Geophysical Insights at SEG 2017 in Houston, Booth #301
Using Attributes to Interpret the Environment of Deposition - A 1 day course. Taught by Kurt Marfurt, Rocky Roden, and ChingWen Chen
Dr. Kurt Marfurt and Dr. Tom Smith featured in the July edition of AOGR on Machine Learning and Multi-Attribute Analysis
Rocky Roden and Ching Wen Chen in May edition of First Break - Interpretation of DHI Characteristics using Machine Learning
Seismic interpretation and machine learning by Rocky Roden and Deborah Sacrey, GeoExPro, December 2016

Interpretation via SOM to reveal sub-seismic resolution

Houston Geological Society (HGS) North American dinner covering seismic facies in the Eagle Ford.

An excerpt:

I'm going to show you how the attributes were selected for the self-organizing maps that are going to be the "result sets" we call them. I'm going to use some jargon here and I want you to let me know if I lose you at any point because there are some key points I'll try to do my best to be clear about. And then I'm going to show you the conclusions before we even get started so that you'll know where we're headed. Principles are applicable. Ok, I have a pointer to use from time to time. So we used, eventually, about 216 square miles of that merged 3d I referred to its preset time migration. 

...read more >

Self Organizing Maps and the Eagle Ford

By Patricia Santogrossi, Geophysical Insights
February 2016

Full Presentation Text...

 

This analysis was on a very large merged 3d that was offered to Tom by a bud, a friend of his, mostly for research purposes. It covers two counties in south-central Texas and the original volume is almost 900 square miles but we sectored out a piece of it. That was elapsed area, not all live traces but we sectored out an area of that for me to work. The results presented here are really collateral to this envisioned research project. Be advised, it was not an exploration or exploitation project per say but so much came out of just using the software in a thoughtful manner and I will explain some guidelines and then see what we could see.

 

Although, the Eagle Ford was the focus of all the work. I also changed the title up a little bit because this work resulted in subseismic resolution stratigraphy and facies for the Eagle Ford group that even started my colleagues who had been working with the software for three or four years by the time I came on board about a year ago in November 2014 So anyway, I want to remind you that the principles here in are completely transferable to conventional plays and that because they're very, they're quite fundamental.

 

So I'm going to do a brief overview and then the talk will be in four parts. If we start to run terribly long we can cut it short. If you're not satisfied with how many well calibrations you see in it we can look at that stuff on a side bar afterwards. I did try to up the anti there on that particular part. So in terms of the overview, we're going to talk about the project basis real quickly. The tools that are in use with paradise. I'm going to give you guidelines to multi-attribute analysis because a lot of people have not a clue how to apply multi-attributes successfully. Helps to have a really good piece of software to work with.

 

I'm going to show you how the attributes were selected for the self-organizing maps that are going to be the "result sets" we call them. I'm going to use some jargon here and I want you to let me know if I lose you at any point because there are some key points I'll try to do my best to be clear about. And then I'm going to show you the conclusions before we even get started so that you'll know where we're headed. Principles are applicable. Ok, I have a pointer to use from time to time. So we used, eventually, about 216 square miles of that merged 3d I referred to its preset time migration.

 

They gave me 0 to 5 seconds and importantly the data was sampled at 2 milliseconds sample interval. I'll explain why that's important in a bit. I was given four mapped horizons ultimately. I re-picked on every line in this huge volume on the Eagle Ford Shale and the Buda because they were going to be part of my delimiting of the dataset that was focused on in the end. I have used the upper Eagle Ford pic itself only for illustration purposes. And the Austin Chalk came in later so I didn't... it was never picked on every line and you'll see the difference it makes in some of our illustrations.

 

They're less pleasing in a way but they still work. I had over 300 wells to work with, tight logs, that were less than satisfying only in that they didn't include enough key type logs to satisfy the kind of work that I like to do. But I was given horizontal borehole montages. There are only nine calibration wells but that was kind of sufficient to capture most of the zonation variation that I discovered. These were pilot holes, vertical pilot holes, and with all the bells and whistles. And then, importantly, part of the research project was... which we have completed the first phase of... and we're not going public with it entirely but I'll show you some of the high level results.

 

We were given 274 of the horizontal wells with a spreadsheet of numerous non-seismic attributes. These were geologic and petro physical measures from these wells and it was phenomenal how well they worked with the observations I already made. All right, so the tools that we use in Paradise are basically three. We do, originally we do, principal component analysis and this is where the key attributes that you want to combine in your self organizing map can be selected.

