Jeffrey Yarus is well-known throughout the petroleum industry as a leader in training and mentoring. He is the co-founder of Quantitative Geosciences LLP, a consulting firm that specializes in data analysis and geostatistics for the petroleum, environmental and mining industries. He has worked for Amoco, Marathon, GeoMath (subsidiary of Beicip-Franlab), Roxar, Inc. and Knowledge-Reservoir Inc. He also served as a professor at the Colorado School of Mines.
Jeffrey has served as AAPG’s Chair of Geological Computing Committee, Publications Chairman, and is currently the Chairman of the Reservoir Development Committee. In 1995 he co-edited the AAPG volume on ‘Stochastic Modeling and Geostatistics’ and in 2000 he co-edited a second AAPG volume on ‘Geologic Information Systems’.
Jeffrey was very enthusiastic about this interview. Though suffering from jet-lag after his long flight from Australia, he was very responsive to the questions we posed and we enjoyed the lively discussion.
Jeffrey let me start by asking you about your educational qualifications and your work experience.
I began my college career at the College of Wooster in Ohio, outside of Cleveland, where I was born. I majored in geology; that was from 1969 to 1973. During that time I also spent one year at the University of Durham in England, again studying geology. From there I went to Michigan State, where I began my graduate studies under the auspices of Dr. Robert Ehrlich, who first got me involved with statistics and geology. I followed Ehrlich to the University of South Carolina, finished my Masters degree and then finished my Ph.D.
Please tell us about your work experience also.
I began my career with Amoco Production Company in 1977 and worked for Amoco more or less until 1980 or 81, I can’t remember the exact dates. I then took advantage of the end of the oil boom, moved to Denver and ultimately started a small company called Diversified Resources. I was with Diversified Resources for a number of years until everything fell apart in the oil business and then joined Marathon Oil Company at their Petroleum Technology Centre in 1988 as a Senior Scientist. From Marathon Oil Company I spent a year with GeoGraphix and then moved to Houston and spent a few years with Beicip-Franlab and Geomath. I left Geomath in 1998 and joined Roxar (then called Smedvig) as a Vice President. From Roxar, I joined a consulting group, today called Knowledge Reservoir Inc., where I was a partner. Finally, in 2001 I started Quantitative Geosciences with Richard Chambers, my partner. That’s the short summary, a typical string of jobs constituting a geological career!
You quit Amoco to become a consultant?
Yes, until I joined Marathon Oil Company.
How was that move?
It was an interesting move. It was the end of the oil boom that we had back in the early 80’s and it was an exciting time to be an entrepreneur; drilling a number of oil wells and putting together drilling programs. It was very exciting.
Now obviously you had an interest in teaching as you joined CSM and the University of Houston. So why did you not continue teaching as such?
I have always enjoyed teaching. I think that being involved in the University on a professional basis keeps your skills honed and I like being in front of students who I find to be quite challenging. The only reason why I haven’t done it on a full time basis is the low salary. However, I enjoy watching the light bulbs go on, as you explain things to students. I think that my practical experience is really what makes it more interesting to the students., So, I think the best way to be a teacher and involved with the University is to do it while you are a practising professional.
So how do you like co-managing Quantitative Geosciences?
Fantastic! Quantitative Geosciences – it’s an interesting story. My partner Richard and I actually met at a conference. Do you remember the company Zycor?
Yes, the company associated with contour mapping packages.
That’s right, we met at a conference years ago that Jim Downing had put together and we were each giving a paper on geostatistics – the only papers on geostatistics, I might add! At the same time I was involved heavily with the AAPG Publications Committee and they had asked if I would be interested in putting together a volume on this new thing called “geostatistical modeling”. You may recall that in 1986 it hit the oil industry after a small SPE conference that had taken place in Grindenwald, Switzerland, and began to blossom. By the early 90’s many of the companies were getting involved with the Stanford Centre of Reservoir Forecasting which was run by Andre Journel at the time. I was asked by the AAPG to see if I could put together a volume on stochastic modeling and geostatistics. I wanted to do that very much because I was in charge of bringing this technology to Marathon Oil Company and saw this as an opportunity to get to know the players. I needed a partner to share the workload and Richard was the only other guy I met who knew anything about it. He and I presented our papers on Kriging at the Zycor meeting and that single event essentially started our professional association. We actually went to Graduate School together at Michigan State, but really didn’t know each other, mainly because he was a few years ahead of me. In any event, after working on the book together, we know that one day we would start a company, and we did.
