Evaluation and Modeling Soil Salinity Using Remote Sensing, Regression Model and Random Forest

Nowadays, soil salinization is one of the world’s major threats that reduce soil productivity by intensifying the process of desertification and land degradation. Since laboratory analysis is mostly time consuming and costly, especially in large scales, attempts have been made to study soil salinity using remote sensing techniques in recent years. The present study assessed the potential of remote sensing in predicting soil surface salinity in the east of Lenjan                 City. Salinity reference points were identified based on analyzing 50 randomly selected surface soil samples. Satellite indices including DVI, NDVI, EVI, MSAVI, SAVI, RVI, NDWI, SI1, SI2, SI3 and SBI were derived from a Landsat-8 satellite image (path and row of 164 and 37) acquired on September 13, 2019. These indices along with three topographic indices of elevation, slope and topographic wetness index (TWI) were introduced to the Multiple Linear regression and Random Forest models. The linear regression model was built using band 6, RVI, NDVI, elevation and TWI with a p-value of 0.049. In the Random Forest model, band 7, slope, band 5 and elevation were among the most important parameters. The r2 value of this model was 0.21. The results of this study showed that topographic indices had also great importance in salinity prediction. Moreover, comparison of the results indicated that Random Forest had a higher accuracy than the regression model for determining salinity in the study area.