Remote sensing and statistical analyses for exploration and prediction of soil salinity in a vulnerable area to seawater intrusion

Soils along the Egyptian coast are vulnerable to environmental degradation and soil salinity problems. The main objective of this study is to identify and rapidly predict salt affected soils using remote sensing data and multivariate statistical analysis. To achieve this objective, the Operational Land Imager 8 (OLI) of Landsat imagery acquired in March 2022 was processed through the Maximum Likelihood classifier to assess Landscape features and to produce Normalized difference salinity index (NDSI) and normalized difference vegetation index (NDVI). Water and soil samples were collected from 13 field sites as ground truth data and to investigate representative physical and chemical properties. Linear regression model was used to predict soil salinity while soil parameters were mapped using Inverse Distance Weight (IDW) in ArcGIS 10.5. In order to explore the soil salinity content using VNIR-SWIR spectra, this work investigated the potential of Partial least squares regression (PLS regression) and SVM (Support vector machine). For simulating salinity in the investigated area, a total number of 65 different sites were identified considering that almost 75 % (50 sites) were used to develop the model and 25 % (15 sites) for validation of the established model. The results indicated that EC levels of water samples are not suitable for irrigation (> 3 mS/cm). Majority of the collected soil samples represent saline-alkaline soils. NDSI ranged from -0.83 to 0.57 with mean of -0.25. Based on the variance of components, 90 % of data were obtained from the first three PCA. The PLS model's R² score of 0.763 and extremely low p value indicates how well it predicts soil salinity. SVM model R², on the other hand, is 0.719. Further, Ca⁺⁺ and Mg⁺⁺ are the main significant parameters selected in the predicted model. This shows that remote sensing data and multivariate analysis are very important tools to map spatial variation and predict soil salinity. The developed model for salinity considered both the spectral retrieved parameters and lab analyses of cations giving higher accuracy.