Estimation of infiltration rate and deep percolation water using feed-forward neural networks in Gorgan Province

摘要

The two common methods used to develop PTFs are multiple - linear regression method and Artificial Neural Net work. One of the advantages of neural networks compared to traditional regression PTFs is that they do not require a priori regression model, which relates input and output data and in general is difficult because these models are not known. So at present research, we compare performance of feed - forward back - propagation network to predict soil properties. Soil samples were collected from different horizons profiles located in the Gorgan Province, North of Iran. Measured soil variables included texture, orga nic carbon, water saturation percentage Bulk density, Infiltration rate and deep percolation. Then, multiple linear regression and neural network model were employed to develop a pedotransfer function for predicting soil parameters using easily measurable characteristics of clay, silt, SP, Bd and organic carbon. The performance of the multiple linear regression and neural network model was evaluated using a test data set by R 2 , RMSE and RSE. Results showed that artificial neural network with two and five ne urons in hidden layer had better performance in predicting soil hydraulic properties than multivariate regression. In conclusion, the result of this study showed that both ANN and regression predicted soil properties with relatively high accuracy that show ed that strong relationship between input and output data and also high accuracy in determining of data.