Spatial distribution of soil salinity using auxiliary variables and hypercube sampling method in Meybod

Digital mapping is a suitable alternative method to the traditional methods. In this method, soil salinity correlated to the environmental variables and then soil salinity predicts in the other locations. At present research, based on thehypercubemethod, the locations of 73 soil samples selected and then sampled. Electrical conductivity was measured in the saturation paste of soil samples. Then using artificial neural network (ANN) the relationship between ground point data and environmental variables (terrain attributes and Landsat 8 image data) was calculated and applied to the other parts of area. Sensitivity analysis indicated some environmental variables had more influence on prediction ANN model including normalized difference vegetation index (39.51%), soil-adjusted vegetation index (27.60%) and slope (5.80%), respectively. Moreover, the cross-validation implied high performance of ANN model to predict soil salinity (R2=0.57 and RMSE=17.40 dS/m). Our results, overall, showed that remote sensing data and digital elevation model and ANN had acceptable performance to predict soil salinity and hence it is recommended the same methodology in the future.