Validation of Satellite Rainfall Estimates over Equatorial East Africa

Abstract

Rain gauge data sparsity over Africa is known to impede the assessments of hydrometeorological risks and of the skill of numerical weather prediction models. Satellite rainfall estimates (SREs) have been used as surrogate fields for a long time and are continuously replaced by more advanced algorithms and new sensors. Using a unique daily rainfall dataset from 36 stations across equatorial East Africa for the period 2001–18, this study performs a multiscale evaluation of gauge-calibrated SREs, namely, IMERG, TMPA, CHIRPS, and MSWEP (v2.2 and v2.8). Skills were assessed from daily to annual time scales, for extreme daily precipitation, and for TMPA and IMERG near-real-time (NRT) products. Results show that 1) the SREs reproduce the annual rainfall pattern and seasonal rainfall cycle well, despite exhibiting biases of up to 9%; 2) IMERG is the best for shorter temporal scales while MSWEPv2.2 and CHIRPS perform best at the monthly and annual time steps, respectively; 3) the performance of all the SREs varies spatially, likely due to an inhomogeneous degree of gauge calibration, with the largest variation seen in MSWEPv2.2; 4) all the SREs miss between 79% (IMERG-NRT) and 98% (CHIRPS) of daily extreme rainfall events recorded by the rain gauges; 5) IMERG-NRT is the best regarding extreme event detection and accuracy; and 6) for return values of extreme rainfall, IMERG, and MSWEPv2.2 have the least errors while CHIRPS and MSWEPv2.8 cannot be recommended. The study also highlights improvements of IMERG over TMPA, the decline in performance of MSWEPv2.8 compared to MSWEPv2.2, and the potential of SREs for flood risk assessment over East Africa.