为探究获取高精度土壤湿度变化数据的新方法，选取2005年1月至2012年12月为研究时段，利用GRACE重力卫星反演降尺度得到的空间分辨率为025°×025°的陆地水储量变化数据和GLDASNOAH陆面模式模拟的地表径流、积雪深以及实测地下水位等数据，基于水量平衡原理估算东部季风区6个主要流域的土壤湿度变化数据，反演深度范围可达0~200 mm。从时空相关性、均方根误差、[JP2]趋势度等方面，与AMSRE、ERAInterim、MERRA2、ASCAT、GLDASNOAH、GLDASMosaic、GLDASCatchment共7种土壤湿度产品进行对比验证。结果表明：估算的土壤湿度变化数据优势明显，在珠江流域表现最好，与上述产品的皮尔逊相关系数分别为070、079、086、085、081、050和051；与MERRA2、ASCAT、GLDASCatchment的单变量Moran′s I较为相近，分别为067、080和074，且与其他产品的均方根误差均较小，大部分控制在025~030；各种产品的土壤湿度变化在黄河流域表现一致性较差，皮尔逊相关系数均在041以下，变化趋势也不尽相同。

Soil moisture (SM) is significant in hydrological cycle with its frequent interaction between surface water and groundwater.Recently,there are two major methods of field measurement and remote sensingbased retrieval for monitoring SM.Although field SM measurement is widely used in practical applications,it fails to work in large scale areas with complex terrain for its inability to reflect spatial heterogeneity of SM.Moreover,the soil structure would be destructed due to its soil desiccation process.Remote sensingbased retrieval from optical or radar imagery is another major way to acquire SM at a large scale in space.The obstacle to the development and application of this technology is its limited maximum detection depth of 1-5 cm.SM levels affect a range of soil and plant dynamics.Surface SM is the water that is in the upper 10 cm of soil,whereas root zone SM is the water that is available to plantsgenerally considered to be in the upper 200 cm of soil.Consequently,one new method to obtain SM should be developed to overcome the problems above. Based on the principle of water balance,changes in terrestrial water storage (TWS) include three main parts:changes in surface water storage,SM,and groundwater storage.Detailly,changes in surface water storage are dominated by changes in snow depth and surface runoff.Changes in TWS were obtained from downscaled GRACE data,and changes in surface runoff and snow depth were from GLDAS NOAH data.Changes in groundwater from measured groundwater level were used to construct the water balance equation to calculate changes in SM.Then the SM change product was compared with other seven SM products of AMSR-E,ASCAT,ERA-Interim,MERRA2,GLDAS NOAH,GLDAS Mosaic,and GLDAS Catchment.All SM products were resampled to the same spatial resolution of 0.25°×0.25° for comparative analysis.Due to the inconsistent units of the various SM products,all data were normalized to the same range between 0 and 1.Furthermore,five statistical indicators were applied to validate the SM change product proposed.(1) Pearson correlation coefficient was used to verify the correlation of estimated SM change with other SM products in a time series.(2) Spatial auto-correlation coefficients (univariate Moran′s I) was used to verify how well the estimated SM change variability matches the spatial distribution of other SM change products.(3) The bivariate Moran′s I was used to reflect the correlation between the changes of estimated SM and other SM products in each grid point.(4) The root mean square error (RMSE) of the estimated SM change was compared to that of other SM products to assess the dispersion degree of all products.(5) The Mann-Kendall trend test was used to assess the trend of the time series of various SM change products and to verify the consistency of the trends for the estimated SM change with that of other products. Through the research,the following findings were made.Generally,the estimated SM changes showed a good temporal agreementwith each SM product in research area except the Yellow River basin and the Haihe River basin.This may be caused by the significant difference in the soil discrepancy characters during the permafrost and nonpermafrost periods in these basins,which affected the SM changes detected using microwaves such as AMSR-E and ASCAT.This discrepancy may amplified to some extent by the different methods of the discrepancy correction in different SM products.In terms of spatial distribution,a drying process was showed from the upper and middle reaches of the Yangtze River basin to the Huaihe River basin in spring;in summer,SM showed a south-wet,north-dry state;in autumn,the extent area of SM reduction gradually expanded southwards to the Pearl River basin;in winter,SM increased significantly in the upper reaches of the Yangtze River basin,and showed a slight increase in the lower reaches of the Yellow River basin and the Haihe River basin,while the rest of the research area were in the process of drying out.By analyzing the validation indicators,it is found that the estimated SM variation performed best agreementwith other products in the Pearl River basin,witch Pearson correlation coefficients are 0.70,0.79,0.86,0.85,0.81,0.50,and 0.51 with AMSR-E,ASCAT,ERA-Interim,MERRA2,GLDAS NOAH,GLDAS Mosaic,and GLDAS Catchment.The RMSE with other products is low,mostly controlled between 0.25 and 0.30.And the univariate Moran′s I was similar with [JP2]MERRA2,ASCAT,and GLDASCatchment (0.67,0.80,and 0.74),the bivariate Moran′s I was also[JP] similar with ERA-Interim,MERRA2,and ASCAT (0.70,0.70,and 0.77).The estimated SM changes showed poor consistency in the Yellow River basin,with Pearson correlation coefficients are all below 0.41,and the MK tests all show a near opposite trend to the other products,whitch may be the combined effect of retrieval errors in GRACE data due to the mountainous terrain in the western part of the Yellow River basin,the difficulty of land surface model data to accurately describe SM transport in loess soils,and the underrepresentation of measured groundwater data. The following main conclusions are made:(1) One method for estimating SM based on GRACE TWS was proposed.The SM obtained could keep more agreement with the connotation of SM in hydrology.It can overcome the limitations of field observations due to topography and number of stations,which may reflect spatial heterogeneity of SM of the whole study area.In addition,traditional remote sensing observation techniques can only detect the surface SM changes,while the proposed method can estimate SM of all soil layers above groundwater level.However,the estimation method does not consider the changes of biological water and the influence of human activities,which may introduce some uncertainty.(2) The estimated SM changes showed high spatial and temporal correlation with all other products in the Pearl River basin,which indicated that the estimation method is highly applicable in this basin.But in the Yellow River basin,the consistency of SM variation between the estimated SM changes and that of others products are not so good,which indicates that there are many other factors influencing SM variation in this basin and the uncertainty is high.Herewith,the research on SM estimation in the Yellow River basin needs to be strengthened in the future.