BRENNEN

Industries / Energy, water and waste / Oil and gas extraction / Oilfield services / Well plugging and abandonment

Finding and assessing old wells

This is the work of locating wells that were never recorded or were recorded in the wrong place, and judging from records and surveys which ones pose a risk.

6

Milestones achieved, from 1985 to 2026.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1985

    Magnets find buried casing

    US rules for underground injection set an area of review around each proposed injection well, inside which old wells that could carry waste upward had to be found. With money from the Environmental Protection Agency, the US Geological Survey flew magnetic surveys over four test areas near Oklahoma City and one in Colorado; a magnetometer senses the disturbance that steel well casing makes in the Earth's magnetic field. The team concluded that 95 to 98% of the wells in the five areas probably showed up, that the aircraft found more wells than air-photo study did, and that it also flagged more features that were not wells.

    Lessona magnetic survey finds steel, so a well whose casing was never set or was pulled out leaves no signal.

    pubs.usgs.govpubs.usgs.gov

    Credit

    Frank Frischknecht and colleagues, US Geological Survey; US Environmental Protection Agency

  2. 2016

    Helicopters survey old fields

    The National Energy Technology Laboratory (NETL) flew a helicopter carrying two magnetometers over 259 hectares of the Salt Creek oil field in Wyoming, in production for more than 100 years. The survey located 93% of the visible wells, and 20% of the wells it found had been unknown or recorded in the wrong place. A methane sensor flown with it traced four large plumes, and only one came from a leaking well. In forested Pennsylvania, NETL reported in 2023, the surveys found two to five undocumented wells for every recorded well drilled between 1880 and 1950.

    Lessonthe earliest wells were often cased with wood, which gives no magnetic signal; NETL found some of them as round dips in drone lidar scans, laser maps of the ground's shape, where soil had sunk into the rotted casing.

    osti.govapi.semanticscholar.org

    Credit

    Richard Hammack, Garret Veloski and colleagues, NETL

  3. 2020

    Lidar finds forest wells

    Lidar is a laser scanner flown over the ground that measures its shape, even under trees. NETL researchers combined an airborne magnetic survey and lidar over the Pioneer Run watershed in Oil Creek State Park, Pennsylvania, where hundreds of wells were drilled in the mid-1800s, and georeferenced two photographs of the oil field from the late 1800s to match. Lidar picked out 290 places on the ground, 86% of them possible well sites. Of the wells confirmed on the ground, 62% had never been plugged. The authors describe this as the first use of lidar together with an airborne magnetic survey to find wells. When the NETL team tested 210 old wells in and around the park for methane, 21 were releasing it, and all but one of those lacked concrete plugs, National Geographic reported in 2022.

    osti.govnationalgeographic.com

    Credit

    Patricia Saint-Vincent, Natalie Pekney and colleagues, NETL

  4. 2021

    Gas explodes in Wheatley

    On 26 August 2021 an explosion in the centre of Wheatley, Ontario, injured 20 people and destroyed two buildings. It came during the third leak of hydrogen sulfide, a poisonous gas, in as many months, and a mix of hydrogen sulfide and methane from one or more old wells beside the site was the suspected cause. In December 2021 crews clearing the site found a second gas well. A year later the investigation into the exact cause was still not complete.

    Lessonwells drilled before modern records can sit under a town, and the first sign may be gas in a building.

    amp.cbc.catheglobeandmail.com

    Credit

    Municipality of Chatham-Kent; Ontario Ministry of Northern Development, Mines, Natural Resources and Forestry, whose petroleum inspectors responded

  5. 2022

    Wells beside buildings counted

    After Wheatley, The Globe and Mail analysed the 26,674 wells in Ontario's public Petroleum Well database. It counted 15,257 abandoned wells, 6,210 of unknown status and 7,424 wells that may have no owner. Using building outlines from OpenStreetMap, it found 1,319 wells closer than 75 m to a structure, the minimum distance Ontario allows between a well and a building, counting only wells whose recorded position was accurate to 50 m or better.

    Lessonrecorded well positions in Ontario range in accuracy from 1 m to 1,000 m, so any distance count leaves out the worst-placed wells.

    theglobeandmail.com

    Credit

    The Globe and Mail

  6. 2026

    Drones find 31 wells

    A team led by Jade Boutot of McGill University flew a drone with a magnetometer over 12 square kilometres of hilly forest in northwest Pennsylvania. Field checks led by the state regulator confirmed 31 undocumented wells, and methane was found leaking at three of them. Ground magnetometers at 34 other drone signals pointed to 15 more wells buried out of sight. The drone's methane sensor saw raised methane over the forest but could not tie it to a source. The authors call it the largest drone survey of its kind.

