Industries / Energy, water and waste / Power generation / Geothermal
Earthquakes, water and gases
This work deals with the side effects of taking heat and fluid from underground: small earthquakes, sinking ground, drained hot springs, and the gases that come up with the steam.
6
Milestones achieved, from 1950s to 2018.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1950s
Wairakei's ground sinks
Production testing at Wairakei, in New Zealand, began in the 1950s, and level surveys have tracked the ground ever since. In the field's main subsidence bowl the ground has sunk 15 m in total, and parts of Wairakei, Tauhara and Ohaaki fell by more than 3 m, or faster than 100 mm a year. The worst sinking was traced to weak, clay-rich layers 100 to 300 m down, altered by hot fluids, that compact as the pressure in them falls. Regional councils now require regular survey reports, checked by independent review panels, as a condition of each field's consent.
Lessonweak shallow layers, not the deep reservoir, set how far the ground sinks, and only decades of surveys show it.
Credit
Chris Bromley (GNS Science) and colleagues at Energy Surveys, Central Surveys and Contact Energy
1997
Wastewater refills The Geysers
The Geysers in northern California is a steam field whose output fell fast after its 1987 peak as steam pressure dropped (see Power plants). To refill it, a 46 km pipeline began delivering treated wastewater and lake water from Lake County in 1997, about 22 million liters a day, and a 64 km pipeline from the city of Santa Rosa followed in 2003 with about 30 million liters a day, by Lawrence Berkeley National Laboratory's figures. Calpine says output has since held steady at about 725 MW. The laboratory's review found that small earthquakes there closely track injection, and that quakes above magnitude 4 became more frequent in the two years after the Santa Rosa water arrived, raising concern in nearby communities.
Lessonputting water back can save a reservoir and also make it shake more, so both need watching.
Credit
Calpine; Lake County; the City of Santa Rosa; Ernest Majer and colleagues at Lawrence Berkeley National Laboratory
2006
Basel's traffic light trips
A traffic-light system fixes in advance what happens when ground shaking crosses each threshold: green, carry on; amber, pump with caution; red, stop. Julian Bommer and colleagues designed one for a 2003 stimulation (pumping water at high pressure to open cracks in hot rock) at the Berlín field in El Salvador, and Basel's Deep Heat Mining project in Switzerland adopted it with a red threshold of 5 mm/s of ground velocity, against Berlín's 60 mm/s. Six days into pumping into crystalline rock 4 to 5 km under Basel, a magnitude 2.6 quake led the operators to cut and then stop injection on 8 December 2006. About five hours later a magnitude 3.4 quake was felt across the city and cracked plaster; claims totaled CHF 6 million, and the project was ended in 2009 after a risk study. At St. Gallen in 2013 the traffic light called for a stop, but gas rising in the well forced the crew to keep pumping, and a magnitude 3.5 quake followed.
Lessonquakes can keep coming after pumping stops, and well control can override the light, so a threshold alone does not make a project safe.
Credit
Julian Bommer and colleagues (traffic light); the Swiss Seismological Service (monitoring)
2017
Some plants emit like coal
Geothermal plants emit far less CO2 than fossil plants on average, but not all of them. A 2017 review for the World Bank by Thráinn Friðriksson and colleagues gave 122 g of CO2 per kWh as the best global average, from a 2002 survey, and recent national averages of 34 g/kWh for Iceland, 104 for New Zealand, 107 for California and 330 for Italy. It also found plants at Mount Amiata in Italy and at several sites in Türkiye emitting more than 500 g/kWh, some more than 1,000 g/kWh, on par with coal.
Lessona plant's carbon footprint depends on the gas in its own reservoir, so it has to be measured plant by plant.
Credit
Thráinn Friðriksson, Almudena Mateos, Yasemin Orucu and Pierre Audinet (World Bank)
2017
Pohang's magnitude 5.5
Korea's first enhanced geothermal system (EGS) project at Pohang pumped water at high pressure five times into crystalline rock about 4 km down, to open cracks between two wells. On 15 November 2017 a magnitude 5.5 earthquake struck beneath the site (5.4 in some estimates), the most damaging in South Korea since instrument records began in 1905. Studies in 2018 by Kim Kwang-Hee and colleagues and by Francesco Grigoli and colleagues pointed to the injections, and in 2019 William Ellsworth and four co-authors reported that injection into one well had activated a previously unmapped fault that triggered the quake. It is the largest known earthquake induced at an EGS site. Korean courts have split over the state's liability, and the case was before the Supreme Court in October 2026.
Lessonan unmapped fault near the well can turn small, managed quakes into a damaging one.
Credit
Kim Kwang-Hee and colleagues; Francesco Grigoli and colleagues; William Ellsworth, Domenico Giardini, John Townend, Shemin Ge and Toshihiko Shimamoto
2018
Helsinki stays under magnitude 2
In June and July 2018 the energy company St1 pumped 18,160 m³ of fresh water over 49 days into crystalline rock through a 6.1 km well in Espoo, near Helsinki, under a limit set by local authorities: a quake of magnitude 2.0 would stop the project. A network of 24 borehole seismometers located quakes in near real time, and when activity grew the engineers paused pumping, varied the wellhead pressure between 60 and 90 MPa and added rest periods. The largest quake was magnitude 1.9, and the team published its catalog of quakes on GFZ Data Services.
Lessonmonitoring that feeds straight back into the pumping plan can keep quakes under an agreed limit.
