BRENNEN

Industries / Energy, water and waste / Oil and gas extraction / Oilfield services / Saltwater disposal

Earthquakes

This work finds out when pumping salt water underground sets off earthquakes, and what regulators and operators do to make them stop.

7

Milestones achieved, from 1966 to 2022.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1966

    Quakes traced to a well

    Small earthquakes began near Commerce City, just north of Denver, in spring 1962, where earlier records showed probably none. In November 1965 David Evans, a Denver consulting geologist, proposed that they were triggered by waste fluid pumped into a 12,000-foot disposal well at the US Army's Rocky Mountain Arsenal, with charts showing that the frequency of quakes tracked the rate of injection. A US Geological Survey (USGS) report of March 1966 found the epicenters clustered in an area about 6 by 3 miles around the well, 4.5 to 5.5 km deep, although the surface geology showed no faulting. Healy, Rubey, Griggs and Raleigh published "The Denver Earthquakes" in Science in 1968.

    Lessonrock that looks quiet at the surface can hide a fault that injected fluid will move.

    pubs.usgs.govdoi.org

    Credit

    David Evans; J. H. Healy and colleagues at the USGS

  2. 1976

    Quakes switched by pressure

    At the Rangely oil field in Colorado, researchers raised and lowered the fluid pressure on a fault by alternately injecting water into wells and pumping it back out, and the number of earthquakes rose and fell with it. They had predicted from laboratory tests on the rock and a stress measurement in the field the pressure at which the fault would slip, and the experiment confirmed it.

    Lessonthe pressure on a fault, not the volume pumped as such, decides whether it slips.

    doi.org

    Credit

    C. B. Raleigh, J. H. Healy and J. D. Bredehoeft, Science, 1976

  3. 2008

    Gas field quakes in China

    In the Rongchang gas field in western Chongqing, more than 1 million cubic meters of unwanted water were injected at 2.6 to 2.9 km depth from July 1988. By the end of 2006 the injections had induced more than 32,000 earthquakes recorded at the surface, including two of local magnitude (ML) 5 or more and 14 of ML 4 or more. The busiest earthquake phase matched the period of highest injection rates, and seismicity fell when the rate was lowered.

    Lessonlowering the injection rate lowered the earthquake rate.

    doi.org

    Credit

    Lei and colleagues, Journal of Geophysical Research, 2008

  4. 2012

    Ohio rewrites disposal rules

    Twelve earthquakes of magnitude 2.1 to 4.0 struck Youngstown, Ohio, from March 2011, all within a mile of the Northstar 1 disposal well, which had been drilled 200 feet into the hard basement rock below the sediments. The state had one seismometer in the area until Columbia University's Lamont-Doherty Earth Observatory set up more on 1 December 2011; the well was shut on 30 December, a day before the largest quake. The Ohio Department of Natural Resources (ODNR) then required, for new deep disposal wells, a full set of geophysical logs, a review of known faults and a seismic monitoring plan, continuous pressure monitoring sent to the state, and an automatic shut-off above a set pressure. A later study by Kim located 109 quakes along a fault in the basement rock and concluded the well had induced them.

    Lessona regulator needs the geology logged and a seismometer nearby before trouble starts.

    yosemite.epa.govdoi.org

    Credit

    Ohio Department of Natural Resources; Lamont-Doherty Earth Observatory; Kim (Journal of Geophysical Research, 2013)

  5. 2016

    Largest quake during cutbacks

    On 3 September 2016 a magnitude 5.8 earthquake struck near Pawnee, the largest recorded in Oklahoma, on a basement fault no one had mapped. It came while Oklahoma was cutting disposal volumes to stop a surge in earthquakes that is largely attributed to deep injection of oilfield wastewater. By the authors' count, earthquakes of magnitude 3 and above were fewer in 2016 than in 2015, but the total energy released was higher.

