Downscaling Of GRACE Water Storage Based On Multi-source Hydrological Data

With the successful launch of the GRACE gravity satellite in 2002,remote sensing observation has become the main means of retrieving total water storage at this stage.However,the processing strategies and calculation models used by different research centers when processing the original satellite observation data are different,which brings uncertainty to the final water storage calculation results.At the same time,the coarse spatial resolution of the data greatly limits the application of GRACE satellites in small and medium scale.Compared with satellite observation data,the models can separately describe the precipitation,surface runoff,soil moisture,groundwater and other components in the water cycle process,and can obtain hydrological data with higher spatial resolution.However,the accuracy of the model depends largely on the artificially constructed hydrodynamic process.In addition,the model is often based on a single point operation,without considering the correlation between the spaces,and cannot fully reflect the real change of terrestrial water storage.Therefore,by combining the complementary characteristics of satellite observations with hydrological models and land surface model simulation data,this paper achieves the spatial downscaling of GRACE water storage and reconstructs the monthly high-resolution terrestrial water storage data(0.25° × 0.25°).The main research content includes:(1)Terrestrial water storage data fusion and uncertainty analysis.In this paper,the existing terrestrial water storage data are divided into three types of GRACE satellite observation data,land surface assimilation data,and global hydrological model data according to the data source and calculation principles.Then GRACE and global hydrological model data are fused by weighted average and introducing the threecornered hat(TCH)method to perform uncertainty analysis on the three types of data and the fusion results.(2)GRACE terrestrial water storage data downscaling.Based on the results of uncertainty analysis,introducing the idea of point surface fusion,satellite observation and model simulation data are used to constrain each other to select training points.Generalized regression and BP(Back Propagation)neural network are used in the coarse resolution(1° × 1°)to establish a regression statistical relationship between GRACE terrestrial water storage data and hydrological parameters such as precipitation,air temperature,soil moisture.Then combining high-resolution hydrological parameters to reconstruct monthly-scale high-resolution water storage data(0.25° ×0.25°).(3)Comprehensive accuracy evaluation of the downscaling results.This article first uses TCH to analyze the uncertainty of the downscaling results.Secondly,the Chinese Yangtze River Economic Belt and the United States are used as the experimental areas,combined with special hydrological events,and the long-term water storage changes and visual qualitative analysis are carried out on the downscaling results.Finally,combined with the underground monitoring well data provided by the United States Geological Survey,site verification was performed in the California area.The results show that: 1)The downscaling result has the highest accuracy in the uncertainty analysis;2)In the long-time analysis,the downscaling result reduces abnormal fluctuations compared to the original GRACE and models data;3)In the visual analysis,the downscaling result can accurately reflect the spatial distribution and size of the signal;4)In the verification of the monitoring well site,the downscaling result has improved in correlation,root mean square error and average absolute error.The research in this paper shows that the use of multi-source hydrological data can effectively improve the quality of GRACE water storage data.