 

This is sort of the rule of thumb that comes from experience, my own and others that you generally tend to select 2 to 5 of the prominent attributes. Here's where five are selected from 1 to 4, maybe even five sometimes of the Prague prominent eigenvectors. This is an eigenvector. You'll punch on that, you'll see what comes up, and I generally pick the ones are at least in double digits. And then I'll pick however many I'm going to pick.

 

This is an older example so it's actually from a tight-carbonate project where I picked fewer. But in this case, and in every case, I always check, we have a little cumulative button here, check to say that the attributes that I've selected, the number of them represent or comprise over 60% of the data. In other words they apply to at least 60% of the data. Because you don't want too few statistic, you need the right number, not too many and not too few statistics, in order to get a good answer.

 

Machine Learning

Then we go and take the selected attributes that we have and I'm going to show you the list that I used in this work and we make a self-organizing map. These are called Kohonen maps, if you are familiar with that terminology. This is a learning machine and it learns what the data has in it and it classifies the data in clusters. And it classifies them in the number of clusters that you suggest. It's called a topology. So, mine are going to be largely eight by eight. A topology of eight by eight, which produces 64 neurons or it's going to go out and classify the data into 64 little partitions. That can be too many, although eight by eight is a real good place to start, but for some problems or some geology that could be too many. You get it cut up into too many fine pieces and you can't observe the right thing or it can be too few so it's a lump or split kind of notion.

 

I went to five by five and found out that I didn't get a very big change in what I saw so I was very satisfied with my result. But this shows the geologic features anomalies and it does enable us to do very specific calibration. Yes sir, Tom? No, I started with eight by eight. I went to check what would happen with five by five and I still got remarkable resolution. So no point in doing that.

 

Yes You make these selections, you coat it up and, depending on how larger dataset is and how many attributes you got in there, it may take... this project was 10 square miles and they took minutes. I think the longest I ever ran took 13 minutes. This one was much bigger and so it is an hour and a half, you know sometimes I would purposefully run things overnight so I wouldn't have to wait for them, that kind of thing. We also have interactive 2D color maps. So our neurons will be laid out given a color map and I used mainly just a couple of different color maps here and try to help you keep abreast of when and why I changed the color bar.

 

All right, guidelines for multi-attribute analysis. We are collecting these as we go and we're learning we have the nomenclature. Your seismic data that you're given or you have, would be your base survey and if a lot of attribute generators require 32 bit seismic data and we have found that you get a much better resolution figure at least sampled to two milliseconds.

 

If you have more than 4 millisecond data you can resample it in your interpretive software down to two mils. We have some wild men who want to go down to half a mil. So you know once they get power they go crazy. So anyway we run all the attributes from the same base survey. Okay?

 

You don't want to mix apples and oranges; we do not combine attributes from different classes. So we do the instantaneous attributes or geometric we rarely use wavelet but wavelet are very dominating. And so if you combine them with anything else they'll wipe everything, you'll never see your other stuff because they will be dominating. Anyway, the reason is that these attribute types have different data distributions that can be incompatible so they won't give you a good answer. We suggest you use a suitably focused geologic interval that contains whole genetic units, and that's me talking genetic units, with horizon delimiters so that you return coherent statistics.

 

Principal Component Analysis and Interpretation

You don't want to cut across you know a facies band or a geometry that's you know you'll get only get half of it. You want to get the whole thing so you can see the systematic variation. And we used judiciously the 2D color map to produce greatly enhanced images. The other thing to remember is what I said about picking on every line I'll try to guarantee that ok so I'm mainly working with the instantaneous attributes and I first worked with them in 1982 but I only had three then. Now we have 16. And so the attributes for the PCA is put the whole class of those attributes in them. And these attributes are particularly useful for stratigraphic and facies determinations.

 

What the PCA does is dimensionality reduction. It gave me back nine attributes in three different eigenvectors. Three each, it turned out. And there they are, not in any particular order except alphabetic. But this is my base survey. Real part is highlighted because it's almost another copy of your base survey but the base survey is included in this list because it's used to prune out null values. You don't want one of your neurons being classified null. Just a waste of good neuron. Ok so... they are different SEGY volumes, absolutely. Ok you know these are all separate. These can be generated you know outside paradise. We generate some inside paradise.

