Streamline-based interpretation of 2D GPR reflection images

Abstract Ground penetrating radar (GPR) reflection imaging is a key geophysical method to investigate near-surface environments. Under the basic assumption that processed GPR data represent electrical impedance contrasts, the main goal of interpreting GPR images is to derive a structural model, for example, outlining major geological boundaries and layers, respectively. Often, such interpretations use strategies known as horizon tracking and facies mapping to generate line drawings and zonal (facies) models, respectively, where the identification and selection of key reflection events and patterns typically rely on manual input of the interpreter. Thus, these strategies can be regarded as laborious, subjective, and non-reproducible. To address these shortcomings, we present a novel, streamline-based approach for the interpretation of 2D GPR reflection images. In our proposed workflow, we first calculate the gradient structure tensor (GST) to characterize local structures in GPR images. Then, using eigenanalysis of the GST, we derive a steady-state flow velocity field to compute streamlines aligned with the structures imaged by GPR. We introduce automated strategies for seeding and for aborting the streamlines in discontinuous image regions (e.g., abruptly ending reflection elements) which represents the basis for an efficient reproducible interpretation. Using synthetic GPR data simulated for a realistic sedimentary environment and a field data example collected across Holocene sediments, we demonstrate and evaluate the applicability of our approach. Our results show that streamline-based line drawings and facies models comprise the structural details present in the input GPR data and provide reasonable subsurface models similar to those obtained from a laborious manual interpretation. We conclude that streamline-based approaches allow to interpret GPR reflection images in a more automated, objective, and reproducible fashion and, thus, represent a promising approach for different typical fields of application.