Industries / Energy, water and waste / Carbon capture
Underground storage
Injecting CO2 into porous rock deep underground, below a tight layer that keeps it from rising, and showing that it stays there.
6
Milestones achieved, from 1996 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1996
Sleipner stores its CO2
Since 1996 Statoil, now Equinor, has removed CO2 from natural gas produced at the Sleipner field in the Norwegian North Sea and injected it into the Utsira formation, a layer of water-filled sandstone, instead of releasing it. The gas held up to 9.5% CO2, which had to be cut to 2.5% to meet sales rules, Offshore magazine reported in 1996. Equinor counts more than 19 million tonnes stored by the end of 2020. The site has been watched with repeated seismic surveys since injection began, and a reference set of its data was released to researchers through the SINTEF-led CO2 DataShare in January 2020.
Credit
Statoil (now Equinor) and partners
2010
A well class for CO2
On 10 December 2010 the US Environmental Protection Agency (EPA) set rules for wells that inject CO2 for long-term storage, creating a new class of well, Class VI. The rules cover site characterization, the area of review (the zone where the CO2 or the pressure it creates could endanger drinking water) and fixing old wells inside it, well construction, monitoring, financial responsibility, care of the site after injection ends, and closure. States can take over permitting once EPA approves their programs: North Dakota in April 2018, Wyoming in 2020, Louisiana in 2024, West Virginia, Arizona and Texas in 2025, and Colorado from 12 November 2026.
Credit
US EPA; state regulators
2011
In Salah stops injecting
From 2004 to 2011, BP, Statoil and Sonatrach injected about 4 million tons of CO2, separated from natural gas, through three horizontal wells into a 20 m thick water-filled layer deep below the Krechba field in Algeria. Satellite radar showed the ground rising above the wells, in a pattern near the well KB-502 that later modelling explained by a deep fracture zone opening. Injection at KB-502 stopped in mid-2007 after CO2 appeared at the wellhead of a nearby old appraisal well, and all injection stopped in June 2011 over concerns about the integrity of the caprock, the sealing layer above. Lawrence Livermore's reviewers reported in 2013 that no monitoring data suggested storage integrity had been compromised.
Lesson"It is useful to deploy multiple, independent monitoring tools," the Livermore team wrote; satellite radar showed changes that the wells alone did not.
Credit
BP, Statoil and Sonatrach; modelers at Lawrence Berkeley and Lawrence Livermore national laboratories
2011
Snøhvit changes layers
At the Snøhvit gas field in the Barents Sea, CO2 is separated from the gas at the Melkøya plant on shore and piped back to the field for injection under the seabed. Injection into the Tubåen formation began in 2008, and shortly afterwards "the pressure rose sharply near the well." In 2011 the operators decided to switch injection to another formation, the Stø. Equinor now reports about 750,000 t of CO2 captured and stored a year.
LessonHow much CO2 a rock layer will take is only known once injection starts, so storage plans need a fallback.
Credit
Statoil (now Equinor) and partners
2016
Basalt turns CO2 to stone
At the CarbFix site in Iceland, at the Hellisheiði geothermal power plant, CO2 was injected into basalt, a volcanic rock. A team led by Jürg Matter reported in Science in June 2016 that over 95% of the injected CO2 had turned into carbonate minerals in less than 2 years. The paper says this contrasts with the common view that such mineral trapping takes hundreds to thousands of years, and calls it the first demonstration of permanent disposal of CO2 as carbonate minerals in basalt. Carbfix says it has injected more than 100,000 t since 2014.
Credit
Jürg Matter, Sigurður Gíslason, Eric Oelkers, Wallace Broecker and co-authors; the CarbFix project
2021
Gorgon falls short
Chevron's Gorgon gas plant on Barrow Island, Western Australia, takes the CO2 that occurs naturally in its offshore gas and injects it into a sandstone formation about 2 km below the island. The system started in August 2019. In November 2021 Chevron said it would buy and surrender 5.23 million greenhouse gas offsets and spend A$40 million on lower-carbon projects in Western Australia to make up an injection shortfall over the five years to 17 July. Reuters reported in May 2022 that the system, designed to bury 4 million tons a year, managed 2.1 million in 2021. Chevron's operations director said the company was learning "how you inject CO2 into the reservoirs."
LessonInjecting CO2 raises the pressure in the rock, and if the wells that draw off the rock's water fall behind, injection falls behind too.
