POST-PROCESSING OF ALADIN WIND SPEED PREDICTIONS WITH AN ANALOG-BASED METHOD

In this paper, different post-processing methods are described and evaluated for deterministic and probabilistic point-based 10-m wind speed forecast over Croatia. These methods are applied to forecasts of operational high-resolution dynamical adaptation model (DADA) run with 2 km horizontal resolution to address the following question: which point-based post-processing method is the best suited for wind forecasting in the operational suite at DHMZ (Meteorological and Hydrological Service of Croatia). The verification procedure includes several metrics computed considering wind speed as continuous, categorical and probabilistic predictand. Those metrics were used to optimize the configuration, and to test both the deterministic and probabilistic prediction performance. This study shows that deterministic analog-based predictions (AnEn) improve the correlation between predictions and measurements while reducing forecast error better than using Kalman filter based predictions (KF), even though KF shows better bias reduction. The best results are achieved when forecasting the mean of analog ensemble or the Kalman filter of the mean of analog ensemble. Probabilistic AnEn predictions are properly dispersive, while having better resolution, discrimination and skill than forecast generated via logistic regression. These results encourage the potential use of AnEn in an operational environment at the location of meteorological stations, as well as at wind farm s