Hassan Rashidi - Allameh Tabataba&#۰۳۹;i University
Navid Raisi - Qazvin Branch, Islamic Azad University

Soil is an essential natural resource for life that provides the required substrate on which plants grow and flourish. One of the challenges for environmental specialists is to accurately estimate and control soil erosion. MPSIAC (Modified model of Pacific Southwest Inter-Agency Committee) is a common model for estimating erosion and sedimentation rate. In this study, we used MPSIAC, regression and artificial neural networks (ANN) to estimate sediment yield in Kordan Basin, a region in Alborz Province of Iran. The erosion and sedimentation data of the region were collated using the opinions of sedimentation experts. A linear regression was performed in Weka software to determine the factors influencing the sedimentation rate. Based on the results and the opinion of the experts, the factors with less impact on the sedimentation were removed. ANN was implemented using NeuroSolutions and Matlab software. The neural network was a Multi-Layer Perceptron (MLP) with one hidden layer and five neurons. The hidden layer consisted of tan-sigmoid activation function, and the output layer had a linear-sigmoid activation function. The algorithm used for training the neural network was Levenberg-Marquardt. The ANN results were superior to that of regression and the Matlab's output was more accurate than that of NeuroSolutions, with a mean square error of ۰.۰۰۹ for sediment yield. Finally, Matlab's neural network was extracted in the form of a function for later applications without the need to further training.

Soil erosion, Artificial neural networks, MPSIAC model, MLP Neural Network