Probabilistic Precipitation Forecasting Based On Bayesian Model Averaging Over Northeast China

Improving the scientificity and reliability of precipitation forecast has been always one of the most important problems in the field of climate statistics.This paper uses the Bayesian Model Averaging(BMA)method to study the deterministic and probabilistic prediction of daily rainfall over Northeast China based on the three ensemble forecast systems of ECMWF,JMA and UKMO from the Interactive Global Large Ensemble Prediction System(TIGGE).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 precipitation classification to improve the adaptability of BMA to moderate and heavy rainfall.Firstly,precipitation events are classified into several grades by FCM method based on the variables of three ensemble model,average temperature,average pressure,average relative humidity and average wind speed variables.Secondly,the Bayesian model averaging method is applied to each level of rainfall to improve the accuracy of ensemble forecast.Finally,the validity and applicability of the proposed method are verified by the deterministic prediction and probabilistic prediction.In this paper,the adaptive sliding window method is used to select different optimal window period lengths considering different climate types of stations.From the deterministic prediction results,the proposed CBMA model is better than the BMA and the three ensemble forecasts for different climate types of stations,and the NSE of most stations is greater than 0.5,which indicates that CBMA has good applicability.In order to further study the applicability of the proposed CBMA,this paper uses the test and evaluation indexes of different magnitude of precipitation forecast to study the prediction effect of CBMA and BMA in light,moderate and heavy rainfall.It is found that the ensemble forecast of each center has a serious situation of false reporting of light precipitation events and missing reporting of heavy precipitation events.Compared with BMA,CBMA has a certain correction effect on light precipitation events,and the correction effect on moderate and heavy precipitation events is obvious.From the probabilistic prediction results,the proposed CBMA method can greatly improve the overall prediction accuracy in terms of the advantages of the probabilistic prediction interval boundaries,and the regions with high precipitation have greater uncertainty than those with low precipitation.By comparing the BMA and CBMA found in percentile of each grading precipitation forecast,the BMA probability forecast range forecast for medium and heavy rain events tend to be more lenient,has great uncertainty,but CBMA can reduce the occurrence of moderate and heavy precipitation events in a smaller range and can significantly reduce the uncertainty of BMA probability forecast.Compared with BMA,probabilistic precipitation forecast of CBMA can better describe the actual situation of rainstorm events and reduce the omission error of extreme events.Therefore,this paper also proposes a rainstorm warning scheme based on CBMA percentile forecast.The results show that the rainstorm forecast of the rainstorm warning scheme is in good agreement with the observed precipitation.