What is your company involved with?
We started Quantitative Geosciences, the name comes from the department that I was in at Marathon Oil Company which eventually disbanded, we thought that we would be primarily involved in consulting on various petroleum issues that involve statistics and numerical applications. Geostatistics is only one part of what we do although it seems to be what we are better known for. Both Richard and I have pretty significant backgrounds in applied statistics. It turned out that more people were demanding basic generic courses in not only statistics but specifically geostatistical applications. As we weren’t really software specific, and we understood the basic principles behind most of the commercial earth modeling software packages, we began a series of lectures and seminars on the principles of applied geostatistics. We find that that’s now about half our business.
What do you consider your most important contribution to your profession?
That’s a tough one. We are pretty small fish in this field of mathematical geology. There aren’t a whole lot of geologists interested in math anyway so we are a bit of an odd breed, but I suspect the most significant contribution we have is that we have learned how to explain these concepts to people that have even a harder time understanding them than we do. So if we’ve had any role at all, particularly in the proliferation of geostatistics and statistics in the geosciences, it’s that we have the ability to explain the topic simply in non-mathematical ways. We have approached it from the practitioner’s point of view as opposed to a theoretical one and I think our industry colleagues appreciate this.
Let me ask you a basic question. What is geostatistics and would you term it as methodology or do you think it is a discipline or what?
Well, first of all geostatistics is a branch of classical statistics. The classical statistician would call it “spatial statistics.” It is a bit of a thorn in their side in the sense it was brought to the forefront by a geological engineer as opposed to a pure statistician. It’s a discipline that evolved in an applied area that has made its way into the domain of classic statistics. The idea is that every point, every piece of data that we have has a geographic location that is not independent of data at other nearby locations. Classical statistics assumes data independence. That is, one sample is completely independent of another. That can’t be so, of course, in our business because it matters whether you drill an offset location north, south, east, or west, so our data are completely dependent upon a geographic coordinate system. Thus, our data samples are not truly independent of one another. Geostatistics introduces a metric that accounts for spatial dependencies and is thus a modification of basic statistical principles.
What drew you to geostatistics? I know you just mentioned someone was responsible for drawing your interests towards this, but to take it up as a profession is an all together different decision.
If you are asking who was responsible, – Robert Ehrlich, my major professor, had a definite propensity towards statistics. He was the first one who introduced me to statistics and geostatistics. Bob insisted that if I was going to study with him, then I had to have a secondary discipline and he mentioned things like chemistry, physics and math, which of course scared the heck out of me. Then he mentioned statistics and I said – well isn’t that math? – and he said “no, no, no, it’s not, it’s better than math.” Bob introduced me to a book by John Davis called “Statistics and Data Analysis in Geology”. He also introduced me to a book written by Michel David, one of Metheron’s Students out of France that caught my interest so I started looking at funny things called kriging and variograms, and the like. From there I actually got to work with John Davis from the Kansas Geological Survey when I was at Marathon Oil Company. John was brought in to help do training courses and I got to be his assistant. Then I became involved with the Stanford Center of Reservoir Forecasting; I was very pleased and honored to meet Andre Journel who I credit a lot with peaking my interest in this particular topic.
Could you tell us a little bit about the historical development of geostatistics and how it found its application in the oil and gas industry?
That’s a well known story. I am not sure if I’m necessarily the best one to tell it—
There are some people who may not even know anything about it, so that will be good.
There is a guy still around and still very active intellectually. He’s been a regular participant at some of the specialized geostatistical meetings that we have. His name is Danie Krige. Dr. Krige, of course, is credited with “Kriging” although he didn’t give it that name. The technique now referred to as “Kriging” was dubbed by George Matheron a number of years afterwards, but Danie Krige and a colleague named Herbert Sichel, both from South Africa, were involved in the mining business, principally uranium and gold, I believe. They realized that standard regression techniques weren’t really working well for estimating ore values because geological data are not classically “independent” of one another. Independence of samples is a prerequisite for classical regression. They dismantled the regression equation and modified it by introducing information by way of matrix algebra that informed the interpolation system about surrounding data samples. That really was the beginnings of what George Matheron later coined, The Theory of Regionalized Variables. Matheron, a professor at the Paris School of Mines, recognized the importance of what Krige did and went on to form the Centre of Geostatistics. Matheron is credited with developing many of the original theoretical underpinnings of what is now known as “geostatistics.” His academic progeny have gone on to establish various geostatistical centres throughout the world.
