F. I. Louis, K. C. Makropoulos and I. F. Louis

Abstract: In seismology the accurate mapping of faults and fault zones plays an important role as a crucial step for characterizing the earthquake potential of an area. Cross-hole seismic surveys for engineering site investigations can provide high resolution images of subsurface seismic velocity capable of delineating tectonic structures such as vertical or dipping faults juxtaposing different geological formations or step-like structures with good accuracy. The result of velocity reconstruction in travel time seismic tomography is restricted by factors such as ray aperture and distribution. These physical limitations cannot be completely surmounted even in cross borehole acquisitions where the ray coverage is the best one possible. The present paper studies the effects of staggering normal grids as a tool to increase resolution and reduce inversion vagueness and instability. Truncated Singular Value Decomposition as a mathematical tool of least demands on the final solution is utilized, and the inversion scheme is examined with respect to the number of available data. Numerical simulations of a model featuring a fault-like structure are performed and the resulting recovered images are compared against straight grid inversions. Reconstructed images using staggered grids provide a smoothed version of the true model since they basically operate as a moving average filter on the model. On the other hand the final tomograms of the synthetic tests, as a result of shifting the grid, show high accuracy and resolution since both the geometry of the fault and the velocity values throughout the model are better determined. Conventional grid inversion fails to image properly regions of reduced ray coverage, and in general a considerably blurred image is generated. By staggering the grid we can enhance image quality and reduce possible nonuniqueness of solutions without imposing any constraining conditions such as smoothing or damping, retrieving in that way all possible information from our data without the risk to lead to a preferred result.