Patterson, Matthew. Simulations of three-dimensional water flow with discrete classes of soil structure. Retrieved from https://doi.org/doi:10.7282/t3-b3yg-y623
DescriptionSimulation of water flow in a soil volume requires the characterization of the soil hydraulic properties in three dimensions. Such characterization is expensive and requires destructive sampling. The implementation of a non-destructive method to represent variations of soil structure in space could potentially reduce sampling expenses without compromising the accuracy of flow simulations. The overall objectives of this dissertation were to investigate the reduction of soil complexity from a continuum to discrete classes, each assumed to have uniform hydraulic properties, and test the impact of such a simplification on the simulation of water flow.
Electrical Resistivity Tomography (ERT) was used to define variations of soil properties within the volume of interest. This technique allows the generation of three-dimensional distributions (i.e., tomographs) of electrical resistivity (ER). A clustering method was used to reduce the spatial variation of ER values to three classes. Two of the classes contained ER values that were correlated to each other and represented regions with ER greater or smaller than the average ER. The third class contained non-correlated ER values. Each of these cluster types were assigned uniform soil hydraulic properties. A three-dimensional numerical model was used to assess the impact of clustering on simulated water flow.
In the second chapter of this dissertation, clusters of simulated ER data from a previous study were used to characterize the impact data inversion (required to produce the tomographs) may have on simulated water flow. The simulated datasets consisted of a total of 100 realizations of spatial distributions of ER values clustered into two sizes. Each of those realizations consisted of the raw (r) and the inverted (i) data. The clusters were assigned hydraulic properties obtained from a laboratory core experiment previously conducted at Rutgers University, and used to generate spatial structures in a flow model. Each realization was used to simulate water flow over ten flow rates. Correlations between model outputs of the r- and corresponding i-datasets were quantified with the normalized mutual information. The inversion process homogenized the spatial structure of the model domain, smoothing out the smaller clusters. Gradients in water content and velocity at cluster edges were sharper at higher flow rates and it was concluded that the soil hydraulic properties used in the clusters likely influenced these interactions.
The third chapter simulated the laboratory experiment conducted at Rutgers on an undisturbed soil core, with spatial structural units defined by clusters of electrical resistivity data measured several times over the course of the experiment. Water retention was measured on thirty soil cores extracted from an experimental lysimeter. The water retention model parameters for each core were estimated by assuming that the pore system had unimodal or bimodal distributions. A third representation of water retention was derived from the texture and bulk density measured in each core. Clusters were assigned spatially-averaged parameter sets defined by median values or similar-media scaling. Spatially-variable outflow from the lower boundary of the lysimeter was simulated through a three-dimensional domain that replicated the Rutgers lysimeter at five flow rates. The main finding of this work was that the spatial distribution of the hydraulic properties is more important than the method used to define those properties.
The fourth chapter applied the clustering framework to the field scale by comparing clusters of electrical resistivity derived from a measurement at the beginning of the experiment and water content values simulated over a period of 150 days. The soil hydraulic properties of the experimental volumes were characterized by 121 soil core samples and by field measurements of pressure potential and water content. The spatial distribution of the hydraulic properties in the model domain was defined by maps of scaling factors interpolated through Ordinary Kriging. Maps of simulated water content were clustered and compared to clusters of electrical resistivity measurements from the field. The dynamic nature of the simulated water content made it difficult to characterize consistent spatial clusters and made comparisons between electrical resistivity and water content clusters uncertain.