Multi-time Scale Analyses And Nonlinear Dynamic Prediction On Characteristics Of Runoff And Sediment Discharge In The Yangtze River Basin

In this paper,the Hilbert-Huang transform(HHT)technique of empirical mode decomposition was applied to carry out multi-time scale analyses on the annual and monthly average runoff,monthly sediment discharge and the relevant meteorological factors at the representative stations of the Yangtze River Basin.First,we conducted a multi-time scale analysis for the annual average runoff at Yichang,Hankou and Datong Stations in the Yangtze River Basin to develop nonlinear prediction models.Second,the multi-time scale analysis was used for the past 60-year monthly average runoff at three representative stations of the upper,middle and lower reaches of the Yangtze River Basin(Yichang,Hankou,Datong).The analyzed results indicated that the HHT method was able to identify the various time-series characteristics at different time scales and their close relationships with heavy rainfall-induced extraordinary flooding events occurred in 1954 and 1998.Third,the multi-time scale analysis was also applied for monthly Sea Surface Temperature(SST)and Southern Oscillation Index(SOI),in order to identify the key time scales of these meteorological factors affecting on Yangtze River runoff.Finally,we applied multi-time scale analysis for the time series of monthly sediment discharge in the past 60 years at the three representative stations located in the Yangtze River Basin to identify their relationships with average monthly runoff at different time scales.Major results are summarized as follows:1.The multi-time scale analysis of the HHT method was employed for the annual average runoff at Yichang(1900-2016),Hankou(1952-2016),and Datong(1951-2016)stations.The results showed that the annual average runoff at three stations was able to be decomposed into five/four different IMF(Intrinsic Mode Function)components.Among these components,there are two major fluctuation cycles:3 and 6-7 years,which can explain 75%-77%of the total variance within the dataset.2.The decomposed IMF components were used in this paper to develop multi-time scale nonlinear prediction model for Yichang,Hankou,and Datong station's annual average runoff.There are two prediction models:Model I using all decomposed IMF componenets and Model ? using all components except for the high frequency component C1.The modeled results indicated that the nonlinear prediction model was successful at replicating the quantity of annual average runoff.The simulation results from Model ? are usually better than Model ?.Using the relative error less than 20%as a mesurment,Model ? can capture 85%-93%of the total historical events for these three stations.3.The monthly average runoff in the past 60 years at three representative stations of the Yangtze River Basin(Yichang,Hankou,Datong)can be decomposed into six different cyclic components and one trend term.The decomposed IMF component C2 representing the seasonal variation of about one-year cycle is the most important one,which can explain 84.48%-86.35%of the total variance within the dataset.In addition,during the years of 1954 and 1998 when the extraordinary flooding events occurred in the Yangtze River Basin,both the annual cyclic component C2 and combined low-frequency components(C3+C4+C5+C6)showed abnormally high values.In the future work,when the rainfall-related flood prediction models are developed,these decomposed IMF components can be selected as more effective predictors to impove prediction accuracy.4.Monthly Sea Surface Temperature(SST)in ENSO's four key SST areas(Ni(?)ol+2,Ni(?)o3,Nino4,Ni(?)o3.4)and Southern Oscillation Index(SOI)from 1951 to 2016 can be decomposed into six different cyclic IMF components and a trend term.The results showed that the SST decomposed IMF component C2 representing the seasonal variation of about one-year cycle has a good relationship with the corresponding component decomposed for the monthly mean runoff at Yichang Station.The SST decomposed seasonal variations in Nino1+2,Nino3,and Nino3.4 generally occur about 3-5 months earlier than the runoff.However,there is no good relationships between the decomposed IMF components corresponding to SOI and the monthly mean runoff.5.The monthly average sediment discharge rates in the three representative stations of the Yangtze River Basin(Yichang,Hankou,Datong)in the past 60 years can be also decomposed into six different cyclic IMF components and one trend term.Among them,the decomposed component C2 representing the seasonal variation of about one-year cycle is most important,which can explain the largest variance within the dataset,ranging from 63.48%to 67.32%across three stations.There is a very good relationship between monthly sediment discharge rates and monthly mean runoff at the three representative stations in the Yangtze River Basin.Especially,both the seasonal variation patterns of the decomposed one-year cycle component C2 are very consistent and synchronous.The results from this study showed that the HHT multi-time scale analysis technology is very effective to decompose hydro-climatological time series.The decomposed results can provide more effective and better prediction factors as well as theoretical and practical bases for future development of hydrological prediction models.In the future work,when the non-linear runoff prediction models are developed,the HHT multi-time scale analysis technology can identify the time series characteristics of the decomposed IMF components on different time scales.Furthermore,the HHT multi-time scale analysis technology can decompose the relevant meteorological elements to determine runoff-related decomposed IMF components as more effective runoff prediction factors.Eventually,it is helpful to improve the accuracy of future hydrological prediction models.