Statistical postprocessing of the WRF output for 10 m wind speed over north and northwest of Iran

In this study, four statistical methods including diurnal cycle forecast correction (DRL), linear least-square corrected forecast (LLS), mean and variance corrected forecast (MAV) and average of the three previous methods (AVG), were used to postprocess the WRF mesoscale model output for wind speed at 10-m above ground level over 85 synoptic stations located in north and northwest of Iran. For this purpose, the WRF model with a horizontal resolution of 15 km was run to prepare 3, 15, 27, 39, 51, 63, 75, 87, 99 and 111hour forecasts from the beginning of November 2011 until 30th of April 2012. By varying the number of days used for training and calculating the mean absolute error (MAE), an optimum number of days were selected as training period separately for each method. The results show that the mean error over all stations dropped to zero or close to zero for all four methods, indicating the removal of the systematic error of WRF direct model output for the wind speed. In general, LLS has a better performance in reducing the systematic error when compared to other methods and improves the MAE of the direct model output between 26 to 42 percent, for different forecast ranges So that in this way, has been reduced the mean absolute error of about 36 percent for all predictions in all studied stations. The highest amount of decrease in MAE was around 65 percent and was obtained for Qazvin, Baladeh and Alasht stations while least improvement of 1 percent was observed for Ardebil, Nir and Khodabande stations.