Remote Sensing Retrieval Of Soil Moisture In Ordos Blown-sand Region Based On SVR

Soil moisture is an important component of surface water cycle and a key parameter of meteorology,hydrology,agriculture and biology.It plays an important role in hydrology forecast,agricultural monitoring,surface carbon cycle,surface water evaporation and vegetation transpiration.In arid and semi-arid areas of Northwest China,precipitation is scarce and evapotranspiration is intense.Soil moisture,as an important ecological factor,affects the energy balance of soil-atmosphere interface.Therefore,it is of great significance to study the spatial distribution characteristics of soil moisture in Northwest China for water resources allocation,drought control and ecological environment monitoring.Based on the full polarization Radarsat-2 radar image data,GF-1,Sentinel-2B optical image data and field measured data,and taking into account the characteristics of surface vegetation and the roughness of the surface,this paper chose different vegetation parameters and used water-cloud model to remove the influence of surface vegetation layer.By comparing the inversion accuracy of soil moisture under different vegetation parameters,the optimum vegetation parameters suitable for characterizing the scattering characteristics of vegetation in the study area were determined.Based on the optimal vegetation parameters,the backscattering coefficient of bare soil on Radarsat-2 polarization image was extracted by using water-cloud model,and the backscattering coefficient database was established by combining AIEM model.The effective surface roughness parameters?S,L?were simulated by LUT method,and the soil moisture support vector regression?SVR?model based on microwave-optical coupling was constructed.In order to further explore the surface scattering information contained in polarization SAR data,different polarization decomposition techniques were used to extract polarization characteristic parameters.After analyzing the scattering characteristics reflected by polarization characteristic parameters and the correlation between polarization characteristic parameters and soil moisture,this paper constructed a soil moisture inversion model based on polarization characteristic parameters by using support vector regression model.By synthetically comparing the results of soil moisture inversion and precision evaluation indexes of the two models,the best inversion model suitable for inversion of soil moisture in blown-sand region was determined.The main research results are as follows:?1?By comparing the simulation results of different vegetation parameters?LAI,NDVI,RVI,VWC?on the scattering characteristics of vegetation,it was found that the ratio vegetation index?RVI?was the most suitable model parameter for describing the scattering characteristics of sparse vegetation in blown-sand region.?2?The accuracy of inversion of effective correlation length by backscattering coefficient was related to the choice of polarization mode and root mean square height.When S=0.7cm,the correlation coefficient between backscattering coefficient and effective correlation length reached the maximum.Compared with other polarization modes,VV polarization has higher fitting accuracy,R²=0.806.?3?Combined with water-cloud model and AIEM model,a SVR model of soil moisture based on microwave-optical was constructed to realize surface soil moisture inversion.The results showed that the inversion accuracy of VV polarization was the highest,and the accuracy evaluation indexes was R²=0.869,MAE=3.99%,RMSE=5.38%.The inversion spatial distribution characteristics of surface water was consistent with the distribution rules of sand dunes and bottomland in the study area.?4?The polarization characteristic parameters were extracted by Cloude-Pottier and Freeman-Durden polarization decomposition techniques,and the correlation between polarization parameters and surface soil moisture was analyzed.The results showed that there was a strong positive correlation between the alpha angle???,second eigenvalue?₂?,third eigenvalue?₃?,Freeman-Durden decomposition of double scattering component?Dbl?and volume scattering component?Vol?,radar vegetation index?RVI?and soil moisture.Single-bounce eigenvalue relative difference?SERD?was negatively correlated with soil moisture;entropy?H?,first eigenvalue?₁?,odd scattering component of Freeman-Durden decomposition?Odd?were weakly positively correlated with soil moisture,while antientropy?A?,double-bounce eigenvalue relative difference?DERD?was weakly negatively correlated with soil moisture.?5?Based on the scattering characteristics reflected by different polarization characteristic parameters and the correlation between them and surface soil moisture,combined with principal component analysis?PCA?dimension reduction technology,the optimal characteristic parameters for inversion of surface soil moisture in the study area were obtained,and the SVR model of soil moisture based on polarization characteristic parameters was constructed.The inversion results showed that compared with the direct combination of different characteristic parameters,the SVR model could be used to invert surface soil moisture.The principal component analysis?PCA?dimensionality reduction method was more suitable for the inversion of surface soil moisture in the study area.?6?By synthetically comparing the inversion results and precision evaluation indexes of the two models,the inversion results of SVR model based on microwave-optical coupling was in good agreement with the field measured data,which is more suitable for the inversion study of surface soil moisture in blown-sand region.