Eric, our First Break readers would like to know you better. Could you tell us about your educational background and some of the key milestones in your professional journey? Looking back, how have your views on major developments in seismic technology and their applications evolved throughout your career?
My M.Sc. education is in Applied Physics, but throughout my studies I found myself increasingly drawn to programming and signal processing rather than experimental physics. This interest led me to join the research group of Prof. Berkhout for my M.Sc. thesis, where I worked on a project in room acoustics. At that stage, I had little interest in the field of seismics. However, after joining that group and observing the seismic research being carried out by fellow students, I became increasingly fascinated by the signal processing and imaging challenges it presented. That experience ultimately ignited my passion for seismics and established the path for my future career.
After earning your master's degree in applied physics, what motivated you to pursue doctoral research on Surface-Related Multiple Elimination (SRME)? How did this research direction emerge, and what convinced you that this approach had the potential to make a significant contribution to seismic processing?
Looking back, I consider myself fortunate to have been in the right place at the right time. At the time, Guus Berkhout had already developed the basic theory for SRME and published this in his 1982 book. He then invited another M.Sc. student to develop a simple, one-dimensional example using spiky wavelets to demonstrate the method’s effectiveness, and those results were incorporated into the 1985 revised edition of the book. Following this, Prof. Berkhout secured a research grant from the Dutch Research Council to further develop the concept and was looking for the right person Ph.D. candidate to undertake this work. I happened to finalize my M.Sc. work around that time and the rest, as they say, is history.
In the early stages of my Ph.D. research, there was quite a bit of skepticism on this topic. Even after successfully demonstrating the method on 2D finite difference data, many believed that it would not be practical for field data due to its sensitivity to the source wavelet and the large computational requirements at the time.
SRME was initially viewed as computationally challenging but later became an industry standard. What were the key breakthroughs that allowed the method to move from theory into routine practice?
Indeed, it was computationally expensive at that time, even for 2D field data. I remember my first 2D field data example running for a couple of weeks, with the input data stored on magnetic tapes, so halfway through the process, a colleague had to manually change tapes while I was away on my vacation. I remember calling from a telephone booth on a French campsite to provide some final instructions.
There are three main elements that guided the success of the SRME methodology: the first is its adaptive implementation, which circumvented the strict requirement of precisely knowing the source wavelet. Also, overall amplitude errors (e.g. while using 3D data in a 2D method) could be dealt with via local adaptation. The second element, which is often overlooked, is proper pre-processing of the data: noise and direct waves need to be properly removed, near offsets need to be accurately interpolated, and source and receiver sampling regularized, as deficiencies in the input data are amplified by the auto-convolution process. The third factor for success was reformulating the non-linear series expansion implementation into an iterative, linear process, which made the implementation of SRME much more efficient and user-friendly.
Your contributions have left a distinctive mark on seismic geophysics. Looking back, what personal qualities, experiences, or circumstances do you believe have been most influential in shaping your career?
Firstly, the fortunate timing I mentioned earlier was a major enabler. If I had completed my M.Sc. one year earlier or a year later, my career path would likely have taken a very different direction. Having said that, I was selected for this Ph.D. position because of my good performance during my M.Sc. thesis, demonstrating my ability to solve challenging problems, especially related to signal processing and algorithmic development. Equally important was my aptitude for translating theoretical concepts into practical solutions.
During the early days of my Ph.D., I did not have access to any commercial seismic processing package. As a result, I developed many of the basic processing steps, such as normal move-out (NMO) correction, CMP gather sorting, data interpolation, and Radon transforms. Writing these algorithms from scratch gave me invaluable practical experience and a much deeper understanding of seismic data processing, insights that have remained with me throughout my career. In fact, my first field data processing was carried out entirely using such self-developed programs. It was only later, we gained access to software packages such as Seismic Unix, which greatly streamlined many processing tasks and allowed us to work much more efficiently.
Reflecting on your professional journey, is there one decision or opportunity you would have approached differently, and one experience or achievement for which you feel particularly fortunate?