    Lessonfewer small quakes does not mean less hazard, because pressure keeps spreading after injection falls.

    doi.org

    Credit

    Yeck and colleagues, Geophysical Research Letters, 2017

  6. 2021

    Texas shuts deep disposal

    The Railroad Commission of Texas (RRC), the state's oil and gas regulator, set up Seismic Response Areas where it found disposal wells likely to be causing earthquakes. It suspended 33 deep disposal wells in the Gardendale area near Midland from 31 December 2021; a magnitude 5.4 quake still struck at the area's edge on 16 December 2022. In Northern Culberson-Reeves, where 15 quakes of magnitude 4 or more had struck since January 2020, operators agreed to cut deep disposal by 68% from January 2022 levels, to 162,000 barrels a day by June 2023, against a goal of no quakes of magnitude 3.5 or more after 18 months. A magnitude 5.4 followed in November 2022 and a 5.2 in the run of quakes that began on 8 November 2023, and the RRC suspended all 23 deep disposal wells there from 12 January 2024.

    Lessoncutting volume works slowly; the operators' plan, posted by the RRC, estimated a lag of 12 to 18 months between injection and earthquakes there, and large quakes kept coming.

    rrc.texas.gov

    Credit

    Railroad Commission of Texas; TexNet, the state seismic network run by the Bureau of Economic Geology at the University of Texas at Austin

  7. 2022

    Disposal quake in Alberta

    On 29 November 2022 a local magnitude 5.59 earthquake struck about 40 km southeast of the town of Peace River, Alberta, with more quakes to March 2023. The Alberta Geological Survey concluded that Obsidian Energy's well disposing of water into the Leduc Formation induced them, and on 23 March 2023 the Alberta Energy Regulator (AER) ordered the company to cut the frequency and size of the quakes, monitor for any above ML 2.0 and place accelerometers within 10 km. A study by Schultz and colleagues (Geophysical Research Letters, 2023) found 3.4 cm of ground deformation and tied the quakes mainly to pressure from wastewater disposal at in situ oil sands (bitumen) operations.

    Lessonthe same finding bears on rules for storing carbon dioxide underground, as the study's authors note.

    aer.cadoi.org

    Credit

    Alberta Geological Survey; Alberta Energy Regulator; Schultz, Ellsworth and colleagues

Watch for

  1. Smarter shut-ins, in simulation

    Kayla Kroll and Elizabeth Cochran built a physics-based model of Permian Basin faults, ran 5,000 simulations and kept the settings that best reproduced the TexNet earthquake catalog from 2019 to 2025. They then tested four other ways of choosing which wells to shut. Two, ranking wells by distance to the modeled faults or by a well-sensitivity score, gave earthquake counts and pressure changes like those of the actual shut-ins while allowing substantially more water to be injected.

    Achieved whenthe method is published in a peer-reviewed journal and a forecast made with it is borne out by later earthquakes, or a regulator uses it and outside seismologists report the result.

    central.scec.org

    Claimed by

    Kroll and Cochran, poster at the 2026 SCEC Annual Meeting, August 2026

Open

  1. 01

    Score Texas's quake response plans

    A public scorecard, refreshed every month, for each of Texas's three Seismic Response Areas (Gardendale, Stanton and Northern Culberson-Reeves). For each area it would show daily deep and shallow disposal volumes, the count of earthquakes of magnitude 3.5 and 4 or more, and each plan's stated goal and checkpoint dates, rebuilt by published code from public files.

    The brief
    Credit · unclaimed
  2. 02

    Forecasts filed in advance

    A prospective test of forecasts for disposal-induced earthquakes in the Permian Basin. Modelers would file their forecasts (counts of quakes of magnitude 3 or more per map cell per quarter for the next 12 months, from public injection data) before a fixed date, and a neutral scorer would grade them against the TexNet catalog when the year ends.

