scontent1

Overautomation and misguided interpretations are my themes - convincing you there is a better way is my hope.

Reinterpretation should be in your mind when you finish this study. My approach to true deconvolution should be clear to you also.

think.gif (20751 bytes)
I

dpaige1@houston.rr.com

Do I know what I am talking about?
Skim through the "credentials" group above if
you have doubts.

This series has 6 thought groups - Each depends on you having looked at the previous ones. Don't assume you know what I am going to say on a subject, since It generally will differ from industry dogma. To change to a new group you must click on the title up above. Otherwise your mouse does the switching.

1. You are in the Over-Automation group. To
proceed, move your mouse over the topics on
the left panel.

Otherwise click above for:
2. Long Leggy Wavelet - describes the main
    problem. If you don't go there forget the rest.
3. Unique Character by Offset - sets the stage
    for the full spiking solution. Also hits AVO.
4. "Inside" Deconvolution - shows why current
    software doesn't cut it.
5. The ultimate answer - Describes nonlinear
    solution, along with recaps.
6. Credentials.

5 or 6 gives access to Q&A for more detai

If you had to use the vertical scroller to read the last paragraph we are in trouble. This series is only checked out on the Microsoft browser under the Windows system. If you are using AOL, get out and invoke that browser (I have to do it at my daughter's house). The thought groups that follow do not have a vertical scroller, so you will miss a little. Let me know if you can't solve the problem, and I will put the scrollers in.

Let's start with results I would have been happy with way back then.

Let me list the criteria:
1. No obvious noise patterns.
2. Independent dip patterns
(see "A").
3. Extensive continuity.
4. Good character variations.

This looks like deep water data. While there is a hint of a deep water repeat, It probably has been subdued by the velocity differences.

Even here there are processing problems. Look at point "B" and follow the data downwards. This is not structural or stratigraphic. Automatic picking would be dangerous here.

What we can say for sure is either the wavelet was short to start with, or the processing was good. It is not really clear why deep water stuff does not seem to have the wavelet problem. We discuss that as we go on.

As good as this is, we can do better. For now let's see some bad stuff.

Great graphics and lousy data. Take some time to look at the lower band of energy on the left hand section (detail below) to see what I mean.

detail.gif (8373 bytes)

Automatic picking
results

If you believe this our relationship is over.

The best event here is coming off the fault face, and they missed it.

The rest of the data is similar to the previous example, with:
1. Very low frequencies,
2. No independent character.

Here the automatic picking just wanders between the low frequency lobes. Personally, I don't think differences in stratigraphy explain why this data is so much poorer than the first I showed.

What I think the problem is - Complex mixing of "leggy" primary reflections.

Long wavelets
(a two bed example).

Two sandsones in shale convolved with a long wavelet. Note the strong lobal energy that does not line up with the beds. Take some time here please.

Event patterns and
over-deconvolution.
Under the decon set of topics I get into what happens when people go too deep using current programs. This is what I see here. This generally leaves the low roll shown above, and tends to bring out noise patterns.

Stare at this a while, looking through the mesmerizing colors.

A. Take your pick between two opposing patterns. Look around the rest of the section to see the same kind of thing. If one is real data, the isn't and vice versa.
B. The same kind of thing. Notice the upturn to the right of the "B". Not picked here, but the same type of event to the right of "C" is used to define a fault.
C. Note the absence of anything (yet an event is picked running parallel to the reflection from a fault face - wishful thinking at best).
D. Nothing again. The deconvolution has erased all onsets, leaving only the very low frequency roll that may well have nothing to do with true reflections.

E. Note the event under "E" colored in aqua. A fault pushes right thru what normally would be picked as continuity (if you took out the coloring). In fact the fault itself can be questioned.
F. The same goes for the imaginary fault between "A" and "F".

General observations - We are just seeing a glimpse of real data here. The unconformity shows up, but it is probably higher in the section in real life. 3D does not eliminate our problem, since noise patterns continue spatially if recording stays the same.

The misconception - Seismic sections are mostly looked on as true mirrors of subsurface structure (except for essentially random noise). When the data is muddled most think the reflections are just not strong enough to make themselves visable.

As we saw in the last few examples, whatever comes through tends to be believed. The first purpose of this series is to show that this is highly dangerous. The second is to show how to solve the problem.

To continue from here, click on a thought group above. Their thumbnails are:

Long & leggy wavelet =	The long wavelet is a most serious source of noise - the wavelet tail masks events we want to see (and produces artificial ones that give us trouble). If you don't study this forget the rest and get out of here (since everything I say hinges on this truth).
Unique character by offset =	Each trace in the field spread has its own unique character, caused by offset (contradicting AVO theory). An important topic as it ties into the inversion discussion.
Inside deconvolution =	My predictive deconvolution and frequency methods are dissected. Important comments on the limitations of both.
The ultimate answer =	Pattern recognition is a big hope in the industry. A discussion of why this is pointless until we solve the basic wavelet problem. Discussion on my nonlinear solution, and an ongoing Q&A where I respond to the expected criticisms (add yours via email).
MyCredentials =	A technical autobiography.