This study evaluates the performance of the Sentinel-2 multi-spectral instrument (MSI) and Landsat-7 Enhanced Thematic Mapper Plus (ETM+) for soil salinity mapping across contrasting agroecosystems in Egypt, with particular emphasis on subsurface salinity conditions (>0.5 m). A multi-stage calibration framework was implemented, in which historical Landsat-5 imagery (1995) was first integrated with field-measured electrical conductivity (EC) data to establish a spectral baseline. This baseline was subsequently applied to Sentinel-2 and Landsat-7 imagery acquired in 2015 and validated using in-situ total dissolved solids (TDS) measurements. Among the evaluated spectral indices, Salinity Index 5 (SI5) demonstrated the strongest relationship with field data and was selected for salinity mapping. Comparative analysis revealed that Sentinel-2 significantly outperforms Landsat-7, achieving a higher predictive accuracy (R-2 = 0.89) compared to Landsat-7 (R-2 = 0.72), primarily due to its finer spatial resolution (10 m) and reduced mixed-pixel effects. In addition, the application of second-degree polynomial regression substantially improved model performance relative to linear approaches, confirming the non-linear nature of soil salinity-spectral relationships. The results further indicate that surface spectral indices can provide meaningful estimates of subsurface salinity under specific environmental conditions. Overall, the integration of multi-temporal satellite data, robust spectral indices, and non-linear modeling provides an effective framework for soil salinity assessment in arid environments. This approach enhances the reliability of remote sensing-based monitoring and supports sustainable land management in salinity-affected regions.