As the general impression goes, geostatistics is highly mathematics oriented and that drives people away. Having said that, it is also true that we see some very interesting and accurate case studies being published; so what I was going to ask you is why do you think a geostatistical approach would be more accurate, first, and what can be done to prevent or help the geophysicists to overcome their reluctance and encourage them to use geostatistics?
That’s one question I find very intriguing. First of all, it is mathematical and, I would like to remind our colleagues, first and foremost we are scientists. We tend to forget that, and today in particular we find ourselves becoming increasingly less scientific and more sophisticated at pushing buttons. Behind all those computer buttons are mathematical equations! You know it wasn’t that long ago, Satinder, when you and I were beginning our careers, there weren’t buttons to push, and we had to think about what we did! So, we are scientists in a discipline that deals with a paucity of data, and data is usually of poor quality. By its very nature, petroleum exploration and reservoir characterization is first and foremost a statistical problem. We are continually asked to make huge conclusions from the small amount of data, so we need to be diligent and to understand that the technology is not to be taken lightly. While many of us are frightened by math – and this may well be true particularly of geologists, it doesn’t excuse us from trying to wrestle with the concepts to be sure they are grounded in reality. The truth is that, at the end of the day, math is a bit of a hoax in the sense of the way it is taught. It doesn’t have to be so difficult. We are heavily reliant upon it for numerous reasons not the least of which has to do with data handling and manipulation in data bases that are now so accessible to us. In order to process this plethora of information, we need to have mathematical techniques that can assist us in data mining and data analysis. But our conclusions must make sense in a geoscientific way. To quote my good friend and lifelong advisor, Dr. Ehrlich, “math is too important to leave to the mathematicians.” Someone is entrusting us to make good decisions so we better understand what’s behind those buttons we are pushing. If it were your 300 million dollars that were being spent, you would care, right?
I must add another point to this and that is the concept of the “80% solution.” It always comes up when we start talking about software – we want the 80% solution. And I tell my participants in classes that I teach that we should drop to our knees every day and thank God we are not in the medical business because we might be removing someone’s kidney when we were supposed to be doing an appendectomy! If you went to your doctor for surgery and he told you he was going to give you the 80% solution, how would you feel? We need to be better than that! I am compelled to remind us again that we are scientists.
You travel a lot and you conduct a lot of projects that entail geostatistical applications. How much do you think is the application of geostatistics to hydrocarbon exploration and do you think it is adequate or do we need to do more?
Well I think we see a lot of different types of projects in our business too. We have expanded into petroleum, mining and environmental as well as to some unexpected disciplines like telecommunications and climatology. With respect to petroleum reservoir studies, we see the use of geostatistics increasing. Does there need to be more of it? Well, as I said earlier, in answer to another question, geostatistics is a discipline that includes a number of methodologies within it. I see that there are methodologies within the geostatistical and classical statistics domain that solve a certain set of questions really well, but they don’t solve every question. Your best guide is your education and your experience. It is not some panacea; it is a set of tools that we evoke when we need to solve problems that they are capable of assisting with. I look at it in a very practical sense.
Oliviere Dubrule conducted an SEG course in geostatistics a couple of years ago, and he works for a French company. Your company also has some sort of alliance with Geovariances, another French company. So I was curious to know why geostatistical research was focused in France and not other places or do we have some groups in other places of the world as well?
