Development of Generalized Polarization Analysis Methods for Regional Array Data

摘要

In this report we present test results on various non-conventional signal processing approaches as applied to data from regional seismic arrays. In particular, an adaptive approach for automatic identification of regional arrivals was tested too (Der et al 1993). This method was designed to circumvent some difficulties that plagued previous attempts of 3-component polarization analyses for regional arrival identification. In particular, the methodology is designed to avoid complications due to signal distortion by near sensor geology and complex arrivals made up of superpositions of multiple independent signal components. The idea is to replace time delay operators and assumed particle motion patterns used in deterministic signal processing with empirically computed sets of transfer functions using a learning set of relatively high S/N events. The application of this approach requires more than three sensor outputs (such as a 3-component combination plus some extra array sensors). We have tested both a frequency domain and a time domain approach to this algorithm. Thus far the frequency-domain version has reached a further stage of development. The methodology appears to be an attractive alternative to conventional array processing methods in quickly identifying and discriminating various arrivals from the same source region, or similar arrivals from different source regions at a small array. We are also addressing the problem of onset time estimation for regional arrivals. Most methods proposed thus far attempt to find the first point where the statistical properties of the seismic trace change. This is difficult or next to impossible with the often emergent regional arrivals. Instead, we are advocating a methodology based on the cumulative sum (CUSUM) of various statistics. Superposing a linear trend on the cumulative sum can make the changes in the trend of CUSUM to become minima of the resulting function.