Hydrological parameter estimations from a conservative tracer test with variable‐density effects at the Boise Hydrogeophysical Research Site

Abstract

[1] Reliable predictions of groundwater flow and solute transport require an estimation of the detailed distribution of the parameters (e.g., hydraulic conductivity, effective porosity) controlling these processes. However, such parameters are difficult to estimate because of the inaccessibility and complexity of the subsurface. In this regard, developments in parameter estimation techniques and investigations of field experiments are still challenging and necessary to improve our understanding and the prediction of hydrological processes. Here we analyze a conservative tracer test conducted at the Boise Hydrogeophysical Research Site in 2001 in a heterogeneous unconfined fluvial aquifer. Some relevant characteristics of this test include: variable-density (sinking) effects because of the injection concentration of the bromide tracer, the relatively small size of the experiment, and the availability of various sources of geophysical and hydrological information. The information contained in this experiment is evaluated through several parameter estimation approaches, including a grid-search-based strategy, stochastic simulation of hydrological property distributions, and deterministic inversion using regularization and pilot-point techniques. Doing this allows us to investigate hydraulic conductivity and effective porosity distributions and to compare the effects of assumptions from several methods and parameterizations. Our results provide new insights into the understanding of variabledensity transport processes and the hydrological relevance of incorporating various sources of information in parameter estimation approaches. Among others, the variable-density effect and the effective porosity distribution, as well as their coupling with the hydraulic conductivity structure, are seen to be significant in the transport process. The results also show that assumed prior information can strongly influence the estimated distributions of hydrological properties.

Hydraulic Tomography: 3D Hydraulic Conductivity, Fracture Network, and Connectivity in Mudstone

“…However, all of these methods require spatial interpolation to generate 3D representations of aquifer properties from one-dimensional (1D) and/or 2D analyses and measurements (Butler 2005;Bohling et al 2007;Deutsch 2007;Castagna and Bellin 2009). When geophysical methods are used, rock physics or petrophysical relations are needed to convert geophysical parameters into hydrologic parameters (e.g., Ellefsen et al 2002;Rubin and Hubbard 2005;Chen et al 2006;Linde et al 2006;Dafflon et al 2011).…”

Estimation of water velocities in boreholes using a new passive flowmeter

“…Here, minimum sinking effects can be noted. Sinking effects have usually been observed through well BTCs in field tracer experiments affected by density-driven flows and were recently discussed in Dafflon et al (2011). On these BTCs, an unexpected reduction and a successive recovery of concentration can be noted, like a negative source (or sink) effect on the transport of tracer.…”

The hydrogeologic information in cross-borehole complex conductivity data from an unconsolidated conglomeratic sedimentary aquifer

“…The variance of ln K is 0.49, more than three times the ln K variance at the well-studied aquifer of Cape Cod (Wolf et al, 1991) and less than twice that of the aquifer at Borden (Woodbury and Sudicky, 1991). However, tracer behavior at the BHRS (Johnson et al, 2007;Nelson, 2007;Dafflon et al, 2011a) is clearly not representative of a homogeneous or simple layered medium in K (or ∅).…”

“…Each identified cluster could include any nonzero proportion of the total number of voxels in the grid and could have irregular and discontinuous boundaries. For selected cluster analyses, using five or six clusters, the nonparametric Kolmogorov-Smirnov test (Conover, 1999) was applied to determine whether the 3D distribution of σ 0 within each identified cluster differs significantly from the distribution within every other cluster, and this was repeated to test for significantly different distributions of σ 0 0 . In all cases, the null hypothesis that the distributions do not differ can be rejected (p < 10 −4 ).…”