Eliminating the influence of moisture to enhance the accuracy of soil salinity estimation using hyperspectral remote sensing

Soil salinization is a global land degradation problem that restricts agriculture and ecological stability. Precise monitoring is essential for effective management. Hyperspectral remote sensing enables rapid soil salinity assessment, but soil moisture spectrally interferes and weakens model accuracy. This study examined soil spectral responses under different salinity and moisture levels. Three preprocessing methods—OSC, PDS, and MSC—were used to reduce moisture interference, and UVE was applied to select sensitive bands. A CNN-based model estimated soil salinity. Results showed that spectral reflectance increased with salinity and decreased with moisture. All correction methods improved prediction accuracy, with OSC performing best at removing water noise. The optimal model reached an R² of 0.72 and RMSE of 7.712. The study confirms the importance of moisture correction in remote-sensing salinity estimation and provides a technical reference for dynamic regional soil salinization monitoring.