 

We will eventually be generating them all inside paradise. We are working on that now. Instantaneous phase is good for discontinuity and continuity, both in structure and stratigraphy. The normalized amplitude is also known as the cosine of instantaneous phase. It gives you the sense of peak and troughs value while the relative acoustic impedance gives us our geobodies that you're gonna see, and imaginary part and real part together, give you envelope which is the total energy kind of like amplitude but it's the total energy. All these are instantaneous which means you got a value at every sample and they are normalized by standard deviation, these attributes, so that you get the results that we get.

 

The other three are present but these five, we have a new utility that allows us to tell, for every neuron, what attribute really went in to compose it. The things I'm showing you today had some combination of these five more than anything else. But I have one geobody that I found late in the game where trace envelope is important. Ok, we're going to get to the pictures here pretty soon but I want to give you my conclusions before we get too far. So we have a terminology and I'll show you this again the upper Eagle Ford morrow can be readily distinguished and separated out.

 

The ash beds at the top of the Eagle Ford shale can be distinguished and readily because they're non-reservoir can be mapped and separated from the reservoir. You may know pepper shale by this company, it was called Basal Clay rich facies or basal clay rich shale and it can be mapped and separated from the underlying Buda and the overlying Eagle Ford Shale target. The Eagle Ford itself had two reservoir types basically, an upper low resistivity and a lower high resistivity. Occasionally when these, when a borehole went lateral from one to the other it would be called mixed. You will hear me say hi res and low res and I apologize. But I do not mean resolution in this case, I'm going to be meaning low resistivity upper or high resistivity lower in the reservoir sense.

 

Here we're separating them into at least six non-layer cake zones. The non-layer cake was a real surprise for a lot of people. And it also includes at least two unrecognized and underdeveloped geobodies. At least one of them we know to be a sweet spot. We know that it has the characteristics and was a target but often they just blew it they just missed it. Because our attributes return back from attribute space back perfectly into survey space we never lose where we are in the scheme of things. The company had trouble seeing faults in the Eagle Ford. We do not. We're resolving them in multiple ways. And then I tried some spectral decomp work and we found out that the selected frequencies that I chose to work with as a base survey had its best results in the Austin Chalk and Buda but also uncovered interesting things in the Eagle Ford.

 

Stratigraphic Results

Ok here we go, Stratigraphic results. I borrowed this... one of things you don't get when you're doing mostly a research project and not really explorations you don't get a column, and sidebar this from article that's in press this is Mike Fairbanks at Apache and two guys from the BEG, Steve Repel and Harry Row. You know who... long time, big-time Eagle Ford players, and it was suitable because it's similar enough so that for my study I could lay in the Buda, the base clay shale, the high resistivity unit, the low resistivity unit. Eagle Ford ash, although I'm going to come back to that, the upper Eagle Ford morrow and the Austin Chalk.

 

And I put the well top colors and this won't make too much difference to you, but I'm showing another feature here. This ash bed still flows but I'm showing a red and gold distinction here because we have an up dip red zone that I’m gonna talk about that down laps and then a gold zone that overlays on on laps. And here is the red zone. This I've done just recently and I don't know how well you can see the distinctions. But I'm going to show you the distinctions between 2 and 2 of these colors with a little animation. Ok, first of all, the lower, deep rusty red zones, lap out down dip and they also reappear over here but they are not present in the middle right there.

 

They are the high resistivity lower part of the redzone. The redzone is only once part of our overall stratigraphy. It's the lowest part on top of the basal clay shale. These next two colors, that was 61 and 62, 63 and 64 neurons are painted on... I have to say this progradational. Aggradational on the top of this up dip are some 63 and 64 or and down here these are what I had identified and calibrated in wells, the Eagle Ford ash. But they're the same if they're the same neuron; captured by the same neurons they are the same facies. There's a similarity there.

 

There is also a change in the sixty-two in the type of attributes that bring it out, that tell you there's sort of an unconformity between this lower and this upper. But this is just fascinating to me. What I don't know, what this helped me understand is all this is clay stone present; understand why I had evidence if you will that this upper part was kind of commensurate with that. Although I don't think it's very much thickness. It's a veneer relatively and I do not know whether this 63 and 64 are coeval in age. I believe that my ash, like I said, it fills lows, and it could be this or it could be this stuff carved into it except it's so flat. I really think it is part of this low resistivity zone. It is low resistivity; I believe that it is part of that low resistivity zone.

