MIT geophysicists are developing lower-cost, more-accurate techniques for imaging and monitoring carbon dioxide (CO2) injected into underground geological formations for long-term storage. Standard methods used by industry to image the subsurface involve tracking sound waves sent into the earth. However, it is expensive to process the gathered seismic data, and differences in survey conditions such as equipment location and weather make comparing data collected before and after CO2 injection difficult. The new MIT technique requires significantly less data processing than conventional methods do, and it incorporates several steps that reduce the confusing effects of data-gathering inconsistencies. As a result, it produces clearer images of the subsurface that focus especially on the area of interest. Initial field tests confirm the viability of the technique.
One approach to mitigating climate change while carbon-free energy options are being developed is to capture CO2 emissions from power plants and other major sources and store them underground, sequestered from the atmosphere. Promising underground formations for storage include deep saline aquifers, depleted oil and gas reservoirs, and unminable coal seams. But widespread adoption of that approach will require reliable techniques for determining the amount and location of the sequestered CO2 and for monitoring it for potential leaks, which could allow the greenhouse gas to escape back into the atmosphere.
“The oil and gas industry has seismic-based methods that we expect could be adapted to perform those tasks,” says Alison E. Malcolm, the Atlantic Richfield Career Development Assistant Professor in the Department of Earth, Atmospheric, and Planetary Sciences (EAPS). The methods involve using large trucks or other sources to send sound waves deep into the earth. How those seismic waves are reflected by underground layers provides information that sophisticated signal-processing techniques can turn into three-dimensional images of the subsurface. “Performing such analyses before and after CO2 injection can help to delineate the spatial distribution of the CO2 in a reservoir,” says Malcolm. “But the high cost of those procedures and the huge computational resources required will likely preclude their use at every sequestration site.”
By adapting conventional seismic-based methods, Malcolm and her colleagues in the MIT Earth Resources Laboratory—Michael Fehler, EAPS senior research scientist, and Di Yang, EAPS graduate student and a 2009-10 Eni-MIT Energy Fellow—in collaboration with Lianjie Huang of Los Alamos National Laboratory, have been developing a new technique that should not only cost less but also produce images that are clearer and more easily understood.
During a seismic survey, the sound waves generated by exploding dynamite, a huge vibrating truck, or some other source travel downward through the geologic layers and are reflected back to an array of receivers located in the region of the source. During their travels through the subsurface, the sound waves move at varying speeds. For example, their velocity is lower through a soft layer than through a more rigid layer. Velocity generally increases with depth, and it changes abruptly at interfaces between layers.
Because of those variations, the timing and amplitude of the returned signals measured by the receivers can be processed to generate images such as the examples below. Those images show the velocity of the seismic wave created by a source as a function of position beneath the earth’s surface. While color represents seismic velocity, it is in essence representing the geologic layers, so the overall image serves as a proxy for a physical model of the subsurface.
In the carbon-sequestration application, the aim is to look for local changes in velocity that occur as a result of CO2 injection. The left-hand diagram below is a “baseline model,” which represents the subsurface before the CO2 is injected. The right-hand diagram is a “time-lapse” model of the subsurface after CO2 injection. In the time-lapse model, there is a semicircular region of anomalous wave speed just below center—a new feature indicating the reservoir of injected CO2.
While this pair of images is useful, a single image showing just the velocity differences between the baseline model and the time-lapse model should—in theory—provide greater focus on the new CO2 reservoir and assist in efforts to characterize and monitor it. The researchers’ new technique is a novel approach to calculating such an image.
To test their technique, they needed to perform parallel tests using the conventional method and their new technique. If they started with seismic data from field surveys to make the comparison, they would have no way of knowing which approach came closer to the “correct” answer. So to begin, they created the models shown above. They next worked backward from those models and numerically generated (estimated) sets of “synthetic” seismic data that could yield those models under field conditions.
Now they were ready to perform their comparison. First they used a conventional method to generate an image of velocity change: They processed the before-injection and after-injection synthetic seismic data to generate a baseline model and a time-lapse model, and then they subtracted the former from the latter.
