Optimizations should converge. When they don't, there is a reason. Either the logic is flawed, or something in the data interferes with our pattern recognition. The fact that my wavelet definition logic was floundering led me to ask for the raw pre-stack data.

And here is the set of pure noise I found. This data is also from the middle east.

We are looking at various sets of refractions. There is little doubt that most of the coherent noise we see throughout this series is caused by near surface, highly reflective beds. This is the worst of the bunch. I haven't given up on uncovering real reflections yet, but the problem is daunting. The positive result is that we know not to trust any interpretation until we solve it. 

Again, the key is to understand the problems before we jump in!

I asked for the gather data and this is what I saw after I had programmed a display capability. Earlier in this presentation I mentioned that explaining the noise cone as either an "air wave" or as "ground roll" did not explain the phenomenon. The reason for this statement can be seen at the center of the cone where low frequency events mimic a real section. The cone is bounded by refractions.

Introducing the omni-present central noise cone (and the refraction phenomenon). The cone is at "A" of course. Some call it an "air wave" and others pass it off as "ground roll". Neither of these descriptions fits what I am seeing on thousands of examples. Where a ground roll (or air wave) would pass after a few cycles, this phenomenon continues as a series of apparently independent, low frequency events.

This, and other observations are made on the basis of literally looking at thousands of cones, using the new software I developed to ease this burden.

The shear wave explanation makes more sense. We know shear waves are present (where particle moveout is lateral rather than vertical). Of course the source of the noise is not as important as the recognition that it is there, and that it is a problem.

To the left of "B" and "C" you can see refractions. Note the variation in the one way velocities. You can see the one at "B" butting up against what is the target energy. We will discuss this interference problem later.

The probable cause of both of these noise forms is  the presence of strong shallow reflectors. Of course refractions (energy traveling horizontally along an interface) occur when the angle of incidence becomes greater than the critical angle. It may well be that deep water results are better because the angle of incidence is lower for any ray path. Since the shear wave energy travels at very low velocities, the increased depth may keep them out of sight.

The presence of more than one wavelet shape within a time zone makes it very hard to do advanced things with the data. Time series mathematics essentially assumes there will be just one. In the picture below it is interesting to look at the interference zones. Even though that data was filtered with a non-linear prediction technique, one can see phase shifts in the target data. I discuss this below under the filtering topic.

Pattern recognition and noise removal. The reason for re-emphasizing this point here is to point to the need for non-linear, "pattern recognition oriented" methods in attempts to remove coherent noise. Because there is generally a spectrum overlap in the wavelets involved, straight, frequency dependent filters are dangerous. Often they will appear to have solved the problem, but the distortions they introduce cripple resolution improvement. There is more on this subject in the next topic.

The second image shows the data that was passed on to the next gather stage.

I use it again later to describe this process.