 

So, like I said, this is something I only got a chance to play with lately. Ok when I started, my first steps, when I started I didn't have my picking all done so I did a SOM just with time delimiters, 1.5 to 3.2 seconds because that was the extent of the height of my Eagle Ford. And I use the default color bar. This is what we call the eight by eight topology.

 

This is the default color bar and you see while the Eagle Ford, well I can tell where the Eagle Ford is. It's not very distinctive. So I went in and I punched two different color changes the purple and the yellow and it reset some nearby colors compared to these reds and suddenly the Eagle Ford is standing out like a sore thumb looking quite unique as I kind of expected that it would.

 

Ok? And that's just from tweaking the colors from our 2D color map. But how unique is it?

Well, was I delighted to see when I turned off all the colors except those that were contained in the Eagle Ford and I also included 1 of the neurons that was an upper Eagle Ford color just for contrast because its green. And this is the sequence of colors in that stack. Low and behold there is a section cut into the Georgetown that has that same facies configuration and there's a piece down here we didn't even know, we couldn't calibrate to, but I suspect is the Pearsall. So handy dandy figure number 2 borrowed from this group. But this even goes back even further to BEG guys.

 

And I added this little guy because the Pearsall has characteristics apparently, just like the Eagle Ford. When I use this statistical band, ok, and this broader statistical band because we're gonna use an arrow over one real shortly, and for the rest of the talk. Those scowers are right here into the Georgetown and you will see some caricaturing that I've done on the Eagle Ford and the upper Eagle Ford morrow are based on clay shale appears to be sort of blanket like through the area. We have the red zone that covers the entire Eagle Ford reservoir area up dip and down dip, it laps down. The gold, I call this gold overlay zone, and there's a gold zone out here and I have a Geo body that occurs in there.

 

Facies Architecture

All high resistivity reservoirs in this zone. The low resistivity part of the gold is above that. Then there's non-conformity in the upper Eagle Ford morrow. Didn't do a lot with the morrow but you're going to see plenty of figures. Ok, now we're going to talk about facies architecture. I'm going to repeat this slide twice because you find as you hear, boy I'm in big trouble, this is good stuff but it takes a little bit to talk about it.

 

The client provided me this piece of section. Conventional seismic, conventional color bar, the Austin Chalk down to the Buda two and a half three depends on how you count your cycles when you've got stuff like this. But not many peak trough cycles, and not much resolution. This section reminded me in the 80's before I started working with instantaneous phase, people used to boost the seismic to get it to look continuous.

 

This looks real continuous but there's nothing resolved there. Nothing like you're going to learn can happen. They told me that the Eagle Ford objective was about 14 mils thick. So that's seven samples here, 108 feet. That's possibly at tuning. Tuning could be a hundred feet it could be a hundred and fifty feet but, anyway so that the entire Eagle Ford group is only 28 mils, interestingly 14 samples, and for by and large it's just kind of crazy but the Eagle Ford and the basal clay are roughly 14 different color bands so I just don't know what to make of it. And they talked about influence. One of the things that we don't have to worry about in our normalized standard deviation world, is tuning and absorption. We can find it with attributes; we can find it with attenuation. But they are not big issues because of the way this data is handled. And you'll see that. You'll see very discreet results. Ok so I think that's all I wanted to do.

 

Notice that Seitel gave us permission to use this data. I haven't said that yet. Okay now we're going to be on the right side. Here's my eight by eight and I used maybe you're familiar with map shade dark, is a color bar from the Kingdom interpretation package that we can migrate into Paradise. And you can do the opposite, we have color bars in Paradise and we can move them out into the software.

 

Map shade dark just happens to work well with an 8x8 and so I've taken it and you'll see some examples of where I've taken it out and used it in SMT so, for example, but at any rate, What you need to know, and you also need to watch this whenever I make this note it's either going to say seismic thickness is a hundred mils or a hundred and ten mils. Because that's what I chose for this new statistical base for focusing in. And with this focusing we got all this detail. This surface behind here is the Buda. It is a horizon in Paradise; there are a couple of color bands beneath this that are the Buda.

