Accuracy of satellite and reanalysis rainfall estimates over Africa: A multi-scale assessment of eight products for continental applications

Continental Africa This study evaluates the accuracy of eight gauge-corrected rainfall products across Africa through direct comparisons with in situ observations for the period 2001–2020. The effect of validation datasets on the performance of the rainfall products was also quantified in ten African countries. Four categorical and five continuous metrics were estimated at multiple spatial and temporal scales as part of the evaluation. Results indicate that the performance of the rainfall products varied in space and time. Evaluation at temporal scales revealed that, on average, most rainfall products showed poor results (KGE < 0.35) at the daily timescale. In contrast, RFE v2.0, ARC v2.0, and MSWEP v2.8 were reliable (KGE > 0.75) at the monthly and annual timescales. Among the rainfall products, the performance of TAMSATv3.1, PERSIANN-CDR, and ERA 5 was relatively poor in capturing in situ observations. Evaluation at various spatial scales revealed mixed results. The ARC v2.0 and CHIRPS v2.0 rainfall products were reliable in detecting no rains (< 1 mm/day) for all 19 spatial scales, indicating a high level of confidence for drought studies. IMERG-F v6B and RFE v2.0 were reliable in detecting heavy and high-intensity rainfall events for all spatial scales. Using the KGE performance metrics at the regional level, MSWEP v2.8 in the Northern Africa region, RFE v2.0 in the Western and Southern Africa regions, ARC v2.0 in Central Africa, and CHIRPS v2.0 in the Eastern Africa region showed better performances at monthly timescale. Moreover, the performance of the gauge-corrected rainfall datasets was reduced when compared with independent validation data (gauge data not used by rainfall products) than dependent validation data. This study provides several new insights into choosing a rainfall product for continental to regional applications and identifies the need for bias correction.