Assessing the impact of the NEX-GDDP CMIP6 bias correction on crop yield simulations across Africa

Abstract Biases in the Coupled Model Intercomparison Project Phase 6 (CMIP6) datasets pose a significant challenge to their effective application in climate impact assessments. Although bias correction techniques have been introduced to mitigate these issues, there is limited evidence regarding their effectiveness in improving crop simulations across Africa. This study evaluated the impact of bias correction applied to CMIP6 data within the NASA Earth Exchange Global Daily Downscaled Projections (NEX-GDDP) dataset on crop yield simulations across Africa. Thirteen NEX-GDDP simulations were assessed using the Global Meteorological Forcing Dataset (GMFD) as the reference. The crop yields of maize, sorghum, and pearl millet were simulated using the Decision Support System for Agrotechnology Transfer (DSSAT) model. DSSAT’s ability to reproduce the spatial distribution of yields was evaluated using GMFD as climate forcing and FAO yield data as the benchmark. Comparative analysis of DSSAT simulations forced with CMIP6 and NEX-GDDP datasets revealed that bias correction significantly improved yield simulations, as reflected in the relative root mean square error (RRMSE), which reduced from approximately 0.35 to 0.20, and percent bias (Pbias), which reduced from approximately − 15% to − 8% across maize, sorghum, and pearl millet simulations. However, performance varied across simulations: seven model yield simulations showed significant reductions in RRMSE (from 0.30 to 0.10) and Pbias (from − 13% to − 3%), together with a lower spatial bias relative to the reference data following bias correction. In contrast, six simulations underperformed, showing little or no improvement, with RRMSE remaining as high as 0.40 and Pbias reaching − 18%. Spatial bias plots further indicated yield declines of up to 700 kg ha⁻¹ in eastern Southern Africa after bias correction. Overall, bias correction improved simulated yields as a result of improvements in key climate variables, particularly precipitation and maximum temperature, which showed reduced Pbias and RRMSE across both the relatively small-bias group (RSB) and the relatively large-bias group (RLB). The overestimation of the minimum temperature (Tmin) following bias correction emerged as the primary factor distinguishing RSB from RLB. These findings underscore the value of bias correction in improving climate-driven crop simulations, while highlighting the need for further refinement to address residual biases, especially in vulnerable regions such as eastern Southern Africa.