 

Self-organizing Maps and the color bar

This is the basal clay shale and with this color bar and this SOM, the basal clay shell is gray like this everywhere and all the wells that I calibrated. Ok, it's just a single gray band. Really distinctive, mostly non-reservoir, so it can be pulled out of the picture. And one of the things we're developing is the ability to map on the neuron color interfaces, which is going to be really powerful. This is that red zone lapping out with the black so you know all the red zone is up dip, I don't know that the example that I'm going to show you... Let me look... it's not that I think, I know. Oh, no, everything's thin. This is the whole thing. The whole Eagle Ford group is only 260 feet. So yes sir, yes ma'am. What? They're thick.

 

They have some thickness. Yeah [different speaker] Tricia has been kind enough to say if you don't understand the principles to let her know and she will backtrack and explain them but please keep regular questions till the end of the talk. [Tricia]

 

I can pull those out Walter. I have examples, I can pull them out. I, you know, I have a hard enough time remembering the neuron numbers sometimes. It's detail. Ok so what do I do with this geobody that I found. One of the things we can do is we can focus the classification on a horizon or we can take that horizon and move it down through a body and that's what I did here. So I took the top Eagle Ford and I moved it 12 milliseconds below the top, its normal position, to intersect that geo body. And now this is where I need to tell you about color bars again because I found this thing with map shade dark. It was rust colored. I had a devil of a time extracting it with the rust. For some reason I think there were too many other things involved that also had a rust color or something. So I went to this color bar where it resolved with 3 neurons, as I'll show you in this red body.

 

So, don't be alarmed, this is still the rust geobody, it's still called the rust geobody by me but it is red. But what I hadn't expected is this flange or backdrop that is 2 other colors. And, think of it like this, I put this little silly thing in here because I'm cutting across where the flanges are, where this whole fairway thing, as I call it, is wide and then I'm going down four more milliseconds and this thing is narrow. To give you an idea of the three dimensionality the thickness variation and also with these wells that they didn't quite get there. See that bore hole is above this red and so on.

 

They kind of missed this and they knew it and so first thing I wanted was a flat map so they could see where they should go looking for it, because it is a high resistivity. So this is just the extraction of that geobody separately and those are the three new neurons that find it. Mostly the sixty read and then the oranges and the other oranges are down, mostly down dip. Those variations are real there's an expander, a slight expander down there those are real features. Ok here's my well calibration. This is from our own software. This is a log track display. This is the feature... oh and by the way, I have been lovingly referring to these as bacon strips. Can you tell why? Okay I just couldn't help it. I go bacon strips! Anyway, they, again we're at 100 mils which says the Buda is below my service which is the Buda pick. You see how smooth that is?

 

It's smooth because I picked it on every line. There is the basal clay shale, there's the rust zone, and this is the deviated borehole. They gave us only three ways to locate the deviated boreholes and the screen is the first perf, purple and mid perf and this rust one is the last perf. So that's how we found these in 3D space from them. This is the top of the Eagle Ford. That's the ash bed underneath it. The upper Eagle Ford is largely magenta and this purplish color. And again the greyish enveloping the rust is high resistivity and the plain gold and the ash beds, which, remember, were 2 neurons, 63 and 64 are low resistivity. Here is the Buda.

 

The pepper shell or basal clay rich shale, very distinctive. This was called the target zone on their... the little go by logs that they gave me. But they missed it entirely. And you won't remember but if these are dark green, these EUR numbers that means the volumes come from high resistivity pay when they're light green, low resistivity pay. This is the low resistivity pay and this is where they recovered all the volumes. So they missed this beauty, which is the thing that's brittle and would frack, and you know. It was their target but they missed it.

 

And these are normally compacted chills. There is nothing too fancy but this was my very best example and so I spent a little time putting that together. All right, one more geobody cropped up. This unfortunately is not calibrated but we have a vertical well nearby it and this geobody I call the yellow-brown geobody and this is their cluster of horizontal wells either went above it into the morrow or below it, into pay that is high resistivity. Now you remember the red that was way out board, this is that red. It is 61 and 62, not the ash beds and they're low down there. And then there's the brown and yellow.

 

Now because there's the same yellow in the Austin Chalk I extracted this geobody using only the brown because the stack everywhere on top. This is a time slice just giving, you know, I wanted to see because these looked scooped, like they had scoop bases, I went out to see where are these nestled. And in a time slice, you can see the brown yellow. This is up section, down section. The gray stripe is the Buda top so now I have that in but I don't have the whole of the Austin Chalk here because this is just the top of the Eagle Ford morrow mapped top.

