Estimation of Soil Infiltration and Cation Exchange Capacity Based on Multiple Regression, ANN ( RBF, MLP), and ANFIS Models

Investigation of soil properties such as cation exchange capacity (CEC) and soil infiltration is an important role in environmental research. The measurement of these parameters is time-consuming and costly. In this study, intelligence-based models [artificial neural networks (MLP and RBF), adaptive neuro-fuzzy inference system (ANFIS), and multiple regression (MR) techniques] are employed as alternatives to estimate the CEC and soil infiltration parameters from more readily available soil data. Two hundred soil samples were collected from soil 0-30 cm deep from two sites of the Ghoshe Region in Semnan Province, Iran. The first site was a flood plain and second site was agriculture land. The input data for models were electrical conductivity (EC), soil texture, lime percentage, sodium adsorption ratio (SAR), and bulk density. To evaluate the performance of these models, the statistical parameters such as root mean square error (RMSE), mean absolute error (MAE), mean error (ME), and coefficient of determination (R-2) were used. Then the results of the intelligence-based models and MR were compared to each other's. The results show that the MLP model was better than ANFIS, MR, and RBF models. Finally, sensitivity analysis was conducted to determine the most and the least influential variables affecting the soil infiltration and CEC parameters. It was found that EC and bulk density have respectively the most and the least effect on soil infiltration, and for CEC they were clay percentage and bulk density, respectively.