Calibration of prediction for rainstorm event caused by typhoon Lekima based on Bayesian Model Averaging
ZHAO Linna1,2, YAO Mengying1, GONG Yuanfa1, MU Xiuxiang3, LI Yitong3, AN Jianyu4
1. Plateau Atmosphere and Environment Key Laboratory of Sichuan Province, College of Atmospheric Sciences, Chengdu University of Information Technology, Chengdu 610225;
2. Institute of Artificial Intelligence for Meteorology, Chinese Academy of Meteorological Sciences, Beijing 100081;
3. Jilin Meteorological Observatory, Changchun, 130062;
4. Shangrao Meteorological Office of Jiangxi Province, Shangrao 334000
Using the products of European Centre for Medium-range Weather Forecasts (ECMWF) ensemble prediction and rain gauge data with interval of 6 hours from July 1 to August12 in 2019, we have established a precipitation probability forecast model (hereinafter referred to as the BMA model) based on Bayesian Model Average, and then conducted a comparative analysis of the deterministic forecast and probability forecast of typhoon Lekima precipitation based on the BMA model with respect to the raw ensemble forecast of ECMWF. The main results are as follow. (1) In general, the BMA model has a good correction to the typhoon precipitation ensemble prediction, and it can improve the dispersion of ensemble prediction effectively. (2) Compared with the raw ensemble forecast, the averaged values of CRPS, MAE and RMSE of the BMA model were reduced by 13%, 34% and 25%, respectively, which improved the reliability of the forecast to some extent. However, the improvement degree gradually weakened as the precipitation grade increased. The BMA model can also reduce the false alarm of probability prediction of rainstorm and heavier precipitation to a certain extent. It can also improve the accuracy of typhoon Lekima rainstorm and heavier precipitation’ s area forecast. (3) In this case, the 25th percentile to the 95th percentile forecast can be regarded as an effective prediction interval. The effective prediction interval revised by the BMA model has a stronger ability to capture precipitation observations, and the ensemble forecast capture rate revised by the BMA model is 13.3% higher than that of raw ensemble forecast.
GONG Yuanfa, et al
.2020. Calibration of prediction for rainstorm event caused by typhoon Lekima based on Bayesian Model Averaging[J].
Torrential Rain and Disasters, 39(5): 451-461.