Abstract:

The corrections using Kalman filter decreases average method can effectively improve the prediction accuracies of modal predictive temperature, however, they can also cause significant negative corrections which make the results inferior to the original model outputs.Based on the bias removed ensemble mean (BREM), October of 2019 to April of 2020 in the optimal sliding training period is chosen to make integration predictions using the results of ECMWF (EC), corrections by Kalman filter decreases average method (EC_COR), and the data of national meteorological center forecast (SCMOC), respectively.The corrected results of BREM on EC_COR predictions are evaluated as well.It is shown that the accuracies in various predictions all appear worse in winter and at night, with a systematic higher temperature deviation from November to the next January.BREM can effectively prevent the negative corrections of EC_COR on EC with better effects than those of any other single method before integration, significantly improving the predictions.In addition, the integration of high-quality predictive productions, which is not limited to the model output predictions or forecast corrections, can obtain better results than a single forecast.

Key words: Multimodal integration, Kalman filtering decrement average method, Surface temperature, Deviation correction