Parameters of infiltration models as affected by the measurement technique and land use

Authors: KARAHAN, GULAY - CANKIRI KARATEKIN UNIVERSITY; Pachepsky, Yakov

Submitted to: Catena
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/26/2022
Publication Date: 5/24/2022
Citation: Karahan, G., Pachepsky, Y.A. 2022. Parameters of infiltration models as affected by the measurement technique and land use. Catena. https://doi.org/10.36783/18069657rbcs20210147.
DOI: https://doi.org/10.36783/18069657rbcs20210147

Interpretive Summary: The measurement technique and the land use are two soil structure-related attributes that are typically available in descriptions of infiltration experiments. We hypothesized that these attributes might be good predictors of different infiltration models' performance and the parameter values in those models. The Soil Water Infiltration Global (SWIG) database assembled in the Institute of Agrosphere in Jülich, Germany, was used. The database encompasses about 5000 experiments all over the world. Texture, measurement method, and land use were known for all experiments; the availability of organic carbon content, bulk density, saturated hydraulic conductivity (Ksat), pH, the electrical conductivity of saturated paste, and initial water content varied. Comparison of the performance of four infiltration models showed that Horton and Mezencev models outperformed all others and that one of these two models could be preferred based on the infiltration measurement method. The machine learning method –regression trees– was applied to find the most influential predictors of Horton and Mezencev models' parameters. The measurement method (40% of cases), the textural class, and the land use were the most influential predictors in 80% of cases for both models. The accuracy of parameter estimates increased when only the subset of measurements with the same method was used to estimate infiltration parameters. Land use, textural class, and organic carbon content dominated as the most influential predictors for the parameters of the Mezencev, whereas land use, textural class, Ksat, and bulk density became most important in the case of the Horton model. Overall, estimates of the infiltration equation parameters can be more accurate if they have been developed for the same measurement method as in the task at hand. The land use category and the infiltration measurement method provide useful surrogate information about the soil structure effect on infiltration.

Technical Abstract: The measurement technique and the land use are two soil structure-related attributes that are typically available in descriptions of infiltration experiments. We hypothesized that these attributes may be good predictors of the performance of different infiltration models, and of the parameter values in those models. The Soil Water Infiltration Global (SWIG) database assembled in the Institute of Agrosphere in Jülich, Germany was used. The database encompasses about 5000 experiments all over the world. Texture, measurement method and land use were known for all experiments, availability of organic carbon content, bulk density, saturated hydraulic conductivity (Ksat), pH, the electrical conductivity of saturated paste, and initial water content varied. Comparison of the performance of four infiltration models showed that Horton and Mezencev models outperformed all others, and that one of these two models could be preferred based on the infiltration measurement method. The machine learning method –regression trees– was applied to find the most influential predictors of parameters of Horton and Mezencev models. The measurement method (40% of cases), the textural class, and the land use were the most influential predictors in 80% of cases for both models. The accuracy of parameter estimates increased when only the subset of measurements with the same method was used to estimate infiltration parameters. Land use, textural class, and organic carbon content dominated as the most influential predictors for the parameters of the Mezencev, whereas land use, textural class, Ksat, and bulk density became most important in the case of the Horton model. Overall, estimates of the infiltration equation parameters can be more accurate if they have been developed for the same measurement method as in the task in hand. Land use category and the infiltration measurement method provide useful surrogate information about the soil structure effect on infiltration.