Microbiomes flag hydrocarbon reservoirs
DNA analytics, machine learning applied to near surface tests for microbial life sensitive to micro-seepage from hydrocarbons below
Netherlands-based Biodentify uses new DNA analytical methods and advanced machine-learning algorithms to identify potential hydrocarbon reservoirs. The science is used to predict oil and gas deposits based on microbiome reactions to micro-seepages of gas molecules.
“What we’re trying to do is to help you avoid drilling a lot of dry holes,” said Robert Chelak, in a presentation to a May 31 technical breakout session at the state Geologic Materials Center in Anchorage. The session focused on the potential for new investigative technologies and machine learning systems to better assist geoscientists and resource companies to meet the challenges of interpreting Alaska geology.
The mix of microbial species in a near surface soil sample is compared to a database to correlate DNA, soil samples and relevant production data from previous drilling, Chelak said. The DNA fingerprint of the sample is an indicator of the presence of vertical micro-seepage to the surface from hydrocarbon accumulations in underlying strata.
“Predicting where reservoirs are by doing DNA fingerprinting, and machine learning,” Chelak said.
“It’s mind blowing; it’s new; it’s innovative; it’s only been around for about two years.
“We want to make sure we’re ... going into areas where there is reservoir, versus shooting seismic or drilling wells in areas where we don’t need to be,” he said. “We want to make the areas that are productive very highly productive; we want to drill the best locations, maximize your profit, and minimize the impact on the environment.
“In Alaska, where you’re opening up for exploration, the environment is a big deal,” he said.
Medical breakthroughThe company’s technology is borrowed from a medical science breakthrough that uses saliva to test for tumors, as opposed to a much more invasive biopsy.
The process was developed in a lab in Holland, Chelak said, adding that one of the company’s founders, Chris te Stroet, postulated that the science could be adapted to the oil and gas industry.
Vertical upward micro-seepage occurrences have been known since the 1930s, Chelak said, adding, “The Russians were looking at this.”
“There are cracks in the subsurface and these bubbles are making their way up,” he said, comparing the action to bubbles rising in a champagne glass.
It can take months to years for those bubbles to rise from a reservoir to the surface, depending on the thickness of the overburden, Chelak said. “These micro bubbles are moving meters per day.
“But the microbial life is much more complicated than just a few species that were known to be hydrocarbon oxidizing bacteria,” Chelak said. “It is necessary to determine the complex composition of microbes - not only those that flourish at micro-seepage, but also those that are eliminated and are therefore found in reduced concentrations above hydrocarbons.”
A one-millimeter soil sample is taken at a one-foot depth from multiple locations in the test area.
“We analyze that over many different sites to come up with the microbial count,” Chelak said.
Of over 340,000 microbes that the company has identified so far, about 200 are affected by hydrocarbon micro-seeps.
The company said its database with over 2,500 samples from both onshore and offshore locations, with related production data on bio markers, is used as modeling input to its proprietary “localized triple loop” computational model.
Six plays in the United States are extensively sampled: Bone Spring (Permian), Bakken, Antrim, Avalon, Lewis, Haynesville and Marcellus. In addition, studies have been carried out both on and offshore the Netherlands as well as offshore Norway.
The sample data is merged with characteristics on productivity, the age of the producing interval, type of play (oil vs. gas), geology and climate and soil type.
Local fingerprintingChelak said when Biodentify first studies a field, the samples are initially matched against the company’s large database to build a map of prospective sub areas of the study area.
On a job in the Haynesville shale, “we were using the fingerprint from all of the other areas of the United States,” he said.
The company then took local samples from producing areas and dry areas and added the data back into the database before re-running the process. Of the 362 samples taken, three were from known producing areas and four were from dry areas.
“The first map was 70% accurate; the final map is about 86% accurate in predicting where the reservoir is,” he said.
“Based on the information that came out of this, we looked at the wells that were drilled in the area,” Chelak said. “Approximately 100 wells were drilled in the area - they would have only needed to drill 46 wells to get the same amount of production over the producing area.
“Wells in this area wells are between $6 million and $10 million; it would have saved about $300 million in drilling costs,” he said.
The savings don’t stop there, he said. Later in the production phase of the field, it is possible to go back and re-survey to see if there are some remaining hydrocarbons that could yet be produced - without having to go back and shoot more seismic.
Machine learningDue to the massive computing power now available combined with machine learning, Biodentify is able to surpass the accuracy and richness of information from a traditional geochemistry study.
“In a strictly geochemistry study, people will take the samples, they’ll put it in a Petri dish, they’ll grow it in a lab, and they’re looking at about five to maybe 10 microorganisms that would grow that would indicate hydrocarbons being there,” Chelak said. “But you can get a lot of false positives with geochemistry; the microorganisms can come back to life, and again they’re only looking at about five to 10.
“We’re looking at 50 to 200, and a majority of those are the ones that find the micro bubbles toxic; 35% of the ones we’re looking for are the ones that are living, that are flourishing off of hydrocarbon molecules that are coming up.
“Things are changing over time; these organisms are feeding off each other; they’re competing with each other; there’s things that happen - there may be a slight spill of some sort that would affect the microorganisms in that particular area,” Chelak said. “There’s climate change, but that will affect all of the things we have programmed into our machine learning (which) can filter those things out.
“We’re trying to find the signal of that 50 to 200 in all of the noise that we have in the background.”