 

And, at any rate, there is the Eagle Ford between those two, and you can see that it's kind of nestled in there. And it was higher up than the rust zone. And you can see that because the red is coming down and the rust is here. Now, we also have another feature in our software that allows us to put a probability filter on top of anything we're looking at. And it comes out white. This is a 10% probability that I have it set to but we can do whatever we want. But when you choose a small probability, the sense is that that area is anomalous for whatever reason. And in this case we believe it's the hydrocarbons.

 

The Buda lights up some places, you can see that on this example back here on the poster. That's Buda stacked below Eagle Ford all brightened up, done by another worker. And anyway but what's neat about this overlay is you can see fault traces that you might not have appreciated. And this break also resolves as a fault trace looking thing. So there's some subtleties in almost every view we make through the volumes. Ok, now I'm going to summarize this bit before we go on to the next section and this is from the research. Through the magic of Disney, we have a gentleman who helped me pull out my rust geobody and my brown geobody so that you can see how they are situated.

 

Remember the Brown is on top. It was put on top here and there literally is... it literally seems to wrap around the rust zone down there. But I had already classified the red zone, which later I broke into two parts because there are two segments over here and three segments over here and they appear to be, given parameters, appear to be slightly different. There is a zone I call gold overlay where the gold overlays the down lapping red. There's a place, this was sort of named later, Gold Gap because there is a time when the Red by this time has lapped out and the gold, it can be narrow or it could be thick, but these wells are all in Gold Gap actually. Only one segment came out rust geobody one. A segment here is a group of the wells.

 

So these numbers of wells add up to that 274 for which I have the tabular data, petro physical and geological parameters that were put into a SOM and sectored out these wells in the same way I had done visually on the color map. And then some, because there was a group... segment 12, that had peace in the red zone and it had peace in the gold overlay zone. So I separated the wells and sure enough their parameters were more consistent where I moved them. So this is how you can have a 4x4 topology, 16 neural clusters but you can get 17 segments out of it because you can observe it. So this is close, but this was done with just 4x4 on this tabular data. And that's about all I can tell you about it but this other number is the number of calibration wells that are in that zone. So I don't have a calibration well for every zone but when I'm depicting them you can see the neighboring zones well enough.

 

Resolution of faults in the Eagle Ford

Okay? Now we are going to talk briefly about resolution of faults and how we doing? Everybody awake? Ok, this part is short. This particular client had used similarity as a singular attribute to try and look at faults and they provided me a really lovely map and so forth but it was on papers, so I went in and across this well which, as it's vertical here and its deviated bore hole going down that line basically, that in line, you can't see an image going across that and sure enough, they didn't deflect a single picked horizon through there because they can see the faults coming. I picked this fault from probably instantaneous phase data. And you can see that it goes forever. But these all look like the cut off right before you get to the Upper Eagle Ford Morrow, so at the base of the Austin Chalk.

 

And so their feeling was that the faults couldn’t be observed in the Eagle Ford. But you can see the Austin Chalk is deviated there, not so much in some of the other positions. And that's the vertical pilot well with gamma ray and a resistivity curve. Ok this is a geometric SOM result or result set from having run geometric attributes and the PCA here suggested a reduction of the 25 input geometric attributes down to 10 from just the first two eigenvectors. 5 from one, 5 from the other. The first 5 were all different varieties of similarity. The second 5 were 5 variations of curvature. So, what's important in this seismic dataset for this kind of geometric attribute are those types and no others. You know, unless I went well beyond the sixty percent and just gathered things. But that wasn't necessary because this well had six faults in it. And they had never seen any faults.

 

This is geometric classification on the Eagle Ford top that's been pushed down into the high res Eagle Ford pay which is probably a little more brittle but we could bring this up and down we can see faults in detail everywhere. This well had two faults in it and I actually did a better resolution of that using spectral decomp which we will see in a little bit. So you've seen that map. So this is the down-dip area squeezed just a little bit.

