Modeling of soil salinity within a semi-arid region using spectral analysis

Soil salinity, is an environmental problem that threatens lands mainly in arid regions. Traditionally, monitoring of salty soils has been a hard task, due to the high expenses of soil samples analysis. Recently, remote sensing techniques provide new alternative to assess and monitor salt affected soil rapidly and over larger areas. The investigation area is a semi-arid region located in southern Tunisia. This study aims to identify classes of soil salinity, explore the potential of multispectral data to discern soil features and patterns of saline soil and predict soil salinity. For this purpose, Landsat 8 data were used to generate nineteen spectral indices. Correlations between reflectance indices and Electrical Conductivity (EC) measured on laboratory, showed that the Short Wave Infrared (SWIR) offers the best correlation with -57 %. Three salinity classes were obtained from image classification, with an overall accuracy 73%. Partial Least Square Regression (PLSR) method was applied to estimate soil salinity. The calibration model gives a moderate coefficient of determination R² =52 % and a RMSE=0.66 dS/m.