Issues during the inversion of crosshole radar data; can we have confidence in the outcome?

One method of assessing the confidence in modeled features is to compare the results from different inversion schemes. I use synthetic traveltimes calculated from a model of an unconfined aquifer to determine the reliability of crosshole tomography. I compare the inverted models from straight and curved ray approximations to wave propagation. I investigate the effects of added random noise, regularization, the starting model, and the reference model on the curved ray inversion method. I also investigate the effects of different grid sizes for the forward model and of limited ray coverage through the earth model. Understanding the effects of these different methods and parameterizations will help place confidence limits on modeled features to more accurately reflect our knowledge of the subsurface. Straight or curved ray approximations to wave propagation resulted in similar models. However, the resolution estimates are substantially different and would lead to different assessments of the model reliability. Comparisons of the different choices for the model parameterization show that the resulting models are similar, indicating that tomography is a robust method. The most important factor to obtain reliable parameter estimates from crosshole radar tomography is to acquire wide aperture, densely sampled data with little noise.