In June 2010, all that Paul Hsieh knew about the Deepwater Horizon oil spill in the Gulf of Mexico was what he had read in the news. Then Marcia McNutt, director of the U.S. Geological Survey (USGS) at the time, left him a voice mail.
“Monday morning, I got a phone message from the previous Friday,” recalled Hsieh. “[McNutt] called and said, you know, ‘Can you come to Houston to help us?’”
Hsieh, a hydrologist now retired from USGS, was researching groundwater contamination at the time. He knew that an oil rig in the Gulf had exploded and sunk and that every day for 2 months, tens of thousands of barrels of crude oil had been flowing freely into the sea. Oil had begun washing up along 1,700 kilometers of the Gulf Coast, threatening delicate marshes and estuaries in four states.
Hsieh, whose research focuses on mathematical modeling of how groundwater interacts with the surrounding geological features, had never worked with oil before, but he would come to play a key role in the effort to mitigate the Deepwater Horizon spill. The geoscience community responded quickly and effectively, with many scientists contributing extraordinary efforts. Among them was Hsieh, whose solitary work one night in July hugely influenced the capping of the leaking well.
The U.S. government, led by the Nobel Prize–winning physicist who happened to be secretary of energy at the time, Steven Chu, had assembled a diverse group of scientists to help find a way to mitigate the unprecedented environmental crisis. The alternative was to rely solely on experts at BP, the company ultimately responsible for the disaster. Many of the scientists, including McNutt, had set up shop near BP’s offices in Houston, where they could monitor what was going on in the Gulf.
“The way I think of us is the staff,” said Hsieh, who made several trips to Houston in the weeks after McNutt’s call. “We were often given assignments to calculate something or evaluate something or read through what BP had done and evaluate it.”
The first thing they needed to figure out was how fast the oil was coming out of the well. As an expert in modeling underground reservoirs, including how various conditions in and around a reservoir affect the rate of flow through a well, Hsieh was well suited to the task.
Hsieh also had an outstanding reputation: USGS hydrogeologist Steve Ingebritsen described him as “a zealous and unselfish collaborator, motivated entirely by the goal of achieving high-quality science” when nominating him for an AGU Ambassador Award in 2014.
“Paul came recommended as the person who had the best ability to do modeling of flow in reservoirs,” said McNutt, now president of the National Academy of Sciences. “As a team of one, he could do the work when push came to shove.”
A Reservoir Under Pressure
By the time Hsieh was brought on, all attempts to stop the flow of oil had failed, and the army of oil industry engineers and government scientists had settled on using a relief well—drilled at an angle to cut the original well off at the base, then injected with a plug of cement—to stop the flow of oil once and for all. But it would be several months before the relief well would be completed.
In the meantime, Hsieh and several colleagues were tasked with verifying BP’s assessment of the oil reservoir into which the Deepwater Horizon had been drilling, a part of the informally named Macondo prospect. (A prospect describes a site where hydrocarbons have the potential to accumulate in a reservoir.) The Macondo prospect sits nearly 4,000 meters below the seafloor, deep enough that the oil there is heated to about 116°C by radioactive decay in Earth’s interior. It lies beneath a basin called the Mississippi Canyon, and—particularly relevant to Hsieh’s assignment—it’s under a good deal of pressure.
Most oil reservoirs sit within permeable layers of rock. Just as a wet sponge squeezed between your hand and a table leaks water onto the table, the reservoir is squeezed by the weight of the rock above it, and the pressure can reach equilibrium as liquid escapes into the surrounding formation.
But for millions of years, the Mississippi River and its ancestors have dumped sediment from the continent onto the floor of the Gulf—all kinds of sediment, from coarse sand to fine silt. Over time, the silt has been compressed into dense layers of rock, effectively sealing reservoirs like Macondo under an impermeable barrier. As the river keeps piling on sediment, pressure within the reservoir has nowhere to go.
So when the Macondo well blew, oil didn’t just leak into the ocean; it was ejected at a rate of more than 50,000 barrels per day.
Soon after Hsieh’s first trip to Houston, the question of whether the well could be contained from the top, like replacing the cap on an exploding soda bottle, arose.