    The brief
    Credit · unclaimed

Open 01 · the brief

Score Texas's quake response plans

Credit · unclaimed

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

What to build

A public scorecard, refreshed every month, for each of Texas's three Seismic Response Areas (Gardendale, Stanton and Northern Culberson-Reeves). For each area it would show daily deep and shallow disposal volumes, the count of earthquakes of magnitude 3.5 and 4 or more, and each plan's stated goal and checkpoint dates, rebuilt by published code from public files.

Why it is still open

The RRC says its staff "will use available public data and trends (for example, earthquake count, earthquake magnitude, and disposal volume)" to judge each plan, but its page lists actions, not a running evaluation. Woo and Ellsworth (Bulletin of the Seismological Society of America, 2023) evaluated the two Midland Basin areas with a catalog for 2020 and 2021 only, before the later magnitude 5 quakes. Breton, Shensky and Savvaidis (2024) automated the data preparation but used paid B3 Insight and IHS data alongside public sources. A 2025 conference abstract by Chevron scientists says deep-disposal cuts since 2022 brought a declining trend in Culberson County, which is a company's own analysis.

rrc.texas.govdoi.orgdoi.orgseismosoc.secure-platform.com

Built with what exists

The TexNet earthquake catalog download; the TexNet Injection Volume Reporting Tool's daily well data; the RRC's well lists and boundary shapefiles for each area; InjectA, an open Python package by Yangkang Chen and colleagues for pulling daily injection data (github.com/chenyk1990/InjectA); pandas and geopandas; GitHub Actions for the monthly rebuild.

  1. First step

    Download the Northern Culberson-Reeves shapefile and well list and the TexNet catalog from 2020, count the quakes of magnitude 3.5 or more inside the boundary each quarter, and plot them against the 162,000 barrels-a-day target.

  2. Done looks like

    All three areas covered monthly from January 2020 to the latest month, every figure traced to a public file, the rate of magnitude 3.5 and larger quakes given for the year before and after each RRC action, and six monthly rebuilds in a row run without hand edits.

Open 02 · the brief

Forecasts filed in advance

Credit · unclaimed

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

What to build

A prospective test of forecasts for disposal-induced earthquakes in the Permian Basin. Modelers would file their forecasts (counts of quakes of magnitude 3 or more per map cell per quarter for the next 12 months, from public injection data) before a fixed date, and a neutral scorer would grade them against the TexNet catalog when the year ends.

Why it is still open

Forecasts made in advance exist only one model at a time. In 2016 Cornelius Langenbruch and Mark Zoback of Stanford (Science Advances) forecast that Oklahoma's earthquake rate would fall significantly by the end of 2016 after mandated cuts in saltwater injection; it was published before the outcome but not filed alongside rival models or scored by a neutral party under rules set beforehand. The Permian models found so far are fitted to the past. Kroll and Cochran (2026) matched the 2019–2025 catalog; Caltech and ExxonMobil researchers (2025 abstract) and Chevron researchers (2025 abstract) calibrated their models to past seismicity. The USGS's 2016 one-year hazard forecast for the central and eastern United States assumed earthquake rates would stay the same and called itself "the first step in developing an operational earthquake forecast". Searches for this brief found a pseudo-prospective test (a test on past data run as if in real time) only for a geothermal field, Hengill in Iceland.

pmc.ncbi.nlm.nih.govcentral.scec.orgseismosoc.secure-platform.compubs.usgs.govdoi.org

Built with what exists

pyCSEP, an open-source toolkit for building and evaluating earthquake forecasts (github.com/cseptesting/pycsep); the TexNet catalog and injection data; Zenodo to timestamp each filed forecast.

  1. First step

    Write the test rules and a baseline forecast, a smoothed-seismicity model built from the 2023–2025 TexNet catalog, for the Delaware and Midland basins, and post both on Zenodo.

  2. Done looks like

    At least three forecasts from separate teams filed with timestamps before the test year starts, scored with pyCSEP at its end, and the scores, forecasts and code published.