Conditional stochastic inversion of common-offset ground-penetrating radar reflection data

Abstract

We present a stochastic inversion procedure for common-offset ground-penetrating radar (GPR) reflection measurements. Stochastic realizations of subsurface properties that offer an acceptable fit to GPR data are generated via simulated annealing optimization. The realizations are conditioned to borehole porosity measurements available along the GPR profile, or equivalent measurements of another petrophysical property that can be related to the dielectric permittivity, as well as to geostatistical parameters derived from the borehole logs and the processed GPR image. Validation of our inversion procedure is performed on a pertinent synthetic data set and indicates that the proposed method is capable of reliably recovering strongly heterogeneous porosity structures associated with surficial alluvial aquifers. This finding is largely corroborated through application of the methodology to field measurements from the Boise Hydrogeophysical Research Site near Boise, Idaho, USA.

High-resolution velocity estimation from surface-based common-offset GPR reflection data

“…A search of the phase rotation angle that maximizes the kurtosis when applied to this minimum-phase wavelet is then used to obtain the final mixed-phase GPR source wavelet. The practical validity of this source wavelet estimation procedure was recently demonstrated by Xu et al (2021). Note that the effects of minor dispersion in the GPR data are, at least in part, accounted for in the sense that an effective wavelet that best fits the considered dataset in its entirety, rather than the true emitted GPR source signal, is estimated.…”

“…The comparison with the reference velocity model (Figure 2a) is quite favorable, which clearly illustrates the potential benefits of the proposed diffraction-and reflection-based velocity estimation approach. In this context, is important to emphasize that the former can only resolve the smooth large-scale velocity structure and, hence, entirely misses the presence of the thin bed (e.g., Yuan et al 2019) whereas, on its own, the latter requires coincident borehole information for calibration and recovery of the large-scale component of the velocity structure (e.g., Schmelzbach et al 2012;Xu et al 2021).…”

“…The surficial aquifer consists of late Quaternary fluvial deposits dominated by gravel and sand, and is underlain by a layer of red clay at ~20 m depth (Barrash & Clemo 2002). The depth of the groundwater table at the site varies seasonally between ~2 m and ~4 m. Over the past two decades, the BHRS has been extensively utilized for the testing, validation, and improvement of a wide variety of geophysical and hydrogeological characterization methods (e.g., Tronicke et al 2004;Bradford et al 2009;Dafflon et al 2009Dafflon et al , 2011Hochstetler et al 2016;Xu et al 2020Xu et al , 2021.…”

“…Zeng et al (2015) and Liu et al (2018) adopt similar approaches to estimate the distribution of soil water content and to characterize buried archaeological remains, respectively. Xu et al (2021) also assume a convolution model for the GPR traces, but combine stochastic simulation with simulated annealing optimization in order to generate velocity realizations that honor the GPR measurements and borehole porosity log data along the profile. Forte et al (2013; assume a locally 1D layered subsurface structure and use picked reflection amplitudes to recursively estimate the GPR velocity in a series of identified subsurface layers, in which the velocity is assumed constant.…”

Fast 3D ground penetrating radar simulations for glaciers

“…In the same way, surface multiples, which are events that are reflected twice in the subsurface with a reflection at the air-ice interface in between before being recorded, will also be absent. To our knowledge, such secondary scattering is rather rare in GPR data, as evidenced by the great success of single-scatteringbased algorithms for the processing and analysis of GPR data (Irving et al, 2009;Schmelzbach et al, 2012;Xu et al, 2020Xu et al, , 2021Liu et al, 2022).…”