Yellowstone Caldera Chronicles is a weekly column written by scientists and collaborators of the Yellowstone Volcano Observatory. This week’s contribution is from David Shelly, seismologist with the U.S. Geological Survey.
The Yellowstone region generates a lot of earthquakes. Although most of these earthquakes are too small to be felt by humans, they can be studied using data from Yellowstone’s dense, high-quality network of seismometers. This network, located both within and surrounding the national park, records extremely subtle ground motions from up to thousands of (mostly tiny) earthquakes every year. Many of these earthquakes are routinely detected and located by University of Utah Seismographic Stations (UUSS), but others are recorded only weakly and aren’t detected on enough stations to be included in the earthquake catalog.
Earthquakes at Yellowstone are not distributed evenly in space and time—rather, they can occur in bunches and are often focused on specific areas. The patterns of how earthquakes occur can tell us a great deal about processes active well beneath the surface. Although routine real-time seismic monitoring does an excellent job of revealing many of these patterns, others only become apparent when applying specialized analyses on previously recorded data. Such techniques can detect many more earthquakes than routine processing and locate these earthquakes much more precisely, enhancing our view of the their causes.
While some individual time periods of particularly high rates of seismicity (like during earthquake swarms) have been examined previously, a study (https://doi.org/10.1126/sciadv.adv6484) published earlier this year in the journal Science Advances took a longer-term view. Rather than focusing on a time period of just a few days, weeks, or months, the new research analyzed 15 years of seismic data, from 2008 through 2022. Starting with raw, continuous seismic waveforms (records of ground motion) from the network of seismometers around Yellowstone, the effort took advantage of recent developments in artificial intelligence, applying a machine-learning approach to sift through this vast dataset and identify seismic phase arrivals (P and S waves) from nearby earthquakes. This technique detected very small earthquakes that are not identified by the routine system.
Map view of relocated Yellowstone seismicity recorded during 2008-2022, colored by time, on the left. White line gives the outline of Yellowstone caldera. The A-A’ cross section through Yellowstone Lake is shown at the right and illustrates how seismic swarms that are distinct in time relate to one another. Adapted from Florez and others, 2025 (https://doi.org/10.1126/sciadv.adv6484).
Map of seismic stations in the Yellowstone region, with numbers of channels indicated by number and sensor type by color. Inverted triangles indicate stations operated by University of Utah Seismograph Stations (UUSS), and squares indicate stations operated by other agencies.
After computing magnitudes, the resultant earthquake catalog for 2008–2022 contained 86,276 events, more than 10 times as many as many as included in the routine earthquake catalog. This is not really a surprise, as it is clear that lots more earthquakes happen in the Yellowstone region, and really worldwide, than can be located by current real-time seismic analysis, but these events are all very small—typically less than M1.5. Earthquake locations then were improved by utilizing a recently produced model of three-dimensional subsurface seismic velocities and by applying precise relative relocation techniques that carefully compare the timing of seismic waves among nearby earthquakes. In total, 67,433 earthquakes were successfully relocated.
This catalog provides a new window into longer-term earthquake behavior at Yellowstone. One striking feature is the long-term connection between short-term earthquake swarms. Although individual swarms typically last days to weeks (and occasionally months), this study found that swarms separated in time by many years may occur adjacent to each other in space. Understanding this behavior is still a topic of ongoing research, but one likely explanation is related to fluids (primarily water) at depth. Fluid movement in the subsurface may trigger an earthquake swarm. But what if the fluid moves some distance and then stops? Adjacent swarms separated by years might indicate the reactivation of dormant fluids. An example of such behavior is swarm activity near the northern end of Yellowstone Lake in 2020-2021, which was just to the south of the 2008-2009 swarm. Other zones, such as the Maple Creek area just outside of Yellowstone caldera to the northwest, have also exhibited spatially adjacent swarm activity with a substantial pause in between individual swarms.
Earthquakes at Yellowstone are generated by a fascinatingly complex interplay of tectonic, hydrothermal, and volcanic processes. These new results help us see the bigger picture of Yellowstone seismicity, connecting smaller snapshots of activity from previous detailed, but short-duration studies of individual swarms. Consequently, we now have a better understanding of the relations between earthquake swarms that may be separated by many years.
Although Yellowstone surely has more surprises in store, long-term investments in monitoring are paying off. Seismic data, together with other valuable geophysical and geochemical datasets, are continuing to unveil active volcanic, hydrothermal, and tectonic processes hidden deep beneath the surface. As we build a longer and longer record of this behavior, our understanding of the Yellowstone system will continue to sharpen.