Yellowstone, a popular tourist destination and equally popular television programme, was the first national park in the United States. And bubbles beneath it – to this day – is one of the most seismically active networks of volcanic activity on the planet.
A new study published in the High Impact Journal on July 18th Advances in scienceWestern Engineering Professor Bing Lee and his collaborators used machine learning to review historical earthquake data for the Yellowstone Caldera over 15 years at the Santander University Industry (Santander University) in Colombia and the US Geological Survey. The team was able to retrospectively detect and assign magnitudes to approximately 10 times more seismic events or earthquakes than previously recorded.
Calderas like Yellowstone Park, spanning parts of Wyoming, Idaho and Montana, form a major depression or hollow when the volcano erupts, emptying the magma chamber below it, leading to the collapse of the land above. This is different from the volcanic crater formed by outward explosions.
The Yellowstone Caldera historic catalogue includes 86,276 earthquakes from 2008 to 2022, significantly improving previous understanding of volcanic and seismic systems through better data collection and systematic analysis.
An important finding in this study is that more than half of the earthquakes recorded in Yellowstone were part of the earthquake swarm – a group of small, interconnected earthquakes that spread within a relatively small area for a relatively short period of time. This is different from aftershocks, which are small earthquakes that follow larger main shocks in the same general area.
“Although Yellowstone and other volcanoes each have unique characteristics, we hope that these insights can be applied elsewhere,” says Li, an expert on liquid-induced earthquakes and rock mechanisms. “Understanding earthquake patterns like earthquake swarms can improve safety measures, inform the public better of potential risks, and even guide the development of geothermal energy from the dangers of areas with heat flow.”
Melt detector
Earthquakes were commonly detected by manual inspection by trained experts prior to applying machine learning. This process is time-consuming and expensive, and often detecting fewer events than machine learning could. Machine learning has in recent years triggered a data mining gold rush as seismologists revisit historical waveform data stored in data centers around the world and learn more about current and previously unknown seismic regions around the world.
“If you had to do an old school with someone who was manually clicking all this data looking for an earthquake, you couldn't do it. It's not scalable,” Li said.
This study also shows that earthquake swarms beneath the Yellowstone caldera occurred along relatively immature and coarse fault structures compared to more typical mature fault structures found in areas such as Southern California and just outside the caldera.
Roughness was measured by characterizing earthquakes as fractals. Fractals are geometric shapes that exhibit self-similarity and appear to be similar on different scales. Fractal patterns, first visualized by Benoit Mandelbrot in 1980, can also be found in shorelines, snowflakes, broccoli, and even in the branches of blood vessels. Fractal-based models targeting roughness and regularity were able to characterize these earthquake swarms. Researchers believe it was caused by a sudden liquid mix with slow moving groundwater.
“For the most part, there is no systematic understanding of how one earthquake causes another in a herd. We can indirectly measure the space and time between events,” Li said. “But now we have a much more robust catalog of seismic activity under the Yellowstone Caldera, and we can apply statistical methods that will help us to quantify and find, study, and see that new herds we have never seen before can be quantified and learn from them.”
