Development of an Ensemble Based Model to Estimate the Tortuosity Values of Soil from Reported Models on Tortuosity for Various Porous Media.

Diffusive tortuosity (τ) is generally expressed as a function of porosity (ϕ). Most of the experimental results on τ and ϕ reported fewer datasets to show that such models more or less describe observed data. In this study, an ensemble modeling approach is proposed to estimate τ exclusively from ϕ without the requirement for any other soil properties. Seven ensemble approaches were assessed using 7 different analytical expressions for τ(ϕ) and 100 measured τ-ϕ pairs. The unconstrained multiple linear regression (MLR_UC) was the best ensemble approach for estimating τ. Out of 119 different combinations of individual τ(ϕ) models examined, the combination of three models were identified as the best ensemble model. The three models are the regular geometry-based Yu and Li (2004) and Du Plessis and Masliyah (1991) models and the percolation-based Ghanbarian et al. (2013) model. Soil naturally shows both geometric and fractal behaviour – a reason for which ensemble approach may have succeeded to estimate τ exclusively from ϕ.