Temporal And Spatial Evolution Attribution Analysis Of Runoff In The Xijiang River Basin And Medium-term And Long-term Forecasting

It is an important basic premise for scientific management and flood control and drought relief to grasp the spatio-temporal evolution of runoff and make attribution analysis,and carry out accurate runoff forecast.Since runoff time series is a typical nonlinear and non-stationary series,it is difficult to fully identify its evolution law and direct prediction accuracy is low.In this paper,the Extreme point Symmetric Mode Decomposition(ESMD)method is used.Based on the advantages of its adaptive,baseless and stable processing technology and the identification of large-scale circulation and nonlinear trends,the runoff evolution attribution analysis and medium and long-term runoff forecast were carried out.The research results are as follows:In terms of temporal and spatial evolution characteristics and attribution analysis of runoff,based on the daily runoff data of 7 stations in the Xijiang River Basin in the past 60 years,integrated ESMD,M-K test,wavelet analysis and other methods,the temporal and spatial evolution characteristics of runoff were analyzed from multiple time scales of the inter-decadal,inter-annual,seasonal and intra-annual.Then,the effects of human activities and climate change on runoff change were analyzed based on the method of double accumulation curve and the method of slope change rate of accumulation.The results show that:at the inter-decadal scale,the runoff in the Xijiang River Basin alternated between abundant and dry and varied greatly.The annual and seasonal runoff series showed double cycles of interannual and decade.The annual,summer,and autumn runoff showed a downward trend,and the overall spring and winter runoff showed an upward trend.The abrupt changes of annual and seasonal runoff increased after 1980,especially from 2000 to 2010.After 2000,except the tributary Liuzhou station,the annual distribution of runoff tended to be uniform,the variation range of monthly average runoff decreased,the concentration period moved forward.Except for the downstream Wuzhou station and the tributaries Liuzhou Station and Guigang Station,climate change is the main factor affecting the runoff change,and the influence of human activities is dominant in the runoff change of the other four main stream stations.Among the two factors of climate change,the contribution rate of rainfall to runoff change is greater than that of evaporation at other stations except Tian ’e Station.In terms of medium and long term runoff forecast,the ESMD-PSO-LSSVM-BP combined forecast model was constructed based on the adaptive,baseless and smooth processing technology of ESMD method,combined with the characteristics of PSOLSSVM method with large fluctuations,complex features and strong adaptability,and the advantages of BP neural network’s nonlinear mapping ability and approximation to any nonlinear function.The annual and monthly runoff at Tiane,Qianjiang and Wuzhou stations were predicted.The results showed that:The prediction effect of the combined forecasting model based on ESMD method is significantly better than that of the single forecasting model,while the combined model of ESmd-PSO-LSSVM-BP has the best performance among all models,with the lowest prediction error and the highest degree of fitting with the measured runoff series,indicating that the smooth processing technology based on ESMD method,after stabilizing the nonlinear and non-stationary original series,using different prediction methods to predict the components with different frequencies can significantly improve the precision of runoff forecast,which opens up a new idea for improving the precision of mid-term and long-term runoff forecast.