Changes in African extreme precipitation and storm structure since 1981 revealed by satellite observations

Quantifying past changes in extreme precipitation is crucial to understand the climate response and improve our projections. Due to the limited data availability, information about Africa is of particular interest. We combine high-resolution satellite estimates (CHIRPSv2) with an innovative approach for the detection and attribution of trends in extremes (both annual maxima and rarer events, such as the 100-year return levels) to investigate changes in daily precipitation extremes and storm structure occurred over Africa since 1981. Scale-dependence is explored by comparing trends detected at the local (0.05° resolution) and meso- (1°) scales. The statistical model was validated using a gauge-based dataset (GPCC) before application to satellite estimates. 

Roughly ~20% of the continent experienced significant (p=0.05) changes in annual maxima at both scales. Decreasing trends are observed in the central portion of the continent, and increasing trends in the Sahel and some districts in southern and eastern Africa. Storms tended to become spatially smoother, with faster decreases at the local scales (median=13% faster for annual maxima, 14% for 100-year return levels) and faster increases at the mesoscales (17% for annual maxima, 16% for 100-year return levels). The 100-year return levels increased 33% (25% at the mesoscale) faster than annual maxima and decreased 43% (45%) faster.

The model explains 89% (91% at the mesoscale) of the variance in the observed significant trends. Changing proportions between heavy and mild events explain 25% (38%) of this variance, changes in the overall intensities 13% (21%), and changes in the number of wet days 4% (12%). About ~25% of the area experienced significant trends in at least one model parameter, although no significant trend could be detected in the maxima. Censoring annual maxima, the model still explains 77% of the variance in their trends, suggesting it could be effectively used in situations in which observed/modelled extremes are not trusted.