[Objective] The variation characteristics of desert soil moisture were analyzed to provide a theoretical basis and methodology for remote-sensing monitoring of soil water content in the arid desert of Southern Xinjiang Uygur Autonomous Region.[Methods] The desert soil moisture modeling indicators were constructed based on Landsat 8 data. An optimal 26 spectral index, land surface temperature (T_s), and digital elevation model data (DEM) were selected as modeling factors, and the soil water inversion model was constructed using the partial least squares regression (PLSR), support vector machine (SVM), and random forest (RF) algorithms. After model validation and comparison, the spatial distribution of soil moisture in Kongtailike was retrieved using the optimal model.[Results] ① The temperature vegetation dryness index, NR, GLI, and other 26 preferred spectral indices, as well as T_S and DEM, were significantly correlated with soil moisture. They could be used as indicators for remote-sensing modeling of desert soil moisture in the arid area of Southern Xinjiang. ② Among the three models, the R² of calibration and validation based on the RF model were 0.93 and 0.91, respectively, and the RPD of validation was 3.90, which was the highest. The PLSR model accuracy was the second best, and the SVM model accuracy was the lowest. ③ The surface soil moisture in the study area was retrieved by the RF model, and the characteristics of soil moisture distribution in different ground classifications were noticeably different, especially in the salt crust region.[Conclusion] The comprehensive use of spectral index, environmental factors, and terrain factors could result in the inversion of the soil water content in arid areas with a higher accuracy, providing scientific value for the desertification and ecological environment control in this area.