RainRunner: A Deep Learning satellite rainfall retrieval model for West Africa

Food and economic safety in West Africa rely heavily on rainfed agriculture and are threatened by climate change and demographic growth. Accurate rainfall information is therefore crucial to ensure safety against these challenges. However, existing rainfall models fail to accurately represent the highly variable and sparsely monitored West African rainfall distribution. Satellite rainfall products show a poor correlation with ground-based rainfall measurements and literature suggests that atmospheric aerosols are partly to blame for this poor performance.

To address this challenge, we propose a Deep Learning (DL) model that utilizes satellite water vapor (WV) and Thermal Infrared (TIR) observations in conjunction with temporal information for satellite rainfall retrieval in West Africa. We leverage the TIR and WV channels of the Meteosat Second Generation satellite to develop a DL model for satellite rainfall detection. Our results indicate that incorporating WV data into the DL framework enables the detection of strong convective motions typically associated with heavy rainfall. This is particularly relevant in regions where convective rainfall is dominant, such as the tropics.

Furthermore, the WV data facilitate the identification of dry air masses advected from the nearby Sahara Desert, which often create discontinuities in precipitation events over our study area. The ability to detect such dry air masses is a significant advantage of our proposed DL model, as it aids to reduce false alarms and rainfall overestimation as compared to methods that rely only on TIR data.

Our DL model achieves a robust performance in rainfall binary classification, with fewer false alarms and lower rainfall overdetection (FBias < 2.0) than the state-of-the-art Integrated Multi-satellitE Retrievals for GPM (IMERG) Final Run. Our findings suggest that incorporating WV observations and temporal information in a DL framework efficiently complement TIR observations and enhances the accuracy of satellite rainfall retrieval in West Africa.