RAIN-RATE ESTIMATION FROM SEVIRI/MSG AND AMSR-E/AQUA. VALIDATION AND COMPARISON BY USING U.K. WEATHER RADARS

The satellite rainfall estimation algorithm here proposed is based on a statistical approach (Artificial Neural Network: ANN): it needs only radiation satellite measurements as input data and provides, at 5 km of spatial resolution, surface rain-rate classification onto five classes of precipitation: [less than 1/32] mm/h (no rain), [1/32, 0.125] mm/h (slight rain), [0.125, 0.5] mm/h (slight/moderate rain), [0.5, 2.0] mm/h (moderate), [more than 2.0] (heavy rain). The algorithm works for U.K. area, daytime and summer season and adopts a cascade method where at first a rain no-rain classification is computed and then a similar yes-no classification is computed for the other pair of classes. It has been developed and validated with the use of U.K. weather radar rainfall estimates for both SEVIRI (on the geosynchronous Meteosat-8 satellite) and AMSR-E (on the low earth orbit AQUA satellite) sensors. To assess the performance against radar rainfall estimation some skill indicators are computed: the Equitable Threat Score (ETS) and BIAS are used for pair of classes of precipitation whereas the Heidke Skill Score (HSS) is used for the four raining classes. The validation procedure (over U.K. area and for June, July and August 2004, at noon time), shown that the nine channels SEVIRI classifier provides performances very close to the ones provided by the twelve channels AMSR-E classifier. The advantage of using AMSR-E measurements is more evident when only sea area is considered. The analysis also show which are the best sets of channels (among the nine SEVIRI channels and the twelve AMSR-E channels) that give the most important contribute to the above performances.