92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Monday, 23 January 2012
Neural Ensemble Bayesian Nowcasting of Geostationary Multispectral Imagery for Hydro-Meteorological Applications
Hall E (New Orleans Convention Center )
Michele de Rosa, "Sapienza" University of Rome, Rome, Italy; and F. S. Marzano and A. Eleuteri

The modern Numerical Weather Prediction (NWP) models, used to predict the weather conditions, work on large scale both in time and space and they are initialized few times a day. On the other hand, events, like thunderstorms, develop on small scales because they last from a few minutes to a few hours and from a few meters to some kilometres. For these reasons, it is clear that the NWP models are insufficient in order to achieve a good prediction of the extreme meteorological events and it is necessary to use another kind of models, which are able to give high resolution (in space and time) predictions with a given degree of trust; in other words to make nowcasting. The objective of this work is to propose, develop and validate a new predictive model, based upon satellite observation, which is able to make nowcast over an Area Of Interest (AOI). In particular the satellite source used is the Meteosat Second Generation (MSG) satellite. Its high temporal and space resolution gives the possibility to satisfy some of the requirements needed to nowcast the development of extreme events and, in a more wide view, gives the possibility to monitor the environment in an efficient way and to plan actions in case of dangerous events. This work will discuss a new kind of model and it will show one of its possible applications: rain field reconstruction. The prediction model is based on the ensemble framework, which describes computation by a set of simpler cooperating models, while the rain field estimator is based on the neural networks framework. The results, about a real time working system, is presented and discussed.

Supplementary URL: