Modelling the Relationship between Physical Characteristics of Watershed and the Runoff and Sediment (Case Study: Garmab Watershed, Razavi Khorasan Province)

Achieving the detailed information about flood and sediment, due to practical problems and the lack of hydrometric and sediment gaging stations is very costly and difficult. Therefore, the present research is sought to determine the significant relationships between the physical parameters of Garmab watershed and various components of flood and sediment. For this purpose, all input information such as area and perimeter of watershed, the watershed maximum, minimum and average altitudes of sea level, main stream length,main stream slope, watershed slope, form factor and drainage density variables, annual erosion and specia erosion (m3/km2/y) and flood with the return periods of 2, 5, 10, 25, 50 and 100 years (m3/s) collected. Then, the relationship between 8 estimated parameters of flood and sediment and 10 physical parameters of watershed performed using the multivariate regression analysis and neural network methods. The study of the effective inputs in the multivariate regression method showed that the area and slope variables of watershed are the effective parameters in flood production and the area, minimum and average height of watershed and the main stream length variables are the effective parameters in the sediment production at Garmab watershed. Also the radial basis function network (RBF) for the predicted values of specific erosion and multi-layer perceptron network (MLP) for the predicted values of the estimated annual erosion, respectively, with the maximum values in the coefficient of determination equal to 0.98 and 0.99 and the minimum values in the root mean squared error equal to 0.16 and 253.56, selected as the most efficient models. But for the predicted values of flood with the return periods of 2, 5, 10, 25, 50 and 100 years, the multivariate regression model according to the maximum value of the coefficient of determination, the minimum values in the root mean squared error and the absolute relative error, selected as the best model.