Research On Runoff Prediction In The Upper Han River Basin Integrating CMIP6

In recent years,due to rapid climate change,frequent floods and droughts have occurred,which has had a significant adverse impact on China’s social stability and economic development.It is urgent to strengthen corresponding water resource management,and therefore,it is necessary to make runoff predictions under climate change conditions.The Global Climate Model(GCM)is currently the most important tool for climate change prediction.In order to promote the development of climate models and improve the scientific understanding of the Earth’s climate system,the World Climate Research Program(WCRP)Coupled Simulation Working Group(WGCM)launched the Coupled syntype Intercomparison Program(CMIP)among GCMs.CMIP6 is its sixth and latest comparison program.In this study,Taylor chart method was used to evaluate the simulation performance of precipitation,temperature and evaporation in the upper reaches of Hanjiang River under the historical scenarios of GCM climate models in CMIP6,and 10 models with the best comprehensive simulation ability were selected,and Bayesian average method was used to combine them into a collective model(MME).Use multiple deep learning models to predict runoff in the upper reaches of the Han River basin,select the most suitable model for the region,and explore the climate and runoff changes in the upper reaches of the Han River basin in the future(2023-2100)through MME’s multiple shared socio-economic paths(SSPs).Based on the above research content,the main results obtained in this article are as follows:(1)The geographical detector is used to analyze the meteorological factors that affect runoff changes in the upper reaches of the Han River,select the factors with the largest impact factor,use bicubic B-spline interpolation quantile mapping spatial decomposition downscaling model to reduce downscaling of CMIP6 model corresponding to the selected factors,use Taylor chart method to select the 10 comprehensive optimal models after downscaling,and use Bayesian average to collect the optimized models,The results show that the meteorological factors that have the greatest impact on the runoff changes in the upper reaches of the Hanjiang River arc rainfall,evaporation and temperature,and the impact factors are 0.8193,0.3016 and 0.2906,respectively.The corresponding models have significantly improved the simulation ability of the study area after downscaling processing.At the same time,the correlation coefficients between the rainfall,temperature and evaporation of the syntype MME and the observation data are 0.859,0.982 and 0.958,respectively,better than the single model.(2)Long and short term memory network model(LSTM),gated cycle unit model(GRU),time convolution network model(TCN),CNN-LSTM and GNN-GRU combined model are selected to forecast monthly runoff series of Yangxian,Shiquan and Ankang hydrological stations respectively,and the optimal model is selected according to the prediction results.CEEMDAN algorithm and KPCA algorithm are introduced to reduce the noise of meteorological series,The processed model prediction results are compared with the single model prediction results to select the best.Finally,the feasibility of the MME+deep learning model is verified by using the MME scenario data as input.The results show that the TCN model is superior to other models in predicting the monthly runoff series in the study area.The Nash efficiency coefficients(NSE)of the prediction results at three hydrological stations reach 0.753,0.723 and 0.725 respectively.The prediction results of the mixed model are superior to the single model,Among them,CEEMDAN-KPCA-TCN has the best prediction results,with NSE reaching 0.784,0.815,and 0.835,respectively.Using this model with MME as input,the Nash efficiency coefficient(NSE)reaches 0.708,0.738,and 0.727.(3)Taking the meteorological data of MME under different scenarios after downscaling as the model input,the change trend of rainfall,temperature,evaporation meteorological series and watershed runoff series was analyzed respectively.The results showed that rainfall,temperature and evaporation in the watershed showed an increasing trend under the four future scenarios of ssp126,ssp245,ssp370 and ssp585.The average annual precipitation in the future was 865.09mm,880.24mm,854.83mm and 890.17mm,respectively,The average annual temperature is 12.44℃,12.68℃,12.89℃ and 13.43℃ respectively,and the average annual evaporation is 548.9mm,516.8mm,558.65mm and 596.60mm respectively;The future monthly runoff of the watershed is mainly concentrated in summer and autumn,with the maximum average monthly runoff in September and the minimum in January.The maximum future monthly runoff is 3984.661m3/s,which occurred in September 2076 under the ssp585 scenario and the minimum monthly runoff is 93.80m3/s,which occurred in December 2078 under the ssp 126 scenario;The annual runoff series showed an insignificant downward trend in the ssp370 scenario,while all other scenarios showed an upward trend.The maximum annual runoff was 862m3/s,which occurred in the ssp585 scenario in 2076 years,and the minimum was 366m3/s,which occurred in the ssp245 scenario in 2071 years;The annual runoff sequence undergoes mutations in 2099 under the ssp 126 scenario,2099 under the ssp245 scenario,2065 under the ssp370 scenario,and 2097 and 2099 under the ssp585 scenario;Under the ssp126 scenario,the annual runoff sequence in the study area has a variation period of 9 years,a variation period of 24 years under the ssp245 scenario,a variation period of 19 years under the sp370 scenario,and an variation period of 11 years under the ssp585 scenario.The annual runoff shows an alternating cycle of abundant dry abundant during the cycle.