    Lessonreading magnetic maps still takes expert judgment, so each candidate needs a field check.

    api.crossref.orgblogs.edf.org

    Credit

    Jade Boutot and colleagues at McGill University, Pennsylvania's Department of Environmental Protection, three US national laboratories, Sawback Technologies and the Environmental Defense Fund

Watch for

  1. Burial depth from drones

    A team at Lawrence Berkeley National Laboratory reports a model that turns the strength of a drone's magnetic signal into an estimate of how deep a well's casing is buried, and of the highest a drone can fly and still detect it. It was calibrated at two sites in California and Oklahoma and then used at Hillman State Park, Pennsylvania, where it pointed to one undocumented-well candidate and gave depths for three known wells that agreed with other evidence.

    Achieved whena team outside the authors uses the model on a new survey and records or digging confirm its depths at a stated number of wells.

    api.crossref.org

    Claimed by

    Jiannan Wang, Yuxin Wu and colleagues, Lawrence Berkeley National Laboratory, September 2026

Open

  1. 01

    Score the lidar well finders

    A public test set for spotting old well sites in lidar: high-resolution elevation tiles cut around wells whose positions were checked in the field, and as many tiles with no well, split by county so that any method is scored on ground it has never seen. Add a baseline model and its scores.

    The brief
    Credit · unclaimed
  2. 02

    Reopen the 1985 Oklahoma survey

    A table of every casing-like magnetic anomaly on the 1985 US Geological Survey maps of four test areas near Oklahoma City (Arcadia, Horseshoe Lake, Moore and Oklahoma City), placed on a modern map and matched to today's state well records and orphan well list, each marked as a recorded well, an orphan well or unmatched.

    The brief
    Credit · unclaimed

Open 01 · the brief

Score the lidar well finders

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test set for spotting old well sites in lidar: high-resolution elevation tiles cut around wells whose positions were checked in the field, and as many tiles with no well, split by county so that any method is scored on ground it has never seen. Add a baseline model and its scores.

Why it is still open

NETL found that rotted wood casing leaves a round dip that shows in high-resolution lidar, and at Oil Creek lidar picked out 290 possible sites, but the features were picked by people. A Carnegie Mellon University project with NETL proposed automatic lidar detection in 2018; no results or shared data from it were found. Public image sets for finding wells do exist, but not for lidar: a 2024 set of 120,948 aerial photo tiles from the US National Agriculture Imagery Program, 58,882 of them containing a documented orphan well, and the Alberta Wells Dataset of more than 213,000 wells in medium-resolution satellite images, presented at the ICML machine learning conference in 2025. Searches of the web, Crossref, Zenodo and GitHub in November 2026 found no public lidar test set of this kind.

osti.govengineering.cmu.edupmc.ncbi.nlm.nih.govarxiv.org

Built with what exists

Free lidar from the US Geological Survey's 3D Elevation Program; state well records whose coordinates were taken by GPS when the well was plugged; QGIS; PyTorch with a small image-segmentation model such as U-Net.

  1. First step

    Download 1 m lidar for one Pennsylvania county with many plugged wells, cut 100 tiles around wells with GPS coordinates and 100 without, and count by eye how many show a mark.

  2. Done looks like

    At least 2,000 field-checked well sites and 2,000 empty tiles from at least three states, a second person's check of 200 labels, and a baseline's precision and recall reported on counties held out from training, all under an open license.

Open 02 · the brief

Reopen the 1985 Oklahoma survey

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A table of every casing-like magnetic anomaly on the 1985 US Geological Survey maps of four test areas near Oklahoma City (Arcadia, Horseshoe Lake, Moore and Oklahoma City), placed on a modern map and matched to today's state well records and orphan well list, each marked as a recorded well, an orphan well or unmatched.

Why it is still open

In all the Oklahoma areas but Moore the 1985 survey found as many possible casings as the records held wells, or more, and the team could not match anomalies one to one with wells because the recorded locations were inaccurate. Web searches in November 2026 found no later comparison of these anomalies with current records.

pubs.usgs.gov

Built with what exists

The flight-line and contour maps in USGS Open-File Report 85-614B; the Oklahoma Corporation Commission's nightly well file and weekly orphan well list, both free downloads; QGIS for fitting the old maps to section corners; Python for the matching.

pubs.usgs.govoklahoma.gov

  1. First step

    Fit the Moore test area's contour map to its section corners and overlay the commission's well file.

  2. Done looks like

    Every casing-like anomaly in the four areas placed to within a stated error and matched or not to a current record, a second person's re-check of 10% agreeing on at least 90%, and the unmatched list published and sent to the Oklahoma Corporation Commission.