NN5: A neural network based approach for the downscaling of precipitation fields – Model description and preliminary results

SummaryA collection of one year daily forecasts with the MM5 mesoscale model is used to investigate the possibility to downscale hourly precipitation fields from a horizontal grid spacing of 27 km to one at 3 km. The downscaling is performed using a multi-layer Neural Network built with information of terrain, land use and predicted precipitation at the four adjacent grid points of the MM5 coarse grid. Results obtained for a domain of complex topography show that the proposed technique produces realistic downscaled precipitation fields. Emphasis is given to the possible application of the methodology to the coupling of hydrological and meteorological models or for downscaling coarse scale climate model precipitation fields to hydrological catchment scales.