Presentation is loading. Please wait.

Presentation is loading. Please wait.

Why determining an exact waveform is next to impossible. We start at the recording point with these facts – 1. The seismic continuum consists of overlapping.

Similar presentations


Presentation on theme: "Why determining an exact waveform is next to impossible. We start at the recording point with these facts – 1. The seismic continuum consists of overlapping."— Presentation transcript:

1 Why determining an exact waveform is next to impossible. We start at the recording point with these facts – 1. The seismic continuum consists of overlapping individual primary reflections and noise. Each reflection coefficient produces an independent primary event (tops, bottoms and in-betweens). 2. Each event travels independently, not mixing until the recorder forces it artificially. 3. Downward traveling wave shapes evolve with time, friction and inertia causing trailing lobes to form, with the initial higher dominant frequency lobes being diminished in amplitude. 4. The two way travel time increases greatly as we move to the outside traces. This means the evolved shape of each primary can change significantly between offset stations. 5. The spacing between individual primaries arriving at a station changes (the effective velocity is dependent on travel time). 6. Because of these factors, the primaries are going to mix differently at the recorder, meaning that each offset will have a unique character (in contradiction to the original stacking concept). Event polarity will depend on how they mix. Just for emphasis, the hand drawn down wave at the left is probably not that far off what might be, and I use it to remind all that this is what existing deconvolutions try to compress. When Applied before stack they further confuse the wavelet picture. And pay attention to the reflection coefficients in the center, noting that their amplitudes are dependent on the overlying lithology, as well as on the lithology of the new bed. This is why undo faith on attributes from non-integrated inversion results is misguided Trying to get perfect answers makes linear approaches unstable, so they have to back off. On the next slide I expand on the non-linear alternative.

2 While non-linear inversion approaches can differ drastically, the one I describe uses advanced pattern recognition to compute reflection coefficient guesses. While it wants the signature to be as clean as possible, it still can locate the primary events when the signature is muddled by noise. The true power of statistical optimization was unknown to the author until he started pushing it hard. In his original predictive deconvolution he had used the autocorrelation extensively to get a guess at wave shape. Then, one day, working on the new inversion, he decided to see if he could improve the initial wavelet guess. The fresh idea here was to use it to make an initial convolution pass, record the spike guess timings and go back to use this information to compute a new wavelet (repeating this process as long is it was improving the ability to explain the trace energy). It soon became evident that each new wavelet did a better job. Thousands of development hours later the system was showing him how powerful the statistical power was. The key here is to gently lift off the energy explained by the last spike guess. The new philosophy was to ignore the time it took and just concentrate on how deep the system could go. The results were continually surprising. These are the kinds of things one can do when one does not answer to anyone else. Of course tests (that allowed the system to exit the loop if the improvement was not significant) were inserted. Happily, results were never worse. The operating theory of this inversion is to keep making wavelet and spike position guesses until the original trace energy is explained to the limit of system ability. The logic consists of 3 layers of iteration. The first (open ended) level loops through consecutive wavelet guess runs, starting with an initial wavelet that is computed using autocorrelation like logic The 2 nd layered level loops through the selected set of stacked traces, and the 3 rd through the per/trace optimization. Here, at the third level, is where the coefficients are calculated (the same waveform being used for all). The system subtracts the pertinent energy (associated with each spike) from the current working trace. If, at the end of the major loop no improvement has been made, the system exits this phase of the operation. If improvement has been made, a new wavelet (see below) is computed, and the original, untouched trace is loaded back into the work area. This “return to the original data” keeps the system entirely honest about what it is doing. To compute the new wavelet for the next major pass, the system moves through the previous spike guesses, adding data from their effective spans to a summation vector.It then formalizes the new wavelet from this vector. Each guess is displayed during the run, and watching the shape develop is an education in itself. Obviously driving logic tricks were developed to push the convergence, but the system is as honest as it can get. It can display the spiked output, and if you understand what you are looking at, it is impressive. Since potential users did not like the complexity of multiple interface interpretation the developer went to the integration processes and soon became convinced that it in itself was a major contribution. So let us go on to see some of the great well matches achieved.

