Comparison of Linear and Nonlinear Pedotransfer Functions with Artificial Neural Networks in Prediction of Surface Fractal Dimension

Investigation of soil hydraulic properties like soil moisture retention curve and unsaturated hydraulic conductivity plays an important role in study of environmental researches in which their spatial and temporal variability led to development of indirect methods in prediction of these soil characteristics. Therefore، in this study indirect methods have been used in order to estimate surface fractal dimension to predict soil moisture curve. One parameter linear and nonlinear regressions were developed and compared to artificial neural networks by using readily available parameters like soil clay content، water content at permanent wilting point، cation exchange capacity and soil porosity. In the training step of regression analysis and neural networks، 97 measured soil samples and in the testing step 24 rest of soil samples with Petersen et al. (1996) database were used. The calculated values of RSE and RMSE showed that neural networks with seven neurons in the hidden layer are able to estimate surface fractal dimension from the easily available parameters more accurate than the other models.