Of course, it really started in South Africa. To say that it started in France would be to some extent inaccurate. However, the theoretical underpinnings of the technology were embellished in France, the Paris School of Mines, Centre of Geostatistics. Inevitably, you find a lot of people who went through the French School system that are perhaps more familiar with the technology because it took root there very early on. But certainly, there are other centres that have developed. I mentioned a couple of them earlier on; one certainly is Stanford, there is one here in Canada, the Centre of Computational Geostatistics, through Clayton Deutsch who was a student of Andre Journel. There is the Norwegian Computing Centre, which of course is heavily involved in the use of geostatistical principles. We find a lot of smaller Universities and places beginning to use or to teach geostatistics – more than just a course but actually use it for research and help in developing those research projects. The University of Texas, The University Oklahoma, Colorado School of Mines, The University of Houston, The University of Tulsa, Heriot Watt University, Imperial College, are just a few of the Universities that come to mind immediately. So I think we see it growing and I think that it will continue to grow because it has such wide application, well beyond the geosciences.
Could you tell us about some of the research directions in the field of geostatistics?
Yes, I think there are lots of problems for us to still solve in the field of geostatistics. As I said, it is not a panacea. One underlying principle that often causes trouble in the reservoir business is the geostatistical requirement referred to as stationarity; that the average value of the property being measured, along with its variance, doesn’t change regularly in some way as we move across our area of investigation – particularly within the neighborhood of points that we are dealing with. In other words, we don’t like to see “trends.” I don’t like to use that word very much because people confuse its meaning. The meaning in a geostatistical sense, for example, would be a surface where the elevation is consistently rising, or a body of sand that is consistently thickening in some direction; those are the kinds of things that I am talking about. Of course, this situation is extremely common in reservoirs. We have excellent ways of overcoming non-stationarity but they are not necessarily simple to implement inside of software. One area of improvement would be to find easier ways of identifying and handling horizontal and vertical trends. This is kind of a large, perhaps philosophical piece of work that needs to be addressed, but crucial to the software user.
On a more immediate basis, we still have issues with modeling fractures with geostatistical methods and we still have issues with modeling facies. Fracture modeling is nontrivial and not available in a number of the commercial packages and needs to be. Some have implemented fracture modeling methods, but a great deal of work still needs to be done particularly with respect to modeling carbonates.
We do have some basic facies modeling techniques that have been around for a while, and there are two new algorithms in the process of being commercialized as we speak. One algorithm out of Stanford that is based on what we call a “multiple point statistics” algorithm which does away with direct variogram construction (a two-point statistic). Users often have a hard time with variograms. In addition, this algorithm takes advantage of selective anisotropies (i.e. continuity directions that are say, north, but not south) which is rather unique as far as I can tell. Another algorithm from the Centre of Geostatistics is called the Plurigaussian technique, which also allows for complex definitions for facies boundary conditions, including the ability to embed sub-facies within others, or to have cross cutting facies. Both of these are in some stage of commercial release in some software packages. They allow users more flexibility, and this is good.
Another direction of geostatistical modeling that shows great promise is the use of geostatistics in process-based modeling. Building depositional systems from the ground up as if they were taking place dynamically using time as one of the dimensions is already working at the university research level in the Paris School of Mines. Finally, another important area to mention is the idea of integrating production data with our fine and up-scaled resolution reservoir models. There is work being done through the Institute of French Technology and the Centre for Computational Geostatistics. Some of this is commercially available, but there needs to be more available to users so high resolution and their up-scaled equivalents will guarantee history matching. There is much more, but those are a few of the immediate areas of research that come to mind.
How much do you think the final results depend on the choice of the variogram that you make?
Yes, that is an interesting question too. The variogram, first of all, is extremely robust. Small changes in the variogram model do not result in large changes in the static model which my partner Richard and I now like to call high resolution geological model. The variogram should be ultimately completely intuitive. It should be based on the geology. If you were to tell me, Satinder, that we had marine bars that were 200 meters long and 50 meters wide, and oriented North 27 degrees east, that is the basis of a variogram model. So, although it’s the mathematical rendition of such a conceptual piece of information, it should be intuitive. If it isn’t intuitive, then I would suspect your variogram model probably can’t be trusted. Does it make a difference in the choice of variogram models? Well, there are certain things for sure that will effect your results, but if you think that you are going to create large changes in your volumetrics by extending the range of your variogram a little, you may be sorely disappointed. I think changing the structural or stratigraphic model is going to be a whole lot more significant than changing the range on your variogram.
