Koestel, Johannes
- Department of Soil and Environment, Swedish University of Agricultural Sciences
Research article2014Peer reviewed
Koestel, Johannes; Jordá, Helena
Preferential flow and transport has to be taken into account to successfully predict solute transport through the vadose zone. The relative 5%-arrival time of inert tracer can serve as a measure for the strength of preferential transport. As direct measurements of solute transport are not practical at large scales, soil susceptibility to preferential flow and transport has to be estimated using proxy variables. In this study we investigated how well the relative 5%-arrival time of inert tracer could be inferred from soil properties, site factors, scale and hydrologic conditions for 442 breakthrough curve experiments on undisturbed soil columns under steady state irrigation. Using a random forest as a global regression tool, we found a coefficient of determination of 0.561 in a ten-fold cross-validation. When predicting relative 5%-arrival times on a completely independent benchmark dataset of 149 experiments we obtained a still reasonable coefficient of determination of 0336. When the soil columns had not been sampled from the same site and soil horizon, the random forest was able to rank the experiments correctly according to their relative 5%-arrival time, apart from one exception. Our study demonstrates that soil susceptibility to preferential flow and transport occurring under steady state initial and boundary conditions is to a large part predictable from proxy variables. We furthermore found evidence that the prediction performance should be considerably increased if information on the water saturation state during the experiment could be included into the random forest. An investigation of the importance of the predictors for estimating the relative 5%-arrival time yielded that the clay content was fundamental. Next important were the ratio between clay content and organic carbon, the lateral observation scale and whether the column had been slowly saturated from the bottom prior to the experiment or not. Flow rate, soil management and bulk density were found useful to further refine the predictions. A caveat has to be given that the investigated dataset includes few experiments on large columns and no experiments under natural transient hydrologic boundary conditions, since such experiments are scarce. Availability of such experiments is crucial to account for additional important preferential flow transport mechanisms caused by hydrophobicity, instabilities at infiltration fronts or funneling at soil horizon boundaries. (C) 2013 Elsevier B.V. All rights reserved.
Solute transport; Preferential flow; Random forest; Pedotranfer function; Breakthrough curve; Machine learning
Geoderma
2014, Volume: 217, pages: 144-160 Publisher: ELSEVIER SCIENCE BV
Soil Science
DOI: https://doi.org/10.1016/j.geoderma.2013.11.009
https://res.slu.se/id/publ/56295