Multi-sensor Satellite Precipitation Estimate for Hydrogeological Hazard Mitigation

High-impact meteorological events have in the last decade received increasing interest and considerable efforts are constantly undertaken to mitigate their effects on human activities and environment. Several projects addressing different aspects of the risk mitigation strategy have been financed in Europe, and PROSA (Prodotti di Osservazione Satellitare per l'Allerta Meteorologica - Satellite products for meteorological alert), funded by the Italian Space Agency (ASI), represents the Italian attempt to solve the meteorological side of the hazard mitigation scheme. It is devoted to design, develop, test and demonstrate a prototype system dedicated to the innovative dynamic characterization of meteorological parameters at the ground by means of satellite data. This work is part of PROSA and the main objective is the implementation and optimization of three di�erent satellite precipitation estimation algorithms. The algorithms are based on Artificial Neural Networks and correlate multi-sensors satellite data, in the Visible, Infrared (from the European Geostationary satellite Meteosat) and Microwave bands (from polar orbiting satellites), to the precipitation rate at ground. The ANNs are set up as classification problem and use rain-gauges data as true values of precipitation at the ground for the training, testing an validation of the techniques. The work is divided in three main steps: the first version of the algorithm gives a binary classification of satellite pixel as rain and no-rain classes, with seasonal and day-time characterization of the precipitation maps. The second version gives a quantitative estimate, classifying the rain-rate in five precipitation intervals. Finally, the last version provides precipitation maps with quantitative values expressed in mmh^-1, and also explicitly uses microwave data. To reach the main objective several sensitivity studies and intermediate goals have been pursued, in order to refine and tune the technique. The sensitivity to precipitation of the infrared channels with respect to the seasonal cycle and the impact of the visible channels on the estimates have been assessed. The relationship between the probability of precipitation, output of the neural network, and the rain-rate, as measured by rain-gauges, has been established for warm and cold months, and the optimal way to ingest in the algorithm the microwave estimates has been defined by analyzing the di�erent performances of microwave and visible-infrared techniques. Finally, the results have been critically discussed in comparison with other algorithms taking part of the PROSA system.