Looking back, there were two occasions where I wish I had pursued the ideas further. The first occurred in the early 1990s when my Ph.D. student Nurul Kabir, who, unfortunately, passed away earlier this year, and I were working on near-offset interpolation using the parabolic Radon transform. Due to computational advantages, we decided to continue with the linear least-squares approach, although we were already considering a non-linear approach, incorporating sparsity constraints. We recognized that such a non-linear approach had the potential to improve the resolution in the Radon domain and, thereby, the interpolation performance. This work was carried out around 1994. Less than a decade later, sparse Radon transforms emerged as a major break-through, confirming the promise we had already envisioned.
Similarly, the second instance involved imaging with multiples. By the mid-1990s, we had demonstrated that incorporating multiples into the imaging process could significantly improve subsurface illumination. However, we struggled to overcome the problem of crosstalk artifacts. I realized that once a multiple had been used for imaging, it should be removed from the data, such that it could not be imaged again at an incorrect location.
But I could not get my head around it. Again, within about a decade, the least-squares migration process became the solution, where we would re-model the seismic data from the image and suppress crosstalk in an iterative fashion. Perhaps the timing was not right for either of these ideas.
Nevertheless, I believe our early work on imaging with multiples made an important contribution. At a time when multiples were widely regarded as noise to be removed, we recognized that they also contain valuable signal that can be exploited for imaging. Although our work was not the complete solution, I believe it helped lay the groundwork for what has since become accepted practice in seismic imaging.
Marine seismic data provide relatively favourable conditions for multiple prediction because the water surface acts as a strong and predictable reflector. Land seismic data, however, involve complex near-surface effects such as scattering, internal reflections, and mode conversions.
What are the major challenges in extending data-driven multiple prediction methods from marine to land environments, and why is careful data conditioning so important?
Even more than on marine data, thorough pre-processing is essential for success on land. I routinely maintained two versions of the seismic data: one processed with regular, conservative techniques, and another subjected to aggressive pre-processing. The latter was only used for the multiple prediction process and discarded afterward, while the predicted multiples were subtracted from the data that had undergone gentler pre-processing. I discovered that any residual noise in the pre-stack data, even which might stack out in the final imaging, interferes with multiple prediction process by creates a ‘haze’ of noise in the predicted results. Consequently, careful adaptive subtraction, with a choice of methods, is the second enabler for successful land data applications.
You have often suggested that multiples should not simply be viewed as unwanted noise but as a potential source of subsurface information. How can interpreters adopt this perspective, and what future opportunities do you see for exploiting multiple energy in seismic imaging and characterization?
From an interpretation perspective, the final seismic image should ideally show no evidence that it was created using primary and multiple reflections. Instead, the interpreter should simply benefit from the improved reflector continuity, and reduced acquisition footprint that multiples can provide. This represents the ideal scenario.
However, as with most advances, the benefits come with potential trade-offs. The extended illumination provided by multiples, especially in situations where primary illumination is, e.g. for OBN acquisitions, may be accompanied by crosstalk artifacts from multiples in the migrated image. Such crosstalk will manifest itself in different ways than the traditional residual multiples in standard seismic imaging results, which may be less recognizable by interpreters. For example, crosstalk energy might appear at shallower depths than the associated primary reflection rather than below it, making it more difficult to identify and interpret correctly.
However, as our industry seeks to optimize seismic acquisition geometries and reduce survey acquisition costs, especially for the energy transition applications, I truly believe that multiples should be considered as an integral part of the seismic wavefield, rather than as unwanted noise. When properly incorporated into the imaging workflow, multiples provide valuable additional illumination that can compensate for reduced acquisition density, potentially allowing fewer shot points or receiver locations without compromising image quality.
With increasingly sparse acquisition geometries, such as 3D VSP and sparse Ocean Bottom Node surveys, how does the treatment and exploitation of multiples differ from that in more densely sampled conventional seismic datasets?
