Evaluating the performance of new CORDEX-Africa regional climate models in simulating West African rainfall

Abstract

This study examines the performance of the outputs of three RCMs (CCLM, REMO, RegCM) that participated in CORDEX phase 2 to dynamically downscaling three CMIP5 GCMs at a resolution of 0.22° × 0.22° in replicating West African rainfall from 1970 to 2005. Using different statistical methods and climatological description at seasonal and annual scale, these nine models, downscaling institution-based ensembles mean (CCLM-ENSEMBLE, REMO-ENSEMBLE, RegCM-ENSEMBLE) and the grand ensemble mean (ENS-RCM), were compared to the ensemble mean (ENS-OBS) of the two observations. In simulating the spatial distribution of mean seasonal rainfall over West Africa, outputs of the three CMIP5 GCMs dynamically downscaled by RegCM and RegCM-ENSEMBLE consistently performed better than other models, just like ENS-RCM with little bias. HadGEM-CCLM, HadGEM-REMO, HadGEM-RegCM, MPI-CCLM, and CCLM-ENSEMBLE showed better performance in replicating annual rainfall cycle evident by the two rainfall peaks simulated over Guinea Coast and a single peak over the Savannah and Sahel, unlike the grand ensemble mean. In MAM and SON seasons, the statistical performance evaluation gives a better agreement between ENS-OBS and most of each model, downscaling institution-based ensemble mean and ENS-RCM with r ≥ 0.89. It is further observed that some individual models and ensemble mean from different downscaling institutions outperform the grand ensemble in replicating mean seasonal rainfall over the three climatic zones and entire West Africa in DJF and JJA seasons. Moreover, the annual statistical evaluation revealed a poor performance with low correlation values over the Guinea Coast and better performance over the Sahel by all the models and various ensemble mean. Similar to what is observed with ENS-OBS trend assessment, only MPI-REMO adequately simulates the observed decreasing and increasing trend that is not statistically significant over the south and north of 10° N, respectively. These show that ensemble mean may reduce or amplify uncertainty in climate models. Hence, these results accentuate that using ensemble mean will reduce uncertainty in climate models in projecting rainfall over West Africa and over different periods which may not be necessarily true.