Picking noisy refraction data using semblance supplemented by a Monte Carlo procedure and spectral balancing

Abstract

Abstract In crustal refraction experiments, the crucial deeply refracted and head wave arrivals often have a low signal-to-noise ratio. A method to aid in the picking of noisy refraction data is presented which is applicable to any branch of a seismic section whose waveform is approximately invariant throughout the branch. The technique exploits the spatial correlation of arrivals and is based on the lateral coherency which results if the refracted arrivals are aligned by applying appropriate time shifts to each trace of the branch. The alignment of arrivals occurs iteratively and is accomplished through a cross-correlation of each trace with the stack of the section of the previous iteration. The iteration yielding the section with the highest degree of lateral coherency (semblance) is used to extract the travel-time pick of each trace. The pick, plus a possible d.c. component, is the negative of the time shift required to achieve arrival alignment. Two modifications can improve the performance of the picking routine. To prevent a cycle skipping problem, a Monte Carlo technique is implemented in which the cross-correlation function is transformed into a probability distribution so that the lag corresponding to the maximum cross-correlation is most probably selected. Second, to increase the coherency of the arrivals, a spectral balancing technique is applied in either the time or frequency domain. The picking routine is applied to both a synthetic and real data example, and the results suggest that the routine can be applied successfully to data with a signal-to-noise ratio as low as one. Also, the Monte Carlo procedure together with spectral balancing increases the final semblance over that obtained with the unmodified method.