This is a very good point: the sparser the acquisition, the more we must rely on combining primaries and multiples. Conversely, in a full ‘carpet’ acquisition of sources and receivers, all illuminating wave paths are already captured in the primary reflections, with multiples only carrying redundant repetitions. Such fully sampled datasets are ideally suited for multiple attenuation techniques, such as SRME for surface multiples, and Marchenko methods for internal multiples. As the source and receiver sampling becomes sparser, with VPS and OBN data representing extreme cases, we become increasingly reliant on using multiples to ‘infill’ the missing primary illumination.
However, while decreasing the source and receiver density, For the full-wavefield methods such as FWI and full wavefield migration, reducing acquisition density also changes the optimal acquisition strategy. Instead of uniformly sparse sampling, it can be advantageous to adopt asymmetric acquisition geometries, with dense sampling on one side (either sources or receivers) and sparser sampling on the other. From the perspective of conventional primaries-only imaging, such geometries may appear suboptimal, since traditional acquisition design is based on symmetric sampling principles. But when multiples are treated as an integral part of the imaging process, these primaries-only design concepts no longer apply. Consequently, incorporating multiples requires re-inventing acquisition design.
In academia, the emphasis is often on extracting the maximum amount of information from the data, for example, pushing full-waveform inversion to recover the finest possible detail. In industry, however, the objective is typically to reduce uncertainty sufficiently to support sound drilling, development, and investment decisions. How do you reconcile these two perspectives in your own work? Can you share an example where a simpler geophysical solution ultimately delivered greater value than a more advanced approach?
I have encountered situations with multiple removal on field data where a simple predictive deconvolution filter performed much better than any of the more advanced SRME approaches. This reinforces the need for a toolbox approach; rather than assuming that the most sophisticated method is always the best, we should determine which solution is best suited to a particular problem.
Modern AI-driven geophysics often faces the “black-box” challenge: highly accurate predictions may come with limited physical understanding. What is your perspective on this issue? For example, if a machine-learning workflow produced a facies model that improved drilling decisions but lacked clear geophysical interpretability, what validation would you require before trusting it?
We should aim for AI outputs that not only provide predictions but also measure of uncertainty. An AI-generated result, by itself, unless extensively validated, will have limited meaning. Our current efforts in using AI, especially for reservoir characterization, focus on integrating uncertainty estimation.
One of the strengths of well-designed neural networks, is that they can scan the solution space and automatically characterize the stochastic distribution of the outcomes. Unlike many traditional deterministic methods, which often rely on simple Gaussian assumptions, neural networks have the potential to capture much more complex, non-Gaussian uncertainty distributions. If we can fully leverage this potential, I believe that AI can provide very useful outputs. However, we are still in the early stages of developing such approaches, and they need to be tested and verified.
In seismic processing, I am less critical of AI, as many deterministic processes, such as denoising and interpolation, are fundamentally based on feature-recognition, a domain where human evaluation is highly effective. Therein these areas, we have seen that AI-based solutions can outperform the deterministic methods, which are often based on loose, intuitive physical assumptions such as local plane waves or lateral continuity.
Your research has highlighted the close relationship between acquisition design and processing/imaging strategies. How has this changed our thinking about seismic survey design? Do you see optimized or non-uniform acquisition geometries becoming practical in field operations? What challenges remain?
As I mentioned earlier, incorporating multiples into the imaging workflow should fundamentally change the way seismic acquisition surveys are designed. This is especially important for energy transition applications, where budgets are often much tighter. By levering all available information within the seismic data, we can optimize survey designs to target specific subsurface zones of interested. This approach can be very helpful for monitoring applications, such as monitoring the evolution of the CO2 plume, detecting potential leakage, or, more importantly, conclusively demonstrate that no leakage is occurring.
A remaining challenge is achieving the right balance the between noise and signal. In land seismic surveys, acquisition geometries are often designed not only to provide adequate target illumination but also to suppress the strong near surface noise and generate enough signal-to-noise ratio at the target level. Excessively sparse acquisition may preserve the desired target illumination using multiples, but may not be able to attenuate noise sufficiently.
For field operations, however, I do not see too many obstacles: OBN nodes are already deployed at sparse grids, while maintaining dense source sampling. For land data it could mean that we leave dense receivers intact and have irregular shot patterns, a practice already being utilized for simultaneous-source operations.
