Seismic data is information extracted from vibrations traveling through the Earth, and it underpins far more of modern science and safety than most people realize. While earthquakes are the most familiar source, seismic waves are also generated deliberately with controlled sources and passively from ambient noise, ocean waves, glaciers, traffic, and even animals. The recordings from these vibrations reveal everything from the planet’s deep interior to the movement of subway trains, and researchers are finding new uses for them at a striking pace.
What Seismic Data Actually Records
At its core, seismic data captures how vibrations move through material. When energy is released, whether by an earthquake, an explosion, or a truck passing over a bridge, it radiates outward as waves. Different types of waves travel at different speeds and behave differently when they hit boundaries between rock layers, water, ice, or air. By recording the arrival times, amplitudes, and frequencies of these waves at one or more sensors, scientists can work backward to figure out what happened at the source and what the waves passed through on the way.
A few physical properties matter most. Seismic energy and peak vibration strength decay exponentially with distance from the source, and the dominant frequency of the signal drops as the waves travel farther. These predictable relationships let researchers estimate how far away a source is and how much energy it released, even from a single recording station.
How Seismic Data Is Collected
Traditional seismic networks rely on seismometers and accelerometers installed on the ground or the ocean floor. These instruments convert ground motion into electrical signals, which are digitized and transmitted in real time or stored locally. For earthquake monitoring, governments operate permanent networks of stations spread across seismically active regions. For oil and gas exploration or engineering surveys, crews deploy temporary arrays of sensors and generate their own seismic waves with vibrator trucks, air guns, or small explosive charges.
One of the more exciting developments is distributed acoustic sensing, or DAS. Instead of placing individual sensors at fixed points, DAS turns an entire fiber-optic cable into a continuous string of thousands of virtual microphones. A laser pulse is sent down the fiber, and tiny imperfections in the glass scatter light back. When the cable is disturbed by a passing vibration, the scattered light shifts in a measurable way. Researchers have used a 20-kilometer stretch of seafloor fiber-optic cable offshore of Monterey Bay, California, to extract waves from ambient ocean noise and build a high-resolution image of the sediments beneath the seabed, all without any active source at all.1Scientific Reports. Utilizing distributed acoustic sensing and ocean bottom fiber optic cables for submarine structural characterization Because telecommunications companies have already laid vast networks of fiber across continents and ocean floors, DAS could dramatically expand global seismic coverage without installing new hardware.
Urban monitoring has introduced another approach. In one study covering the Guangdong-Hong Kong-Macao Greater Bay Area in China, an array of over 6,200 geophones blanketed a 60-by-60-kilometer area, capturing ambient seismic noise at high resolution.2Earth and Space Science. High‐Resolution Monitoring of Urban Activities With Nodal Seismometers in the Guangdong‐Hong Kong‐Macao Greater Bay Area, China These dense, portable “nodal” arrays are increasingly affordable and can be deployed for weeks or months to capture detailed pictures of both the subsurface geology and the human activity above it.
Mapping What Lies Underground
One of the oldest and most commercially important uses of seismic data is imaging the subsurface. In reflection seismology, a controlled source sends waves downward, and sensors on the surface record the echoes that bounce back from rock boundaries. The timing and strength of these reflections reveal the depth, thickness, and composition of underground layers. This technique is the backbone of oil and gas exploration, but it is also used to study fault systems, groundwater reservoirs, and geothermal resources.
After the devastating 2016–2017 earthquake sequence in central Italy, researchers used previously unpublished seismic reflection profiles, combined with surface geological data, to reconstruct the subsurface geology near the magnitude 6.5 Norcia mainshock. The profiles helped clarify the fault structures responsible for the earthquakes and improved understanding of the regional seismotectonic framework.3Tectonics. Seismic Reflection Profiles and Subsurface Geology of the Area Interested by the 2016–2017 Earthquake Sequence (Central Italy) Work like this feeds directly into hazard maps and building codes.
Seismic monitoring is also becoming central to carbon capture and storage. When CO₂ is injected underground into depleted reservoirs or saline aquifers, operators need to verify that the gas stays where it is supposed to. Laboratory demonstrations using “4D seismic” methods, which repeat 3D surveys over time, have shown that the approach can successfully track a gas plume’s evolution and image the top of the storage layer.4International Journal of Greenhouse Gas Control. Demonstration of seismic monitoring of CO2 storage using a 3D acoustic laboratory As carbon storage projects scale up worldwide, seismic data will likely be the primary tool regulators rely on to confirm that stored CO₂ is not leaking.
