Bias reduction in decadal predictions of West African monsoon rainfall using regional climate models

The West African monsoon rainfall is essential for regional food production, and decadal predictions are necessary for policy makers and farmers. However, predictions with global climate models reveal precipitation biases. This study addresses the hypotheses that global prediction biases can be reduced by dynamical downscaling with a multimodel ensemble of three regional climate models (RCMs), a RCM coupled to a global ocean model and a RCM applying more realistic soil initialization and boundary conditions, i.e., aerosols, sea surface temperatures (SSTs), vegetation, and land cover. Numerous RCM predictions have been performed with REMO, COSMO‐CLM (CCLM), and Weather Research and Forecasting (WRF) in various versions and for different decades. Global predictions reveal typical positive and negative biases over the Guinea Coast and the Sahel, respectively, related to a southward shifted Intertropical Convergence Zone (ITCZ) and a positive tropical Atlantic SST bias. These rainfall biases are reduced by some regional predictions in the Sahel but aggravated by all RCMs over the Guinea Coast, resulting from the inherited SST bias, increased westerlies and evaporation over the tropical Atlantic and shifted African easterly waves. The coupled regional predictions simulate high‐resolution atmosphere‐ocean interactions strongly improving the SST bias, the ITCZ shift and the Guinea Coast and Central Sahel precipitation biases. Some added values in rainfall bias are found for more realistic SST and land cover boundary conditions over the Guinea Coast and improved vegetation in the Central Sahel. Thus, the ability of RCMs and improved boundary conditions to reduce rainfall biases for climate impact research depends on the considered West African region. Key Points RCMs reduce Sahel rainfall bias of GCM predictions but raise Guinea Coast bias Coupled atmosphere‐ocean RCMs improve both Atlantic SST and WAM rainfall biases Improved SST, vegetation, and land cover boundary conditions reduce some biases