Utilizing ANN for improving the WRF wind forecasts in Turkiye

The accurate estimation of wind speed is crucial for wind energy production, given the exponential relationship between wind and power. However, this is a challenging task due to the stochastic nature of meteorology. In this study, the Weather Research and Forecasting Model (WRF) with a 9 km spatial resolution was used to simulate hourly wind speed values for Turkiye, using Global Data Assimilation System (GDAS) 0.25 degrees boundary data. Several statistical metrics, including Root Mean Squared Error (RMSE), Mean Bias Error (MB), Index of Agreement (IoA), and Pearson correlation, were used to evaluate the performance of the WRF model. The WRF model, which used CONUS parametrization and was supplied with GDAS boundary data every 6 h, operated for 17,520 h in a 1-month consecutive run. The ANN model, which has Hecht-Nielsen (2n + 1) topology, was used to perform hindcasting of the WRF model. The input layer of the ANN model used temperature, pressure, and wind speed values obtained from WRF. The analysis was done spatio-temporally for 2 years and presented with seasonal and annual performance values. After applying the ANN model to the WRF model, which had initial values of MB of 1.42, RMSE of 2.26, R of 0.51, and IOA value of 0.02, the new MB, RMSE, R, and IOA values were found to be 0.04, 0.96, 0.56, and 0.60, respectively. Therefore, it can be concluded that the ANN model improved the WRF model's wind speed prediction performance in Turkiye by 11% on average, relatively.