Evaluation of soil salinity by analyzing Landsat-8 images and field Observations (Case study: Behesht-e- Gomshodeh, Fars province)

Soil salinity is considered as one of the potential environmental hazards. The purpose of this study was to find the best index and the most suitable relationship for estimating soil salinity and its mapping using remote sensing data. At the first step, random sampling was performed using fishnet method and surface soil electrical conductivity (EC) measurements. Then, the threshold levels (92%, 95%, and 98%) were applied to the output images of each indicator. The methodology included using the least squares fitting (LS-fit) technique and principal components analysis (PCA) for halite and gypsum minerals, determining the correlation between the output of indices and ground data, and performing clustering and factor analysis between EC and output images. In order to select the best model derived from Landsat-8 band combinations and the amount of salinity, collinearity test, Durbin-Watson test, and backward multivariate regression were employed. The Cohen‘s kappa coefficient was also applied to evaluate the multivariate regression formed by Landsat-8 bands. The performance of the indicators was evaluated based on four criteria of root mean square error (RMSE), mean bias error (MBE), mean absolute error (MAE) and R-squared (R2). The results of the factor analysis showed the smallest distance between the EC, salinity index (SI) and brightness index (BI). The SI with an amount of 0.89 had the highest Pearson correlation with EC. In the dendrogram diagram, SI index with EC was placed in a cluster, and the RMSE, MBE, MAE and R2 values of the SI index were estimated to be 0.16, 0.11, 0.12, and 0.76, respectively. Compared to the rest of the indicators and linear, multivariate regression (with Cohen‘s kappa coefficient of 60%,), the SI index has provided better outcomes.