Estimating seepage losses from lined irrigation canals using nonlinear regression and artificial neural network models

Abstract

The Slide2 model was used to estimate seepage losses from canals after validation considering different canal geometries, lining thicknesses, and lining materials. SPSS was used to develop three models: NLR, MLP-ANN, and RBF-ANN. MATLAB software was used to write down the script code for the ANNs. Results showed that seepage losses were highly increased when the liner had high hydraulic conductivity, while with the increase of lining thickness, a noticeable reduction in seepage losses was obtained. The canal's side slope had a minimal effect on the seepage losses. Moreover, the MLP-ANN and RBF-ANN models performed better than the NLR model with determination coefficient (R2) of 0.996 and 0.965; Root-Mean-Square-Error (RMSE) of 1.172 and 0.699; Mean-Absolute-Error (MAE) of  0.139 and 0.528; index of agreement (d) = 0.999 and 0.991, respectively. The NLR model had lower values of R2 = 0.906, RMSE = 1.198, MAE = 0.942, and d = 0.971. Thus, ANNs are recommended as a robust, easy, simple, and rapid tool for estimating seepage losses from lined trapezoidal irrigation canals.