Ashley Francis of Earthworks Environment and Resources Ltd. U.K. was the February 2005 CSEG luncheon speaker. The topic of his well-attended talk was ‘Limitations of Deterministic and Advantages of Stochastic Seismic Inversion’. Following are excerpts from the interview we recorded with Ashley after his lecture.
Can you begin by telling us something about your early education and your work experience?
Okay, well I started with a joint honours degree in Oceanography and Soil Science from the University of Wales in 1984. I really went there to spend three years enjoying rock climbing but at the end of it, I did manage to come out with a Geophysics Degree and I joined Seismograph Service Ltd. (SSL) when they were still a small, independent seismic company and I worked for them for seven years in the Borehole Seismic Division. I spent four of those years working overseas. I went to Johannesburg in 1987 and to Muscat, Oman, in 1990.
So what sort of work have you been doing all along?
Up until 1991 I was almost exclusively working in Borehole Seismics, VSP Data, log calibrations and synthetic seismograms but when I went back to the U.K. in 1991, I left SSL and I joined a couple of small consultants at J. Arthur & Associates and that’s when I started to work on more general geophysical methods, including seismic inversion and geostatistics.
From the time you spent abroad and your experience in Borehole Seismology, in your opinion how much VSP technology was being used in different companies?
VSPs were available at that time but I think people probably didn’t find as many applications as they had anticipated.
Yes, that is my impression as well. The use of VSPs has been quite restricted over the last three decades. So, how did you decide to take up geophysics as a profession?
I didn’t actually. I went to University to do Zoology and when I got there I didn’t like it and I really enjoyed my first year at sciences. I did a geology course and I did solar science and so I changed. I was always good at physics. I had very good physics qualifications in High School so it wasn’t difficult to change to geophysics. I have never regretted it. It’s been good.
So what is the state of geophysics in universities in the U.K.? Is the enrolment quite satisfactory or is it declining?
I think it is probably declining. Recently, we’ve been trying to recruit geophysicists and we find we get a lot of applications from people who’ve got geology degrees but relatively few with a degree in geophysics. Maths and Physics are somewhat in decline in teaching in the U.K., which is a problem I think, and demographically, we have a problem in the U.K. and in the Oil Industry in general with young people not going into the more numerate degree programs.
Are the professional societies or others taking some steps?
I don’t think that they are, not that I am aware of anyway. I think there is a general acknowledgement in the U.K. that more needs to be done to encourage people to take maths and physics courses but these courses are probably declining whilst perhaps less strongly scientifically numerate disciplines are on the increase.
Yes, that is the state of affairs all around, even the Americas. In Asia and the Far East, of course, I think it is a little different because there is still enrolment in the universities and employment in the oil companies.
You covered a lot of ground at your talk today but for the benefit of the members who were not able to attend our luncheon, I would like to ask you some questions on inversion and get your responses. What are the different types of inversions and which one would you recommend under a given circumstance?
I think there are two parts to that question. One is the approach that you take to inversion and that is partly what the talk was about today, to discuss the differences between deterministic and stochastic inversion. But there are also different algorithms as well. I think, largely speaking, that the algorithms are perhaps not as critical as choice of methodology and how you think about the inversion process. Different contractors who supply inversion products obviously prefer to talk up their own algorithms and I think that’s quite natural. In terms of what I talked about today, deterministic inversion is really where you have thicker layer problems and stochastic inversion is where you are looking at thinner layer problems, more details in the reservoir and why the uncertainty has an important impact in your reservoir predictions.
You also mentioned in your talk about coloured inversion; what do you mean by colour inversion and how is it used?
Well, we quite like colour inversion as a q.c. tool and it is a very useful seismic attribute, really. Colour inversion is simply shaping the amplitude spectrum of the seismic data to the amplitude spectrum of the well logs. BP developed the method some years ago and it has been in public demand for some time. I like coloured inversion as a starting point inversion because it gives you the best from your seismic data without having any model. In that sense, it’s quite robust and a useful product.
There is no low frequency trend of any sort incorporated there, so is that why you prefer it?
As an interim product, it’s a means of providing interpreters with probably the best type of seismic data to use for general interpretation but as soon as they want to be quantitative about what’s in the reservoir, then they need a deterministic or a stochastic inversion.
How much of each is being used these days?
They come and go in popularity as they have success or failure. I think the market is dominated by deterministic inversion, which has been quite popular since the 1980’s. Stochastic inversion schemes really started being published as methodologies in the mid 1990’s on the back of the explosion in geostatistics technology. The problem with stochastic inversion is the speed, which has tended to restrict its application for realistic sized problems and I think that has put people off.
Yes, in geostatistics, one has to account for the horizontal and vertical variation in the form of a variogram. Based on the available literature and the courses that have been taught, there is uncertainty involved in the choice of the variograms. So how do you think that can help inversion where you use variograms?
