Documentation

Summary

This product, developed at Chalmers University of Technology (Sweden), provides probabilistic retrievals of precipitation over Africa by an artificial neural network. The network consists of a fully convolutional, quantile regression neural network, trained using multi-task learning and nearly four years of collocations of SEVIRI level 1.5 data and the GPM DPR and GMI Combined Precipitation L2B. The network infers the precipitation probability distribution for each pixel in a re-gridded SEVIRI image, using all infrared channels and the satellite zenith viewing angle. This enables retrievals independent of the time of day. Only retrievals for the last seven days are stored.

Full documentation

This product will likely be documented in a scientific article or a technical report. In the meantime, the following documents are related:

• Hee, L.: Rain over Africa. An application of quantile regression neural networks to retrieve precipitation from geostationary satellites, Chalmers ODR, https://hdl.handle.net/20.500.12380/305472, 2022.
• Pfreundschuh, S., Ingemarsson, I., Eriksson, P., Vila, D. A., and Calheiros, A. J. P.: An improved near-real-time precipitation retrieval for Brazil, Atmos. Meas. Tech., 15, 6907-6933, https://doi.org/10.5194/amt-15-6907-2022, 2022.