Evaluating the Performance of Artificial Neural Network Model in Downscaling Daily Temperature, Precipitation and Wind Speed Parameters

Numerous studies yet have been carried out on downscaling of the large-scale climate data using both dynamical and statistical methods to investigate the hydrological and meteorological impacts of climate change on different parts of the world. This study was also conducted to investigate the capability of feedforward neural network with error back-propagation algorithm to downscale the provincial segmentation of Iran (30 provinces) on a daily scale. This model was proposed for the downscaling daily temperature, precipitation and wind speed data, and it was calibrated and verified by using the daily outputs derived from the National Center for Environmental Prediction (NCEP) database including air temperature, air pressure, absolute and relative air humidity, wind speed and direction, and data for the base period (1982-2001) at the selected synoptic station in each province. Correlation and root mean square error (RMSE) coefficients were used to analyze the performance of the proposed models. These criteria indicated the high accuracy of the proposed models in downscaling of daily temperature parameter rather than precipitation and wind speed parameters.