A Study Of Watershed Runoff Prediction Based On The Selection And Bias Correction Of Global Climate Models

Global warming has become a widespread consensus in the scientific community and the international community.To predict the impacts of future global warming,Global Climate Models(GCMs)are used to make assessments of climate change impacts.Meteorological variables such as precipitation,maximum and minimum temperatures are commonly used in combination with hydrological models for runoff response assessment.Compared with Phase 5 of Coupled Model Inter-comparison Project(CMIP5)published in 2013,Intergovernmental Panel on Climate Change(IPCC)in recent years has added the sharing economy hypothesis to future scenarios,and set higher resolution and more detailed underlying surface conditions in Phase 6 of Coupled Model Inter-comparison Project(CMIP6).How will these changes affect the prediction of the hydrological response to climate change? It is still worthwhile to investigate.In particular,there are a variety of models in CMIP5 and CMIP6,and how to select a reasonable model for runoff response assessment needs to be studied urgently;meanwhile,the variables from GCMs such as precipitation and temperature have low resolution and large bias,and the correlation structure is not reasonable.Therefore,in this study,21 climate models were selected in CMIP5 and CMIP6 separately,and a comprehensive evaluation indices system was constructed for climate variables such as precipitation,maximum and minimum temperature,and a preferred combination of GCMs was proposed after evaluating the performance of climate variables from each GCM.Meanwhile,based on the gamma distribution and a distribution-free shuffle method,we propose a bias correction method that considers the multi-dimensional properties of climate variables from GCMs,and conduct a runoff response assessment in 343 watersheds in the mainland of China,and characterize the changes of predicted runoff in the future period(2071-2100)relative to the historical period(1976-2005)under the high emission scenarios of CMIP6 and CMIP5.The main research work and conclusions of the paper are summarized as follows:(1)The simulation effects of GCMs output precipitation and temperature variables on the extremes are evaluated in CMIP5 and CMIP6.The performance of 12 extreme precipitation and extreme temperature indicators simulated in CMIP5 and CMIP6 are evaluated using mean deviation,relative root mean square error,annual variation fraction,Taylor fraction and quartiles.The results show that extreme temperature and precipitation intensity increases from the historical period(1976-2005)to the future period(2071-2100)under the high emission scenario,and CMIP6 multimodel average significantly increases more than CMIP5.The simulation effect of CMIP6 extreme climate indicators is better than that of GCMs in CMIP5,and the intensity of temperature and precipitation increased more,and the uncertainty is smaller.(2)A combination selection method of GCMs for climate variables is proposed.Based on multiple temporal and spatial indicators,a comprehensive evaluation index system of climate variables from GCMs is established,and a method for selecting the combination of GCMs for three variables(precipitation,maximum and minimum temperature)is proposed,based on the comprehensive evaluation and ranking from multiple temporal and spatial indicators,the best combination of GCMs with precipitation,maximum and minimum temperature in the mainland of China is selected.For the problem of large uncertainties of variables such as precipitation and temperature from GCMs,the effects of multiple GCMs combinations are quantified from the characteristics of temporal trends and spatial distribution variability.The results show that the GCMs in CMIP5 and CMIP6 perform differently for precipitation,maximum and minimum temperatures in the mainland of China using 12 spatial and temporal indicators and comprehensive evaluation formulas.The best performing GCM for the precipitation in CMIP5 is CCSM4,the maximum and minimum temperature is Can ESM2,while the best performing GCM for precipitation in CMIP6 is INM-CM4-8,the maximum temperature is IPSL-CM6A-LR,and the minimum temperature is ECEarth3.In terms of the best performing combinations of GCMs,the number and composition of the best GCMs combinations differed for different variables.The best combinations for all the three variables are CSIRO-Mk3.6.0 and Can ESM2 in CMIP5,and Had GEM3-GC31-LL,Nor ESM2-MM and IPSL-CM6A-LR are the best combinations for all the three variables in CMIP6.(3)A multi-dimensional bias correction method for climate model output variables is proposed.Five distributions,including exponential(EXP),gamma(GAM),Weibull(WBL),mixed index and generalized Pareto(EXPP),and mixed index(MEXP),were selected to simulate the precipitation on rainy days in 343 watersheds,and the gamma distribution was found to be the most effective in simulating the mean and extreme values of precipitation.We propose a multi-dimensional bias correction method for the output variables of the climate model by using the gamma distribution to correct precipitation and the normal distribution to correct temperature,and using a distribution-free shuffle method to recalibrate the corrected precipitation and temperature.And compared with linear bias correction and DBC,the multidimensional bias correction method for climate variables is the best for the correction of mean and extreme precipitation,which corrects the probability distribution and extreme value of precipitation on the one hand,and reconstructs the correlation structure between temperature and precipitation on the other hand.Among the four typical basins,the watershed controlled by Hankou station in Yangtze River is the best simulated,the watershed controlled by Wuzhou station in Zhujiang River is the second,the watershed controlled by Kachun station in Yarkant River is the third,and the watershed controlled by Imin Ranch station in Imin River is the worst,which is related to the selection of hydrological models,area of controlled watershed,and geographical location.Based on the ranking of GCMs,the simulation effects of different weighting combination methods,including multi-model averaging,optimal GCMs,linear weighting method,and exponential weighting method,in the runoff response were comprehensively analyzed,and the options of different weighting combination methods of GCMs in the hydrological response to climate change were discussed.Using the performance of simulated annual mean daily runoff and annual maximum 1-day runoff as indicators,the exponential weighting method combination performs best overall.(4)343 watersheds in the mainland of China were selected,and a study on the response of runoff to climate change under CMIP5 and CMIP6 high emission scenarios in future periods relative to historical periods was carried out,based on a combination of the multidimensional bias correction method for GCMs output variables and the exponential weighting method.The results showed annual mean and maximum flows are predicted to increase in most watersheds in both CMIP5 and CMIP6;the difference between CMIP6 and CMIP5 predicted future annual runoff is limited,especially in the south,and in the north CMIP6 predicted runoff increases more than CMIP5.The GCMs in CMIP6 predict less uncertainty in temperature than CMIP5,especially for the historical period.For most watersheds in southern and northeastern China,model uncertainty in CMIP6 GCMs is smaller than CMIP5 for both annual mean daily flow and annual maximum daily flow,while the opposite pattern is observed in central China,consistent with corrected precipitation.