Abstract
Cation Exchange Capacity (CEC) is an important characteristic of soil in terms of nutrient and water holding capacities and contamination management. Measurement of CEC is laborious and time-consuming. Therefore, CEC estimation through other easily - measured properties is desirable. In this study, PTFs for estimation of cation exchange capacity from basic soil properties such as particle-size distribution, organic carbon, percentage saturation and pH were developed and validated using artificial neural network (ANN) and multiple-linear regression methods and the predictive capabilities of the two methods was compared using some evaluation criteria. Total of 200 soil samples was divided into two groups as 165 for the development and 35 for the validation of PTFs. Accuracy of the predictions was evaluated by the criteria of coefficient of determination (R2) and the root mean square error (RMSE) between the measured and predicted CEC values. Clay (%), OC (%), SP and sand (%) predicted CEC better than other models with an R2=0.81 and RMSE=3.05 cmol.kg-1 when a neural networks model with one hidden layer and seven nodes was used. The R2 and RMSE varied from 0.66 to 0.69 and from 4.26 to 4.69 for regression, and varied from 0.78 to 0.81 and from 3.05 to 3.29 for ANN, respectively. Results showed that neural networks predictions is better than regression-based functions.

Key words: Artificial neural networks, Cation Exchange Capacity, Chaharmahal - Bakhtiari