Global inversion of GPR traveltimes to assess uncertainties in CMP velocity models

• Published in: Near surface geophysics (Online) 2014-08, Vol.12 (4), p.505-514
• Main Authors: Hamann, Göran, Tronicke, Jens
• Format: Article
• Language: English
• ISSN: 1569-4445; 1873-0604
• DOI: 10.3997/1873-0604.2014005

ABSTRACT Velocity models are essential to process two‐ and three‐dimensional ground‐penetrating radar (GPR) data. Furthermore, velocity information aids the interpretation of such data sets because velocity variations reflect important material properties such as water content. In many GPR applications, common midpoint (CMP) surveys are routinely collected to determine one‐dimensional velocity models at selected locations. To analyse CMP data gathers, spectral velocity analyses relying on the normal‐moveout (NMO) model are commonly employed. Using Dix’s formula, the derived NMO velocities can be further converted to interval velocities which are needed for processing and interpretation. Because of the inherent assumptions and limitations of such approaches, we investigate and propose an alternative procedure based on the global inversion of reflection travel‐times. We use a finite‐difference solver of the Eikonal equation to accurately solve the forward problem in combination with particle swarm optimization (PSO) to find one‐dimensional GPR velocity models explaining our data. Because PSO is a robust and efficient global optimization tool, our inversion approach includes generating an ensemble of representative solutions that allows us to analyse uncertainties in the model space. Using synthetic data examples, we test and evaluate our inversion approach to analyse CMP data collected across typical near‐surface environments. Application to a field data set recorded at a well‐constrained test site including a comparison to independent borehole and direct‐push data, further illustrates the potential of the proposed approach, which includes a straightforward and understandable appraisal of non‐uniqueness and uncertainty issues, respectively. We conclude that our methodology is a feasible and powerful tool to analyse GPR CMP data and allows practitioners and researchers to evaluate the reliability of CMP derived velocity models.