AI reveals hidden patterns in 15 years of Yellowstone earthquake swarms
A 15-year analysis of Yellowstone’s seismic record, published in Science Advances, has uncovered how thousands of small earthquakes cluster in time and space, revealing complex interactions between hydrothermal fluids and fault systems beneath the caldera.
The Grand Prismatic hot spring in Yellowstone National Park is sourced from a magma chamber beneath it. The bright colours are produced by hydrophilic bacteria in the mineral-rich water. Credit: Bing Li
A new long-term study by the Yellowstone Volcano Observatory and the U.S. Geological Survey (USGS) has revealed that the ground beneath Yellowstone National Park is far more dynamic than previously understood.
By using artificial intelligence to reexamine 15 years of seismic data, researchers uncovered how thousands of tiny earthquakes interact through time and space, forming a complex, living network of movement deep within the Earth.
This effort marks a major advance in understanding the forces that shape Yellowstone’s unique combination of volcanic, hydrothermal, and tectonic activity.
From 2008 to 2022, scientists analyzed continuous ground motion recordings collected from one of the world’s most sophisticated seismic networks. The system surrounds the park and detects even the faintest vibrations, often from earthquakes too small for humans to feel.
By applying advanced machine learning models to this enormous dataset, the researchers detected over 86 000 earthquakes, an order of magnitude more than in the routine catalog used for daily monitoring. These discoveries have expanded the known seismic history of Yellowstone by revealing countless small events that were previously invisible. Each one represents a small but significant pulse in the planet’s internal rhythms.
Through this data, Yellowstone’s subsurface appears as a restless landscape in constant motion. Earthquakes are not distributed evenly but instead cluster in bursts known as swarms, where hundreds or thousands of events occur over days or weeks.
Some of these swarms have been linked to the movement of hydrothermal fluids, others to the slow readjustment of fault zones, and a few to deeper volcanic processes. The AI-based catalog now shows that many of these bursts are connected across years or even decades, telling a much longer story of Yellowstone’s evolving crust.
This study, published in Science Advances in July 2025, demonstrates how long-term, high-resolution monitoring combined with artificial intelligence can illuminate processes far below the reach of direct observation. The findings mark a turning point in how scientists study subtle seismic behavior in active volcanic regions around the world.
Mapping Yellowstone’s deep and shifting structure
The new seismic catalog provides an unprecedented look at how earthquakes outline the anatomy of the Yellowstone Caldera.
Most of the 86 000 recorded earthquakes occurred at depths between 1 and 4 km (0.6 to 2.5 miles), forming dense, linear clusters that trace fault systems beneath the park. Beneath about 8 km (5 miles), seismic activity nearly disappears, suggesting the presence of partially molten rock that absorbs energy rather than fracturing.
Researchers achieved this precision using a three-dimensional model of seismic wave velocities beneath Yellowstone. This model accounts for how different rock types and temperatures alter the speed of seismic waves, allowing scientists to locate earthquakes with accuracy measured in hundreds of meters rather than kilometers.
The refined results reveal intricate fault corridors, some trending northeast across the caldera and others running along its western rim near Hebgen Lake.
Within these structures, scientists observed distinct differences between areas inside and outside the caldera boundary. Inside, earthquakes tend to migrate upward through rough, immature faults. This vertical movement reflects the rising motion of pressurized fluids, primarily hot water and gases, within Yellowstone’s active hydrothermal system.
Outside the caldera, by contrast, the faults appear smoother and more stable, with earthquakes showing little vertical migration. These zones likely represent older, more mature faults responding to lateral shifts in crustal stress.
The use of deep learning tools such as EQTransformer and PhaseLink enabled the detection of subtle seismic phases, small compressional and shear waves, that traditional methods often miss. More than 4.7 million individual phase arrivals were automatically picked from continuous waveforms.
By combining these detections with the advanced velocity model, researchers relocated 67 000 events with remarkable precision, mapping a complex and finely layered crust that bridges the worlds of tectonics and volcanism.
This level of detail provides a new foundation for studying Yellowstone’s subsurface. The data reveal not just where earthquakes happen, but how their patterns shift in depth and direction through time. Each cluster becomes a trace of movement, showing how the park’s internal plumbing continues to evolve.
Swarms that return after years of silence
One of the most remarkable findings is the long-term connection between separate earthquake swarms. While individual swarms typically last only a few weeks, the new analysis shows that swarms years apart often occur in nearly the same place.
For example, the 2020–2021 swarm near the northern end of Yellowstone Lake occurred immediately south of the 2008–2009 sequence, separated by more than a decade of quiet. This recurrence suggests that fluids within the crust may move in short bursts, activating one area before pausing and reappearing nearby years later. Another recurrence occurred at Maple Creek, northwest of the caldera, where major swarms in 2017 followed earlier activity after nearly a decade-long lull.
Such behavior hints at a dynamic system of underground reservoirs where water and gas migrate slowly through fractures in the rock. When these fluids encounter sealed zones of lower permeability, pressure builds until it forces the rock to crack, producing bursts of seismicity.
