عنوان مقاله [English]
The conventional methods for estimating the suspended load in rivers are using the sample of debit-sediment and fit them into exponential patterns, power model, neural networks, or frequency distribution table. In high flood flows these models cannot be extrapolated and their estimations are unrealistic. This is due to the bounded water power to carry suspended load (physical properties). Multivariate Adaptive Regression Spline (MARS) model are proposed in this paper to pass this limitation. MARS is a piecewise linear spline model that has better performance in extrapolations. The MARS exponential and power models were fitted on debit-sediment data of 23 stations khorasan razavi province in this article. The results were compared in two modes of interpolation and extrapolation and base on statistical and physical criteria. Statistical criteria models included the R2adj, sum of squares error (SSE), generalized cross validation (GCV), and diagnostic residual. The physical criteria included realistic estimation of the sediment in two modes of interpolation and extrapolation. The results showed the superiority of MARS model to other practices.