Turning Echoes Into Velocity Models
Raw seismic recordings are not photographs. Converting wiggly traces into an actual image of the subsurface requires knowing how fast seismic waves travel at every point underground. Building that velocity model is one of the hardest problems in geophysics, and the technique called full-waveform inversion (FWI) represents the current state of the art. FWI works by simulating what the seismic recordings should look like for a guessed velocity model, comparing the simulation to the real data, and then adjusting the model to reduce the mismatch.5Geophysics. Velocity model building by 3D frequency-domain, full-waveform inversion of wide-aperture seismic data This process repeats thousands of times until the synthetic data closely matches what was actually recorded.
FWI sounds straightforward in principle, but in practice it is plagued by a problem called cycle skipping, where the algorithm latches onto the wrong alignment between predicted and observed waves and converges on a false solution. Solving this requires careful choices about the starting model, the way misfit is measured, and which frequencies to invert first.6Geophysical Prospecting. Velocity model building from seismic reflection data by full‐waveform inversion Recent work has explored physics-driven machine learning approaches that use generative adversarial networks to guide the inversion, producing velocity models that score well on structural similarity and signal-to-noise metrics for complex geological benchmarks.7arXiv. Seismic full-waveform inversion based on a physics-driven generative adversarial network The computational cost of FWI remains enormous, but hardware improvements and algorithmic shortcuts keep pushing the boundaries of what can be imaged.
Earthquake Early Warning
Seismic waves travel fast, but not as fast as electronic signals. That gap is the basis of earthquake early warning (EEW) systems, which detect the first-arriving compressional wave from an earthquake and send an alert to people and infrastructure farther away before the more damaging shear and surface waves arrive. Even a few seconds of warning can be enough to slow a train, open a fire station door, or prompt someone to drop under a desk.
Italy recently deployed the first EEW system dedicated to its high-speed rail network. The system uses a dedicated network of accelerometers to detect ground motion in real time and predict peak ground acceleration. A probabilistic decision-making module evaluates the incoming seismic data and dynamically updates alerts as the initial wave spreads. Instead of shutting down an entire rail line, the system targets specific segments for operational restrictions like deceleration or halting, minimizing unnecessary disruptions.8Nat. Hazards Earth Syst. Sci.. The first Earthquake Early Warning System for the high-speed railway in Italy: enhancing rapidness and operational efficiency during seismic events
Machine learning is accelerating these systems further. A system called E3WS can detect, locate, and estimate the magnitude of an earthquake from just three seconds of the initial wave recorded by a single station. Its magnitude estimates are virtually unbiased and more accurate than existing single-station methods. By updating its estimates every second, the system tracks the earthquake’s source as it unfolds, providing faster alerts than traditional networks that wait for data from multiple stations.9Journal of Geophysical Research: Solid Earth. Earthquake Early Warning Starting From 3 s of Records on a Single Station With Machine Learning Those additional seconds could make a meaningful difference for automated safety systems.
Understanding How Earthquakes Work
Beyond issuing warnings, seismic data is the primary tool for understanding the physics of earthquakes after they happen. A key step is determining the “moment tensor,” which is essentially a mathematical description of the forces that produced the earthquake. It tells seismologists whether a fault slipped horizontally, vertically, or at some oblique angle, and it captures the earthquake’s magnitude and the geometry of the rupture.
