The performance of regional climate models driven by various general circulation models in reproducing observed rainfall over East Africa

Abstract

Regional climate models (RCM) are commonly used to downscale the coarse resolution general circulation models (GCMs) to produce climate variables at spatially high-resolution grids. The quality of the downscaled data depends on the skills of both GCMs and RCMs. In this study, 10 GCMs are used to constrain the boundary and provide initial conditions of three RCMs. A total of 18 GCM-RCMs combinations are employed to produce simulations over East Africa (EA). The accuracy of simulated rainfalls is evaluated with respect to Climate Research Unit (CRU) rainfall to identify the best GCM-RCM combinations. Bias, root mean squared error (RMSE), correlation coefficient, and MAE-based model skill score have shown that MPI-REMO, MIROC-REMO, MPI-RCA4, IPSL-RCA4, CCCMA-RCA4, MOHC-CCLM, MOHC-REMO, and CNRM-RCA4 during spring season; ICHEC-REMO, MIROC-REMO, MOHC-REMO, MIROC-RCA4, CSIRO-RCA4, and MPI-REMO during autumn season; CSIRO-RCA4, MIROC-RCA4, CCCMA-RCA4, MIROC-REMO, CNRM-RCA4, and MOHC-RECA during boreal summer; and ICHEC-REMO, NOAA-RCA4, MOHC-REMO, MOHC-CCLM, MIROC-REMO, MPI-REMO, and IPSL-RCA4 during boreal winter season are the best performing GCM-RCM combination. It is also evident that the skills of the models are better in autumn than their skills in boreal spring and summer. Moreover, summer rain in EA is the most difficult for models to simulate. Comparison of annual mean with the CRU rainfall shows that MPI-REMO, MIROC-REMO, CSIRO-RCA4, MOHC-REMO, CCCma-RCA4, IPSL-RCA4, and CNRM-RCA4 are also the best GCM-RCM combinations as observed from strong significant spatial correlation, as well as low bias, RMSE, and positive skill score as high as 0.7. Therefore, the GCM-RCM combinations that exhibit superior performance over EA in most seasons as well as in capturing observed annual mean are CCCMA-RCA4, MIROC-REMO, MPI-REMO, IPSL-RCA4, CSIRO-RCA4, MOHC-REMO, and MIROC-RCA4. The difference in skills between models as well as variation of the same model skill both spatially and seasonally implies the role of several factors such as local topography, vegetation, and surface type as well as robustness of model physics in capturing small scale processes such as mesoscale convection in boreal summer (e.g., over Ethiopian highlands).