Seepage Forecasting in Earth Dam Using Multivariate Adaptive Regression Spline Method (MARS), Case Study: Eyvashan Dam

The accurate design of the geometry and the sealing layers of the earth dam to obtain the optimum structure are important from a practical and economic point of view. Due to the complexity of the analytical methods, the numerical method is the best practical solution for measuring the mechanical behavior of the dam including seepage and slope stability. In this study, reliability analysis of seepage and slope stability of Eyvashan dam has been performed using Multivariate Adaptive Regression Splines (MARS). The Eyvashan earth dam was modeled in Seep/w and Slope/w Software's and the model was validated by observational data of instruments. Then, forecasting of seepage and slope stability was investigated by MARS method. For this purpose, 214 data series from Seep/w numerical analysis and 300 data series from Slope/w numerical analysis were used to establish the MARS model. The model was evaluated and validated by these data. The MARS model showed high performance in modeling process with determination coefficients of 0.98 and 0.97 and the root mean square error of 0.0132 and 0.0160 for seepage discharge and safety of factor of slope stability, respectively. Also, in this study, the effective parameters in seepage and slope stability such as permeability of materials, internal friction angle of soil, specific weight of materials and upstream water level were sensitized to determine the best combination of modeling input's structure. The parameters of the upstream water level 29.5% and the permeability of the core 12.5% had the greatest effect on the seepage and internal friction angle of shell with 69.7% had the greatest effect on the factor of safety. Also, suitable combinations of input parameters were presented and a comprehensive relationship was obtained to accurately estimate the seepage discharge and factor of safety in different modes of layer arrangement and their characteristics