How it Works (Propagation)

At each seed point, the propagator extracts a window of values from the volume. The size of the window above and below the seed point is controlled by the settings.

It then follows these steps for each current seed (or pick):

  1. For each neighbouring location
    • It cross-correlates the seed waveform and the seismic waveform for the range of allowable shifts.
    • It finds the TWT shift that gives the highest correlation.
  2. It chooses the neighbour with the the highest correlation.
  3. It adds this point to the horizon with the shift calculated from step #1.
  4. When stored, it includes the seed point it came from. It uses this seed point to compare its neighbours on the next time through the cycle.
  5. If a point is reached by picks from different seeds, it uses the shift calculated from the one with the highest correlation.

This is repeated until there are no neighbours that exceed the stop Threshold (%).

Implications of the algorithm

This has a few important implications that are not always obvious:

  1. When determining a pick and there is more than one picked neighbour, distance from the seed has no bearing on the next pick.
  2. It only uses seeds used by one of its neighbours, i.e. there must be a path all the way back to the seed following picks that used that seed.
  3. It does NOT compare every seed with every point.
  4. The propagation does NOT track peaks or troughs. It tracks the shape of the waveform, and will place a pick at the TWT where the correlation is highest.
  5. If snapping is on:
    • It snaps manually added seeds to the chosen event type.
    • After finding the highest correlation TWT, it adjusts the pick to match the chosen event type.
  6. By its nature, the propagator does not match amplitudes. It picks the best match to the shape of the wave, not the amplitude.
  7. For small windows, the correlation is higher and a higher threshold is required.
    • e.g. If your window only includes a single peak, then it can find high correlations to that simple waveform everywhere, despite the fact that a human would say that they are obviously not the same event.
  8. Similarly, for large windows the correlation is lower and a lower threshold is required.
  9. The Search (allowable shift) distance is per trace.
    • e.g. Using a Search (m) of 20 ms allows jumps of 20 ms from one trace to the next and 100 ms over 5 traces. This is probably way too large. Use the smallest value that will pick successfully.

Depth vs. Time

Because it is a waveform propagator, it avoids using depth data if it can. If you have selected a depth volume but time/depth conversion is configured, then it will operate in time. If you select a depth volume and time/depth conversion is not configured (or the velocity does not cover the propagation area), then it will operate directly in depth.

Not getting good results?

If it is making too many picks that are not on the event

Is the window length (Above/Below) big enough to capture a sensible waveform?

  • If the window length is too small, then you may be trying to match a tiny piece of a waveform  which may appear everywhere.

Is the allowable shift (Search) too big?

  • If this value is large, then you are giving the propagator a lot of freedom to make big jumps between traces.  If it is "cycle-skipping", then you may be allowing it so much freedom that it is jumping to another event with a better mathematical match. Generally use the smallest shift that you can get away with (the default is 5 ms).

If it is making way too few picks

Is your window length (Above/Below) too big?

  • In other words, is it capturing a big waveform that does not remain consistent from trace to trace? Look at the size of your window relative to the size of the waveform.

Is the waveform changing rapidly?

  • If you cannot identify a common waveform — if the sidelobe energy is changing, polarity reversing, etc — then you may have to play additional seeds in those different areas. In clean data, you can often pick an entire horizon with a single seed, but difficult data may require more seeds.

Is it simply very noisy data?

  • If the signal is dominated by the noise, then there simply may not be a lot that the propagator can do. Try constraining it to smaller areas, or manually pick a lattice, and use those points as seeds. The propagator will fill the holes.