BP had maintained for weeks that capping the well was a bad idea. Without a release of pressure at the wellhead, engineers thought, oil would leak through openings in the damaged well casing and into the surrounding rock formation. From there, it could end up pushing up through the rock and eventually find its way to the seafloor.
The constantly gushing well was an unprecedented environmental disaster, but it would pale in comparison to such a widespread underground blowout. There would be no way to contain the spread of oil, which could flow from multiple places. The natural gas mixed with the petroleum could liquefy the seafloor into a kind of hydrocarbon quicksand. Eventually, the pressure in the wellhead would have equalized with the water pressure at that depth and the flow would have stopped, but by then the reservoir could have emptied as much as 4 times more oil than it already had, Hsieh now estimates.
But then BP engineers changed their minds. Now, they said, capping the well could work. Or rather, they could try it, and use pressure measurements taken in the crucial first hours to determine whether the oil could stay shut in.
Hsieh and his colleagues were asked to double-check that analysis.
Modeling the Reservoir
Until the end of the 20th century, well integrity calculations were done analytically. The equations were necessarily simple, assuming the reservoir was box shaped and the permeability around it was uniform. Later, complex computer algorithms were developed to model the reservoir itself, making it easier to account for all kinds of factors, like the varying permeability of the formation or the relationship between an irregular shape in the reservoir and the location of a well.
Hsieh had never put together a model for an oil well, but he knew that the same hydrological principles that governed water reservoirs would apply to petroleum reservoirs.
His team spent several weeks determining what the current pressure of the reservoir was likely to be, given how long and at what rate it had been draining into the Gulf. The idea was that once the well was capped, that baseline pressure could be compared with the pressure readings at the top of the well.
The team members settled on two key numbers. If they saw anything above 7,500 pounds per square inch (psi) (about 51,710 kilopascals at the well head), the well was sound—there was no leak in the casing—and the full pressure of the reservoir was pushing up on the well cap. The cap could stay in place, and the oil spill would be over. Less than 6,000 psi would mean the pressure was being released somewhere else, and oil eventually would find its way to the surface through that breach. In that case, the cap would have to be opened again to relieve the pressure, spilling oil into the Gulf for another 2 months until the relief well could be completed.
The problem was the no-man’s-land between 6,000 and 7,500 psi. There weren’t enough data to say conclusively what a pressure reading in that zone would mean, and all the scientists could do was hope the capped pressure they measured wouldn’t fall in that range.
So, of course, it did.
Determining Aquifer Support
On 15 July, hours after the cap was in place, well pressure had risen to 6,600 psi and was only creeping up. There was a decision to make. “For the first time in almost 3 months, there was no oil flowing into the Gulf of Mexico,” said McNutt. “And everyone was devastated at the thought that we might have to open that up again.”
The safest thing would have been to let the oil continue to flow and wait for the relief well. The potential consequences of an underground blowout weren’t exactly known at the time, but it was clear that that scenario was unacceptable. “If it actually hydrofracked to the surface through other channels, there was no way we could control that,” said McNutt. “We would have actually actively made a bad situation worse.”
But maybe the well was sound, and there was an explanation for the low pressure readings. If someone could make a convincing argument for the latter, the cap could stay in place, at least until more data could fill in the uncertainties. The group needed to come up with an answer within 24 hours, before a potential underground oil leak would have time to reach the seafloor.
To do that, scientists and engineers would needed to know whether it was plausible that the initial threshold of 7,500 psi was wrong. Could the pressure in the reservoir be lower? All they could do was plug the known conditions into a computer model and fiddle with the variables until the model produced out a pressure reading like the one they were seeing on the wellhead. They now had the extra parameter of how fast the pressure had changed after the well was shut, and it might be just enough to narrow the possibilities.
The only person on the government science team with modeling experience was Hsieh, who wasn’t even in Texas at the time.
It was already evening at his office in Menlo Park, Calif., by the time Hsieh was tapped to build the pressure model, and he needed to present his results first thing in the morning, Central time. Knowing he wasn’t going home anytime soon, he got to work. He would have preferred to have many months to arrive at a publishable model, with plenty of discussion with colleagues and, eventually, peer review of the finished results. But Hsieh had only hours, by himself.
