Remote sensing (RS) has been widely applied to map soil salinity in landscapes where salinity exhibits strong spatial contrasts, characterized by high electrical conductivity (EC) values. However, its effectiveness in regions dominated by low EC values remains less understood, particularly in irrigated agroecosystems where salinization processes differ from natural dryland settings. This study evaluated RS-based models within the Riverhurst Irrigation District, Saskatchewan, where both irrigation induced salinity and naturally occurring salinity occur, and where the majority of EC values fall within the 0-2 dS/m range. Vegetation and salinity indices derived from 30 m Landsat 8 imagery, together with geomorphometric variables from a 5 m LiDAR-derived digital elevation model, were used to model soil salinity for three depth intervals (0–30, 30–60, and 60–90 cm) using Random Forest (RF) and Support Vector Machine (SVM). Model evaluation on independent test dataset showed that the SVM outperformed RF, achieving a higher coefficient of determination (R2) of 0.77 and a lower root mean square error (RMSE) of 0.48, compared to RF (R2=0.66, RMSE=0.51) .
Gowera et al. (Wed,) studied this question.