| Soil moisture plays a critical role in the earth system,which are not only crucial for the vegetation,hydrology,land-air interaction,but are also an important source of predictability of weather,climate and agricultural drought.Compared with in-site observation and satellite remote sensing that have incomplete spatiotemporal coverage,Land surface model(LSM)driven by meteorological data serves as an alternative method for spatiotemporally continuous soil moisture estimation under physical and dynamic constraints.However,LSM simulations are subject to the uncertainties from meteorological forcing data,model structure and model parameters,it is necessary to improve the accuracy of soil moisture products by improving the meteorological forcing and LSM,and even by using data fusion method with the observations.Limited by in-situ observations,the contribution of meteorological data and land surface model to soil moisture simulation has not been systematically quantified in a large area,and there is a lack of comparison at different depths.In addition,how to use advanced machine learning algorithm to integrate observation and soil moisture simulation,and further reduce the simulation error needs to be explored.By validating against soil moisture observations over2090 in situ stations,this study aims to investigate the effects of model and meteorological data on soil moisture simulations at different depths.Moreover,random forest algorithm is used to generate soil moisture fusion product.The results emphasize the role of advanced model and data fusion method in improving soil moisture estimation.The main conclusions of this paper are as follows:(1)The development of LSM has a more significant effect on soil moisture simulation than high-resolution meteorological forcings.The Conjunctive Surface-Subsurface Process version2(CSSPv2)model was driven by China Meteorological Administration(CMA)Land Data Assimilation System(CLDASv2.0)for soil moisture simulation over China.The validations over 2090 stations during 2012–2017 showed that CLDASv2.0/CSSPv2 soil moisture simulation performed better than ERA5 and GLDASv2.1 reanalysis products,with an increased correlation of 26%–68% and reduced errors of 14%–24% at the daily scale.By replacing the meteorological data with ERA5 and GLDASv2.1,the contribution of CSSPv2 and CLDASv2.0to soil moisture simulation was estimated.CLDASv2.0/CSSPv2 only increased the correlation by 5%–35% and decreased the errors by up to 9% when compared with ERA5/CSSPv2 and GLDASv2.1/CSSPv2.In contrast,ERA5/CSSPv2 and GLDASv2.1/CSSPv2 soil moisture simulations increased the correlations from their alternative reanalysis LSMs by 17%–63%,and decreased the errors by up to 18%.The influence of the LSM was more obvious over semiarid regions,such as northern China.The influence of meteorological forcing was more significant for soil moisture simulations at the surface layer,while the LSMs played a more critical role for the middle and deep layers,especially during the cold season due to freeze-thaw processes.(2)The soil moisture fusion product could significantly reduce the simulation biases,but the improvement of correlation is limited.Considering the influence of topography,precipitation,surface temperature and soil texture,the random forest model was developed and applied to integrate the CLDASv2.0/CSSPv2 simulations with observations to generate a set of multi-layers(0?10cm,10?40cm and 40?100cm)and daily soil moisture product from 2012 to2017 over continental China.The results based on independent in-site observation showed that the fusion product reduced the root mean square error by 6%-29%,and reduced the bias by98%-99%.However,the improvement of correlation and the reduction in unbiased root mean square error were less than 6%.This might be because of the ability of CLDASv2.0/CSSPv2 to capture the soil moisture dynamics,but with systemic biases.The improvement of the fusion product is better in the deep layer,and in the northwest River basin,Yangtze River basin,Pearl River basin. |
PDF Full Text Request