Credit
St1 (Tero Saarno); Grzegorz Kwiatek, Georg Dresen, Marco Bohnhoff and colleagues at GFZ; fastloc GmbH
Watch for
Haute-Sorne's quieter rock
Geo-Energie Jura says a six-day stimulation test in July 2025 injected 430 m³ of water at 3,800 m depth, set off about 3,400 microquakes, the largest of local magnitude 0.3, raised the rock's permeability more than a hundredfold, and showed the rock there to be more than 60 times less seismically reactive than Basel's. The Swiss Seismological Service monitored the surface independently for the canton. In October 2026 the project was still waiting for the Jura government's permission to drill a second well and stimulate a full reservoir.
Achieved whenthe full multi-stage stimulation is carried out without felt quakes, as reported by the Swiss Seismological Service or the canton's independent expert group.
Claimed by
Geo-Energie Jura, part of Geo-Energie Suisse, October 2025
Capturing nearly all the gas
Carbfix dissolves in water the carbon dioxide (CO2) and hydrogen sulfide (H2S) that come up with the steam at Reykjavík Energy's Hellisheiði plant in Iceland, and pumps them into basalt, where they turn into minerals. It says that between 1 July 2025 and 30 June 2026 its Silverstone capture plant at Hellisheiði dissolved 99.6% of the CO2 and all of the H2S that entered the capture system, more than 19,000 tonnes of CO2, and sent it into basalt through old wells now used for reinjection. The figure covers gas entering the capture system, not necessarily all the gas the power plant releases.
Achieved whenan independent verifier or Iceland's environment agency publishes the plant's total emissions and the share captured and stored.
Claimed by
Carbfix, with ON Power, September 2026
Open
01
Ground motion at every field
A public map and table of how fast the ground has risen or sunk since 2015 over every producing geothermal field in the world, measured by the Sentinel-1 radar satellites, set beside each field's production and reinjection where those are public, and checked against ground surveys.
The briefCredit · unclaimed02
One scorecard for quake forecasts
An open benchmark that replays at least five past geothermal stimulations hour by hour, with injection records and earthquake catalogs in one format, and scores three or more published forecasting methods on how early and how well they warned of the largest quakes.
The briefCredit · unclaimed
Open 01 · the brief
Ground motion at every field
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public map and table of how fast the ground has risen or sunk since 2015 over every producing geothermal field in the world, measured by the Sentinel-1 radar satellites, set beside each field's production and reinjection where those are public, and checked against ground surveys.
Why it is still open
Radar studies of geothermal ground motion are done one field or one region at a time. A study of Sarayköy in Türkiye, presented in 2025, measured local sinking of up to 0.7 m a year, and a 2023 study fed radar maps and injection data at The Geysers into a machine-learning forecast. Lawrence Livermore National Laboratory researchers wrote in 2018 that Sentinel-1 and open software would allow routine processing over geothermal regions "in the Western U.S. and worldwide", yet a November 2026 search found no worldwide comparison. Ready-made ground-motion maps now cover Europe, through the European Ground Motion Service, and North America, through NASA's OPERA displacement product.
isprs-archives.copernicus.orgosti.govosti.govland.copernicus.eujpl.nasa.gov
Built with what exists
European Ground Motion Service and OPERA DISP-S1 data; Sentinel-1 images from the Alaska Satellite Facility for other regions; open radar processing software such as GMTSAR, MintPy or LiCSBAS; Global Energy Monitor's geothermal power tracker for field locations; New Zealand's published level surveys as a check.
First step
Download the OPERA ground-motion product over The Geysers and compare it with the radar maps in the 2023 study of that field.
Done looks like
Ground-motion rates for at least 100 producing fields, published with code; at least five fields compared with independent level or GPS surveys, agreeing within 10 mm a year.
Open 02 · the brief
One scorecard for quake forecasts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open benchmark that replays at least five past geothermal stimulations hour by hour, with injection records and earthquake catalogs in one format, and scores three or more published forecasting methods on how early and how well they warned of the largest quakes.
Why it is still open
The Induced Seismicity Test Bench (Király-Proag and colleagues, 2016) scored two models on two projects, Basel 2006 and Soultz-sous-Forêts 2004, and its authors note how few projects that is. Later work tests one or two sites at a time: a replay of stage 3 of the 2022 Utah FORGE stimulation (Lanza and colleagues, 2026), and a deep-learning forecaster trained on one Geysers well pair and the 2024 FORGE stimulation (Bi, 2025). A November 2026 search found no shared benchmark across projects. Open data now exist for several more: curated FORGE injection records for 2022 and 2024, published in February 2026; a machine-picked catalog of the April 2024 FORGE quakes; and the Helsinki catalog.
arxiv.orgdoi.orgzenodo.orggdr.openei.orggdr.openei.orgpmc.ncbi.nlm.nih.gov
Built with what exists
Those datasets, plus Swiss Seismological Service catalogs for Basel and St. Gallen where released; open forecasting codes, including ETAS models (statistical models of how quakes trigger aftershocks) and Bi's transformer code; pyCSEP for scoring; Python.
First step
Line up the curated April 2024 FORGE injection record with the machine-picked catalog of the same stimulation, hour by hour, and rerun one published model on it.
Done looks like
At least five stimulations in one open format; at least three methods scored with the same tests; for each project, the hour at which each method first gave at least a 10% chance of passing the project's red-light magnitude, set against the hour it was actually passed, or not.