENT = 10111 km2). Within DSSAT's 95% confidence area for future suitability, the mean average yield increase is 17.8%. Within the region jointly identified as suitable by 95% of DSSAT and 95% of MAXENT scenarios, yield increases averaged 22.0%. The model discrepancies regarding future maize suitability relate to crop CO2 responses-DSSAT directly incorporates CO2, but MAXENT does not. However, DSSAT may overestimate water-use efficiency gains, and thus future suitability/yield. Nevertheless, these results suggest that South African maize production may be less impacted than previous climate impact studies suggest, and potential future losses could be offset by technology gains. Results also suggest that empirical models may tend to overestimate negative climate change impacts.