Implementing the data processing sequence

Because we determine only one set of PEF's for a given processing effort, we improve the ability of the PEF's to model their respective parts of the data by processing only half the traces at one time (in this case, the positive offsets). Therefore, the noise is more easily modeled since all coseismic energy dips in the same direction. We include the three nearest negative-offset traces so that the final result (after ``losing'' traces to convolution with PEF's) includes all of the positive offset traces.

Data processing begins with the selection of suitable windows for determination of PEF's 332#332 and 338#338. With these windows chosen, we can test different parameters for the rest of the processing, namely the sizes of the two PEF's, the value of the weighting factor 12#12, and the number of iterations for the inversion.

Because the signal of interest is horizontal, it is logical to use a signal PEF 332#332 that has only one element in the vertical direction. This assumption is confirmed by experimentation, as taller PEF's are less effective at modeling non-horizontal energy. We find that a width of 4 is a fair trade-off between the better resolution of a wider PEF and the smaller number of traces lost to convolution afforded by a narrower PEF. The size of the noise PEF 338#338 is more arbitrary, but we find that dimension 4x3 yields satisfactory results.

12#12 is best determined experimentally, though the development of a more rigorous approach would improve the versatility of the processing sequence. Interestingly, a broad range of 12#12 values results in essentially the same final result. We find 362#362 to be a reasonable value for the data shown here. Finding a reasonable criteria to define the end of the iterative inversion process is another aspect of this processing sequence that deserves more attention. We have simply stopped the inversion after 350 iterations.