A quantitative investigation of the use of ground-penetrating radar in hydrogeology

Ground-penetrating radar (GPR) has become a useful tool for non-invasive imaging of the subsurface. However, the qualitative nature of current methods for the interpretation of GPR makes it difficult to use in groundwater modeling applications, especially for problems where accounting for uncertainty is important. In this thesis, the need for quantitative interpretations is addressed using observational, theoretical, and computational approaches that explore how complex subsurface heterogeneities are expressed in GPR data. This understanding of how radar samples the earth is exploited to suggest quantitative methods that can be used for interpreting radar data.; In the first part of this thesis, radar facies analysis is examined as an approach to determine the large-scale architecture of the subsurface using GPR surface reflection data; it is often assumed that radar facies can act as a valuable proxy for defining hydrologic facies, given that both types of facies are related to lithology. In the approach explored here, artificial neural networks are used to probabilistically segment the subsurface into radar facies based on characteristic signatures of the radar data. Specifically, radar texture---the pattern of reflections within a window of radar data---is used to discriminate between different radar facies.; In the second part of this thesis, the nature of the relationship between dielectric constant, determined by GPR surveys, and water content, important in hydrologic investigations, is investigated. Using a stochastic averaging approach that accounts for the way radar averages over heterogeneity it is demonstrated that field-scale dielectric constant-water content relationships are not necessarily equivalent to those measured in the laboratory. As a result, a numerical analog method for building field-scale rock physics relationships that accounts for heterogeneity, the physics of sampling, and geophysical survey design is proposed. In synthetic studies, it was found that the accuracy of water content estimates obtained using the field-scale approach was significantly improved using the field-scale approach compared to that obtained using traditional rock physics concepts.