Full-waveform approaches to moment tensor inversion, which match the entire shape of recorded seismic waves rather than just the first-motion direction, provide more comprehensive constraints on complex earthquake sources.10Seismological Research Letters. NA‐GCAP: A New Full Moment Tensor Inversion Method Based on the Renewed GRTM and NA Method A persistent challenge, though, is that these methods depend on knowing the Earth’s velocity structure, and errors in that structure can bias the results. Recent work using Monte Carlo simulations to characterize the uncertainty introduced by imperfect 3D Earth models has shown that incorporating structural error estimates into the inversion significantly improves accuracy.11Geophysical Journal International. Towards a new standard for seismic moment tensor inversion containing 3-D earth structure uncertainty Researchers studying the 2013 Lushan earthquake sequence in China developed a method that uses 3D velocity models directly in the inversion, breaking seismograms into body-wave and surface-wave segments and allowing time shifts to account for modeling imperfections.12Physics and Chemistry of the Earth, Parts A/B/C. Seismic moment tensor inversion using 3D velocity model and its application to the 2013 Lushan earthquake sequence
Telling Explosions From Earthquakes
One of the highest-stakes applications of seismic data is nuclear test monitoring. Underground nuclear explosions produce seismic waves that look a lot like earthquakes on a seismogram, and the Comprehensive Nuclear-Test-Ban Treaty Organization operates a global network of monitoring stations specifically to detect and identify them. Analysts distinguish explosions from natural earthquakes using several discriminants: the ratio of body-wave to surface-wave magnitudes, spectral ratios, signal complexity, and the corner frequencies of different wave types.13Pure and Applied Geophysics. Seismic Discrimination Between Nuclear Explosions and Natural Earthquakes using Multi-Machine Learning Techniques Explosions tend to produce relatively strong compressional body waves and weak surface waves compared to earthquakes of the same size, because they radiate energy outward symmetrically rather than along a fault plane. Machine learning classifiers trained on known explosions and earthquakes are now being layered on top of these traditional methods to improve discrimination accuracy, especially for smaller events near the detection threshold.
Imaging the Deep Earth
Seismic data from large earthquakes recorded at stations worldwide provides the only direct window into the planet’s deep interior. Global seismic tomography, which works on the same principle as a medical CT scan but using earthquake waves instead of X-rays, has revealed broad features like mantle plumes and subducting slabs. But the resolution of tomography is limited by the uneven distribution of earthquakes and stations. More advanced multichannel methods, including network waveform stacking, array processing, and 3D migrations, can detect weak signals from structures that tomography misses entirely. These techniques have identified both large- and small-scale features with velocity variations ranging from a fraction of a percent to tens of percent, reshaping how geophysicists think about deep mantle convection and the planet’s thermal and chemical evolution.14Annual Reviews. Deep mantle seismic modeling and imaging
Listening to Glaciers and Ice Sheets
Seismic data is increasingly used to study ice. Glaciers generate their own seismic signals, called icequakes, as ice slides over the bedrock or fractures under stress. By analyzing icequake source mechanisms, researchers can infer the shear strength of the bed beneath the ice and the physical processes governing glacial sliding, information that is extremely difficult to obtain any other way.15Journal of Geophysical Research: Earth Surface. Icequake Source Mechanisms for Studying Glacial Sliding
On an Alpine plateau glacier in Switzerland, seismic sensors captured tremors generated by subglacial water flow, icequake bursts, and water pouring into vertical shafts called moulins. By combining the seismic analysis with hydrological models of subglacial drainage, researchers tracked the gradual development of a branching drainage network beneath the ice and found that hydrofracturing preceded a glacial lake drainage event. The fracturing showed up in the seismic record before the lake level dropped, suggesting it could serve as an early warning signal for outburst floods, though the authors cautioned against generalizing that finding to all glacier-dammed lakes.16The Cryosphere. Glaciohydraulic seismic tremors on an Alpine glacier As glaciers respond to warming temperatures, this kind of monitoring is becoming more urgent.
Marsquakes and Planetary Seismology
Seismic data is no longer confined to Earth. NASA’s InSight lander, which operated on Mars from 2018 to 2022, carried a sensitive seismometer that recorded the first-ever Martian seismic record. Using arrival times of body waves from 17 marsquakes, researchers constrained both the quake locations and the one-dimensional average interior structure of Mars.17PubMed Central. Marsquake Locations and 1-D Seismic Models for Mars From InSight Data The data revealed that Mars has a liquid core larger than expected, a thick lithosphere, and a crust with distinct layering. These findings carry implications for understanding how rocky planets form and evolve, and they validated the principle that a single well-placed seismometer can unlock fundamental planetary science. Future missions to the Moon and potentially Europa may carry seismometers of their own.
