Probabilistic Near‐Real‐Time Retrievals of Rain Over Africa Using Deep Learning

Abstract We introduce Rain over Africa (RoA), a public retrieval algorithm providing near‐real‐time precipitation estimates over the African continent. The retrievals are based on Meteosat thermal infrared observations. Therefore, rain can be monitored constantly, minutes after input data dissemination. Despite this low latency, RoA accuracy is comparable to estimates requiring hours or more to obtain. Consequently, RoA is of particular interest where a rapid response is critical, such as for disaster preparedness. RoA retrievals employ a convolutional and quantile regression neural network: the latter enables detailed case‐specific descriptions of the retrieval uncertainty. Four years of data from the calibration satellite in the GPM mission were used as training and evaluation labels. With this setup, limitations in earlier near‐real‐time retrievals for Africa were overcome. Moreover, the RoA network runs on regular workstations. With a 30‐km effective resolution, RoA retrievals over land are more timely and detailed than the established IMERG precipitation estimates. RoA is also applicable over the surrounding ocean regions, maintaining a similar performance. However, there IMERG exhibits a better effective resolution, at least for its more favorable conditions. Additionally, RoA's probabilistic nature enables addressing the inherent uncertainties of satellite precipitation retrievals by using probabilities of exceeding precipitation thresholds. Further assessment reveals similar diurnal cycles between RoA and IMERG, although IMERG shows some instability. Visual inspection of rain evolution patterns also indicates that RoA is more consistent. Finally, an annual mean analysis including CHIRPS estimates shows regional differences among the three, with no clear outlier behavior for RoA.