On April 5, 2024, a magnitude 4.8 earthquake struck near Tewksbury, New Jersey, surprising East Coast residents. More than 180,000 people reported feeling ground shaking — the largest number of respondents to the U.S. Geological Survey (USGS) online “Did You Feel It?” reporting tool ever. Millions of people from Virginia to Maine felt the motion, including numerous reports from across the New York metropolitan area, about 65 kilometers away from the epicenter.

Curiously, damage from the earthquake wasn’t as extensive as might be expected near the epicenter for an event of this magnitude. A paper published in The Seismic Record noted that for a magnitude 4.8 earthquake in the central and eastern U.S., we’d expect people to experience a shaking level of VII on the Modified Mercalli Index (MMI) scale, which corresponds to very strong shaking. Yet, a maximum intensity of VI was “hardly reported.” Another paper, also published in The Seismic Record, reported the partial collapse of a 264-year-old pre-Revolutionary War structure about 4 kilometers south of the earthquake’s epicenter. But, damage elsewhere, per the second paper, was generally restricted to drywall cracks or objects falling from shelves, corresponding to moderate shaking.

One way that scientists help protect people before an earthquake strikes is to simulate how the ground might move in various earthquake scenarios using computer-generated models. To know whether a model is accurate, scientists can compare their predicted results to real data collected from stations in the region of interest. However, for the central and eastern U.S., this process is especially tricky because earthquakes large enough to feel — with magnitudes high enough to cause damage — are relatively rare.

One important input into these simulations of shaking is a model of Earth’s velocity structure. In a new study published in Seismological Research Letters, a team of scientists led by Oliver Boyd of the USGS used a three-dimensional seismic velocity model to better constrain the Tewksbury event. The application of the three-dimensional model could also help people prepare for a future earthquake in the region.

What’s a seismic velocity model, and why is it necessary?

A seismic velocity model is just that — a model of the speed of seismic waves, and how those velocities change from layer to layer in Earth’s subsurface. Velocity can depend on a number of factors, including density, temperature, porosity, fluid content, and more. So, seismic velocity models provide a way to consider the physical nature of the rocks through which seismic waves travel.

A one-dimensional seismic velocity model varies only with depth. This means that each horizontal layer is assigned homogenous properties. Imagine a layered sheet cake with no icing on the edges. This kind of model often serves as a useful starting point. It also means quicker calculations when you need them.

For instance, velocity models are required in order to calculate the locations of earthquakes. Because the USGS must rapidly calculate earthquake locations in near-real time, it uses one-dimensional velocity models specific to the epicentral region via a program similar to HYPOINVERSE2000. This provides a rapid and relatively accurate estimate for an earthquake’s location that’s later refined by analysts who carefully select arrival times of seismic waves from additional seismic stations. Then, says Boyd, “It is possible that later studies will use 3D velocity models to further constrain the earthquake location.”

Three-dimensional seismic velocity models incorporate lateral changes along with depth. Imagine a grid of cubes in which each cube has a different P-wave or S-wave velocity. In this way, different rock types can be modeled as sitting next to each other, as you’d expect with basins that juxtapose hard basement lithologies next to the softer, sometimes unconsolidated sedimentary rocks that fill them.

Using a three-dimensional seismic velocity model can lead to more accurately determined earthquake locations, as Boyd says. Moreover, these models can also facilitate more realistic shaking estimates relative to a one-dimensional model, particularly where the subsurface or topography is complicated.

Modeling motions

To predict just how much shaking might happen in future quakes, scientists often turn to ground-motion models. In this kind of modeling, scientists can select various earthquake scenarios — for instance, choosing a location, depth, magnitude or other factors that will affect a seismic source— and calculate how seismic waves will change as they propagate from the earthquake’s source to different sites of interest.

These changes to seismic waves as they travel from source to site are called path effects. Site effects refer to how the location of the seismic station (or structure) influences the seismic waves at that location. Velocity models provide a way to account for both path and site effects — how the physical nature of the subsurface and surface affects seismic waves.

Statically speaking, ground-motion models can account for source, path and site effects in two general ways. Ergodic ground-motion models use spatially averaged source, path and site effects instead of location-specific terms. Conversely, non-ergodic ground-motion models do not average these effects, but instead use terms that specifically account for them.

“Incorporating lateral velocity variability [via a three-dimensional seismic velocity model] causes earthquake ground motions to depend on where the sources and sites are located,” Boyd explains. “With a one-dimensional seismic velocity model and with ergodic (independent of position) ground-motion models, earthquake ground motions do not depend on where the sources and sites are located or on the source-to-site [direction].” In other words, because there’s no lateral velocity variability in a one dimensional model, distance matters, but direction and locations do not.

