Abstract: Accurate estimation of soil moisture content is of great significance for eco-environmental conservation and sustainable development in arid areas. In Juyanze, the southeast of Ejina banner, the western Inner Mongolia, 4 drought index models, including Temperature Vegetation Drought Index (TVDI), Perpendicular Drought Index (PDI), Normalized PDI (NPDI), and Soil Moisture Monitoring Index (SMMI), were established based on 3 Landsat-8 images and 150 in-situ soil moisture samples from different soil depths. Model accuracy and applicability were compared and verified, and the optimal model of TVDI was used to retrieve soil moisture content from 2015 to 2017. Five land use types, namely sandy land, saline-alkali land, bare land, vegetation and tidal flats were classified based on random forest classification and their differences in soil moisture content were analyzed. The results showed that all the 4 drought indices were negatively correlated with the measured soil moisture content. The highest fitting accuracy was observed between the 4 drought indices and the surface soil moisture. Meanwhile, the fitting accuracy decreased with the increase of soil depth. The accuracy of TVDI model was higher than that of the other models, particularly in the surface soil, with a R² of 0.76. Discrepancies of the average soil moisture content were observed in different land use types, and generally the soil moisture content was the lowest in the sandy land, followed by saline-alkali land, bare land, and vegetation, the highest in the tidal flats.