Workflow and Strategy

The best results with AIMBAT are obtained if you understand the unique, non-linear workflow that it prescribes. Unlike more traditional top-down processing, AIMBAT is more of an iterative process whereby parameters are constantly adjusted to gradually determine the optimal settings. The exact strategy for a particular event may differ, but there are some general guidelines below that we recommend following.

Without AIMBAT

Multi-Channel Cross-Correlation¹ (MCCC) relies on narrow time windows, focused on the initial arrival of the targeted phase in order to yield high quality results. Manually picking the phase arrival on each seismogram individually is a very time consuming task, which is highlighted by the stacked cards in the flowchart below:

flowchart TD
  A@{ shape: circle, label: "Start"} --> B>Import seismograms];
  B --> C[Select suitable filter parameters]
  C --> D[Choose high quality seismograms to use for MCCC]
  D --> E@{ shape: processes, label: "Individually pick phase arrival for seismograms 1...N"}
  E --> F[Choose time window for MCCC]
  F --> G>Run MCCC to align seismograms]

With AIMBAT

AIMBAT² stacks all input seismograms (aligned on the picked phase arrival) and operates on that stack instead of individual seismograms. This allows picking the phase arrival once for all seismograms simultaneously, and then improving it iteratively before running MCCC. Note that both the ICCS algorithm, as well as adjusting AIMBAT parameters are iterative processes.

flowchart TD
  A@{ shape: circle, label: "Start"}
  A --> B>Import seismograms containing initial picks t0];
  B --> F
  E[Adjust AIMBAT parameters];
  E --> F>Run ICCS with initial/updated parameters]
  F --> G[Inspect results of alignment];
  G --> H{"Continue
          with
          MCCC?"}
  H ---->|Yes| M>"Run MCCC for final alignment"];
  H -->|No| E;

Strategy

AIMBAT does not prescribe a single strategy for picking/updating processing parameters. That said, some general principles to follow are:

1. Only change one parameter at a time, and then run ICCS to see the effect of that change on the alignment
2. Snapshots are immediate and use no storage, so create them often (and don't skip adding a comment that describes the snapshot).
3. Don't get too distracted by individual seismograms that are not well aligned or seem of poor quality. Trust the algorithm to deal with those.

ICCS running modes

Points 1 and 2 above are pretty self-explanatory, but point 3 deserves a bit more explanation. The ICCS algorithm has two flags that can be set before each run:

• Autoflip: If set to True, the algorithm will toggle the flip parameter for seismograms that are negatively correlated with the stack (i.e. the maximum absolute correlation coefficient is negative).
• Autoselect: If set to True, the algorithm will automatically set the select parameter for seismograms that are poorly correlated with the stack to False, but it will also toggle previously deselected seismograms back to True if they become well correlated with the stack.

The flip parameter determines whether a seismogram is flipped in polarity as the data are prepared for the stack and following cross-correlation. The select parameter determines whether a seismogram is included in the stack; however, all seismograms are cross-correlated with that stack regardless of their select status. This means that all seismograms, even with select=False, can self-recover if they fit the stack better after changing parameters.