基于改进的贝叶斯模型平均的方法和应用研究

Methodological and Applied Research on a Modified Bayesian Model Averaging

基于交互式全球大集合预报系统TIGGE的ECMWF、JMA和UKMO集合预报数据,运用概率统计中的贝叶斯模型平均对东北地区24小时累积降水量的预报技术进行后处理.已有研究发现,BMA预报对弱降水事件是准确的,但对中等、强降水事件的能力有限.因此,提出基于模糊C均值聚类的分类贝叶斯模型平均(Categorized Bayesian Model Averaging,CBMA).结果表明,CBMA基于FCM算法,考虑BMA在不同量级降水的参数不确定性,提高BMA在中等、强降水的适用性,具有较好的预报效果.

In this paper,the Bayesian model average method(BMA)is applied to process the prediction technology 24-h accumulated precipitation over Northeast China based on the ensemble forecast system outputs of ECMWF,JMA and UKMO of the Interactive Global Large Ensemble Prediction System(TIGGE).Previous studies have found that the prediction of BMA method is accurate for light precipitation events,but limited for moderate and heavy precipitation events.Accordingly,the paper proposes a Categorized Bayesian Model Averaging(CBMA)model based on Fuzzy C-means clustering(FCM).The results show that CBMA,based on FCM algorithm,takes into account the parameter uncertainty of BMA in different magnitude of precipitation,and thus improves the applicability of BMA in moderate and heavy precipitation,and has a better prediction effect.