The result is shown in the left-hand figure below. Here color is the change in wave velocity (as a function of position) from before the CO2 injection to after. The semicircular velocity change inside the reservoir is evident but not detailed or distinct. In addition, there are many small changes away from the reservoir that make interpretation confusing: Do they reflect small-scale, isolated geological structures that affect velocity—or might they instead be caused by isolated pockets of CO2?
The researchers next performed the same analysis using their technique—“double-difference waveform inversion”—which was first suggested by Huang and is now being refined and applied by Malcolm and Fehler’s team. The first step was to calculate a baseline model in the conventional way—by processing the before-injection synthetic seismic data. But rather than then generating the time-lapse model, they next subtracted the baseline data from the time-lapse data, thereby producing a new data set of the changes in velocity. By processing those data, they directly generated the desired outcome—a model showing changes in velocity in the subsurface.
The right-hand figure above shows their result. The image inside the reservoir is now clearer, and there are fewer scattered velocity changes elsewhere. “You’ll see that we’ve recovered fairly well this kind of semi- circular change in velocity,” says Malcolm. “By using the difference between the before and after data sets, we focus in on the change in the reservoir with less noisy scatter outside it.”
Why does the double-difference technique work so much better? One challenge here is that conditions change between the data-gathering expeditions. “This is not a lab experiment where you’re sitting inside and you can control all of the parameters,” says Malcolm. “You’re outside. It might rain during the first survey and not the second, or there might be intermittent noise from active drill rigs.” Also, it is difficult to get the receivers and—especially—the source in exactly the same locations. (The source of their test data is a 4-meter-long, 60,000-pound truck. Relocating the central vibration pad in precisely the same place can be tricky.) All of those inconsistencies introduce systematic error in the data—not just within the reservoir, where changes have actually occurred, but also outside it.
In addition, there are small-scale structures throughout the subsurface that cause changes in velocity but are difficult to resolve with certainty in a model. Translating seismic data into a model involves “waveform inversion,” an optimization process that identifies the model that best approximates most of the wave events in the seismic data. However, as in any optimization process, there are many models that could fit the data. With the conventional approach, the two independent inversions may not resolve those small-scale structures in an identical manner. Subtracting the two models therefore yields apparent changes in velocity that result from those different interpretations. “The subtraction produces a ‘noisy’ image that reflects the poor resolution of each independent inversion,” says Malcolm.
Using differences in the data rather than the models enables Malcolm and her team to take steps to reduce those sources of inconsistency. “The data really see the earth, not whatever our model is,” says Malcolm. “The only thing that actually changed in the earth…is in the reservoir region.” They therefore infer that changes in the data for regions well away from the reservoir must be caused by changes in the survey conditions, such as weather or equipment location. Based on that knowledge, they create a filter that removes that systematic error from the entire data set—outside as well as inside the reservoir.
Likewise, they know that the small-scale structures remain unchanged and that they need not be resolved “accurately” because the focus is on the reservoir. So when interpreting the time-lapse data, they impose the same small-scale structure that was obtained from the baseline data in regions away from the reservoir. “So we’re not looking at different models for the outside-the-reservoir region,” says Malcolm. “We’re only looking at different models for the inside-the-reservoir region because that’s where the changes we’re interested in are located.”
The researchers have begun field tests of their double-difference technique. Already they have applied it to seismic data collected before and after injection of CO2 into a reservoir in an oil field in Texas located several thousand meters underground. Initial results show that the double-difference approach produced cleaner images of the CO2 reservoir than conventional methods of analysis did—and with far less data processing. And in related work, the MIT team is developing a novel system that can detect when CO2 may be leaking from a reservoir and raise an alarm that further study should be undertaken. Together, their techniques to verify the location of injected CO2 and to ensure that it remains safely underground will help support the adoption of subsurface CO2 sequestration as an industrially viable process for mitigating climate change.
This research was supported by a seed grant from the MIT Energy Initiative and by the National Energy Technology Laboratory of the US Department of Energy through a subcontract from Los Alamos National Laboratory. Further information can be found in:
D. Yang, M. Fehler, A. Malcolm, and L. Huang. “Carbon sequestration monitoring with acoustic double-difference waveform inversion: A case study on SACROC walkaway VSP data.” Proceedings, Society of Exploration Geophysicists 2011 annual meeting, San Antonio, Texas, pp. 4273– 4277.