Predicting ESP and SAR by artificial neural network and regression models using soil pH and EC data (Miankangi Region, Sistan and Baluchestan Province, Iran)

Monitoring exchangeable sodium percentage (ESP) and sodium adsorption ratio (SAR) variability in soils is both time-consuming and expensive. However, in order to estimate the amounts of amendments and land management, it is essential to know ESP and SAR variations and values in sodic or saline and sodic soils. Thus, presenting a method which uses easily obtained indices to estimate ESP and SAR indirectly is more optimal and economical. Input data of the current research were 189 soil samples collected based on a regular networking approach from Miankangi region, Sistan plain, Iran. Then, their physicochemical properties were measured. Results showed that SAR = 3.8 × ln(EC) + 22.83 × ln(pH) – 44.37, (R ² = 0.63), and ESP = 3.98×ln(EC) + 36.88(pH) – 56.98 (R ² = 0.78) are the best regression models for estimating SAR and ESP, respectively. Moreover, multilayer perceptron (MLP), which explains 95–97% of parameters of soil sodicity using EC and pH as inputs, was the best neural network model. Therefore, MLP could be applied for ESP and SAR evaluation with high accuracy in the Miankangi region.