3D seismic refraction traveltime tomography at a groundwater contamination site

Abstract

We have applied traveltime tomography to 3D seismic refraction data collected at Hill Air Force Base, Utah, in an approximately 95 ϫ 40-m area over a shallow ͑Ͻ 20 m͒ groundwater contamination site. The purpose of this study is to test the ability of 3D first-arrival-time data to characterize the shallow environment and aid remediation efforts. The aquifer is bounded below by a clay aquiclude, into which a paleochannel has been incised and acts as a trap for dense nonaqueous phase liquid ͑DNAPL͒ contaminants. A regularized nonlinear tomographic approach was applied to 187,877 first-arrival traveltimes to obtain the smoothest minimumstructure 3D velocity model. The resulting velocity model contains a velocity increase from less than 300 to 1500 m/s in the upper 15 m. The model also contains a north-southtrending low-velocity feature interpreted to be the paleochannel, based on more than 100 wells in the area. Checkerboard tests show 7.5-10 m lateral resolution throughout most of the model. The preferred final model was chosen after a systematic test of the free parameters involved in the tomographic approach, including the starting model. The final velocity model compares favorably with a 3D poststack depth migration and 2D waveform inversion of coincident reflection data. While the long-wavelength features of the model reveal the primary target of the survey, the paleochannel, the velocity model is likely a very smooth characterization of the true velocity structure, particularly in the vertical direction, given the size of the first Fresnel zone for these data.

Borehole seismic monitoring of CO2 storage within a saline aquifer at Ketzin, Germany

“…This means the inversion will not attempt to overfit the data for any iteration, but will only iterate towards a model for which the χ 2 between the observed and predicted data ideally reaches one (χ 2 = 1, Zelt et al (2006)):…”

“…From a free starting value λ 0 , the algorithm reduces the trade-off parameter systematically by a chosen factor λ r . "The systematic reduction of λ stabilizes the inversion by constraining the long-wavelength model structure in the initial iterations and by allowing finer model structure in later iterations" (Zelt et al (2006)). …”

“…In order to treat the underdetermined part of the solution, FAST uses regularized inversion, in which at each iteration, an objective function, which "measures a combination of the data misfit and constraints of the structure of the model, is minimized in the least square sense" (Zelt et al (2006), Zelt and Barton (1998)). The objective function φ(s) minimized at each iteration can be written as…”

“…The testing of λ values is stopped if the maximum number of values to test is reached or if a model is found which "predicts the picked travel times within an accuracy consistent with their assigned uncertainties" (Zelt et al (2006)). This means the inversion will not attempt to overfit the data for any iteration, but will only iterate towards a model for which the χ 2 between the observed and predicted data ideally reaches one (χ 2 = 1, Zelt et al (2006)):…”

“…Therefore, a linearised iterative approach is applied using a Taylor series expansion in which a starting model is required and the model and raypaths are updated over a series of iterations" (Zelt et al (2006)). The first input parameter for the shell script is the maximum number of non-linear iterations, these are iterations in which the model is updated and new raypaths are calculated.…”