Once released, the system quiets again as the fluid pathways reseal. This stop-and-go process creates a rhythmic pattern of swarms that are spatially linked but temporally separated.
The study found that over half of Yellowstone’s earthquakes belong to these swarm-like families rather than typical mainshock-aftershock sequences. Swarms are especially common near hydrothermal areas such as Yellowstone Lake and Norris Geyser Basin, where hot water circulates through shallow fractures. The upward migration of earthquakes within the caldera further supports the role of ascending fluids as a primary driver of seismic activity.
These recurring swarm patterns demonstrate how Yellowstone’s crust stores and releases energy on timescales much longer than any individual episode. The return of activity to the same fault zones after years of rest suggests a slow cycle of pressure accumulation and release that shapes the park’s ongoing evolution.
The interplay of magma, fluids, and fault networks
Beneath Yellowstone lies a complex interplay between molten rock, groundwater, and brittle faults. The shallowest 4 km (2.5 miles) of crust hosts most earthquake swarms, while deeper regions contain a partially molten reservoir that feeds the park’s famous geysers and hot springs.
Between 4–8 km (2.5–5 miles) lies an aseismic zone—an area of limited earthquake activity—interpreted as a transition between brittle rock and ductile, fluid-rich material.
The July 2021 swarm under Yellowstone Lake offered a striking example of this layered structure. It featured both deep and shallow clusters separated by an aseismic gap about 3–4 km (1.8–2.5 miles) thick. The deeper earthquakes originated near 9 km (5.6 miles), while the shallower ones occurred above 4 km (2.5 miles).
Scientists interpret this as evidence of gas and fluid transfer from the partially molten region upward into cooler, brittle rock, where the fluids triggered fault movement.
Within the caldera, these vertical migration patterns dominate. Earthquakes expand upward as fluids rise through immature fault systems, carving complex, rough surfaces. Outside the caldera rim, however, migration is mostly lateral, as pressurized fluids move sideways into cooler zones, activating more mature, planar faults.
This difference helps explain why swarm behavior inside Yellowstone differs from that in surrounding tectonic regions.
Understanding these interactions is essential for assessing the park’s ongoing evolution. The connection between heat flow, fracturing, and fluid movement determines not only how earthquakes cluster but also how the hydrothermal system renews itself.
It explains why geysers shift activity zones and why deformation patterns change from uplift to subsidence over time. The new study shows that these processes are continuous, even during periods that appear quiet at the surface.
These observations reinforce the idea that Yellowstone’s seismicity is not a sign of imminent volcanic activity but of an active and self-regulating system. The crust behaves like a breathing organism, pulsing with cycles of pressure release driven by its internal fluids.
Why long-term monitoring matters more than ever
The new 15-year catalog is more than a record of past activity; it’s a guide to future understanding. By linking decades of data, scientists can now view Yellowstone’s seismicity as a connected story rather than a series of isolated episodes. This broader perspective reveals how fault zones evolve, how hydrothermal systems shift, and how stress redistributes across the region.
Long-term monitoring also refines hazard assessment. While none of the observed patterns suggest magma movement that could precede an eruption, they provide vital clues about how pressure accumulates and releases in the crust. Tracking these slow processes can help identify early signs of hydrothermal changes, ground deformation, or larger earthquake sequences.
The USGS and the Yellowstone Volcano Observatory emphasize that continued investment in seismic monitoring is paying off. Each year adds to a growing record that sharpens the image of Yellowstone’s underground dynamics. These efforts not only improve local safety but also serve as a model for volcano monitoring worldwide.
The application of artificial intelligence to this long-term dataset demonstrates a new era in seismology. As AI tools become more capable, they can detect patterns too subtle for traditional analysis, revealing how small quakes interact over time. This combination of human expertise and computational power will continue to transform how scientists study the deep Earth.
Ultimately, Yellowstone’s restless ground offers a reminder that even in apparent calm, the forces shaping our planet never stop. The park’s hidden earthquakes tell a story of energy transfer, heat, and motion—a story now being decoded with the help of technology that listens to the Earth more closely than ever before.
References:
1 Long-term dynamics of earthquake swarms in the Yellowstone caldera – Manuel A. Florez et al. – Science Advances – July 18, 2025 – DOI: 10.1126/sciadv.adv6484 – OPEN ACCESS
2 Patterns of earthquakes over time in Yellowstone highlight the complexity of seismic swarms – Yellowstone Volcano Observatory – December 1, 2025
I’m a science journalist and researcher at The Watchers, contributing to the Epicenter edition, where I cover peer-reviewed scientific research and emerging discoveries across Earth and space sciences. With a background in astronomy and a passion for environmental science, I’ve worked in shark and coral conservation in Fiji, conducting reef and shark-behavior research, contributing to mangrove restoration, and earning PADI Open Water and Coral Reef Certifications. I bring a blend of scientific rigor and storytelling to illuminate the discoveries shaping our planet and beyond.
Commenting rules and guidelines
We value the thoughts and opinions of our readers and welcome healthy discussions on our website. In order to maintain a respectful and positive community, we ask that all commenters follow these rules.