Abstract: Soil salinization severely hinders agricultural development in arid regions and threatens sustainable use of land resources．Effective monitoring for soil salinity dynamics is crucial for achieving agricultural sustainability and maintaining ecological stability．Saline soil areas in western Jilin Province were selected as study region．Data sources included 111 field-measured soil samples collected in 2024 and Sentinel-2 remote sensing imagery．Environmental variables closely related to soil salinization, such as vegetation, topography, climate, and groundwater, were selected．Extreme Gradient Boosting Decision Trees（XGBoost）algorithm was employed to assess importance of spectral bands and environmental variables, while redundant features were removed．Soil salinity estimation models were then constructed using Gaussian Process Regression（GPR）, Elastic Net Regression（EN）, Random Forest Regression（RF）, and XGBoost．Results demonstrated that estimation model constructed using filtered spectral bands and environmental variables （Strategy II） performed exceptionally well, with a coefficient of determination（R²）of 0.80, a root mean square error（RMSE）of 5.83 dS/m, and a relative pervent deviation（RPD）of 2.11．By integrating environmental variables and spectral information, their roles in soil salinity accumulation process were investigated, and providing a scientific basis for soil salinity prediction．