Well the variogram is an important part of the choice but it is also in a reservoir model as well. It’s the horizontal variogram range that is difficult to select. We approach it by looking at slices from the colour inversion so we try to get some of that information from the seismic data. Otherwise you are restricted to using the well tie as you would in a reservoir model. I think the key thing when you are building either the reservoir model or the stochastic seismic inversion is to say “well, if this is an uncertain parameter then I should experiment with a short-range or a long-range and see whether that has any effect on the volumemetric predictions for the reservoir or anything else that might be important in my reservoir development.” If it has an important impact, clearly it is a sensed parameter. If it has very little impact, you can probably put it to one side and say “it’s not critical by choice”. The best thing is to test it.
In your talk today you did not mention anything about another type of inversion that has been in use but involves a lot of pre-stack data, that’s the joint-inversion. What are your thoughts on using that sort of inversion?
Our scheme is also capable of doing a pre-stack joint inversion. We’ve had some success using P-S, P-P joint inversions; they are very useful. You can go for a full pre-stack inversion but, when you do that, the dimension of the problem becomes very large so implementing it in a stochastic frame would be extremely ambitious. Certainly in principle, it’s not a problem but just a computational issue at the end of the day. The methodologies using very wide angle data to look at density terms are still in their infancy and I think that probably you need to be very careful about acquiring very wide angle data. The density term is a very small term in inversion so you need very wide-angle data and I think generally it’s fairly sensitive to noise.
In AVO, or the Lambda-Mu-Rho type of analysis, we compute the IP and IS separately from the P and S reflectivity and then compute our LMR attributes. So from that standpoint isn’t it much more accurate and beneficial to have a joint inversion?
You can also simplify the problem a lot. I think geophysicists have historically had a desire to think of it as a geophysical problem but at the end of the day it’s not geophysics we are after, it’s reservoir properties. So an alternative way to think about the problem is not to think in terms of IP, IS or LMR, which are essentially physical parameters. We’ve had some success with using an extended elastic inversion, which is a generalization of a pre-stack into an offset gradient method. It predicts the reservoir properties directly and in that sense you are getting away from having to say “it’s a geophysical problem” and say “it’s a reservoir problem”. At the end of the day, it’s the reservoir properties that we are interested in, not the rock physics itself, but saturations, fluid type, lithology and porosity, so we’ve found the extended elastic inversion to be pretty good as an alternative pre-stack type approach.
You are referring to the extension that Whitcombe and others came up with.
Yes, this is the Whitcombe methodology
So what do you say about the future of seismic inversion. Do you foresee a bright future or any particular type of inversion taking over?
Certainly from my perspective I see future seismic inversion coming from reservoir modelling. There has been a lot of success with reservoir modelling software packages using geostatistics to take the well data framework to build stochastic reservoir models. I think those models are limited because they only use the seismic horizons as a constraint in building the models and what we should be looking at is making these models in such a way that the full convolution of those models when converted from lithology and saturations and porosity into impedance should also be consistent with the seismic trace. I think we have driven it very hard from the attribute concept and now we should think from the other side. The reservoir model is important to us and we should think about how we can make the reservoir model constrained by the full seismic volume, amplitudes, convolution of response, pre-stack, etc.
In your talk, the methods that you mentioned and talked about are all local optimization methods. We also have the global optimization technique. Is global optimization the way to go; what are your comments on that?
Which approach you take to some extent depends on which contractors you talk to. To my mind, the real key is in understanding the uncertainty in the process of inversion and the fact that the seismic data is band limited, so all of these methods are simply flavours around the minimization technique. The key thing the stochastic inversion approach brings is being able to understand how non-unique that solution is and it gives you the opportunity to see whether that nonuniqueness is important in your reservoir model. But it’s possible to formulate a stochastic inversion in both the simulated annealing framework and in a sparse framework and in a model-based framework. It is possible to do a stochastic inversion in all of those frameworks.
You have your own consulting firm. How did you decide to start your own consulting company in preference to doing a regular job with a company?
As I mentioned earlier, I went through two consulting firms in the U.K. and then in 1996 I joined Lasmo, which was an independent oil company. I was with Lasmo until 2001 in the technical services group. It was a good time and I got a chance to see a lot of Lasmo’s assets all around the world. But Lasmo was taken over by ENI AGIP in 2001 and at that time every employee was offered a severance package. That seemed to me a good time to go and do my thing so I took the money and ran, basically.
So, do you find it more satisfying now, running your own firm?
Well, it’s nice being able to turn up in the office in jeans with holes in and to take my dog to work. But it’s not so good for the heartstrings when things aren’t going so well and you worry about where the next piece of work is coming from. So it has its good and bad. It can be stressful from a business side but it obviously gives me a lot of freedom. I can turn up, and the guys who work for me they can turn up, in jeans and shorts and we are all kind of relaxed and have the radio on, which I couldn’t do at Lasmo for sure.