Looking ahead, do you see opportunities to combine physics-based inversion frameworks with machine learning or AI to optimize acquisition geometries? Could AI assist in designing surveys while remaining constrained by physical principles?
This is especially true when designing sparse monitoring surveys, where subsurface knowledge is already available, allowing designs to be tailored based on overburden and acquisition noise already known from the base survey. Such optimization can become highly non-linear, creating a role for AI to map these complex relationships, provided it can also quantify uncertainty. Naturally, the input data for such neural network approach must be grounded in physics. I can envision an AI-agent conducting specific synthetic experiments to extract optimal case-specific insights and finally proposing the optimum acquisition design.
Your research interests have expanded to include challenges related to the energy transition. Could you describe some of the problems your group is currently addressing and the areas where you see geophysics making the greatest contribution in the coming years?
There are two sides to this coin. On the one hand, the seismic imaging algorithms are largely oblivious to the purpose of the survey; they simply strive to deliver the highest quality image. Also, multiples are an inherent part of the recorded seismic wavefield, generated primarily in the overburden, regardless of the geological target. I often compare it to medical ultrasonic imaging: imaging an unborn baby is not much different from imaging the heart or liver. In the same way, many seismic imaging techniques are fundamentally application independent.
On the other hand, several emerging applications introduce new requirements. For instance, leakage of CO2 does not have a clear counterpart in fossil-fuel exploration. Similarly, the cyclic injection and extraction of hydrogen in depleted gas fields or aquifers is completely different from any prior monitoring application, let alone the fact that the interaction of these small hydrogen molecules inside geologic formations is completely different from any hydrocarbon molecule. Consequently, certain aspects must be re-invented. Rock-physics models for hydrocarbons are not always useable for describing the hydrogen situation. Another interesting challenge arises in seismic site characterization for offshore wind farms. Here, the used wavelengths are so short that the water surface can no longer be considered a simple reflecting mirror. As frequencies push beyond the kilohertz range, conventional SRME breaks down, and we need to extend it so capture the water-wave effects.
Over the past few decades, the relationship between academia and industry has evolved considerably. As Program Director of the DELPHI Research Consortium, supported by several industry sponsors, how do you see the role of university–industry collaboration changing in today's environment? How can professional societies help maintain dialogue and foster innovation between academia and industry?
We observe that the consortium model for geophysical research has worked exceptionally well over many decades. Delphi started in 1990, building on predecessors established in Delft in the early 1980’s, that replicated the success of even earlier US consortia. However, several factors make maintaining such collaborative research models increasingly difficult today.
Firstly, the exploration teams in the energy industry have reduced considerably over the last decade. Reportedly, the current industry research spending is only 40% of what it was during its peak ten years ago.
Secondly, our focus on energy transition is not yet aligned with the energy sector’s priorities, which remain heavily tied to fossil fuels—presenting distinct challenges, as mentioned earlier.
Thirdly, a growing cultural shift toward free resources, driven by the open-source community and a stronger emphasis on open academic publishing, has reduced the immediate drive to join formal consortia.
To me, this reflects short-term vision. Membership in an academic research consortium offers value that extends far beyond getting technical reports and prototype software. It provides early updates on new developments, deeper insight into theoretical frameworks, and getting direct access to experts who can assist with complex problems or help identify prospective interns and new employees. Above all, it offers a space to encounter those occasional unconventional ideas that just might work. In the long run, as energy transition applications become more important, establishing an early presence in these initiatives will offer a decisive strategic advantage.
Many successful researchers are influenced by mentors and scientific environments. Which individuals, ideas, or experiences have had the greatest influence on your scientific thinking?
Looking back, I realize how fortunate I was at the beginning of my Ph.D. journey; not only was a high-impact topic waiting for me, but I also had the privilege of working with two giants of our community as mentors and supervisors. Prof. Guus Berkhout brought visionary thinking, out-of-the-box approaches, and strong physical intuitions, while Dr. Kees Wapenaar, offered deep theoretical insights, a gift for explaining complex concepts in understandable language and a drive to serve the community with mathematical descriptions of physics. Later,
Prof. Dries Gisolf joined us from Shell to lead our group for ten years during the 2000’s, providing yet another vital perspective, grounded in industry practice and the bridge between seismic imaging and reservoir characterization. Collaborating with these individuals allowed me to develop into the broad geophysicist I believe I have become today. Too often, I hear their voices echoing in my mind when now mentoring my own Ph.D. students.
