Evaluation of CMIP6 models in simulating seasonal extreme precipitation over Ethiopia

Historically, Ethiopia has experienced recurrent droughts and floods, which may intensify due to climate change. This study has evaluated the performance of 45 models from the Coupled Model Intercomparison Project Phase 6 (CMIP6) in simulating ten extreme precipitation indices against corresponding indices from the Enhancing National Climate Services (ENACTS) during short rainy (February–May, FMAM) and main rainy (June–September, JJAS) seasons for the period 1981–2014 over Ethiopia. Ensemble mean of the top-ranking models are also evaluated against ENACTS in reproducing extreme indices over five Agro-ecological zones (AEZs) of the country. The Taylor Skill Score (TSS) was used to rank the performance of the individual CMIP6 models for JJAS and FMAM seasons with respect to ENACTS while Comprehensive Rating Metrics (RM) were used to compute the overall ranks of the models. Our results show that most CMIP6 models reasonably captured the spatial distribution of the seasonal extreme precipitation indices even though they could not reproduce the magnitude of indices, especially in the highland and high rainfall areas of the country such as Northwest and west parts of the country. However, the biases in lowland and low rainfall regions, such as the eastern and northeastern parts of the country, are smaller compared to other areas. More than 30 CMIP6 models underestimated the extreme indices with the exception of consecutive wet days which is grossly overestimated in the highland and high rainfall areas specifically in western parts of the country. Additionally, EnseMean in the tropical and desert AEZs performs particularly better in simulating extreme indices compared to other AEZs. The ensemble mean of the top-ranking models (EnseMean) generally outperformed both individual models and ensemble of all models in the representation of observed extreme indices across all metrics and seasons. Moreover, the performance of individual models is subject to variation based on the season, and the selected extreme indices. It is also noteworthy that their performance is relatively less influenced by horizontal resolution. Further evaluation, focusing on teleconnections such as ENSO and IOD, is a crucial next step for evaluating models and creating a sub-ensemble.