Performance evaluation of artificial neural networks in statistical downscaling of monthly precipitation (Case study: Minab watershed)

Assessment of the impacts of climate change on water resources has been obtained significant attentions in the past decade. This paper assesses the climate change impacts on precipitation in the Minab basin, in the Hormozgan province in Iran. Two monthly precipitation downscaling methods were proposed based on multi-layer perceptron (MLP) and radial basis function (RBF) neural networks. The downscaling models were calibrated and validated using the large scale climatic parameters (predictors) derived from National Center for Environmental Prediction (NCEP)/ National Centre for Atmospheric Research (NCAR) reanalysis data set for downscaling monthly precipitation in the Minab basin in Iran. Pearson correlation was employed to choose the predictors among the NCEP/ NCAR reanalysis data set and final predictor combination for each station is assigned. The results of the downscaling models revealed that the MLP model produced more accurate and consistent results by downscaling the large scale climatic parameters compared to the RBF model. The proposed model can be reliably utilized for developing future projections of precipitation using the general circulation models outputs which can be employed also as the inputs in hydrological models.