Nearly instantaneous probabilistic retrievals of Rain over Africa

How much has rained in the last minutes anywhere in Africa? Contrary to Europe or North America, a dense network of weather radars providing this information is not available on the African continent. A new product, Rain over Africa, aims to find an answer to the question by retrieving rain rates from Meteosat geostationary infrared images and making them available to the public within minutes from satellite downlink. By using geostationary observations, rain retrievals with a resolution of 3-5 km and 15 min update time can be offered. Machine learning is at the core of Rain over Africa. The GPM DPR and GMI combined precipitation L2B product was exploited to train a convolutional neural network. The trained model outputs a pixel-wise rain rate distribution free from traditional assumptions, enabling not only point estimates such as an expected value, but also non-Gaussian error estimates or likelihoods of extreme events by computing tail probabilities. Moreover, the Rain over Africa retrievals compare similar to the IMERG Late Run product, but can offer additional statistics at a finer spatiotemporal resolution, with a product latency of few minutes instead of hours. Further details on the model and its performance, characteristics of the Rain over Africa product, how to access the data, and data availability, combined with a product outlook, will be given in this presentation.