Abstract:In order to further study the interaction between shallow groundwater, vegetation and soil of arid and semiarid regions, the database of the Sentinel-1A, Landsat images, soil moisture and the groundwater depth were utilized to quantitatively analyze the information of soil moisture and groundwater depth in the study area by the model of support vector machine (SVM) regression algorithm. Furthermore, the comparison of optical remote sensing and microwave remote sensing collaborative inversion in soil moisture and groundwater depth was also analyzed. By the survey of soil moisture and groundwater depth in the study area, the results indicated that the highest accuracy in SVM model was the correlation between soil water content in 0~10cm and groundwater depth. The accuracy of temperature vegetation drought index (TVDI) was improved, through the C calibration model. It was feasible to invert the groundwater depth by SVM model with different parameters. For single factor modeling, the model constructed by TVDIMSAVI had the highest accuracy and the R2 of modeling set was 0.74, the value of RMSE was 4.66%, and the R2 of verification set was 0.70, the value of RMSE was 4.65%, compared with only single factor (σ0soil or TVDI), σ0soil and TVDIMSAVI combination work with the highest model accuracy, R2 was 0.86, and RMSE was 4.16%, the R2 of verification set was 0.92, and RMSE was 2.73%. The results of the optimal model parameters were used to retrieve the soil moisture and groundwater with good accurate. The average relative error of groundwater was 8.23%, which was better than the previous results of the study area of 18.06%.