Comparison of Linear Regression and Neural Networks to Estimate the Wetting Dimensions on Sloping Lands

The aim of this study was to compare the linear regression and neural networks methods in estimation of the wetting dimensions in the drip irrigation systems on sloping lands. Experiments were performed with a constant flow rate of 4 L.hr-1 with five irrigation duration times of 4, 6, 8, 10 and 12 hours on the lands with sloping of 0, 5, 15 and 25 percent on a silty loam soil in Moghan Fathali plain region at four replications. The Results of the estimation of the wetting front depth by means of statistical indices of R2, EF and RMSE, using multi-layer perceptron neural networks were 0.98 and 0.98 and 1.07 cm, respectively and using multiple linear regression method were 0.93 and 0.93 and 2.1 cm, respectively. The comparison results of these mentioned methods for estimating area of wetted soil profile using statistical indices of R2, EF and RMSE, were 0.99, 0.99 and 22.16 cm2, also 0.93, 0.93 and 74.77 cm2, respectively. So, the multi-layer perceptron neural network was more suitable for the estimation depth of the wetting front and area of wetted soil profile than the multiple linear regression method. However, Results of comparison between multi-layer perceptron neural networks and multiple linear regression methods for the estimation of the wetted soil surface area using statistical indices of R2, EF and RMSE, were 0.99, 0.99 and 18.22 cm2, also 0.90, 0.90 and 126.44 cm2 respectively, showed that the multiple linear regression was more appropriate than the multi-layer perceptron neural networks.