This study proposes a methodology for selecting global climate models (GCMs) from the Coupled Model Intercomparison Project phase 6 (CMIP6) on the basis of their ability to simulate characteristics of mean and extreme precipitation over North-East Africa and Arabia. The seasonal climatology, annual rainfall cycles, and spatial and temporal variability (as documented by five ETCCDI indices) of twenty-five GCMs have been assessed against rain gauge observations and seven gridded rainfall products. Most of the GCMs simulate reasonably well the climatology of mean rainfall (annual and seasonal totals and number of rainy days). Large discrepancies are found in the reference products for some indices related to rainfall intensity (SDII, P95 and R95ptot), which is a major concern for the validation of GCMs. For these indices, we evaluate whether historical CMIP6 simulations fall within the uncertainty range of the rainfall estimates. Ten CMIP6 models are finally retained based on their ability to reproduce the geography and seasonality of mean and extreme rainfall. They tend to have a higher spatial resolution, although there is no systematic relationship between resolution and skill. The selected CMIP6 models perform better, not only at the regional scale (by construction), but also, and more meaningfully, at the local scale of the Republic of Djibouti particularly for the March-to-May rainy season. In a companion paper, the projections of the selected CMIP6 models will be used to study future changes in mean and extreme precipitation over North-East Africa and Arabia.
This study provides a first analysis of future changes in mean and extreme precipitation over Northeast Africa and the Arabian Peninsula. To this aim, we exploited projections from ten selected CMIP6 models (in part I) under various SSP greenhouse gases emission scenarios for the mid to late 21st century. We found a north-south differentiation in future changes in total precipitation for the JF and MAM seasons, with decreases in the north and moderate increases elsewhere, although model uncertainties are high, particularly for MAM. In contrast, the JJAS and OND seasons show larger positive changes with less model uncertainty. These increases in mean precipitation will be accompanied by an increase in the intensity and frequency of extreme precipitation events. In addition, the JJAS (+6.7%) and OND (+4.5%) seasons will contribute more to cumulative annual precipitation, while the JF (-1.3%) and MAM (-9.9%) seasons will experience a reduction. Over Djibouti, where the selected models are shown to perform well, downscaled and bias-corrected CMIP6 data using the CDF-t method indicate in addition that the return period of intense precipitation events (≥ 80 mm/day) causing documented flooding will decrease from 5 years historically to 1.4 years by the end of the 21st century under the SSP5-8.5 scenario. This robust result indicates the need to strengthen flood adaptation measures in Djibouti. Furthermore, similar downscaling exercises are recommended for other sub-regions in Northeast Africa and Arabia, given the consistent trend towards higher intensity rainfall.