Characterization and Projection of Extreme Precipitation Events over Africa

Abstract

In this study, we analysed trends of daily precipitation-based indices in the ensembles of Global and Regional Climate Models (GCMs and RCMs) for historical and future climate projection. The specific objectives of the work were to investigate the variability and trend of extreme precipitation, assess models capability for extreme precipitation studies over Africa and make projections of future extreme precipitation. First, we evaluated the ability of the state-of-the-art GCMs and RCMs to reproduce the mean and spatial characteristics of extreme precipitation indices over the Africa domain. In particular, the extent to which CORDEX (COordinated Regional Downscaling Experiment) adds useful details on the performance of CMIP5 (Coupled Model Intercomparison Project Phase 5) multimodel ensemble was investigated. Comparison of the present day simulation was performed with two precipitation observation datasets, the high resolution TRMM (Tropical Rainfall Measuring Mission) and coarse resolution GPCP (Global Precipitation Climatology Project), to evaluate models strengths and weaknesses. Trends of changes in extreme precipitation indices in the 21st century under the most extreme IPCC (Intergovernmental Panel on Climate Change) emission scenario (RCP8.5 - Representative Concentration Pathway), and projected by ensembles of both CMIP5 GCMs and CORDEX RCMs were also examined. Eight indices generated from absolute (1mm) and percentile (95th) based thresholds as defined by the Expert Team on Climate Change Detection and Indices (ETCCDI) were computed for seventeen CMIP5 GCMs and six CORDEX RCMs (for twelve downcaling experiments) for each year during the historical (1975 - 2004) and future (2006 - 2099) periods, before long-term means and multimodel ensembles were applied. For comparison purposes, both the validation datasets and model outputs were interpolated onto the GPCP grid (100 km) through a bilinear interpolation. Statistical evaluation metrics including mean bias - MB, standard deviation, centered rootmean- square error (RMSE) and correlation were performed over three sub-regions (Sahel, Northern East Africa and Central Southern Africa) having different characteristics of the annual cycle of rainfall.

Our results suggested a good consistency between GPCP and TRMM in producing annual mean and frequency of extreme precipitation events over Africa despite few inconsistencies. However, their associated intensities largely differed from one another with GPCP data displaying drier bias. Overall, multimodel ensembles simulations overestimated the frequency of extreme precipitation events and underestimated their intensities. The results further showed that CMIP5 exhibits wet bias of precipitation events and drier bias of precipitation intensity than CORDEX, and that CORDEX produces precipitation magnitude within the range of the observations and more in line with the higher resolution TRMM data. This illustrates the added value achieved with the higher resolution CORDEX multimodel ensemble for the simulation of such events, and points towards the use of these RCMs ensemble to study extreme precipitation for a better assessment of climate change over Africa. Simulations of percentile-base precipitation indices (R95, R95p and R95ptot) in the 21st century provided a consistent picture of a predominant shift towards a more frequent and more intense precipitation in most parts of the continent, in response to increase radiative forcing as in the RCP8.5. Whilst for the 1mm threshold-base precipitation indices, the simulations predicted less frequent and more intense precipitation events. The increasing trends are larger and more spatially consistent over the tropics than extra-tropical regions, and are spatially more heterogeneous in the CORDEX ensemble simulation than the CMIP5 ensemble for intense precipitation, whilst for the precipitation events the magnitude is higher in the CMIP5 ensemble. This study concluded that the ensemble of high resolution CORDEX models have high ability to reproduce extreme precipitation characteristics over the African domain. They improved on the CMIP5 GCMs ensemble simulations, and closely documented the likely changes in extreme precipitation during the 21st century.