 

This is using the color bar that I call the rust zone extraction color bar and I tweaked one of the colors to be white in the core of this just so you could see it better and I turned off the enveloping 60 reds. So 58 was turned white and 60 reds were turned off although you can't tell it. They were turned off in this view. So were a bunch of other colors as you can see but this is just another way to look at that zone and then squeezing it a little bit I could accentuate sort of simple looking normal faults but there are a number of compressional faults in here. And I probably could've picked faults in here for days. I mean every time I think why didn't I pick this red white place you know or something else but suffice it to say we can see faults in the Eagle Ford with just... this is the instantaneous SOM, the original, not geometric, not those things that are supposed to... But when you when you're working with instantaneous measures you're seeing inherently high frequency and you're seeing continuity and discontinuity. All right this is just a close-up of another really spiffy fault and here again, bacon strips and the matt shade dark. This is the red, the 61 62 down lapping and breaking like scooting away until it finally down laps. And it down lapses this is a sort of gold gap, this is gold overlay and then we're down in the rust zone, high resistivity where it's grey and then just a lil bit of low resistivity stuff there and again the ash beds in that down dip position as we saw them early on. We're right in the middle of the dataset.

 

This is near some of the best wells in the area. Most of these are like 250,000 to 300,000 barrels of oil equivalent. This one is a thousand barrels of oil equivalent. Some of the wells that were in this vicinity. Again we're on the Buda horizons, we're seeing the basal clay shale. A hundred mils there. So I think that's the last one there. Now we're going to talk about spectral decomp. This is some of the coolest stuff that I don't purport to understand completely.

 

Spectral Decomposition

Yet spectral decomposition is also a useful tool for seeing sub seismic they say. This is the conventional wisdom. I think it's true but I wonder about anything that's normally based on amplitude. Because amplitude is inherently not discreet. Amplitude can change with so many different kinds of stacking. You can have silt on silt and it will make amplitude in some parts of the Gulf of Mexico. But it does give us an opportunity to unravel the signal into constituent frequencies and here with our multiple normalized instantaneous attributes without tuning problems and other things, it's really the envelope components that help quantify thickness variability and the phase components that help us detect the lateral. This is my shorthand.

 

Stratton structural discontinuities. We are resolving faults stradel variations in the Buda and Eagle Ford and unconformity and carsts in the Austin Chalk. So let's look let's look at the examples. And this is part of the testing that I went through 12 to 15 hertz and found sub bands for the linear sub bands to be giving me the response I was looking for. Detail in stratigraphy, not amplitude. At 24 hertz for the linear and for the active sub bands, 26.5 hertz. So I took each of these sub bands and made them a base survey, re-ran all of the instantaneous attributes again and got some of these results. Ok this is my illustration of why you wouldn't want to leave your data un-picked on every line. Because you're going to get a footprint.

 

This Austin Chalk was given to me; it was picked on every 5th line, which is plenty to get a good result. These are, again, what we call ghosted horizons of the classification on the Austin Chalk horizon. This one at the linear trace sub band, which is all I proceeded with, at 24.7 hertz and 26.5 hertz. Pretty close and here we see this fault with an orientation unlike any others. It's very straight and narrow. But I think it's real. There are other things I wanted to make faults but with this kind of quality of data it's a little more difficult and I believed I sort of saw it but maybe not as well on this sub band. So, this is typical variability. When you change a statistic or change any part of it, you're going to get changes. Ok this is where it gets more interesting. I'm down to my last handful of examples here but suffice it to say for this one and I've got the extraction color bar on again.

 

These are cross lines instead of in lines. We're in this vicinity and we're looking at neighboring wells, 3 and 4. And in both cases the bottom of the Eagle Ford is identical. And then I turn off a couple of colors in here and everything goes away at well three and hardly anything goes away at well 4. These colors are unique to well 4's area. So you see how subtle variation can occur very quickly. At the Buda, in this color bar this purple is now the basal clay shale, and that's both views, and this orange and red colors in both views but it looks very different in these well locations.

 

This looks scooped and with the judicious turning off of these two colors you see that they're isolated and scooping looking when this just looks like railroad tracks over here at this well. Don't know what it means but you can't interpret what you can't see. So you need to see it, calibrate it, if it's important to you. This was less important to me but learning about spectral decomp was very important to me. I had never worked with it before. Ok this is now a time slice, and by the way that was previous one was just the 24.7 hertz, and this is just the instantaneous SOM on spectral decomp 24.7 base survey, and at this depth, 2.5 seconds, we're looking at the up section is down, the down section is up, and we're seeing a couple of things. First of all, remember the purple-orange-red we just saw in the 3 and 4 well.

