Multiobjective Waveform Inversion of Multioffset Surface Ground-Penetrating Radar Data

Full-waveform inversion (FWI) of multioffset surface ground-penetrating radar (GPR) data has the potential to provide high-resolution subsurface electromagnetic models. Conventional GPR-FWI uses the $l_{2}$ objective function, which is more sensitive to near-offset data and early arrivals. In this study, we propose a random multiobjective waveform inversion (R-MOWI) to balance the contributions of near- and far-offset data, as well as early and late arrivals. R-MOWI adopts a multiobjective framework that integrates the $l_{2}$ , envelope, normalized, and autogain control (AGC) objective functions. R-MOWI randomly selects an objective function for each radargram at each iteration. Our 2-D synthetic examples demonstrate that R-MOWI outperforms the four single-objective FWIs in reconstructing permittivity and conductivity models by leveraging multiple objective functions to constrain convergence. R-MOWI mitigates overestimation by offering greater flexibility in the search space. Furthermore, it exhibits robust performance and facilitates uncertainty estimation of the inversion results. In summary, this study highlights the effectiveness of R-MOWI in processing multioffset surface GPR data.