Tsunami Detection From the Seafloor
Offshore seismic and pressure-gauge networks are proving critical for tsunami early warning. Japan’s S-net and DONET systems combine ocean-bottom seismometers with pressure sensors on cabled networks stretching across the seafloor near subduction zones. The challenge for these near-fault instruments is separating the tsunami’s water-pressure signal from the seismic shaking of the earthquake itself, because both arrive almost simultaneously at sensors close to the source. A method that compares pressure-gauge records with co-located seismic data can strip away the earthquake signal, revealing the tsunami signal earlier and more reliably than previous detection schemes that either waited for the shaking to stop or depended on sensors farther from the source.18Journal of Geophysical Research: Oceans. Early Tsunami Detection With Near‐Fault Ocean‐Bottom Pressure Gauge Records Based on the Comparison With Seismic Data Ocean acoustic waves excited by seafloor uplift during megathrust ruptures could also be used to rapidly quantify shallow slip and improve local warnings.19Earth and Planetary Science Letters. Constraining shallow slip and tsunami excitation in megathrust ruptures using seismic and ocean acoustic waves recorded on ocean-bottom sensor networks For coastal communities living close to fault zones, those extra minutes of lead time can save lives.
Tracking City Life Through Ground Vibrations
Seismic instruments are sensitive enough to pick up far more than earthquakes. In urban settings, the ambient noise recorded by seismometers is dominated by human activity: vehicles, construction, factories, trains. What was once considered unwanted interference is now a data source in its own right. The Greater Bay Area study mentioned earlier found that spatial patterns of seismic noise correlated strongly with population density and transportation routes, and the researchers were able to extract specific signals corresponding to vehicle speeds, traffic flow, subway operations, and even ship movements.20Earth and Space Science. High‐Resolution Monitoring of Urban Activities With Nodal Seismometers in the Guangdong‐Hong Kong‐Macao Greater Bay Area, China
Other work has pushed this further, using advanced signal processing to count individual motor vehicles and even detect runners passing near sensors, highlighting intense spatial heterogeneity in activity levels within a single kilometer.21PubMed Central. Quantifying Urban Activities Using Nodal Seismometers in a Heterogeneous Urban Space During the early COVID-19 lockdowns in 2020, seismologists worldwide noticed a dramatic drop in high-frequency ambient noise, effectively measuring the reduction in human movement through ground vibrations. Urban seismology is a young field, but it offers a privacy-preserving alternative to cameras and cell-phone tracking for understanding how cities function.
Machine Learning and the Future of Seismic Analysis
Across nearly every application described above, machine learning is reshaping how seismic data is processed. One of the most labor-intensive steps in earthquake seismology is “phase picking,” identifying the exact moment each type of wave arrives at a sensor. Traditionally this was done by hand or with rule-based algorithms. A recent approach called AI-PAL trains a deep learning model using detections from an established rule-based picker, creating a self-supervised workflow that generalizes to new regions and earthquake types without requiring manually labeled data.22Journal of Geophysical Research: Solid Earth. AI‐PAL: Self‐Supervised AI Phase Picking via Rule‐Based Algorithm for Generalized Earthquake Detection Similar machine learning pipelines are being developed for noise classification, fault mapping from reflection data, and real-time event discrimination.
The broader trend is a shift from manual or semi-automated workflows to end-to-end systems that can ingest raw data and deliver actionable outputs with minimal human intervention. For earthquake early warning, this means faster alerts. For exploration, it means cheaper surveys. For monitoring volcanoes, glaciers, and carbon storage sites, it means continuous automated surveillance at scales that would have been impractical a decade ago. The seismic data itself has not fundamentally changed, but the ability to squeeze information out of it is advancing fast.
Animals That Use Seismic Vibrations
Humans need instruments to detect seismic signals, but many animals have evolved to sense and even produce ground vibrations as a primary mode of communication. Substrate-borne vibrations are considered one of the oldest communication channels in the animal kingdom, used alone or alongside visual, auditory, and chemical signals.23PubMed. How do animals use substrate-borne vibrations as an information source? Insects use leaf and stem vibrations for mate location and courtship. Spiders sense the vibrations of prey struggling in their webs. Elephants are thought to detect low-frequency ground vibrations through their feet over distances of several kilometers.
Among mammals, one of the best-studied examples is the blind mole rat, a solitary subterranean species that communicates by drumming its head against the tunnel ceiling. Paired animals frequently produce these vibratory signals during encounters, and the behavior appears to assist in mate selection and social recognition.24Animal Behaviour. Seismic signalling as a means of communication in a subterranean mammal Vibration provides information used in predator-prey interactions, recruitment to food, mate choice, competition, and parental care across a range of organisms from insects to elephants.25Integrative and Comparative Biology. Vibration and Animal Communication: A Review Studying these biological seismic systems not only deepens understanding of animal behavior but also occasionally inspires engineering approaches to vibration sensing and signal extraction.