Science1 distinct publisher3 min readPublished
Bei Wang's team searched backwards from each magnitude 3 quake and nearly always found a smaller precursor. That number measures how often a warning existed after the fact. A shutdown rule needs a different number: how often such a warning would have meant anything in advance.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
Start with what this design can and cannot deliver. The team began from mainshocks that had already happened and looked backwards in space and time [5], so 92 percent answers a retrospective question: given a magnitude 3 event, was something smaller detectable nearby first [6][19]. A traffic light asks the forward version, which is the chance that a damaging quake follows a handful of small events under an active pad. Answering that requires counting the swarms that ended in nothing, and those are not in this denominator. The authors put it more gently, noting that foreshocks may generate false positives as well as false negatives [8].
The catalogue arithmetic hints at why the false positives are the expensive half. Seventy-seven qualifying mainshocks across ten years is about 7.7 a year [21], and they account for roughly 0.11 percent of the events the network recorded over that span [22]. Small earthquakes near injection wells happen all the time, and only a rare few of them are worth halting operations for.
Then there is the spread inside the 92 percent. A protocol has to pick a tripwire, and the paper offers no basis for choosing one. Set it at a single detected event and the yellow light fires close to continuously in an active play, with each pause costing pumping time. Set it at a swarm and you skip the mainshocks that arrived behind one small quake, since the observed precursor counts ran from one event to about 700 inside the same five-kilometre, five-day window [7].
Wang and colleagues offer three routes from injection to mainshock: fluid progressively weakening the main fracture so that it creeps, strain building across the region until it fails suddenly, or a domino sequence of slip on several fractures [12]. Ryan Schultz of ETH Zurich, who was not part of the work, notes these are ideas about earthquake-earthquake interaction that date to the 1990s and remain hypotheses, and that different geological systems can favour different triggering mechanisms [13]. Absent a mechanism, 92 percent is a property of the Western Canada Sedimentary Basin measured over one decade [1], not a constant to hand to an operator somewhere else.
The source material contains its own illustration of that limit. The magnitude 5.6 event near Peace River in 2022, the one large enough to shake a nearby community, was initiated by wastewater injection during oil extraction rather than by fracking [18], while all 77 mainshocks analysed here are fracking-linked [4]. Whatever foreshocks do around fracking pads, this study did not characterise them around the injection activity that produced the basin's most consequential recent quake.
Schultz's summary is that the paper mainly raises a problem, quantifying how often operations can jump from green straight to red with no intervening warning [15], and that improving the protocol starts with defining what damage counts as unacceptable and setting the red light some distance behind it [16]. That ordering is right, and it puts the foreshock statistic where it belongs, as an input rather than the calibration. The honest reading is narrow and still worth having: in this basin, over this decade, most damaging fracking quakes announced themselves first [1], and nobody yet knows how to distinguish an announcement from background chatter [8].</body_markdown> </invoke>
Ranked by verification strength, evidence, and original report placement.
Analyses of a decade of seismic data from Alberta's Western Canada Sedimentary Basin show that 92 percent of earthquakes induced by fracking were preceded by smaller foreshocks, reported August 27 in Science.
The work was led by Bei Wang, a geophysicist at Zhejiang University of Technology in Hangzhou, China, and his colleagues.
Wang and his team analyzed seismic data from western Canada from 2014 to 2024, which included about 70,000 earthquakes.
Of those events, the team identified 77 mainshocks of at least magnitude 3 that were linked to fracking.
The team then hunted for foreshocks, smaller quakes that might have occurred within 5 kilometers and five days of each mainshock.
Foreshocks occurred before 71 of the 77 fracking-linked mainshocks.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Canva's forecast cut turns model routing into a product line item1 distinct publisher
science
Graz group measures the geometry a single-molecule collision has to hit1 distinct publisher
science
Sleep tech turns rest into evidence, and employers are the ones who will need rules1 distinct publisher
science
Two cases a year: a Louisiana amoeba death tells northern states little about their own lakes1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One peer-reviewed decade-scale catalogue study, one basin, retrospective design
The core numbers come from a Science paper analysing ten years and roughly 70,000 events in a single basin, with an explicit event-selection rule (magnitude 3+, fracking-linked) and a stated foreshock search window (5 km, 5 days), plus commentary from an independent seismologist. That is solid for a first quantification. It is bounded by being one study, one region, one reporting outlet, retrospective by construction, with no reported false-alarm base rate, lead-time distribution or catalogue-completeness threshold, and with the underlying operator pumping data not openly available.
Underlying protocol already in wide regulatory use; this study's implications not yet acted on
The practice the paper evaluates is genuinely deployed: the article reports that regulatory agencies around the world run traffic-light protocols and that foreshocks are the primary way operators learn the light has turned yellow, which is real-world usage rather than a proposal. Adoption is held below high because the cluster gives no counts of jurisdictions, operators or shutdown events, and contains no evidence that any regulator or operator has changed thresholds in response to the 8 percent no-warning finding.
Careful reporting, but a retrospective recall figure reads as prospective warning reliability
Mildly overstated rather than inflated. The 92 percent was obtained by searching backwards from known mainshocks, so it measures how often a precursor existed in hindsight, not how often a detected precursor implies an imminent mainshock; the cluster supplies no false-alarm base rate, no lead-time distribution and foreshock counts varying from one to about 700, all of which weaken any read as alert reliability. The gap is small because the source itself foregrounds the 8 percent no-warning case, the false positive/negative caveat, and an independent expert who reduces the contribution to 'raising a problem'.
Independent expert with disclosed non-involvement; commercial data held privately
The commentary source is disclosed as not involved in the study and pushes against the most flattering reading, which is a favourable signal. Working against verification: the seismic and pumping data that would test the mechanisms is collected by the companies whose operations the protocol constrains and is not reliably open, varying by country and by US state, and no operator or regulator is on record in this cluster. Academic publication incentives around a headline percentage are present but unquantified.
Consistent single-outlet account of one study, key operational numbers missing
The factual spine (sample sizes, search window, 71 of 77, mechanisms, expert caveats) is internally consistent and specific, and includes independent commentary. Confidence is capped by having one publisher and one primary study in the cluster, no direct access to the paper's supplementary detail, no second basin or replication, and no data on false alarms, lead times or catalogue completeness, which are precisely the quantities that would determine what the finding means operationally.