The study of soil characteristics such as the ability to exchange positive ions CEC (Cation Exchange Capacity) play a significant part in study of ecological researches, also it is important for decision concerning pollution prevention and crop management. CEC represents the number of negative charges in soil, since direct method for measuring CEC are cumbersome and time consuming Lead to the grow of indirect technique in guessing of soil CEC property. Pedotransfer function (PTFs) is effective in estimating this parameter of easy and more readily available soil properties, 80 soil sample was taken from diverse horizons of 20 soil profiles placed in the Aljazeera Region, Iraq. The aim of this study was to compare Neural Network model (feed forward back propagation network) and Stepwise multiple linear regression to progress a Pedotransfer function for forecasting soil CEC of Mollisols and Inseptisols in Al Jazeera Irrigation Project using easily available features such as clay, sand and organic matter. The presentation of Neural Network model and Multiple regression was assessed using a validation data set. For appraise the models, Mean Square Error (MSE) and coefficient of determination R2 were used. The MSE and R2 resultant by ANN model for CEC were 2.2 and 0.96 individually while these result for Multiple Regression model were 3.74 and 0.88 individually. Results displayed 8% improvement in increasing R2 and also improvement 41% for decreasing MSE for ANN model, this pointed that artificial neural network with three neurons in hidden layer had improved achievement in forecasting soil cation exchange capacity than multiple regression. So we can conclude that ANN model by use (MLP) multilayer perceptron for predicting CEC from measure available soil properties have more accuracy and effective compared with (MLR) multiple linear regression model.
Modelling Soil Cation Exchange Capacity in Different Land-Use Systems using Artificial Neural Networks and Multiple Regression Analysis
