Detection of Tracer Plumes Using Full‐Waveform Inversion of Time‐Lapse Ground Penetrating Radar Data: A Numerical Study in a High‐Resolution Aquifer Model

Abstract

Imaging subsurface small‐scale features and monitoring transport of tracer plumes at a fine resolution is of interest to characterize transport processes in aquifers. Full‐waveform inversion (FWI) of crosshole ground penetrating radar (GPR) measurements enables aquifer characterization at decimeter‐scale resolution. The method produces images of both relative dielectric permittivity (εr) and bulk electrical conductivity (σb) that can be related to hydraulic aquifer properties and tracer distributions. To test the potential of time‐lapse GPR FWI for imaging tracer plumes, we conducted a numerical tracer experiment by injecting saline water, desalinated water, and ethanol in a heterogeneous aquifer. The saline and desalinated tracers only changed σb, whereas ethanol changed both εr and σb. Tracer concentrations were retrieved from the inverted εr and σb models using information about petrophysical parameters. GPR FWI εr and σb tracer images could recover preferential paths of ∼0.2 m width, while the derived σb structures are smoother. FWI of 50 time‐lapse data sets demonstrated the potential of the FWI to derive spatially resolved breakthrough curves of the saline and ethanol tracer in the image plane between the boreholes. Thus, high‐resolution imaging with GPR FWI of tracers that produce a high permittivity contrast against the background has a great potential for characterization of heterogeneous transport in aquifers.

Adding realistic noise models to synthetic ground‐penetrating radar data

“…The most common approach to simulate noise in synthetic GPR data is to add white Gaussian noise with a specific standard deviation with respect to the amplitude values of the synthetic GPR data (Haruzi et al., 2022; Yuan et al., 2020; Haarder et al., 2011). In this context, we have to emphasize the difference between white noise and Gaussian noise.…”

“…Huisman et al (2003) state that identifying the precise ray-type, ray-path and correct amplitude values is crucial for a reliable estimation of soil water content. More recently, there is also a growing interest in time-lapse GPR measurements, for example, to monitor water flow processes (Allroggen et al, 2020), to estimate water content changes via reflection tomography (Mangel et al, 2020), to identify gas propagation in the subsurface (Yuan et al, 2020), to monitor tracer plume propagation in an aquifer using full-waveform inversion (Haruzi et al, 2022), or to obtain information about further hydraulic parameters and fluid dynamics (Klenk et al, 2015;Jaumann & Roth, 2018).…”

GPR modelling and inversion to quantify the debris content within ice

“…Numerical simulations, both for the sensitivity tests and the forward modelling, were performed exploiting gprMax , version 3.1.6 (Warren et al., 2016), which is an open‐source software designed to simulate the propagation of an EM wave even in heterogenous media, by solving Maxwell's equations in 3‐D using the finite‐difference time‐domain method. The algorithm can handle complex geometries and materials distributions, being highly adaptable to model a wide range of subsurface scenarios in various fields of application, such as archaeology, civil engineering, glaciology, and hydrogeology, among others (e.g., Cheng et al., 2023; Feng et al., 2023; Haruzi et al., 2022; Hillebrand et al., 2021; Pajewski et al., 2017; Schennen et al., 2022). In order to reduce the computational costs due to model discretization, we exploited a specific module for gprMax modelling on GPU (Warren et al., 2018) and performed the inversion on Cineca Marconi 100 cluster with 2 CPUs with 16 cores 3.1 GHz, 4 NVIDIA Volta V100 16GB GPUs and 256 GB RAM per node running on GPUs and parallelized on several nodes.…”