Eric, you are now also serving as Publications Officer for EAGE. How are you enjoying this role, and what vision do you have for EAGE’s publications going forward?
Yes, I enjoy the role of Publication Officer and being able to serve the association. As a scientist and academic, publications are of course close to my heart. Although all the hard work is done by the editors and reviewers of the journals, I try to make sure what we offer in terms of publications is aligned with the community needs. We also try to push forward E-book publications to provide more in-depth information, which can be crucial in times where a lot of scattered information is around. At the same time, EAGE is taking great advantage of AI and has been building “Earthdoc AI” for the community. Testing the pilot version, I realize this tool earns its place within all other chatbots, as it really focuses on our large geoscience archive and provides information in a usable and organized manner.
Scientific research can be exceptionally demanding. Outside your professional activities, are there other interests or hobbies that help you maintain a healthy balance?
As for many of us in academia, our workday does not simply stop at 5pm. Pursuing science, teaching and mentoring students feels more like a mission driven with strong intrinsic motivation, ultimately offering a deep sense of satisfaction. But I make a conscious effort to set aside time each week for exercise such as running, hiking, biking, as well as reading and meeting family and friends, which serves to recharge my internal batteries and return to work with renewed energy and perspective. Also becoming a grandfather recently has taught me to cherish the small moments.
Besides that, I am passionate about music; I play the bass guitar. Over the years I have played in a gospel choir combo and maintained a rock-band with friends. Currently I play in a rock-band largely rooted in our department, where we perform at work-related parties, and I am also part of the EAGE band, performing at most of the recent annual meetings. Making music with colleagues is really enjoyable. Music unites people, something desperately needed in this polarised world, and there is something gratifying about hearing colleagues complement a performance the day after a conference social event, even if those compliments happen while washing hands in the restroom. 😉
What advice would you offer to a young person considering a career in geoscience? Given the rapid technological changes taking place today, what opportunities do you believe make this an exciting field to enter?
First of all, I am truly convinced that we need engineers with a solid technical background, even in this era of AI. Ultimately, AI can only learn from existing data, and that information requires continuous updating. While AI will undoubtedly play an increasing role in educating and training young professionals, it cannot replace the necessity of mastering fundamental principles. Knowledge about the Earth will always be needed, particularly as dense populations in concentrated areas demand massive geotechnical efforts. Additionally, regarding the deeper Earth, we cannot afford to reduce our efforts on understanding it thoroughly. Energy transition will rely heavily on subsurface storage, and we must also comprehend subsurface hazards while securing critical raw materials that are still buried under our feet.
Finally, is there a question you expected me to ask but that did not come up during our conversation? If so, please feel free to pose the question yourself and share your thoughts.
One topic not yet touched upon is teaching. In fact, about 20-25% of my time at the university is devoted on teaching, and it is one of the most rewarding aspects of my work. It is a privilege to pass knowledge on to the next generation and help prepare them for society. At the same time, working with young people is a profound learning experience that keeps you youthful and grounded. The days when I could out-program my students are long gone! It is amazing how the new generation leverages computers, the internet, and AI models to their advantage.
Education itself is evolving, away from the traditional one-way lecture format, and I deeply admire my younger colleagues who adapt to these shifts so fluidly, while for me this takes more effort, though I am eager to learn. In the upcoming semester, I will be teaching a new B.Sc. course on stochastic time series analysis. This presents a fresh challenge for me, and I am currently digging into textbooks that I have not touched in forty years. In many ways, it feels as though the circle has come full turn as I am both a teacher and a student once again. Ultimately, however, every time my colleagues and I successfully graduate a new M.Sc. or Ph.D. student, I feel an immense sense of pride and privilege to be in this position.
