Rapid estimation of relative and absolute delay times across a network by adaptive stacking

Abstract

Adaptive stacking provides a powerful and rapid procedure for estimating the residual patterns across a network of seismic stations. The approach exploits predictions from some propagation model to achieve an approximate alignment of traces, which are then stacked to form a reference trace. Iterative improvement of the alignment, by comparison of the reference trace with each station trace, leads to a direct estimate of the residuals from the propagation model. Our implementation is fast and robust in the presence of significant noise and waveforms of different character. The major difference from earlier forms is the use of a direct minimization scheme for determining the best match between the reference stacked trace and each recorded trace based on an L3 measure of misfit. This approach has the benefit of generating automatic error estimates. For teleseismic applications, the ak135 model has proved to be very effective for selection of the window around the desired phase and in achieving initial alignment. The approach can be applied to both first motion (P) and later phases (e.g. PcP), with extraction of absolute time via the improved signal-to-noise properties of the stacked trace, after full alignment. The new method is illustrated with three teleseismic events from a similar source region that were recorded by the 72 station TIGGER array in northern Tasmania. Despite significant levels of noise and contrasting waveforms produced by the three events, the residual patterns are very similar. Even when large time-shifts are introduced into the records, trace alignment is achieved with rapid convergence of the iterative procedure and excellent recovery of the imposed shifts. These results confirm adaptive stacking as a valuable alternative to data based cross-correlation techniques, particularly when heterogeneous instrumentation is employed.