Quantitative assessment of soil salinity using remote sensing data based on the artificial neural network, case study: Sharif Abad Plain, Central Iran

Land salinization is one of the most important factors in reducing the soil quality of agricultural land. Accordingly, these regions affected agricultural production and ecological development. Therefore, it is important to assess soil salinity driving factors. However, it is difficult to characterize soil salinity using single-factor and linear models. This research was carried out for soil digital mapping using remote sensing to classify saline lands and their spatial distribution in SharifAbad plain. The methodology is based on the differentiation of saline soils by combining the Landsat 8 data, fieldwork, and neural network model for prediction of soil salinity. The Latin hypercube method is based on the stratified sampling method. Based on this technique, 63 samples were selected from 0 to 30 cm of the soil surface. In the ANN model, considered 30% of the soil EC as the validation set and the rest (70%) for the testing set. To model soil salinity, auxiliary variables such as Landsat 8 satellite image of 2016 including 2–5 main bands and band 7, topographic auxiliary data includes DEM, TWI, TCI, and spectral parameters were extracted. The result revealed that the GFF algorithm which, according to R2 and MSE statistics, the best way to prepare a soil salinity map in Sharif Abad plain. The ANN model in most cases satisfies the EC amount less than the real value. We found that the average values of SI5, TCI, and TWI values in the non-saline, the class were lower than the saline class, and the average values of DEM and NDVI indices in the non-saline class were higher than the saline class and showed a statistically significant difference.