3 Proving ADAPS via honest results. No well data is input! The well images you see here were super-imposed after the runs were made. The ability to match well results is the best test for any tool that is pretending to improve resolution. I present these great examples as my answer to those seeking mathematical proof. To appreciate what has been accomplished, one must compare the “before and after”. In these examples I match each final result with the best the owner had been able to do using state of the art processing.. Dominant events that were probably mapped completely change character as a result of the powerful detuning. This is illustrated by the shale play to the right where significant changes are verified by excellent well matches. This is the essence of the ADAPS argument. The best possible answer under the conditions that exist is the goal of all optimization software. In the case of ADAPS we set out to explain the energy on the individual traces using all our seismic knowhow. I am the first to say that these matches are not always perfect. However most make me smile. The ADAPS de-tuning is a two part thing. First, it does a superb job of simulating the reflection coefficients (spikes to us). However, once done, these interfaces are often hard to map, since even when the beds extend for miles, the individual spikes may come and go. ADAPS goes one step farther by integrating the spikes to simulate actual lithology. please note that on all the examples there are more events showing on the input than on our output. In other words the system has completely re- arranged the input energy bands according to it’s idea of true lithology. Please take time to examine the improvement, lobe by lobe. A solution looking for the right problems – I have yet to see the prospect where the ADAPS results weren’t an improvement. Of course when the structure is boring the differences might not seem worth the effort. The really exciting cases are where this ultimate de-tuning brings out structural details and reservoir possibilities that otherwise might be missed. One of these is discussed two slides later. This is a comparison of the straight stack and my final inverted/integrated product, each overlain with the sonic log image. The arrow points to the one event that miss-matches. (Either due to coherent noise or to low frequency removal error). Both sonic Log and my integrated output have to be corrected for low frequency drift. This becomes more difficult with thicker beds. Actually there’s a strong possibility that my output comes closer to the true lithology than this particular sonic log does. It should be noted that inversions have to be very precise (down to individual spikes) for integration to work.

4 What I want to show in these next two slides is how ADAPS uncovered a major lensing phenomenon. While there are traces of it on this input, you will see the major difference de- tuning can make in interpretation when you toggle to the next slide. But before you go forgive another lecture on well matches. Too often interpreters look right past the fine detail, just noticing when major things more or less line up, regardless of polarity. The reason this is so important is that these matches supply proof of the ADAPS inversion and sonic log integration. Since so many lobes are removed by the process, final matches are crucial to industry acceptance. On this input match, most matching lobes are just randomly misplaced, while many show opposite polarity. Those to the right should match with blue events of course. On the finished product you will see remarkable agreement. Please take the time to study this by toggling. This is the lensing area mentioned. It is around the target, so the phenomenon is most important to reservoir evaluation. When I first began the study I was convinced there was faulting. Now we will see that ADAPS de-tuning straightened out the complex overlap. Toggle please T The best the client could do – The unadulterated input stack. Take your time and toggle!

5 And back Now look at the well match! To sell the merits of de-tuning, one has to find an example where it makes a profound stratigraphic difference. This is one of those finds. While I want you to study the fairly remarkable well match before you leave, first pay attention to the stratigraphic pinchout now evident below. A personal vignette – Back in the 50’s, when I was an interpreter for Mobil, I had an offset being drilled on my say so. We were shooting some work to check my interpretation and the paper results were drying in a locked partition. To see them I scaled the wall. Today, I wonder how many still feel the excitement that still drives me in my efforts. The beauty of this well match set me off. ADAPS inverted & integrated output.

6 Another case where ADAPS made lobes point in the right direction. Remember that the results shown to the right are after the inversion computed reflection coefficients, then integrated them to produce the lithology simulation. No interpretive shifts were made to achieve the well log match. As a matter of fact one could not shift the input artificially to achieve the well log match (as at least one reader has claimed). But also notice the enhanced strike slip fault evidence. BeforeAfter

7 Click after noting there’s no way to shift the sonic log on the input to get a decent match. Now lets talk about the polarity changes. The fact that they happen is probably one of the most important things to be learned about seismic attributes. Once again you have to pay attention to the well log match. Remember that the inversion comes first, computing actual reflection coefficients. Then consider that individual lithologic units will produce primaries from both their tops and bottoms.The function of the integration is to assemble these individual spikes back into a picture that resembles the lithology of each of those beds. Since the data is never perfect all we can hope for is an approximation, but that is surely a lot better than what we see on the input. So when you see a polarity change it is because of the way the primacies combined, and it may be vital to reservoir interpretation. For some reason, people have had a hard time with this concept, but once again, trust the “after the fact” well log match. Once you see enough agreement, trust will grow.and the well log checks will not be needed. This gets us into direct reservoir detection.

8 A “post stack” example where parameters were tailored to Vibroseis input. Here I show the results first. Again note the remarkable well match, concentrating on the thickness. Please toggle with straight stack input.

9 The question is: How could anyone have been satisfied with this match?

10 Rocky mountain (post stack) stacked input, line Y Please toggle w results So, let’s play a game. See if you can move the sonic log up and down (in your mind) to get a better match. Then toggle back and forth with the final result to see that the improvement is real.

11 Rocky mountain (post stack) inversion and integration, line Y Please toggle w input

12 Another final with good well match -

13 And still another – (we could go on). You can contact me at dpaige1@sbcglobal.netdpaige1@sbcglobal.net Click here to continue inversion and integration series with an intro to direct reservoir detection that has a new bunch of well matches. Or here to return to base.


Download ppt "Why determining an exact waveform is next to impossible. We start at the recording point with these facts – 1. The seismic continuum consists of overlapping."

Similar presentations


Ads by Google