Inverse modeling of unsaturated flow parameters using dynamic geological structure conditioned by GPR tomography

Abstract

1] A method is presented to estimate flow parameters and geological structure in the vadose zone by combining time-lapse Ground Penetrating Radar (GPR) traveltime tomography and inverse flow modeling. The traveltime tomography is used to determine the spatial electromagnetic velocity distribution in the vadose zone. These time-lapse velocity images are converted to time-lapse volumetric soil water content images using petrophysical relationships. Subsequently, the water content images are used as constraints in the flow inversion. The influence of the tomographic artifacts on the flow inversion is minimized by assigning weights that are proportional to the ray coverage. Our flow inversion algorithm estimates the flow parameters and calibrates the geological structure. The geological structure is defined using a set of control points, the positions of which can be modified during the inversion. After the inversion, the final geological and flow model are used to compute GPR traveltimes to check the consistency between these computed traveltimes and the observed traveltimes.The method is first tested on two synthetic models (a steady state and a transient flow models). Subsequently, the method is applied to characterize a real vadose zone at Oslo Airport Gardermoen, Norway, during the snowmelt in 2005. The flow inversion method is applied to locate and quantify the main geological layers at the site. In particular the inversion method identifies and estimates the location and properties of thin dipping layers with relatively lowpermeability. The flow model is cross validated using an independent infiltration event. Citation: Farmani, M. B., N.-O. Kitterød, and H. Keers (2008), Inverse modeling of unsaturated flow parameters using dynamic geological structure conditioned by GPR tomography, Water Resour. Res., 44, W08401,

Distributed Soil Moisture from Crosshole Ground‐Penetrating Radar Travel Times using Stochastic Inversion

“…By construction, such models have a spatially variable resolution that is significantly coarser than the discretization of the inverse model grid and depend on the type of model regularization being used. The importance of this resolution discrepancy has been largely overlooked in past research—but also in more recent contributions to the hydrogeophysics literature that use geophysical tomograms for hydrologic calibration or property characterization (e.g., Rubin et al, 1992; Hubbard et al, 1999; Chen et al, 2001; Farmani et al, 2008). These resolution characteristics limit the direct use of deterministic geophysical tomograms for hydrologic model construction and predictions because theoretical or laboratory‐derived petrophysical relationships are not applicable to the typically overly smooth tomographic models (e.g., Day‐Lewis and Lane, 2004; Day‐Lewis et al, 2005; Moysey et al, 2005; Linde et al, 2006b).…”

“…Although these different methods provide information about model resolution and parameter uncertainty, they provide limited insight into the underlying probability distribution of the model parameters and their multidimensional cross‐correlation. This information is of utmost importance, particularly if, for example, the soil moisture estimates derived from geophysical tomograms are to be used as “calibration data” in a hydrologic inversion (e.g., Farmani et al, 2008).…”

Software framework for inverse modeling and uncertainty characterization

“…The characteristics e and hence variability e of an SRF can be discerned by the relationship between model parameters, direct, and indirect information. A number of hydrogeological studies have been conducted using SRF analysis (Delhomme, 1979;Carrera and Neuman, 1986;Dagan, 1987;Bates and Townley, 1988;Bellin and Rubin, 1996;Yeh et al, 2002;Kanso et al, 2003;Gallagher and Doherty, 2007;Farmani et al, 2008). This paper introduces an open source inverse modeling framework, called MAD# (pronounced "mad sharp"), focused on the characterization of SRFs using the Method of Anchored Distributions (MAD), a Bayesian inverse modeling technique (Rubin et al, 2010).…”

Falsification and corroboration of conceptual hydrological models using geophysical data

“…A trained geophysicist might have a good understanding about model features that are well resolved, but this information is not always communicated to the final hydrological user. This leads to the obvious risk that the hydrologist overinterprets the geophysical information by treating each model pixel as an independent well‐known truth or ‘data point’ that is either perfectly or approximately known . An opposing risk is that well‐resolved features are ignored or that the models are disregarded altogether if they are presented such that they have little apparent value to the hydrologist .…”

“…This leads to the obvious risk that the hydrologist overinterprets the geophysical information by treating each model pixel as an independent well-known truth or 'data point' that is either perfectly 6 or approximately known. 54 An opposing risk is that well-resolved features are ignored or that the models are disregarded altogether if they are presented such that they have little apparent value to the hydrologist. 55 Over-and underinterpretation of geophysical models, simplified uncertainty assessments and the choices made in translating geophysical models into hydrological properties often affect the utility of geophysics in hydrology.…”