A neural network model for visibility nowcasting from surface observations: Results and sensitivity to physical input variables

A neural network model recently developed for fog nowcasting from surface o bservations is summarized in its features, paying attention to its particul ar learning structure (weighted least squares training), introduced because of the nonconstant errors associated with the estimation of visibility val ues. We apply it to a winter forecast of meteorological visibility in Milan (Italy). The performance of this model is presented and shown to be always better than persistence and climatology. Finally, we introduce a bivariate analysis and a network pruning scheme and discuss the possibility of ident ifying the more significant physical input variables for a correct very sho rt-range forecast of visibility.