- Copyright: © 2010 This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.
Some test studies were performed for comparison of two travel-time inversion schemes for tomographic evaluation of crosshole ground-penetrating radar (GPR) data. The first scheme was a linearized inversion based on Tikhonov regularization (Method 1). In this scheme, ray tracing was not a part of the inversion algorithm and the Jacobian matrix was calculated by numerical differentiation. Travel-time calculations were performed by a finite-difference eikonal equation solver. Model velocity fields were updated by matrix inversion techniques using iterative conjugate gradient solvers. The inversion process was stabilized by a smoothness-constrained regularization. The second scheme was based on a ray tracing algorithm (Method 2) and velocities were updated by a simultaneous iterative reconstruction technique (SIRT) using both straight- and curved-ray approximations. The test studies included synthetic travel-time data sets generated from the models with various velocity distributions. Broyden's update was implemented within Method 1 to expedite the calculation of the Jacobian matrix, and this greatly improved the computational performance. In the tests, the effect of the regularization parameter on the models from Method 1 was examined. Also, how the straight-ray and curved-ray assumptions affected the solutions from Method 2 was illustrated. The effect of the initial velocity distribution on the resulting tomograms was exemplified by the solutions from both Method 1 and Method 2. The velocity tomograms from Method 1 were characterized by smaller travel-time residuals, Euclidean distances and lower errors in the velocity of cells. Also, the convergence rates of the solutions from Method 1 were faster than those from Method 2. Method 1 better imaged the zones with the high velocity contrast than Method 2, and both methods produced similar velocity distributions within the zones with low velocity contrast. Overall, Method 1 yielded better solutions compared to Method 2, and the curved-ray inversion generated relatively better results than the straight-ray inversion.