Soil Moisture In The Desert Of South Xinjiang Arid Zone Based On Multi-source Remote Sensing Data Remote Sensing Monitoring Research

Soil moisture in arid regions is one of the main factors affecting land desertification and plays an important role in the process of vegetation restoration and community succession.How to obtain soil moisture information efficiently,non-destructively and accurately,and on this basis,to conduct research on the spatial distribution characteristics and driving influences of soil moisture in the surface layer（0～20 cm）of deserts is a hot issue at present.Therefore,this study uses the advantages of optical remote sensing Landsat 8 OLI/TIRS and microwave remote sensing Sentinel-1 SAR data to calculate several spectral indices by optical remote sensing with the study area of Aksu region in South Xinjiang Kongtailike,and at the same time carries out relevant improvements on the basis of spectral indices to carry out relevant optical remote sensing monitoring studies,and acquires soil multi-polarization backscattering coefficients by microwave data.The study is based on the acquisition of soil multi-polarity backward scattering coefficients from microwave data,and the study is carried out by multiple linear regression（MLR）,partial least squares regression（PLSR）,support vector machine（SVM）,Random Forest（RF）,Decision Tree Regression（Cubist）,Partition+PLSR and Partition+Cubist models,and other algorithms to construct an integrated inversion model based on optical data（multispectral indices）to study the spatial distribution of desert soil moisture in arid regions.The main findings of the research include the following three points:（1）Remote sensing inversion study of desert soil moisture based on Landsat8 dataThe 26 preferred spectral indices,such as TVDI,NR and GLI,Ts and DEM all reached highly significant correlation with soil moisture and can be used as indicator factors for remote sensing modelling of desert soil moisture in the southern arid zone;Comparing the three models,the R²of the modelling set and prediction set of the RF model were 0.93 and 0.91 respectively,and the RPD of the prediction set was 3.90,which were the highest in all evaluation indexes.The PLSR model had the second highest accuracy and the SVM model had the lowest accuracy;Inverting the surface soil moisture in the study area with the RF model,there were obvious differences in the soil moisture distribution characteristics in different land use classifications,especially in the salt crust area.It is shown that the multi-factor and multi-index model using spectral indices,environmental factors and topographic data can invert the desert surface soil moisture in the arid zone with high accuracy,and the results of the study provide a certain theoretical basis and methodological support for remote sensing monitoring of desert soil moisture in the arid zone of South Xinjiang.（2）Remote sensing inversion of desert soil moisture based on improved spectral indicesthe correlation coefficients of the spectral indices EBSI,ECI,ECal,ENDVI and EPDI were improved by 0.02～0.11,and the correlation analysis and multiple covariance validation were performed.The spectral indices such as EBSI and ECI with improved correlation coefficients and DVI,NDWI and EPDI,which originally reached a highly significant level but the improvement effect was not obvious,were selected as the improved modelling factors.GVMI as improved modelling factors and BSI,CI and Cal as traditional modelling factors to construct improved and traditional desert soil moisture prediction models;After the spectral indices were improved,the R² of linear and non-linear model prediction sets improved by 0.12 and 0.05,respectively,and the RPD values improved by 0.35and 0.49,of which,the RPD value of the improved MLR-II model was 1.83,which can roughly estimate soil moisture,while the RPD value of the RF-II model is as high as 3.12,which can accurately predict soil moisture;the accuracy of the non-linear model is significantly better than that of the linear model,with the R²of the prediction set of the MLR linear model being only 0.59 and 0.71,while the R² of the prediction set of the RF non-linear model reaches 0.86 and 0.91;soil moisture The distribution is influenced by both natural and anthropogenic driving factors,generally showing 0～5%and 5～12%in the northeastern desert,staggered distribution in the southern farmland,and difficult soil moisture evapotranspiration in the northern and central desert-oasis transition zones inhibited by the degree of vegetation cover and surface salt crust,mostly showing 15～20%and>20%.The results reveal that the introduction of thermal infrared（b10）band improvement in the traditional spectral index is effective in enhancing the prediction of desert soil moisture,and also provides a technical methodological basis for drought control and soil conservation research in similar areas.（3）Remote sensing zoning modeling of desert soil moisture based on multi-source remote sensing dataafter correction of the soil multipolar backscattering coefficient due to the influence of surface vegetation by the water cloud model,the R²of its corrected Oh inversion model improved by 0.11 and the RMSE improved by 0.59;after partitioning with NDVI thresholds,in the whole area six,such as?σ_vh,σ_vv and LAI,eight,such as?σ_vh,σ_vv and DVI,in the bare soil area,and six,such as σ_vh,σ_vv and NDVI,in the vegetation cover area,which were highly significantly related to soil moisture,were selected as modeling factors;the accuracy of the five models,such as PLSR,partitioning+PLSR,and Cubist,with the addition of different feature parameter factors,was better than that of the Oh model,while the accuracy of the partitioned PLSR and The prediction set R²of the partitioned PLSR and Cubist models increased by 0.12 and 0.03,respectively,and the RPD increased by 0.53 and 0.49,and the prediction effect of the models changed qualitatively,with the prediction set R² of the partitioned+Cubist model being 0.90 and the RPD being as high as 3.09,which are the optimal values of the models.The soil water content distribution in different sub-districts is most seriously affected by vegetation cover,soil texture and human activities,with bare soil areas having a high soil water content due to the water retention effect of surface salt crusts,high vegetation cover areas having the most obvious water retention effect of crops and the highest soil water content,and low vegetation cover areas having a high/low staggered distribution trend due to human activities.This study provides a scientific reference value for the inversion of desert soil moisture in the arid zone through the synergy of multi-source remote sensing data,while combining the research idea of zonal modelling.