Review of crosshole ground-penetrating radar full-waveform inversion of experimental data: Recent developments, challenges, and pitfalls

Abstract

Heterogeneous small-scale high-contrast layers and spatial variabilities of soil properties can have a large impact on flow and transport processes in the critical zone. Because their characterization is difficult and critical, high-resolution methods are required. Standard ray-based approaches for imaging the subsurface consider only a small amount of the measured data and suffer from limited resolution. In contrast, full-waveform inversion (FWI) considers the full information content of the measured data and could yield higher resolution images in the subwavelength scale. In the past few decades, ground-penetrating radar (GPR) FWI and its application to experimental data have matured, which makes GPR FWI an established approach to significantly improve resolution. Several theoretical developments were achieved to improve the application to experimental data from crosshole GPR FWI. We have determined the necessary steps to perform FWI for experimental data to obtain reliable and reproducible high-resolution images. We concentrate on experimental crosshole GPR data from a test site in Switzerland to illustrate the challenges of applying FWI to experimental data and discuss the obtained results for different development steps including possible pitfalls. Thereby, we acknowledge out the importance of a correct time-zero correction of the data, the estimation of the effective source wavelet, and the effect of the choice of starting models. The reliability of the FWI results is investigated by analyzing the fit of the measured and modeled traces, the remaining gradients of the final models, and validating with independently measured logging data. Thereby, we found that special care needs to be taken to define the optimal inversion parameters to avoid overshooting of the inversion or truncation errors.

Fractal-Constrained Crosshole/Borehole-to-Surface Full-Waveform Inversion for Hydrogeological Applications Using Ground-Penetrating Radar

“…In order to address the diversity of the hydrogeological applications that GPR is used for, numerous surveying strategies and corresponding processing pipelines have been suggested over the years [12]. One of the most popular is the crosshole/borehole-to-surface measurement configuration, aimed at interpolating hydrogeological properties between boreholes [13], [14], [15]. The interpretation of these data is generally based on the geometrical ray theory [16], which utilises the first arrivals and the maximum amplitudes [17], [18] (subject to regularizations and constraints [19], [20], [21]) in order to reconstruct the dielectric properties of the investigated medium.…”

“…Recent theoretical developments have further established FWI as a reliable interpretation tool for GPR acquisitions and especially for crosshole/borehole-to-surface hydrogeological surveys [15]. In particular, source deconvolution [31], [32] and using the ratio of two electromagnetic field parameters [33] have been successfully applied for removing the effects of the unknown source wavelet.…”

“…Numerous evaluations of the forward model combined with time-consuming Jacobian derivations and memory-intense matrix operations require high-performance computers to operate for prolonged periods of time. Another major drawback of FWI is that it fails to provide an accurate estimation of the conductivity [15] and -to our knowledge-no attempt has been made to characterize the full dispersive properties of the investigated medium. Moreover, the final results are sensitive to the starting model [15] and therefore FWI is not applicable in the absence of sufficient initialisation.…”

“…Another major drawback of FWI is that it fails to provide an accurate estimation of the conductivity [15] and -to our knowledge-no attempt has been made to characterize the full dispersive properties of the investigated medium. Moreover, the final results are sensitive to the starting model [15] and therefore FWI is not applicable in the absence of sufficient initialisation. Lastly, FWI greatly relies on the inversion parameters (regularizations etc.)…”

“…Lastly, FWI greatly relies on the inversion parameters (regularizations etc.) [15], and a parameter selection based on trial-and-error is not attainable due to the high computational requirements for each FWI execution.…”