Artificial neural-network technique for precipitation nowcasting from satellite imagery

摘要

The term nowcasting reflects the need of timely andaccurate predictions of risky situations related to the development ofsevere meteorological events. In this work the objective is the very shortterm prediction of the rainfall field from geostationary satellite imageryentirely based on neural network approach. The very short-time prediction(or nowcasting) process consists of two steps: first, the infrared radiancefield measured from geostationary satellite (Meteosat 7) is projected aheadin time (30 min or 1 h); secondly, the projected radiances are used toestimate the rainfall field by means of a calibrated microwave-basedcombined algorithm. The methodology is discussed and its accuracy isquantified by means of error indicators. An application to a satelliteobservation of a rainfall event over Central Italy is finally shown andevaluated.