Modeling and Mapping of Soil Salinity and Moisture Using Spectral and Radar Remote Sensing

Soil salinity is caused by natural or human processes and is a major environmental hazard. There is also a lack of soil moisture which has a negative impact on agricultural activities in mountainous areas where most of the climate is semi-humid. The main purpose of this study is to map soil salinity and moisture located in the western part of Lake Urmia in Iran using Sentinel 1 and 2 satellites along with five neural network algorithms to model soil salinity and moisture. Learning models are multilayer neural networks (MLP-NN), radial basis radiation performance (RBF-NN), Gaussian processes (GP), support vector regression (SVR), and random forests (RF). First, different salinity and soil moisture indices were obtained using different algorithms, then using 60 soil samples at a depth of 5 to 15 cm during a field survey on 06/18/1398 along with the image time. Sentinel 1 and 2 were harvested, precision was performed. In the soil salinity indices used in optical images, the salinity index with an accuracy of 0.96 was the best indicator for estimating soil salinity according to comparison with terrestrial data. The NDWI index also had the highest accuracy for estimating moisture in optical images with an accuracy of R2= 0.89. The accuracy of estimating soil moisture and salinity in radar images was R2=0.80 and R2= 0.89, respectively. The performance of five algorithms for modeling was also evaluated and compared using mean square error (RMSE) and correlation coefficient (R2). The results showed that the GP model had the highest predictive performance (RMSE = 2 and R2 = 0.82) and was better than other machine learning models.