Multi-factor And Multi-time-scale Statistical Prediction Of Summer Precipitation In Northeast Chin

Northeast China is one of the typical climate-sensitive areas in my country.The effect of improving the precipitation during the flood season has always received the focus of the meteorological departments and research departments at all levels.Aiming at the problems of unclear factors such as the influencing factors in the Northeast summer,and the time-scale accommodation of the time-scale scale in the age of age,this article aims to be based on modern statistical methods,from two perspectives from multi-factor and multi-time standards to 1970-2020 The quantitative prediction of summer precipitation is conducted according to the corresponding research and analysis.The main research content is as follows:(1)Using empirical orthogonal decomposition and Mann-Kendall trend test,the spatial distribution characteristics and temporal evolution of summer precipitation in Northeast China in recent 50 years were studied.The results show that:in terms of spatial distribution,summer precipitation in Northeast China not only has the same distribution characteristics of the whole region,but also has the difference of north-south inverse phase and east-west inverse phase distribution,and can be divided into three main rainfall regions.Due to the influence of topography,land and sea,summer monsoon and other factors,the summer precipitation in the three rainfall regions increased successively,and the precipitation variability increased successively.In terms of time evolution,the interannual and interdecadal variation and non-stationary characteristics of summer precipitation in Northeast China are common,and the precipitation in most stations has an obvious upward trend with time change.(2)By singular value decomposition and BP canonical correlation analysis,the prediction factors that have significant influence on summer precipitation in Northeast China were selected.Based on the selected multi-factor and co-factor,the multiple linear regression model and support vector machine model are established,and the prediction effects of the models are compared and analyzed.The results showed that during the independent test period,the MLR model predicted the second-order abnormal precipitation well,with a score of 62.22.ACC of SVM model was higher than MLR,and the score was 66.50 points.In particular,the average ASCR values of the two models during 2016-2018 were 53.33%and 52.66%,respectively,higher than the average 33.30%predicted by the national Conference.This indicates that the physical statistical prediction model with multiple impact factors can improve the prediction performance of summer precipitation in Northeast China.Meanwhile,the overall prediction effect of the nonlinear model SVM is better than that of the linear model MLR.(3)Using wavelet analysis to estimate the time-scale variability of precipitation;The precipitation time series was decomposed by Complete Ensemble Empirical Mode Decomposition with Adaptive Noise.The robustness of CEEMDAN decomposition and the predictability of different time scales are analyzed.The results show that the summer precipitation in Northeast China has the characteristics of multiple time scales.There are four main time scales in the process of precipitation change:50-52 years,20-22 years,12-15 years and 5-8 years,and the corresponding average cycle is 34 years,10 years,14 years and 5 years,respectively.The stability of multi-time scale decomposition is improved by adaptive addition of white noise in the process of CEEMDAN decomposition.The period of the eigenmode component obtained from the decomposition is clear and stable,and the sensitivity to the length of time series is low,and there is basically no mode aliasing phenomenon.At the same time,the components obtained by CEEMDAN decomposition have high correspondence with the precipitation influence factors,and good predictability.(4)A multi-time scale CEEMDAN-SE-SVM-ARIMA statistical prediction model is proposed in this paper in view of the deficiency of traditional models in effectively predicting multi-scale aliasing data.Firstly,the precipitation time series was decomposed into multiple components based on CEEMDAN method,and the component series was reconstructed according to the calculation results of sample entropy of different components.Then,the SVM multi-factor prediction model was constructed for the high frequency component,and the ARIMA model was constructed for the low frequency component and the trend component.Finally,the predicted value of the combined model is obtained by superimposing the predicted results of each component.The results showed that the combined model of CEEMDAN-SE-SVM-ARIMA had a high score in the whole independent test period,with an average P_s score of 81.81 in 5 years.In particular,in 2019,the combined model had a higher P_s score of 83.87,which was 25%higher than the MLR model.It also showed a certain forecasting advantage in the prediction of abnormal precipitation levels,and the number of abnormal levels reached 17,which was significantly higher than other models.In addition,from 2016 to 2018,the average ASCR of the combined model was 69.60%,twice higher than the national conference forecast,but also higher than the MLR model 53.33%and SVM model 52.66%.Therefore,the combined prediction model can effectively improve the forecast effect of summer precipitation in Northeast China,and has a certain ability to forecast the multi-time scale variability of precipitation.