Research On The Response Of Runoff To Climate Change In The Arid Area Of ??Northwest China Based On Nonlinear Methods

The arid region of northwest China(ARNC)is characterized by landforms of mountain-basin interphase,which responds sensitively to the global climate change;additionally,climate change has altered the water cycle processes and accelerated the uncertainty of water resources in ARNC.Meanwhile,the inherent nonlinear and non-stationary characteristics of the climate-hydrological systems demonstrate that the climate and hydrological factors have obvious irregular fluctuations and nonlinear trends.Therefore,in the context of global climate change,revealing the nonlinear response characteristics of the inland river runoff to climate change scientifically is of great significance to clarify the evolution mechanism of water cycle processes in ARNC.Based on meteorological data and hydrological data from typical river in ARNC(Hotan River,Kaidu River and Heihe River)in recent 50 years,the nonlinear decomposition methods such as ensemble empirical mode decomposition(EEMD)and extreme-point symmetric mode decomposition(ESMD)were employed to analyse the spatial-temporal variation of nonlinear trend of climatic and hydrological factors and its relevance at different time scales,and combining the multiple linear regression(MLR)?back-propagation artificial neural network(BPANN)to establish the runoff hybrid forecasting model,revealing the nonlinear response characteristics of runoff to regional climate change at different time scales.In the end,based on the nonlinear decomposition,the BPANN method was used to evaluate quantitatively the effects of climate change and human activities on runoff change in the typical river in ARNC.The main results are as follows:(1)Both the EEMD and ESMD methods can better separate synthetic signals and reflect the nonlinear trends of climate factors in ARNC,with high reliability for the decomposition results.The overall precipitation in ARNC has exhibited an apparent nonlinear upwards trend from 1960to 2015,and it has began to take a turn in the late 1980s.The results of ESMD decomposition show that the significant(P<0.05)main periods of precipitation change are 3 years(IMF1),6years(IMF2)and 11 years(IMF3)at different time scales.The variance contribution rate of the frst intrinsic mode function(IMF1)is the largest,reaching 58%,indicating that IMF1 contains more information with actual physical meaning.Furthermore,there are obvious spatial differences in the nonlinear trends of average annual precipitation:northern Xinjiang has had a mainly rising trend;southern Xinjiang has had mainly rising and decreasing–rising trends;and the precipitation changes in the Hexi Corridor have been quite complicated,exhibiting six types.(2)The significant(P<0.05)main periods of the average temperature decomposed by ESMD in ARNC from 1960 to 2015 are 3 years(IMF1),9 years(IMF2)and 14 years(IMF3),respectively.The variance contribution rate of RES is the largest,reaching 46%.The nonlinear variation trends of temperature can be divided into six types as the same as precipitation,which have obvious spatial differences:northern Xinjiang has had a mainly rising trend;Hexi Corridor has had mainly rising and decreasing-rising trends;and there are six types of temperature changes in southern Xinjiang,and the only one that falls is located in the northern part of southern Xinjiang.(3)The variation characteristics of runoff nonlinear trend from typical river(Hotan River,Kaidu River and Heihe River)in ARNC show that the significant(P<0.05)periods of the summer runoff in Hotan river are 3 years,6 years and 14 years,respectively.The significant(P<0.05)runoff periods in Kaidu and Heihe River are 3 years,14 years,18 years and 2 years,6years,18 years,respectively.Among them,the contribution rates of IMF1 obtained by three rivers are the largest,58%,38%and 36%,respectively,indicating that IMF1 contains the most information with actual physical meaning.In addition,the residuals obtained by three rivers show a nonlinear upward trend.(4)The completeness of nonlinear decomposition results decomposed by ESMD and EEMD for runoff and the 0°C level height(FLH)in Hotan River is obviously higher than that of the wavelet analysis(WA);however,the significance of period of each component decomposed by ESMD is better than that of EEMD and WA.The decomposition results of ESMD show that the variance contribution rate towards the overall summer runoff in Hotan River of the quasi-3and quasi-6-year are the highest,accounting for 59.51%and 30.60%,respectively.The summer runoff has obvious response to FLH at the inter-annual scale:the periods of both the two factors are quasi-3 and quasi-6-year(P<0.05),while it exists slight differences at the inter-decadal scale:the former is quasi-14-year(P<0.05)and the latter is quasi-18-year(P<0.05).In addition,the correlation between the summer runoff and FLH has obvious differences at different time scales,indicating that the influence degree of climate change to runoff is not the same at different time scales.(5)The simulated accuracy of runoff by the single BPANN has a better performance than the linear regression method.Moreover,the hybrid model based on nonlinear decomposition methods of WA?EEMD and ESMD combined with MLR can further improve the simulated accuracy compared with the single MLR.Among them,the correlation between original and simulated series of the hybrid model ESMD-MLR is the highest(R=0.735,P<0.01),and the AIC value of it is the lowest(-6.978).Additionally,the hybrid model combined with BPANN can obviously improve the simulated accuracy compared with the single BPANN and the hybrid model that based on nonlinear decomposition methods combined with MLR,and the simulated effect of extreme points is also optimized.Among them,the simulated accuracy of the hybrid model ESMD-BPANN is the highest(R=0.853,P<0.001,the AIC value is-12.719),which indicates that it can better reflect the nonlinear response processes of runoff to climate change,providing a new reference for runoff forecasting.It can be seen that considering the nonlinear change processes of hydro-climatic system can effectively improve the simulated accuracy.The prediction results of the hybrid model ESMD-BPANN show that the runoff in the Hotan River?Kaidu River and Heihe River show increasing trends in different degrees from 2014 to 2030,with the rate of 0.124×10~8m~3/a?0.043×10~8m~3/a and 0.321×10~8m~3/a,respectively.(6)Based on the nonlinear response characteristics of hydrological factors to climate system,it is of great significance to quantitatively evaluate the impact of climate change on runoff at different time scales.Therefore,based on the nonlinear decomposition of the climate and runoff factors in the typical rivers,the effects of climate change and human activities on runoff were quantitatively evaluated based on BPANN.The results show that the impact of climate change all play a dominant role on runoff at different time scales in Hotan River,Kaidu River and Heihe River,and the total contribution rates are 82.8%,94.6%and 92.8%,respectively.However,the total contribution rates of human activity are only 17.2%,5.4%and 7.2%,respectively.