The site-specific selection of the infiltration model based on the global dataset and random forest algorithm

Authors: KIM, SEONGYUN - ORISE FELLOW; KARAHAN, GULALY - CANKIRI KARATEKIN UNIVERSITY; Sharma, Manan; Pachepsky, Yakov

Submitted to: Vadose Zone Journal
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/18/2021
Publication Date: 5/20/2021
Citation: Kim, S., Karahan, G., Sharma, M., Pachepsky, Y.A. 2021. The site-specific selection of the infiltration model based on the global dataset and random forest algorithm. Vadose Zone Journal. https://doi.org/10.1002/vzj2.20125.
DOI: https://doi.org/10.1002/vzj2.20125

Interpretive Summary: Water infiltration to soil must be predicted to address various environmental and agricultural issues. Many equations were proposed to simulate infiltration. Coefficients of these equations, also called parameters, reflect local soil and vegetation conditions. Obtaining infiltration parameters from measurements is impractical In large-scale projects. The objective of this work was to use the Global Soil Infiltration models are widely used in simulations of soil water balance and water availability to plants. Many infiltration models were proposed. It is not known which model may be more suitable for site-specific conditions. Our objective was to test the assumption that the knowledge of site-specific soil properties, land use, and the infiltration measurement method can be used to predict which model will perform the best. We used data from the largest international database on infiltration to soil, and compared performance of three empirical and three theoretical infiltration models. The empirical models of Horton, Mezencev, and Collis-George performed substantially better than theoretical Philip, Green-Ampt ,and Swartzendruber models. The infiltration measurement method was the lead predictor for the best model selection. The Horton model was predicted correctly more frequently than others. The results of this work can be useful to a large group of environmental professionals who are applying infiltration equations in their projects.

Technical Abstract: Different infiltration models were found to perform better than others in various experiments. The objective of this work is to test the assumption that the knowledge of site-specific soil properties, land use, and the infiltration measurement method can be used to predict which model will perform the best. The performance of three empirical equations (Horton, Mezencev, Collis-George) and three theoretical equations (Green and Ampt, Philip, and Swartzendruber) was characterized by the root-mean-squared error obtained after fitting them to the 4830 cumulative infiltration datasets from the SWIG intentional database. Then the random forest machine-learning algorithm was applied to predict the best model using soil texture, organic matter content, bulk density, saturated hydraulic conductivity, land use, and infiltration measurement method as inputs. Overall, empirical infiltration models performed better than theoretical infiltration models. The Horton, Mezencev, and Collis-George models were the best in 36 %, 24%, and 12 % of cases, respectively. Swartzendruber, Philip,. aAnd Green-Ampt were the best in 11%, 10%, and 7% cases, respectively. The infiltration method was by far the most important predictor. It was followed by the organic carbon content and the land-use type. The land-use type and the organic carbon content were the most important predictors of the model suitability for the datasets obtained with the double-ring infiltrometer, whereas the organic carbon content and soil bulk density were the most important predictors for the datasets obtained with the mini disc infiltrometer. The SWIG database presents the opportunity to select the most suitable infiltration model for site-specific conditions that can improve the accuracy of hydrological models in large scale applications.