A Multi-model Bias-corrected Large-Ensemble for High-resolution Climate Impact Assessment in Sub-Saharan Africa

Abstract

Sub-Saharan Africa (SSA) is increasingly exposed to unprecedented climate extremes, posing critical challenges to water and food security. Hydrological and agricultural climate-change impact assessments commonly rely on downscaled and bias-corrected climate model simulations to drive hydrological and sectoral impact models. In many regions, including Sub-Saharan Africa, existing studies predominantly apply bias correction to single realisations from multi-model climate ensembles, which limits the explicit sampling of internal climate variability and constrains robust quantification of climate-change impacts and associated uncertainty. To account for internal variability in regional climate change projections, single model initial-condition large ensembles (SMILEs) can be used. Across diverse case studies in Europe and North America, different approaches have been developed to downscale and bias-correct SMILEs while preserving internal climate variability. However, these approaches have so far been applied almost exclusively to individual SMILEs, have not been extended to multiple SMILEs within a unified bias-correction framework, and remain unexplored in the SSA context. This study presents the first multi-model, bias-corrected large-ensemble for high-resolution climate impact assessment in Sub-Saharan Africa, using Uganda as a demonstrative case study. The framework integrates six CMIP6 SMILEs (MPI-ESM1-2-LR, ACCESS-CM2, IPSL-CM6A-LR, MIROC6, CanESM5, and UKESM1-0-LL), together providing more than 150 climate simulations sampling both internal climate variability and inter-model structural uncertainty. Bias correction is applied at monthly scale using the CDF-t method, following the ensemble-based and individual-member-based implementations proposed by Ayar et al. (2021). The correction functions are trained over the historical period 1950–2014, using ERA5-Land as the reference dataset, resulting in bias-corrected regional climate scenarios at 8 km spatial resolution. The resulting bias-corrected multi-model large ensemble is intended for use in hydrological and agricultural impact modelling over selected Ugandan catchments to support future analyses of hydroclimatic change, variability, and extremes. Beyond this case study, the framework is designed as a scalable prototype for the future development of a pan-SSA multi-model, bias-corrected large-ensemble climate dataset to support climate-impact assessments and adaptation planning.