“If I were to call up somebody and explain [it] to them, that would have taken so much time,” he said. “It was just something that I had to do.”
First he needed data. BP had not been sharing direct readouts of pressure gauges at the wellhead, fearing that advance knowledge of how the mitigation effort was going could leak out and lead to insider trading. But Hsieh had to know how the pressure had changed over time, so a USGS colleague who was at BP’s headquarters snapped a photo of the pressure curve and texted it to Hsieh.
Hsieh loaded the photo into an Excel spreadsheet and superimposed it onto the graph from a rough model he’d built. Then he adjusted the variables in his model—for example, the compressibility of the rock surrounding the reservoir—until his graph fit the curve in the photo. He had a number of ways to make his curve match the pressure readings at the wellhead, assuming there was no underground leak. He just had to determine which scenario was actually possible.
“The thing that made the biggest difference was the aquifer support,” said Hsieh. Aquifer support describes a phenomenon that sometimes occurs when oil reservoirs are surrounded by a larger region of water, causing higher well pressure than if no water were present.
BP had used seismic reflection, a method similar to sonar, to survey and map the geological vicinity of the reservoir and had determined that the volume of water in the aquifer surrounding the well was 4 times that of the oil. That number had been a determining factor in the science team’s initial estimate of the 7,500 psi baseline.
But interpreting seismic reflections can be tricky. What if the BP engineers had missed the mark, and there was far less water in the aquifer than they thought? Hsieh dialed down the aquifer support to zero, and sure enough, that could easily explain the 6,600 psi showing at the wellhead.
As the night wore on, Hsieh ran the model three times, once in the English units the oil industry uses (barrels of oil), once in metric (cubic meters of oil), and once using different code. He checked and rechecked his math, recalling an ill-fated Mars mission in which engineers’ failure to properly convert units led to the loss of a $125-million spacecraft (see bit.ly/metric-mistake).
“In our field of work, timelines are so spread out that if you make a mistake, you can catch it, there’s peer review and all of that,” Hsieh said. “And here…if you make a mistake, you’re screwed.”
Matching the Model
The Sun came up. Hsieh wrapped up his model, threw together a PowerPoint, and presented his results to the group in Houston via conference call. When he finished, there were no questions—just silence.
“I didn’t know what happened,” he said. “People were just, like, thinking about it.”
Within a few hours, Chu’s scientists agreed with BP to keep the cap in place. Underwater drones would monitor the seafloor for any signs of erupting hydrocarbons; research ships would do daily seismic reflection runs; and at the first sign of an underground blowout, the team would open the well again.
That didn’t happen. As pressure readings continued to come in, they closely matched with Hsieh’s model. The well stayed shut. Nine weeks later, the relief well permanently ended the crisis.
In the end, Hsieh saw his role as that of an accountant, not a decision-maker. He was concerned mostly with getting his model right, not with how it would be used and what the consequences could be. As long as he’d done the calculations correctly, all he had to do was present the likely scenario based on the data he had.
“I tried to kind of take myself out of the picture,” he said. “I think being a scientist, you are kind of trained to do that.”
He added that Earth scientists in particular are used to not being able to give a definitive answer to any question. There are just too many variables, too many unknowns. They so often deal with the invisible and the hidden and therefore only the plausible, the likely.
“The actual working environment and mental condition was something that I am very used to,” said Hsieh, citing plenty of marathon modeling sessions during which he took breaks only to eat and sleep.
Although the future of the Gulf of Mexico’s ecosystems and its vast economy could be affected by decisions based on the accuracy of his analysis, it wasn’t that different from his normal job assessing aquifers and other groundwater resources.
“In a strange way, it could be almost routine.”
Mark Betancourt (firstname.lastname@example.org), Science Writer
Betancourt, M. (2020), Modeling under pressure, Eos, 101, https://doi.org/10.1029/2020EO140138. Published on 25 March 2020.
Text © 2020. The authors. CC BY-NC-ND 3.0
Except where otherwise noted, images are subject to copyright. Any reuse without express permission from the copyright owner is prohibited.