Estimating soil salinity in the dried lake bed of Urmia Lake using optical Sentinel-2B images and multivariate linear regression models

The soil salinity is one of the most common and important factors in soil degradation in arid and semi-arid regions. It is important concern to predict and monitor soil salinity. In recent years, using remote sensing techniques to monitor soil salinity has been significantly developed and various models have been extended for this purpose. Among all the methods, linear regression is the most commonly used due to its simplicity and computational efficiency. In Iran, many lands are faced with a significant increase in salinity. Beaches of Urmia Lake is one of these area. The purpose of the present study was to evaluate the capability of Sentinel multispectral imagery as well as to compare univariate and multivariate linear regression models to estimate soil salinity at 0 to 10 cm depth in the eastern margins of dried lake-bed of Urmia. For this purpose, 38 soil samples (training and test samples) were collected from three different locations A1, A2 and A3 in in this study area with different salinity values at the satellite transit time. Then their electrical conductivity (EC) was measured in the laboratory. The samples were collected near accessible roads by three mapping teams. Of the total number of samples, 28 and 10 sample were considered as training and test samples respectively. Then the Sentinel 2B multispectral satellite image was prepared by resampling simultaneously on October 6, 201 .In this research, eight spectral bands of the Sentinel image (visible and infrared bands) and 17 salinity indices were utilized. Then, in each of the spectral bands and salinity indices, different univariate linear regression models were calibrated to estimate soil salinity. As well as multivariate linear regression models were designed using simultaneous spectral bands and soil salinity indices. The accuracy assessment of both methods was estimated using 10 test samples by the coefficients of determination and Root Mean Square Error parameters. In univariate linear regression model, the best (results) model for estimating soil salinity were presented of narrow infrared band (8a) and BI salinity index with the highest and lowest values of and RMSE based on test samples, respectively. The and RMSE were obtained for band 8a and BI index 0.89, 0.83, 20.85 and 21.33, respectively. Compared to univariate linear regression models, the proposed method in this paper is based on multivariate linear regressions with 7 variables provided the highest accuracy among all multivariate and univariate regression models. The and RMSE based on test samples were obtained for multivariate linear regressions with 7 variables 0.97 and 8.77 respectively. Finally, soil salinity maps of the area were prepared with the best regression models. Model evaluation results showed that multivariate linear regression models increased the accuracy of soil salinity estimation by 11.81% in comparison with univariate regression models. These results showed the potential of multivariate linear regression model and multi spectral sentinel image to estimate soil salinity content in dried lake-bed of Urmia Lake.