Unsaturated soils permeability estimation by adaptive neuro-fuzzy inference system

Abstract

Unsaturated soils permeability ($K_{\text{unsat}}$) is a required parameter when modeling water flow and transport processes in the subsurface. Having highly nonlinear relationship with volumetric water content (θw) and suction (S), the value of $K_{\text{unsat}}$ varies by several magnitudes when moving from clayey to gravel soils. On the other hand, determination of $K_{\text{unsat}}$ is very difficult, costly, and time consuming. Recently, adaptive neuro-fuzzy inference system (ANFIS) has been used for modeling and prediction of such complex and nonlinear problems. Investigated in this paper is the capability of ANFIS for modeling $K_{\text{unsat}}$. The database used in ANFIS modeling is collected from SoilVision. This database contains 4347 $K_{\text{unsat}}$ test records on 245 soil types collected from all around the world; it approximately covers triangular chart defined by US Department of Agriculture System for classifying mixed soils. In order to get the optimum number of ANFIS training epochs and ANFIS structure, trial and error method was used. To check the predictive capacity of the ANFIS model, several statistics such as determination coefficient (R²), Root Mean Square Error, Mean Absolute Error and Variance Account For were calculated. The results demonstrated that the ANFIS model can be successfully applied for prediction of $K_{\text{unsat}}$.