Is my model right?

Ground-motion models can output peak ground acceleration, peak ground velocity, or other metrics that help scientists predict how much the ground might move. But how do you know if the ground-motion model’s results are reasonable? This is where real data from actual earthquakes come in, which is no problem for places like the West Coast of the U.S., where plenty of data from a variety of earthquakes exist.

Not so in the central and eastern U.S., where an earthquake significant enough to cause shaking is also simultaneously somewhat shocking for the populace. Several notable events like the 1811-1812 New Madrid earthquakes and the 1886 Charleston earthquake were, in fact, major catastrophes. But these events occurred so long ago that modern seismic instrumentation was not present to collect ground-motion data. As a result, the central and eastern U.S. have a seismic hazard assessment problem.

Accurate seismic hazard assessment depends on characterizing earthquake sources — their depth, culprit fault, fault orientation and slip direction — as well as factors like how often the earthquakes occur and how large those earthquakes might be. It also requires recording their ground motions to understand how the ground moves in different locations.

In New Jersey, sedimentary structures that can give rise to strange ground motions include basins that are hundreds of millions of years old. For instance, the Paleozoic Appalachian foreland basin sits west of the modern-day Appalachians, extending from Canada to Alabama. The Newark basin formed during the Mesozoic in a rift basin, a stretching of Earth’s skin that originated during the rending of Pangea.

Another way that the geology in this region significantly influences seismic waves, Boyd points out, is the Atlantic Coastal Plain, where soft sediments sit on hard basement rock. This pairing amplifies long-period ground motion and attenuates short-period ground motion.

It’s these arrangements of bedrock and sediment that can be critical to incorporate into velocity models.

Testing models

Boyd and coauthors simulated the seismic wavefield of the 2024 Tewksbury event via computer modeling within a (metaphorical) box 280 kilometers wide, 260 kilometers long, and 77 kilometers deep. They gave their simulations different hypocentral depths, adjusting the depth at which the earthquake could have nucleated. They also explored a range of focal mechanisms (or beachballs), which conceptualize how the fault moved. And, they tested a range of seismic velocity models, including a one-dimensional seismic velocity model developed for a nearby area and the three-dimensional geology-based USGS National Crustal Model. They also tested an ergodic ground-motion model currently used for seismic hazard assessments in the region developed by the Next Generation Attenuation-East Project.

By turning these knobs, they compared predicted earthquake ground motions with observed ground motions, focusing on motion in the 3 to 10 second period range. The observations came from 13 stations located between 76 to 131 kilometers away from the epicenter. The data were downloaded from the NSF National Geophysical Facility data archive.

Refining the Tewksbury event

Boyd and colleagues found that their three-dimensional seismic velocity model better predicts earthquake ground motions compared to the one-dimensional models, at least for ground motions averaged over 3, 5, 7, and 10 second periods. They also refined the Tewksbury earthquake’s depth and fault plane, finding that the event likely nucleated at a depth of four kilometers on a fault plane that dips shallowly to the east.

Especially interesting about their preferred model: Although the model incorporates a geologic framework specific to the greater New Jersey area (as you might expect), the three-dimensional model’s mechanical properties for rock types was calibrated primarily by velocity profile data from the western U.S. With this starting point, the team further adjusted the mechanical properties to better match waveform characteristics, Boyd says. Moreover, below any stratified sedimentary rocks, extrusive volcanic rocks and basement, the geology of the middle and lower crust was modeled as relatively simple, with no lateral variation.

But obtaining a perfect match between predictions and observations may not be appropriate. For example, the team found that they could improve the prediction-observation match by assigning the mainshock a shallower nucleation depth, and shallower fault plane. Yet this, they write, “seems inconsistent with the distribution of aftershocks, which dip closer to 45° to the east.”

In fact, they note that the effects of directivity — when an earthquake propagates with a pulse of energy in a particular direction — could have led to greater ground shaking at certain locations and in certain directions, as established in previous studies. But, because these effects were not considered in the ground motion models explored in this work, the result could be an artificially shallower modeled fault plane.

Plus, the authors point out that a future earthquake — even if it ruptures in the same location — may not move in the same way or have the same stress drop. “Certainly, the orientation of principal stresses leading to earthquake mainshocks is unlikely to change appreciably over a period of hundreds to thousands of years,” they write, but the range of possible earthquakes reacting to that stress, they note “could be broad.”