Ensemble prediction of winter precipitation in China based on Support Vector Machine Method
Winter precipitation in China is of significant importance for agriculture,water resource management,and natural disaster risk assessment.Due to the influence of various meteorological factors,predicting winter precipitation remains a formidable challenge.Enhancing the predictive capabilities is therefore a pivotal focus in contemporary short-term climate prediction research.The study implements Support Vector Machine(SVM)method,with the aim of improving the accuracy of predictions of China's winter precipitation through machine learning techniques.Capitalizing on model data from NCEP_CFS,ECMWF_SYSTEM,BCC_CSM and two other models along with station data,a SVM ensemble prediction model for winter precipitation is constructed and benchmarked against individual models and equally weighted ensemble mean model(AVE).The SVM model exhibits excellent performance in winter precipitation prediction owing to its strong generalization capabilities and proficiency in handling nonlinear problems.The research shows that:(1)The prediction accuracy and stability of the SVM model exhibit considerable improvements compared to individual models and AVE models.The PS score and PCS score of the SVM model notably surpass those of individual member models,with maximum improvements of 8.0(12.6％)and 3.9(7.4％)respectively,compared to the AVE model,which sees maximum improvements of 5.4(8.2％)and 2.1(3.8％)respectively.This enhancement in forecast skills is particularly pronounced in the southwest and northwest regions where observational data are relatively scarce.(2)Spatially,the SVM model markedly ameliorates situations where member models exhibit substantial errors in regions such as Tibet,southwest,east and south China.Maximum error reduction reaches 90.9％(259)and maximum improvement in forecasting skills by 1.13.(3)In the independent sample testing,the PS score and PCS score of SVM model significantly outperform those of individual model and AVE model,with maximum improvements of 10.79(20.3％)and 11.39(27.3％)respectively.Hence,the advancement of the SVM model promises to catalyze further enhancements in the precision and reliability of forecasts for China's winter precipitation.This progress will provide indispensable technical backing for endeavors in meteorological disaster prevention and mitigation,exploitation of climate resources,and associated applications.
PrecipitationSupport vector machine(SVM)Equally weighted ensemble mean modelEnsemble prediction
万方数据
维普
