Incorporating near-surface layering in GPR data inversion for improved surface water content estimates

We analyze the retrieval of soil surface dielectric permittivity in presence of shallow dielectric contrasts using full-wave inversion of off-ground monostatic ground-penetrating radar (GPR) data. Shallow soil layers affect the surface reflection and lead to constructive or destructive interferences which result, respectively, in overestimated and underestimated surface dielectric permittivity values. Synthetic GPR Green's functions were generated for a series of model configurations with different contrasts and different layer thicknesses. Green's function inversions were then performed to retrieve the initial parameters and analyze the sensitivity of the different parameters in the inverse problem. Two global optimization algorithms were used and compared, namely, a genetic algorithm and the multilevel coordinate search. The results showed that, depending on the contrast and layer thickness, it is not always possible to identify the original soil model. This depends in particular on the width of the frequency range. Both algorithms led to comparable results and did not converge properly in all cases, given their standard parameterization. Yet, the proposed method appears to be promising to improve real-time mapping of surface water content using off-ground GPR and full-wave inversion.