Most of what we know about the Earth’s interior comes from following seismic waves. These travel at somewhat different speeds depending on the details of the rock they’re moving through—whether it’s solid or semi-molten, how much water is present, whether it’s fractured or solid material, and so on. Get enough data from enough seismic events, and you can start piecing together a picture of what’s present at different depths below the surface.
In many cases, we can get this data from naturally occurring events like earthquakes. In others, we intentionally create waves using things like explosives, providing the opportunity to do imaging in specific areas without needing to wait for an earthquake. Now, a team of scientists at Penn State suggests there’s a potential option that sits between waiting for an earthquake and triggering your own seismic event: thunderstorms.
Some of the energy carried by thunder enters the Earth’s upper crust, triggering what are termed “thunderquakes.” But, for a variety of physical reasons, the seismic signals are extremely complex, making it difficult to extract clear signals from it. The Penn State team says it has finally constructed a model that can help make sense of this complexity and used it to reconstruct the terrain under the local campus.
Managing complexity
Why are thunderquakes so hideously complex? It starts with the phenomenon that creates thunder in the first place. Lightning creates thunder by forming superheated bubbles of plasma along its path, creating a structure that has been compared to a string of beads. Each of those beads has the potential to generate an acoustic shock wave, leading to a chain of expanding shock waves that trace the lightning’s path through the area, which is anything but a straight line. These waves also have the potential to interfere with each other as they expand. And, while these shock waves first hit the Earth at a single point, they rapidly expand from there, albeit with decreasing power.
Things don’t get less complex once the Earth gets involved. The acoustic shock waves may strike soft soil, hard rock, various forms of human infrastructure, and so on, each of which will affect how energy gets transmitted. Some of the energy gets converted into what are called Rayleigh waves, where the energy is transmitted as a wave that moves along the Earth’s surface. The rest go deeper, forming waves that may move through some combination of loose material or the underlying bedrock.
To extract information about the Earth’s structure, you have to understand what the seismic waves from a thunderclap would normally look like. Which, to an extent, requires modeling all of the above processes. Since each thunderquake is going to be unique due to the different locations and conditions, this model is going to be, at best, an approximation. The fear that any approximation wouldn’t be good enough to generate usable data probably kept people from trying to analyze thunderquakes sooner.
To get their approximation, the team started with a software package called SPECFEM3D Cartesian, which is dedicated to 3D reconstructions of seismic waves. Already, that choice necessitates a few compromises. For example, the software treats the atmosphere as a 3.6km-thick homogeneous layer, even though the atmosphere near a thunderstorm is anything but. The model also updates events at a frequency that’s slower than the waves moving through the Earth-air interface. So, to compensate for that, the researchers simply stretched the top 20 meters of Earth out to cover 200 meters.
These and other factors mean that there were plenty of reasons to think that the model wouldn’t be sufficient to handle real-world data. So, the people who developed it tested it against the real world, using thunderstorms that passed by their campus.
Passing the test
One of the nicer discoveries in seismology has been the realization that the same fiber-optic cables that rush cat pics to your LAN can act as seismometers. And, conveniently, the Penn State campus has a 4 kilometer fiber line that has been set aside for seismic sensing running under the campus. And said campus happens to be located in a part of the US where summer thunderstorms are a regular occurrence.
Two years of data netted them 458 well-resolved thunderquakes, each of which was confirmed using records from the US’s National Lightning Detection Network (something I had not realized existed). These quakes were characterized by multiple signals arriving from different altitudes, as you’d expect from a chain of beads reaching from clouds to the Earth’s surface. Once the signals arrived at the Earth’s surface, things happened quickly: “The impingement of each bubble onto the ground or environment generates a high-energy impulsive wavelet followed by a decaying wave train dominated by surface-wave content lasting one to two seconds.”
From there, the signal spread out and started to interact with the features of the Earth under the campus. Using this data, the team identified four “weak zones,” where seismic signals slow down as they interact with less rigid materials. These can include sediments, fractured rock, or areas with high water content. The Penn State campus happens to sit on a karst formation, where water has slowly altered limestone bedrock, potentially creating a variety of weak spots.
In these cases, the team was able to confirm that these four sites actually have something unusual going on there. This was done using a mixture of radar that measured surface deformation, engineering surveys, boreholes made at the sites, and independent seismic data.
All of which gives the researchers confidence that, despite all the approximations it required, their model is performing reconstructions that are sufficiently accurate to obtain informative seismic data. And the thunderquakes have a number of advantages, including their relative frequency in many areas of the globe, and the fact that they’re best for reconstruction of the areas closest to the surface, which is where almost all of our infrastructure is located.
So, this definitely appears to be a case where we have a model that’s wrong, but also useful.
Science Advances, 2026. DOI: 10.1126/sciadv.aeg8096 (About DOIs).