 

Well, that... you know, one of them laps out and they intertwine and then they end up being yellow and an orange color. An orange-red color. Something, none of the above on the other side. So in the frequency domain, something is changing there that didn't appear to vary using the SOM color bar before on the basal clay shale and the Buda. Ok? It has some meaning. This purple for the basal clay shale becomes the lowest most olive zone over here.

 

The other thing is this very cool compressional fault and you can see the little dimple there? You can see it in the nearby cross line as well. See the little dimple? And that's that same fault. It looks a lot like the one in the bacon strip that I showed you, the major fault. So the Eagle Ford is very... you know its thickness doesn't vary too much. The upper Eagle Ford is... you know I love that the Marl is blue colors so it helps me think about carbonate. The ash bed is not showing up here like it did before.

 

Again, the ash bed in frequency domain may not be that different. But it should be right up there. And I'm not saying that the green couldn't be it. I just haven't typed these to see. I was too busy being fascinated by the apparent carsts and faults and other things. Oh you think? You want to see it? You straight man, you. Oh my! You mean there? Isn't this great? Here I just turned off mainly, there are some variants in the total color combination but by and large this part of the upper Eagle Ford is nearly identical for well six and eight. Not much changed.

 

They're pretty close together. But I turn them off here and didn't turn them off there so that you could see everything and you could see the unconformity of the Austin Chalk going into the... and this is still 24.7. But now we're down here where all of this Buda is that yellow-orange that it got to be as you went up to the northeast. And again these are the deviated wells and so forth associated with the vertical ones when we have that information. Then here's another one. Now we're on the 26.5 so we've switched to a different SOM and as a consequence the colors look a little different.

 

The upper Eagle Ford is now in part of these colors and there's a color that I turned off just to accentuate, in the Eagle Ford itself, just to accentuate this unconformity and the big facies change that happened just inside whatever this feature is. I haven't extracted these so I don't know what they look like in 3D. I mean this is such an embarrassment of riches I could spend my entire life probably working on these things. But part of the reason for presenting is finding out what's been critical and important to you guys who really know the Eagle Ford. My best hope is that you're going to say, God, I've wondered about that and maybe that's what's happening.

 

Ok, so that's that. I think this last one is the one I used to call Odd Feature. Clearly it carst, but here's the conventional in line segment. Here's the interval between the top of the Austin Chalk in the Buda. This is under 110 mils so you're seeing the whole smear. The Buda is in these blue bands. I've turned off some... where did I put them... ok the Eagle Ford shale numbers that are cut off, so 3 of them are here and then the upper Eagle Ford a couple of things have been turned off there. Again you can see the disconformity. But look at that sucker. This dark brown is both outside and inside.

 

I presume that could only be a carst. That isn't a conventional channel or anything. It's got to be a carst of some sort. But it fascinated me and I just kept bumping into these things and so I always capture when I bump. And so that's my last slide. I'm just going to recoup the conclusions. I think you would have to agree that we can see the upper Eagle Ford Marl distinctively. We can see the ash beds distinctly; we can see the basal clay rich. That means we can extract away the non-reservoir from the reservoir.

 

We clearly have a calibrated hold on low resistivity, high resistivity. Mixed happens when you go laterally from high to low. It happens more often... I've never seen so much porpoising as was done in some of these wells. Sometimes pretty much missing everything. But anyway, we can find these zones. We can see these geobodies, we can resolve faults, and we can see things with the spectral decomp. What I like to tell people is if you want a prognosis or calibrate your wells better, or if you want to geosteer better, which was you know that one of the guys from the company that I worked through, when I reported this last time he said... because they needed help, they were geosteering on dip. Because we know exactly where these things are now, you clearly want to be in depth instead of time and we can work in depth we can work on 2D volume, 2D surveys and the like, but if you want to geosteer better, we can get you there.

 

If you know that this is what you want, this high resistivity thing like that example I showed, you won't miss it like they did. Ok and if you want to evaluate your lease hold, this was the deal with the first the data set I worked for a company that bought the software. Do you want to evaluate your leasehold with more confidence? You might want to think about using this tool. Thank you and that was it. And as a token of our appreciation we are giving you the whole state of Texas in the rattlesnake limestone, which is from a secret quarry here in the state of Texas. Thank you.

 

 


Member Login
Welcome, (First Name)!

Forgot? Show
Log In
Enter Member Area
My Profile Not a member? Sign up. Log Out