Credit
Chevron Australia and its partners; Reuters
Watch for
Greensand at full scale
INEOS announced the official start of injection at Greensand Future on 18 September 2026. CO2, mostly from Danish biomethane plants, is liquefied and shipped from Esbjerg to the former Nini West oil field, about 250 km off the Danish coast. The project starts at 400,000 t a year and aims for 4 to 8 million tonnes a year by 2030.
Achieved whenDanish authorities or EU emissions records confirm a full year of injection near 400,000 t.
Claimed by
INEOS, September 2026, as reported by AFP
Gorgon's pressure fix
Chevron says more than 12 million tonnes of CO2 was injected at Gorgon from August 2019 to March 2026, and that it has begun a project to modify four water-producing wells and draw more water out of the storage rock, lowering its pressure so more CO2 can go in.
Achieved whenyearly injection reported to the Western Australian government reaches the design rate of about 4 million tonnes a year.
Claimed by
Chevron Australia, 2026
Open
01
Re-run a permit's model
An independent rebuild, with open-source software, of the computer model a US operator used to draw its area of review, using only the inputs in its public permit file, and a published comparison of where the CO2 plume and the pressure front reach in each model.
The briefCredit · unclaimed02
Old wells beyond the Permian
A public table, one row per well, for every final US Class VI permit outside the Permian Basin: each well the operator lists inside its area of review, the matching record in the state's well database with plugging date and status, and any well in state records or on old maps that the operator's list leaves out.
The briefCredit · unclaimed
Open 01 · the brief
Re-run a permit's model
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent rebuild, with open-source software, of the computer model a US operator used to draw its area of review, using only the inputs in its public permit file, and a published comparison of where the CO2 plume and the pressure front reach in each model.
Why it is still open
EPA's review of the Wabash Carbon Services application says "EPA did not perform independent, duplicative modeling" of the area of review and judged it "not warranted at this time"; the operator's team used the STOMP simulator. A Pacific Northwest National Laboratory poster on its Model Explorer tool names the "need for open-source tools that evaluate CO2 injection models and map the Area of Review." Hemanth Hariharan, Sarah Saltzer and Sally Benson of Stanford estimated area-of-review sizes with a machine-learning model for simplified reservoirs and compared them with permits, but did not rebuild any permit's own model. Searches of GitHub found no rebuilt permit model.
Built with what exists
OPM Flow from the Open Porous Media project, or SINTEF's MRST with its co2lab module; permit applications and area-of-review plans from EPA's Class VI Data Repository (udr.epa.gov) and its appeals docket; the Sleipner benchmark model from CO2 DataShare to test the workflow first; Python.
First step
Download the Wabash area-of-review plan and list every model input it gives and every one it withholds as confidential.
Done looks like
For one permitted project, the open model's plume and pressure-front maps at the end of injection, side by side with the operator's, with every input traced to a page in the permit file and every difference explained; then the same for a second project, with code and inputs public.
Open 02 · the brief
Old wells beyond the Permian
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table, one row per well, for every final US Class VI permit outside the Permian Basin: each well the operator lists inside its area of review, the matching record in the state's well database with plugging date and status, and any well in state records or on old maps that the operator's list leaves out.
Why it is still open
Abandoned wells can act as vertical channels for injected CO2 and brine, and wells plugged before 1953 often used cement that did not set properly, according to a 2023 report by Robert Rossi and Dominic DiGiulio for the Environmental Integrity Project. They counted about 190,000 abandoned wells in Louisiana, about 13,000 of them plugged before 1953 and lying over possible storage rock, but statewide, not inside each permit's area of review. In October 2026 an open data set by Gale, an AI agent run by Matt Attenborough, joined the well lists in 23 Permian Basin monitoring plans in Texas and New Mexico to state records; about 150 of the Texas wells had no digital record at the Railroad Commission. It left sites outside the Permian for later. A 2024 paper by Arbad and co-authors assessed legacy-well risk from public data for Class VI applications in the Illinois Basin, without a well-by-well match.
Built with what exists
Permit files from EPA's Class VI Data Repository and state regulators (Louisiana's SONRIS, the North Dakota Industrial Commission, the Wyoming Oil and Gas Conservation Commission); the open code and coding guide of the Permian data set as a starting point; USGS historical topographic maps to look for wells missing from records; a document-reading AI model to pull well tables out of permit PDFs for a person to check (see the Note "Seven more flagged"); QGIS and Python.
First step
Take the Wabash Carbon Services area-of-review plan in Indiana, extract its well table, and match each well against the state's online well records.
Done looks like
Every final Class VI permit outside the Permian issued in the United States up to a stated date, each well row linked to its source, with counts of wells matched, wells missing from the operator's list and wells plugged before 1953; an outside reviewer re-checks ten permits and gets the same counts.