Soil salinization remains a major constraint for sustainable agriculture under increasing aridity and irrigation intensification. The study aims to develop and evaluate a predictive model for soil electrical conductivity (EC) using multisensor remote sensing data and machine learning algorithms. The study was conducted in the liman ecosystems of the Southern Trans-Urals. Satellite data from Sentinel-1, Sentinel-2, and MODIS were used, including NDVI, NDBI, LST, and spectral bands B2– B12. The Random Forest algorithm was implemented in the Google Earth Engine environment, with model validation based on laboratory measurements of soil EC. The developed model achieved high accuracy (R² = 0.888, RMSE = 2.20 dS/m, MAE = 1.47 dS/m). The most influential predictors were B8 (NIR), B11 (SWIR1), and LSTC. Correlation analysis revealed significant relationships between NDVI (r = −0.57), LST (r = 0.61), and EC. Spatial analysis showed that areas of high salinity are primarily located in lowlands and poorly drained depressions. The results confirm the effectiveness of integrating multispectral, radar, and thermal data for soil salinity assessment. The proposed approach provides reliable mapping of saline soils and can be used for monitoring, land management, and reclamation planning. Future work will focus on expanding the dataset and applying deep learning models to improve predictive performance.
Miftakhov et al. (2026) studied this question.