The combined descending averaging bias correction based on the Kalman filter for ensemble forecast

To aim at the problem with both bias and small spread in ensemble forecast, a combined descending averaging bias correction method is designed based on the original Kalman filter. By using 850 hPa temperature from the regional ensemble dataset of JMA in the WWRP Beijing Olympics Research and Development Project, the optimal weights of the first and second moment are obtained by the weight sensitivity experiments and applied in the combined bias correction. Then, impacts of the combined bias correction are evaluated. The results show that the first moment bias correction largely reduces the bias in ensemble mean, so the forecast quality of ensemble mean is greatly improved. The second moment bias correction has good ability to adjust the spread to close the RMSE of the ensemble mean, improving the reliability and resolution of ensemble forecasts. To this end, a new descending averaging bias correction method is developed to combine the first moment with the second moment bias correction whose respective optimal weights can be applied to the combined bias correction, so as to improve the overall quality of ensemble forecast. However, the contributions of the first and second moment bias correction to the combined bias correction varies in terms of scores. For RPS and Outliers scores, the contributions of the first moment bias correction are 83.75% and 18.83%, respectively. The 83.98% improvement is from the second moment bias correction in terms of reliability, and the contributions of both moments are largely equal for ROC.