Comparing multi-source runoff data in different watersheds across China
The multi-source runoff data serve as a foundation for comprehending the spatial pattern and temporal evolution of surface water resources.Comparing and screening runoff data are prerequisites for their effective application and widespread adoption.However,comparing multi-source runoff data in China remains a gap.To address this,here,we collect continuous natural streamflow data from 82 hydrological stations covering the period of 1961-2014.These stations are situated in nine major river basins across China,each with distinct non-overlapping control areas.We utilize natural streamflow data as the benchmark to compare 33 data sets from four categories,i.e.,data from earth system models in CMIP6,global hydrology models in ISIMIP3a,land surface models in GLDAS and CNRD,and machine learning results from GRUN,in terms of the multi-year mean streamflow and annual streamflow trend.The results indicate that:(1)PBIAS shows that the bias-corrected CMIP6,ISIMIP3a,GLDAS,GRUN,and CNRD can grasp the mean streamflow in most watersheds,and the Taylor diagram analysis integrating standard deviation,root mean square error,and Pearson correlation coefficient shows that CNRD performs best in the Songliao River Basin,Yangtze River Basin,Pearl River Basin,Southeast river basins,and Northwest and Southwest river basins,while the bias-corrected CMIP6 and ISIMIP3a multi-model ensemble mean show excellent simulation in Huang-Huai-Hai River Basin.(2)The multi-source runoff data exhibit accurate representation of the annual mean natural streamflow,while the simulation results for trends are not satisfactory,particularly for CMIP6 and GRUN,as they significantly underestimate the runoff trends in approximately 10 watersheds,resulting in opposite trends compared to the reference data.(3)The simulation also performs poorly in relatively arid regions,highlighting the urgent necessity to enhance the quality of the driving data,refine the model structure,and optimize model parameters to improve the accuracy of the model in simulating the water cycle process in arid regions.This work provides an important basis for data screening to study the spatial and temporal evolution of streamflow and surface water resources in China,and clarifies the possible problems and measures for the updating and development of multi-source runoff data in different watersheds across China.
streamflow comparisonearth system modelglobal hydrology modelland surface modelmachine learning algorithm
万方数据
维普
