A New Data Fusion Neural Network Scheme for Rainfall Retrieval Using Passive Microwave and Visible/Infrared Satellite Data

Abstract

A new data fusion technique based on Artificial Neural Networks (ANN) for the design of a rainfall retrieval algorithm is presented. The use of both VIS/IR (VISible and InfraRed) data from GEO (Geostationary Earth Orbit) satellite and of passive microwave data from LEO (Low Earth Orbit) satellite can take advantage of both types of sensors reducing their limitations. The technique can reconstruct the surface rain field with the MSG-SEVIRI (Meteosat Second Generation–Spinning Enhanced Visible Infrared Imager) spatial and temporal resolution, which means 3 km at the sub satellite point and 5 km at mid-latitudes, every 15 min, respectively. Rainfall estimations are also compared with H-SAF (Hydrology Satellite Application Facility) PR-OBS3A operational product showing better performance both on the identification of rainy areas and on the retrieval of the amount of precipitation. In particular, in the considered test cases, results report an improvement in average of 83% in terms of probability of rainy areas detection, of 45% in terms of false alarm rate, and of 47% in terms of root mean square error in the retrieval of the amount of precipitation.

Keywords

• precipitation
• hydrology
• meteorological satellites
• global precipitation measurement mission
• MSG
• SEVIRI