Multiregion full-waveform inversion for ground-penetrating radar common-offset data

ABSTRACT Ground-penetrating radar full-waveform inversion (GPR-FWI) is a high-resolution inversion technique that quantitatively estimates the subsurface physical parameters (permittivity and conductivity) by matching waveform amplitude and phase. However, the requirement to meet stability conditions forces the grid dimensions during wave propagation to be at the centimeter level, resulting in significant memory usage when discretizing the full radar wavefield. This makes it challenging to apply the FWI method on large-scale 2D or 3D GPR data sets. Furthermore, most commercial GPR systems record common-offset gathers (COG), so most GPR field data do not need to calculate the full model-space wavefield during the FWI iteration procedure. This work develops a multiregion full-waveform inversion approach (MR-FWI) for common-offset GPR data. The core strategy is to divide the model space into multiple grid regions and dynamically adjust these grids according to the requirements. By focusing on a given spatial range of the wavefield and model gradient, the computational steps are optimized, significantly reducing memory consumption and computational cost. A synthetic model and two field GPR data examples validate that the multiregion strategy can reduce the computational burden without sacrificing inversion accuracy. Our MR-FWI is particularly suitable for limited computing resources and large-scale data sets, which provides a robust and high-resolution approach for most field common-offset GPR applications.