Technical ArticleAn Ensemble Modeling Approach for Estimating Diffusive Tortuosity for Saturated Soils From PorosityChakraborty, Poulamee; Das, Bhabani Sankar; Singh, RajendraAuthor Information Agricultural and Food Engineering Department, Indian Institute of Technology, Kharagpur, West Bengal, India. Address for correspondence: Dr. Bhabani Sankar Das, Department of Agricultural and Food Engineering, Indian Institute of Technology Kharagpur, West Bengal 721302, India. E-mail: [email protected] Financial Disclosures/Conflicts of Interest: None reported. Supplemental digital contents are available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.soilsci.com). Received October 15, 2016. Accepted for publication March 13, 2017. Soil Science: February 2017 - Volume 182 - Issue 2 - p 45-51 doi: 10.1097/SS.0000000000000195 Buy SDC Metrics Abstract Apparent diffusion constants in soil are generally estimated by dividing molecular diffusion coefficient for a solute with soil tortuosity (τ) values. Several models have been proposed to estimate τ from soil porosity (ϕ) alone, but most of these models fail when the variability in observed τ-ϕ pairs increases. Pedotransfer functions can be used to predict τ from easy-to-measure soil properties such soil texture, organic carbon contents, and ϕ, but such an approach requires more measurements to be performed than just measuring ϕ. Here, we show that τ may be estimated from ϕ alone using the ensemble averaging approach. We examined seven different analytical expressions for τ-ϕ and seven different ensemble-modeling approaches to estimate τ for 100 pairs of τ-ϕ collected from a wide geographical area. Modeling results showed that the Bayesian model averaging method was the best ensemble-modeling approach for estimating τ from ϕ. Of 119 different combinations of τ (ϕ) models, three models derived considering (1) packing of square-shaped particles, (2) fractal geometry with particles of different sizes, and (3) percolation theory were identified as the best individual models for ensemble modeling. The coefficient of determination (0.67), root-mean-squared error (0.23), and the Akaike information criterion (94.37) values for this ensemble model were better than those when a single model was used for prediction. Inclusion of these three models that are based on both fractal and regular geometrical shapes for particles of different sizes may be a reason for improved performance of the ensemble approach. These results suggest that τ may be estimated from ϕ using the ensemble approach without the need for additional soil data, as is done in a pedotransfer function approach. Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.