Inversion strategy in crosshole radar tomography using information of data subsets

Detecting discrete anomalies, such as cavities or tunnels, is an important application of crosshole radar tomography. However, crosshole tomographic inversion results are frequently ambiguous, showing smearing effects and inversion artifacts. These ambiguities lead to uncertainties in interpretation; hence, the size and position of anomalies can only be interpreted with limited accuracy and reliability. We present an inversion strategy for investigating discrete anomalies with crosshole radar tomography. In addition to the full traveltime data set, we use subsets of specified ray-angle intervals for tomographic inversion. By analyzing inversion results from different ray-angle intervals, a more accurate interpretation of anomalies is possible. The second step of our strategy is to develop a good inhomogeneous starting model from joint interpretation of the inversion results from different subsets. The third step is to invert the full data set using this new starting model and to evaluate the inversion results by analyzing the distributions of mean square traveltime residuals with respect to the ray angles. We use a synthetic model with two discrete anomalies located roughly at the same depth to demonstrate and evaluate our approach. This inversion strategy is also applied to a field data set collected to investigate karst cavities in limestone. From the inversion results of both examples, we show that horizontal smearing of anomalies can be reduced by eliminating near-horizontal rays. A good starting model can be obtained based on the joint interpretation of the inversion results of the different subsets; it leads to a high-resolution final image of the full data set.