Remote Sensing Monitoring of Soil Salinization Based on Bootstrap-Boruta Feature Stability Assessment: A Case Study in the Minqin Lake Region

Summary

Satellite imagery: Sentinel-1 SAR and Sentinel-2 MSI data processed at 10 m resolution, including salinity, vegetation, and moisture indices. Ground data: 144 soil samples (June–July 2024) with measured electrical conductivity (EC) for calibration and validation. Terrain and climate data: DEM-derived terrain features and 40-year climate averages to enhance environmental modeling.

Programming functions: Custom scripts for feature stability assessment, model training, and salinity mapping workflows. Notable Findings Identified 11 stable remote sensing features (e.g., NDSI, MNDWI, CRSI) with consistent predictive power. The Bootstrap-Boruta + RBMO-BPNN approach achieved the highest accuracy (R² = 0.83, RMSE = 1.23), outperforming traditional methods by up to 24%.

Spatial prediction showed 31.5% of the region affected by salinization, with severe cases concentrated along oasis edges. Interpretation & Usage This dataset enables: Reproduction of the feature stability framework and predictive models. High-resolution mapping of soil salinity to guide land management, agricultural planning, and ecological restoration.

Adaptation of the methodology to other arid or semi-arid regions with appropriate environmental adjustments.