When you go to an office, you have to maintain a particular decorum.
That is probably true. I think people don’t bring their dogs to work very often.
Looking back on your geophysical career, would you like to share with us some of the successful landmarks that you have had?
I really enjoyed my time in the consultancy, which has now disappeared. That was a good time for me, a formative time in the 90’s learning geostatistics and seeing how it had a bearing on geophysical work. I did a lot of forward modelling work and starting looking in much more detail at inversions. That was an important time in terms of my career development. I guess the other time was the time I spent in Lasmo. I think it’s difficult to be a good consultant if you have never worked in an oil company. I think you have to understand where all this data goes and what oil companies do with it, in order to really understand what is important. And to me, and to an oil company, it’s volumetrics, reserves, production. It’s not about geophysics and LMR, it’s about the reservoir itself. I guess those are the two key career developments for me.
Your company is engaged in different types of geophysical work. So between this work, training and software development, which do you enjoy most?
I think I enjoy most of it. We do some consultancy work but primarily we are doing service and project work for industry. Some of that is geostatistical, some of it is seismic conversion based. The training can be good. We are certainly teaching younger people. I teach geostatistics at Imperial College in London and that’s good. I get to see a lot of young people who sometimes ask hard questions and sometimes ask easy questions. I am going around doing the training and trying to help people to see different ways of looking at these problems. But at the end of the day, I guess it’s putting the effort into turning what’s in your mind into software is the best. And I am very lucky in that I have a programmer who is brilliant. He doesn’t know any geophysics at all but it doesn’t seem to matter; we just seem to get on pretty well and he delivers something pretty good.
Yes, well that’s good. It comes in very handy. What is this ultra fast stochastic method?
This is the algorithm we have developed. It is based on a fast spreading transfer approach. That’s basically the quickest geostatistical method for simulation. It works pretty well. It’s kind of like an extension in the way kriging goes to conditional simulation geostatistics. We are essentially doing the same in seismic inversion. We are doing it in the frequency domain for speed so we call it ultra fast. We don’t know, it’s always hard to get run time comparisons, but we are estimating between 10 and 50 times faster than the existing sequential methods.
That’s why it qualifies as ultra fast? What do you think of the velocity information that you may obtain out of your inversion? Could that be a good input for pore pressure analysis?
It could be, I mean certainly full pressure prediction is one aspect of seismic inversion and it’s quite widely used. What the stochastic inversions bring is looking at the uncertainty in that prediction. Very often in full pressure prediction you are also looking at constraining the problem with the seismic velocities, and again we can bring that into our inversion models. We can bring them in as a constraint in a colour-coded simulation. So yes, it has an application in full pressure prediction, certainly we are aware of that and we’ve got some products in the pipeline where we are looking at using it for that purpose.
Have you done some projects there to give really accurate pore pressure information?
Not yet with the stochastic inversion. We are still waiting for somebody to come along with the right project.
What are your other interests, apart from science?
Well, obviously the family. I’ve got a young son and a daughter. My wife, Judy Forbes, is a co-director of the company and she handles all the legal and financial sides. I am a rock climber but I also do a lot of fishing and a lot of shooting, pheasant shooting, so not very politically correct in today’s world.
That’s now illegal in England so I better not make any further comments on it. You can still hunt and flush but you can’t kill the fox. Certainly the law is not very popular where I live.
What has been the most difficult challenge in your professional life?
I think probably starting the company. The initial stages, trying to grow the company, it’s difficult and it takes a lot of time, puts a lot of pressure on your family life and certainly trying to recruit new people and build the company up is difficult. So I think that has been the most difficult thing I have ever tried. All the other stuff seems fairly easy by comparison.
So when you recruit people, what do they see your company giving them, is it better prospects, better salaries or stability?
We are a very small company; we only have 4 people, 5 if you include the dog.
You pay him though?
Well, we wonder if we could avoid paying tax if we paid him. Our salaries are competitive but we are not able to pay huge salaries. I think most people enjoy us because what we do is interesting. It is technically and intellectually stimulating and our location is good as well. We are not in London; we are a long way out in the rural part of the west of Britain, so it’s a nice area to live in. So it’s intellectual and it’s the environment.
Do these people come from the area around there?
Not specifically. We are just about to recruit another geophysicist who is actually quite local, so I guess there would be a local connection.
One last question. Keeping in mind your rich experience that you have gained over the last 20 years or so, what words of advice would you have for young people who are planning to go into our profession?
I think there is no substitute for going out and making sure you keep reading the journals and keeping yourself up to date and really bothering to understand what it is that we do. And don’t just restrict yourself to your own discipline. Reach out to some things that are kind of peripheral. I found lots of good geostatistical work in solar science journals, which are not really geophysics at all but gave me a new insight. So I think just cast your net quite widely in what you read and see and use your imagination to apply that to your work.
I thank you very much for giving us this time to come and talk to you. It has been a really interesting experience.