Dynamic bayesian networks for rainfall forecasting

In this paper we deal with the problem of forecasting local rainfall at multiple meteorological stations over the Iberian peninsula. To this aim a dynamic Bayesian network is introduced for relating rainfall to broad-scale atmospheric circulation patterns. In this way statistical historic information gathered at the available stations is combined with numerical atmospheric predictions developed at different weather services, resulting a single consensus prediction. This technique can be considered an hybrid statistical-numerical method for precipitation downscaling (predicting local values based on broad-scale grided predictions), and can be easily adapted to other meteorological variables of interest. Eje: Informática teórica Red de Universidades con Carreras en Informática (RedUNCI)