Comparison of Artificial Neural Network and Regression Pedotransfer Functions for Estimation of Soil Cation Exchange Capacity in Tabriz Plain

Soil cation exchange capacity (CEC) is defined as the amount of positive charge that can be exchanged per mass of soil. Modeling and estimating of CEC is a useful index of soil fertility. Assessing and designing various management scenarios requires having accurate information regarding the soil data bank. In order to estimate the soil CEC, 32 profiles were dug in Tabriz plain, and 131 different samples were collected from different depths and physiochemical experiments such as particle size distribution, organic carbon, pH and CEC of soil samples were performed. Then using seven regression models that were selected based on previous studies, were calibrated and evaluated for the study area. Also seven different architectures of artificial neural networks were designed to predict the CEC of soil and the results of artificial neural networks and multivariate regression models were evaluated using correlation coefficient (R2), root mean square error (RMSE). Results revealed that artificial neural network with R2 = 0.86 and RMSE= 2.14 is better than regression based functions due to the existence of nonlinear relations between the easily available soil properties (independent variables) and the CEC (dependent variable).