Research On Soil Moisture Downscaling Method Integrating Multisource Spatio-temporal-spectral Information

As a key factor in the ecology and climate system,soil moisture plays an important role in the surface land-atmosphere exchange system.Soil moisture affects hydrological processes such as precipitation and surface runoff,and is very helpful for applications such as agricultural production,water resources management,and disaster monitoring.Therefore,it is very important to realize continuous,high-precision,high-resolution,and large-scale observation of soil moisture.With the rapid development of satellite remote sensing technology,real-time and large-scale soil moisture monitoring based on remote sensing technology has become the mainstream way to obtain soil moisture.However,the existing satellite soil moisture products are mainly generated based on passive microwave methods,and the spatial resolution of these products is very low(kilometer level),which greatly limits the application of soil moisture products in various fields.For this reason,downscaling,as a technical means that can improve the spatial resolution,is widely used in the research of obtaining high-resolution soil moisture products,and a variety of soil moisture downscaling methods have been formed.However,there are relatively few comparative studies between downscaling methods.Based on the current research on soil moisture downscaling,this paper takes Oklahoma in the United States as the study area,and SMAP satellite soil moisture products as an example to carry out soil moisture downscaling with different downscaling methods,including these methods based on parametric statistics,spatiotemporal fusion,and spatio-temporal-spectral integrated fusion model.Finally,this paper achieves the high-precision conversion of SMAP soil moisture data from 36 km to 9km spatial resolution and compares and analyzes the downscaling results of different methods.The main research contents of this paper are as follows:(1)Comparison and analysis of the downscaling results of different parameter statistical models.Based on many existing studies,this paper first selects some representative remote sensing surface parameter auxiliary data that are closely related to soil moisture.Then,four widely used statistical models,including multiple linear regression,BP neural network,generalized regression neural network,and random forest,are selected to achieve downscaling of soil moisture based on parameter statistics.In this process,the cross-validation method is used to test the parameters of different statistical models,and the optimal parameters of each model in the study area are obtained to generate soil moisture downscaled results based on different models.Then,the residual correction is performed on these downscaling results,and the accuracy of soil moisture downscaling results generated by different models are compared and analyzed.After that,downscaling result based on parameter statistics with the highest accuracy in the study area is obtained.(2)Research on hybrid downscaling method combining parameter statistics and spatio-temporal fusion model.On the basis of the downscaling method based on parameter statistics,this paper applies the downscaling method based on spatiotemporal fusion to the downscaling of soil moisture in the study area.This article first analyzes the differences between the two different downscaling methods from the perspectives of the data sources used in the downscaling process,the principles and results of the methods,etc.The analysis results show that when the degree of surface change is different,the accuracy advantages of the two methods’ downscaling results are also different.Therefore,with the support of the land classification data in the study area,a hybrid downscaling method based on a weighting strategy is proposed according to different surface conditions,which combines the downscaling results of parameter statistic model and spatio-temporal fusion model at the decision level to generate high spatial resolution soil moisture products with more complete spatial coverage and higher precision.(3)Research on the downscaling method of soil moisture based on the spatiotemporal-spectral integrated fusion model.Inspired by the downscaling method based on parametric statistics and spatio-temporal fusion,this paper combines the characteristics of all data sources used to extend the concept of "spectrum" in remote sensing to the process of downscaling soil moisture for the first time,which is that multi-source remote sensing surface parameter auxiliary data closely related to soil moisture are defined as "spectrum" data reflecting the multi-dimensional characteristics of soil moisture.Furthermore,by combining some previous studies of the spatiotemporal-spectral integrated fusion model,the concept and framework of the spatiotemporal-spectral integrated fusion model in soil moisture downscaling are innovatively proposed,and an integrated model is constructed and solved to complete the downscaling process of soil moisture,achiving the first application and analysis of the spatio-temporal-spectral integrated fusion model in the study of soil moisture downscaling.