Environmental associations in spectral index-based soil salinity detection: A multi-region and multi-sensor...

Soil salinization is a major form of land degradation that threatens agricultural productivity and food security worldwide. Although satellite-based spectral indices are widely used for salinity monitoring, their performance under contrasting environmental conditions remains poorly understood. This study evaluated the capability of spectral indices derived from Landsat 8, Sentinel 2, and PlanetScope imagery to estimate soil electrical conductivity (ECe). Model performance was evaluated across soil clay-content classes and climate-associated regional conditions in three representative regions: southeastern Spain, the Lake Urmia basin in Iran, and the northeastern Nile Delta in Egypt. A total of 290 soil samples were analyzed, and relationships between ECe and spectral indices were assessed using correlation analysis, Mantel tests, and regression modeling. Results demonstrate that salinity prediction accuracy is strongly environment-dependent. In soils with <20% clay, regression models produced moderate accuracy (R² = 0.458-0.506 and RMSE = 0.365-0.378 dS m^-1). In soils with 20-40% clay, model performance declined substantially (R² = 0.211-0.237 and RMSE >5.25 dS m^-1). Climate-associated regional differences were also strongly related to model performance, with the Mediterranean semi-arid region showing the highest model performance (R² = 0.546-0.647), followed by cold semi-arid conditions (R² = 0.309-0.364), while no reliable models could be established under arid climate conditions. Sensor performance depended on the stratification criterion. Vegetation- and salinity-related indices consistently emerged as the most informative predictors. These findings demonstrate that universal remote sensing models for salinity mapping are unreliable across contrasting environments and emphasize the need for environmental stratification in operational monitoring frameworks.