Comparative Evaluation of Different Post Processing Methods for Numerical Prediction of Temperature Forecasts over Iran

In this study we examined the performance of five post-processing methods on WRF model outputs for daily maximum and minimum temperature forecasts in thirty synoptic meteorological stations over Iran. Direct Model Output (DMO) always contains systematic errors which arise mainly from the simplification of the earth topography in the model and deficiencies in the physics of the model. Different methods for post-processing of these outputs are given to remove the systematic errors. The results of the experiments show all methods are successful in removing the systematic errors in the model outputs. Comparing calculated statistical scores like root mean square error, mean absolute error and mean error indicate that Kalman Filtering (KF) and Artificial Neural Network (ANN) methods are better compared to other methods. Due to the importance of specific temperature thresholds in application, we verified the post-processed temperature forecasts for some specific temperature thresholds. The results of some statistical measure such as Proportion Correct (PC), Treat Score (TS) and False Alarm Rate (FAR) showed satisfactory for various thresholds, but better results have been obtained for higher values of maximum temperature and lowest values of minimum temperature.