Evaluation And Downscaling Of Soil Moisture Datasets In The Alpine Mountain Ranges

Soil moisture is a significant factor for hydrological cycle and energy exchange between atmosphere and land surface Obtaining high resolution soil moisture data is essential for hydrological research and water resources management,especially in arid and semi-arid areas such as Northwest China At present,remote sensing technology and model simulations can obtain large-scale soil moisture data.However,the spatial resolution of such large-scale data is coarse,and cannot be directly applied to related hydrological and ecological researches.Thus,it is necessary to use the downscaling method to improve the spatial resolution of the large-scale soil moisture data.Due to the complex topography and surface conditions in the alpine mountain ranges,the spatio-temporal heterogeneity of soil moisture in this area is quit high,and few studies have been done on the spatio-temporal heterogeneity of soil moisture in the alpine ranges.Therefore,it is particularly important to obtain high-resolution soil moisture data to understand the spatial distribution of soil moisture in the alpine ranges.The upper reaches of the Heihe River are typical alpine ranges,and provid water resources for the middle and lower reaches of the Heihe River,the second largest inland river in China.This study evaluates the satellite soil moisture products and downscale such products to high spatial resolution in the upper reaches of the Heihe River Watershed.In this research,SMOS and SMAP satellite soil moisture products and GLDAS global land surface assimilation soil moisture products are compared with the observed soil moisture data in the upper reaches of the Heihe River Watershed to determine the most suitable soil moisture products for the study area.A multiple regression downscaling method and DISPATCH(Disaggregation based on Physical And Theoretical Scale Change)downscaling method are used to downscale the selected soil moisture products to 9km×9km resolution.Finally,the SMAP high resolution(9km×9km)soil moisture products are compared against the observed soil moisture data in the Babao River Watershed,a tributary of upstreat of the Heihe River Watershed to evaluate the applicability of the two downscaling methods in the upper reaches of the Heihe River Watershed.The main findings are as follows:Firstly,the Pearson correlation coefficient,r value of SMAP soil moisture products in the study area is 0.737,RMSE and bias are 0.045 m~3/m~3 and-0.032 m~3/m~3respectively.r value of SMOS soil moisture products in the study area is 0.584,RMSE and bias are 0.062 m~3/m~3 and-0.037 m~3/m~3 respectively.Compared with the SMOS products,the SMAP soil moisture products in the study area have better fitting trend and higher.The SMAP products under six vegetation types have larger r value,and smaller RMSE and bias values compared with the SMOS products.In general,SMAP soil moisture products are less affected by vegetation cover.In the time series stability analysis,the Mean Relative Difference(MRD)range of SMAP soil moisture products is-0.514 to 0.517,standard deviation(STD)range is 0.104 to 0.360.Compared with SMOS products,the SMAP products have larger MRD range and smaller STD values.The SMAP products can better reflect the spatial variability of soil moisture in the upper reaches of the Heihe River Watershed and have stronger time stability.Therefore,the SMAP soil moisture products with fixed angle radiometer are better than the SMOS soil moisture products with multi-angle radiometer in the alpine ranges.Secondly,the accuracy and trend fitting effect of the three GLDAS land surface assimilation soil moisture products of CLM_V1,Noah_V1 and Noah_V2.1 are better in 0~10cm soil layer,but poor in 10~50cm soil layer.The Pearson correlation coefficient,r value of Noah_V2.1 in the surface layer is 0.737,and 0.327 in the shallow layer,the trend fitting effect of Noah_V2.1 is best;RMSE and bias value of CLM_V1 in the surface layer is 0.042 m~3/m~3 and 0.017 m~3/m~3,the accuracy of CLM_V1 in surface layer is highest;And RMSE and bias value of Noah_V1 in the shallow layer is 0.063 m~3/m~3 and 0.042 m~3/m~3,the accuracy of Noah_V1 in shallow layer is highest.Compared with the CLM_V1 and Noah_V1,The CLM 2.0 of GLDAS introduces supplement of the snowmelt water to the surface soil moisture,which is approaching to the objective condition of this research area,so the trend fitting effect and accuracy of the surface soil moisture simulated by the Community Land Model 2.0(CLM 2.0)model in GLDAS are better.The Noah 2.7 model uses10-40 cm soil layers and vegetation as a whole to simulate soil moisture,the average depth of the soil water layer of the vegetation roots in the upper reaches of the Heihe River Watershed is 10-50 cm,and the Noah 2.7 simulates the water consumption of vegetation roots in 10-40 cm is accuracy,so the trend fitting effect and accuracy of the shallow soil moisture simulated by the Noah model are better.Compared with Noah_V1 and Noah_V2.1,the accuracy of Noah_V1 is higher in the surface layer(0~10cm)and shallow layer(10~50cm),and the trend fitting effect of Noah_V2.1 is better in the surface and shallow layers.GLDAS adopted the lower versions of CLM2.0 and Noah2.7 models,which don't take into account the fact that soil moisture is gradually solid state in winter in the study alpine range.Thus,a higher version of the CLM model sneeds to be be used for simulation assimilation in the future.Thirdly,since the SMAP soil moisture products are more suitable for the study area,soil DISPATCH and multiple regression downscaling methods are used to downscale the SMAP soil moisture products to higher spatial resolution(9 km by 9km).In the Babao River Watershed,the Pearson correlation coefficient,r value of DISPATCH soil moisture in the grid scale is 0.353,RMSE and bias are 0.123 m~3/m~3and-0.043 m~3/m~3 respectively;And r value of it in the watershed scale is 0.322,RMSE and bias are 0.093 m~3/m~3 and-0.076 m~3/m~3 respectively.The r value of multiple regression soil moisture in the grid scale is 0.093,RMSE and bias are 0.085m~3/m~3 and-0.008 m~3/m~3 respectively;And r value of it in the watershed scale is 0.171,RMSE and bias are 0.052 m~3/m~3 and-0.010 m~3/m~3 respectively.The DISPATCH method shows better fitting trend while the the multiple regression method shows greater accuracy at seasonal scale,the downscaling effect of multiple regression method is the best in spring,then in autumn,and the worst in summer;while the downscaling effect of DISPATCH method is the best in autumn,then in summer,and the worst in spring.Affected by the brightness temperature data and the quality problems of the SMAP surface soil temperature in the study area,the two downscaling methods show better accuracy but poor fitting trend than that of the high-resolution SMAP(9km×9km) products.