Applications of Multivariate Adaptive Regression Spline (MARS) for Modeling and Explaining Long Term Salinity and Total Dissolve Solid (TDS) river water Case study: Basins of khorasan jonobi

Salinity (Ec) and Total Dissolve Solid (TDS) are considered as quality factor s of drinking water, agriculture and industry. First relation between Debite-Ec and Debite-TDS (rating function) of taken samples doing with the fitted regression models, exponential, power and the neural network. The long-term analysis of Ec and TDS of water river (day, month and annual scales) did with the selected model.  Almost analytical samples taken from  normal flow and small flood. River floods and low floods have important effects in analyzing and explaining the volatility Ec and TDS. Therefore it is need to estimate Ec and TDS of these flood and low flood (extrapolation). Through Ec and TDS are bounded, so current models can not be well in Extrapolation.  Propose of this paper is implementation a new model multivariate adaptive regression spline (MARS) to solve this problem. Also South Khorasan province basins were selected for case study. MARS is a piecewise linear regression model (or nonlinear) that better perform in extrapolation and observe in bound. Three models exponential, power, and MARS (with transform) were fitted on the statistics of Debit-Ec and Debit-TDS stations in South Khorasan  province. Two views of the physical and statistical model are superior selection criteria that MARS show superiority in these models. The results of  models fitting were compared with two modes of interpolation and extrapolation. The MARS model extrapolation represented the reasonable values, while extrapolation of other models represented very big values (infinite). Analysis showed that  the MARS model with transform can be suitable replacement for rating curve modeling of Ec and TDS