Morning Overview

AI combed old sensor data and found 86,000 hidden earthquakes under Yellowstone.

A deep-learning algorithm applied to 15 years of seismic recordings beneath Yellowstone has identified 86,276 earthquakes that standard monitoring missed. USGS scientist David R. Shelly and colleagues reprocessed continuous waveforms spanning January 2008 through December 2022, producing a catalog roughly an order of magnitude larger than what routine detection had captured over the same period. The result redraws the picture of how fluids move beneath one of the most closely watched volcanic systems on Earth.

Why 86,000 hidden quakes change the Yellowstone conversation

Yellowstone typically registers between 1,000 and 3,000 located earthquakes per year, almost none of them felt at the surface. The Yellowstone Volcano Observatory and the National Park Service have long noted that the vast majority of these events are too small to cause damage. Against that baseline, a catalog containing 86,276 earthquakes over 15 years represents a dramatic jump in resolution. Most of the newly detected events were tiny, sitting below the magnitude threshold that human analysts would flag during routine review.

The practical consequence is not a higher eruption risk. Instead, the denser record lets researchers track how earthquake clusters migrate through the crust over days and weeks. Roughly 52% of the cataloged events fell within swarm sequences, according to the study published in Science Advances. That proportion matters because swarm behavior in volcanic settings is often tied to the movement of pressurized fluids rather than the slow buildup of tectonic stress. If swarm migration velocities line up with independent measurements of hydrothermal discharge and GPS-measured ground deformation, fluid-pressure pulses would emerge as the dominant trigger. Testing that link is the next scientific step the catalog enables.

How EQTransformer detected what human analysts could not

The team used a pretrained deep-learning model called EQTransformer to pick P-wave and S-wave arrival times from continuous data streams recorded across the Yellowstone seismic network. EQTransformer was originally described in a 2020 paper in Nature Communications as an attentive model designed for simultaneous earthquake detection and phase picking. Trained on a global dataset, the algorithm can identify faint seismic signals buried in background noise that a human reviewer scanning thousands of hours of waveforms would likely skip.

Earlier work on individual Yellowstone swarms had already shown that careful waveform analysis could reveal fluid-driven earthquake migration. A peer-reviewed study of the 2010 Madison Plateau swarm, for example, used precise relocation techniques to argue that fluids were driving seismicity along the caldera margin. What the new catalog adds is temporal breadth. By covering 2008 through 2022 in a single, consistent pass, the algorithm eliminates the patchwork quality of earlier analyses that examined one swarm at a time with different methods and detection thresholds.

David R. Shelly, identified in the USGS publication record as a co-author and agency scientist, has focused on applying waveform-based detection to volcanic and fault systems. The institutional release accompanying the study noted that the expanded catalog allows researchers to identify swarms that were not previously detected, moving beyond manual review toward a more complete picture of subsurface activity.

Gaps the new catalog does not yet fill

Several questions remain open. The Science Advances paper does not include a side-by-side comparison table matching the new 86,276-event catalog against the official USGS ComCat earthquake listings for the identical 2008 to 2022 window. Without that direct comparison, the precise multiplier over standard detection remains an estimate rather than a verified figure. Whether the full set of relocated hypocenters and waveform picks will be released as a public dataset is also not confirmed in either the journal article or the USGS warehouse entry.

More consequentially, neither the peer-reviewed text nor the accompanying institutional release states whether the additional events change short-term hazard assessments for ground failure, hydrothermal explosions, or magmatic eruption. The Yellowstone Volcano Observatory has consistently characterized the system’s eruption probability as very low on human timescales, and nothing in the new catalog contradicts that position. But the study stops short of formally updating any hazard metric.

The next development to watch is whether independent research groups can cross-reference the swarm migration patterns in this catalog with GPS dilatation records and hydrothermal outflow data from Yellowstone’s geyser basins. If migration velocities track fluid-pressure changes rather than tectonic loading cycles, the finding would strengthen the case that Yellowstone’s frequent swarms are symptoms of a plumbing system flushing hot water through fractured rock, not precursors to volcanic unrest. That correlation test, drawing on datasets already collected by USGS and university partners, is the clearest path from a bigger earthquake list to a sharper understanding of what the caldera is actually doing.

More from Morning Overview

*This article was researched with the help of AI, with human editors creating the final content.