Besides the quality of the seismic data, geostatistical analysis also depends on the boundaries that you set, hard or soft or whatever. That is my impression. Is that right?
Well I think there are a couple of concepts that are important here. One is the idea of hard data and soft data which is always very relative. For example, we might speak of seismic data as being soft data and well data as being hard data. In that instance, and again, it’s a relative sense because core data could be your hard data, log data could be your soft data. In either case, the combination of using two variables instead of one requires the use of a multivariate geostatistical technique such as colocated cokriging or colocated cosimulation. Statistically, we refer to having dependent and an independent variables. You are using the independent variable to predict the dependent variable. So that’s one kind of hardness and softness.
The other thing that came to mind when you asked me that question, was the nugget effect. The nugget effect is kind of an unusual thing. It’s where the slope of the variogram doesn’t appear to go through the origin. It seems to intersect the y axis above the origin.
You have some intercept there.
The other thing is a kind of aliasing which you can appreciate as a geophysicist. The idea is that whatever it is we are looking for, may not be sampled properly. Maybe our well spacing or our seismic resolution is just beyond the resolution of the feature we are trying to see. The variogram tries to pick it up but it can’t. If a nugget effect occurs we have this obligation then to determine its cause; is it irreducible error or is it due to the presence of some geological feature that occurs at a scale that has not been sampled adequately. So those are the two things that come to mind when you talk about this question.
How accurate do you think the analysis would be in the presence of noise in the data, I was wondering if you could comment on that? Would it be any different choice that I would make in terms of boundaries that we have here or you think it is robust enough to look through the noise?
I think there are a number of ways to approach it. I think geostatistics certainly could be one of them. So I wouldn’t rule it out. The presence of noise, if caused by irreducible variance simply gives you larger bounds of uncertainty. So, the method will still result in an estimate, but it will also give you the plus and minus around the estimate. In fact the idea of “error” estimation is often misunderstood by folks trying to use geostatistics. Geostatistics is not really in the business of risk analysis. It is in the business of uncertainty analysis. We deliver a model of uncertainty that can then be used as part of a risk analysis. We will be able to tell you that the value of, say, porosity at a certain location is 12% and it could be plus or minus 1% percent, or it could be plus or minus 10%! So, we will establish those bounds of uncertainty.
Deviating slightly from the technical part, what are your other interests?
Gosh, well I have a great interest in history in general. I love to study history. I am in constant pursuit of intelligent life on earth! I have an interest in gemology and archaeology. Of course I have a family that I’d like to spend a little more time with. My two older boys are pretty much off on their own; my wife Mary and I, still have a teenager at home. My daughter is a dancer, so we have left the soccer and hockey behind for ballet, but we like the arts quite a bit, so this plays nicely into our interests. Other than that we enjoy doing typical family type things together. I have a bunch of other things I’d like to do, I just never have enough time to fully pursue all of them. Perhaps this is part of the joy of owning your own business.
And finally, let me ask you, what would be your word of advice for inspiration to young geophysicists?
I think probably the best words of advice that I could give really, are to remember that we come from a great tradition. The things that we are applying in our day to day activities have a basis in science and that we are in fact, at the end of the day, scientists. This is something to be very proud of. You know, I hear often that what we do is not rocket science, but I beg to differ. I would say that what we do is harder than rocket science. We deal with poorly measured and poorly behaved data, and very little of it! We are asked to make huge conclusions about the occurrence of one of mankind’s most important natural resources with this paucity of poor data, and we are asked to produce results in unrealistic time frames – not to mention that we are obligated to obey ludicrous regulatory reserve estimation constraints. Of course, we are asked to be correct as well. So, I think we are doing more than rocket science and we are doing it under tougher conditions.
Our young geophysicists should take their work seriously as scientists, and not just to learn a piece of software and morph into a sophisticated button pusher. We need to be able to think about what we do.
Another piece of advice I would give is to look strongly towards other disciplines, completely outside of our own domain, to bring new blood, new thoughts, and new ideas into the field of the geosciences. I see cross disciplinary interaction as one of the best ways to help improve the kind of work that we are doing. Let’s see how other scientists in other disciplines are solving their problems. I bet we can learn a lot!
Jeffrey, thank you for giving us the opportunity and the time to sit down here.
No problem. I hope it helps.