How Seismic Analysis Works in Geology and Engineering

Seismic analysis is the broad practice of using earthquake-generated or artificially produced vibrations to understand what is happening underground, to design structures that can survive shaking, and increasingly, to monitor changes in the earth over time. It spans disciplines from oil exploration to skyscraper engineering to planetary science, and the tools involved range from century-old principles of wave physics to deep-learning algorithms trained on millions of seismograms. What ties these applications together is a shared reliance on how waves travel through materials and what their behavior reveals about the stuff they pass through.

How Seismic Waves Map What Lies Underground

The oldest and most commercially significant branch of seismic analysis involves sending waves into the ground and interpreting what bounces back. Seismic reflection surveys have been used extensively to map underground rock layers, especially in oil and gas exploration. The basic setup involves generating a pulse of energy at the surface and recording the echoes that return from subsurface boundaries. Two methods dominate: single-channel profiling, which is relatively cheap and commonly used for shallow reconnaissance or offshore engineering work, and multichannel profiling, which captures more detail by recording reflections at many receiver positions simultaneously.

1Oxford Scholarship Online. Seismic Reflection and Refraction Methods

Reflection surveys produce images that look roughly like cross-sections of the earth, but the resolution depends heavily on how the data are processed. Over the past two decades, a technique called full-waveform inversion (FWI) has pushed that resolution much further. Rather than just tracking when echoes arrive, FWI tries to match the entire recorded waveform, including amplitude and shape, to a computer model of the subsurface. The result is a detailed velocity map that reveals fine-scale geological structure. Researchers have applied FWI to marine streamer data in New Zealand’s Hikurangi subduction zone, for instance, producing high-resolution models of wave speed down to roughly two kilometers beneath the seafloor.

2Journal of Geophysical Research: Solid Earth. Imaging the Shallow Subsurface Structure of the North Hikurangi Subduction Zone, New Zealand, Using 2‐D Full‐Waveform Inversion

FWI’s main drawback is that it can get stuck on wrong answers if the starting model is too far off. One approach to fixing this combines FWI with a related technique called wave-equation migration velocity analysis, leveraging the latter’s reliability to steer the inversion toward the right neighborhood before FWI sharpens the details.

3Stanford Exploration Project. Full waveform inversion by model extension – Section: Abstract

Noise and data gaps also degrade the results. Recent work on sparse regularization methods has shown that combining sparsity constraints with regularization can suppress noise and fill in missing low-frequency information, improving the accuracy and stability of FWI compared to older regularization approaches.

4Journal of Geophysics and Engineering. Improving full-waveform inversion based on sparse regularization for geophysical data

Designing Buildings to Survive Earthquakes

For structural engineers, seismic analysis means predicting how a building will respond when the ground shakes. The most common tool is response spectrum analysis, which takes a graph representing the maximum accelerations a structure would experience across a range of natural frequencies and applies it to a computer model of the building. Software packages like SAP2000 and ETABS are workhorses of this process. In one study of a high-rise bank building in Indonesia, response spectrum analysis confirmed that the structure met the country’s seismic safety standards, achieving a 90% modal participation mass ratio as required by Indonesian building code.

5Journal of World Science. Interpretation of SAP2000 Analysis Results Using the Design Response Spectrum Method in Earthquake Resistance in the Structural Performance of High-Story Buildings According to SNI 1726:2019

Response spectrum analysis works well for regular structures under idealized conditions, but real earthquakes are messier. For buildings that span a large footprint, like bridges or industrial facilities, the ground motion at one support point may differ from the motion at another because seismic waves take time to travel between them. A modified response spectrum method accounts for this by adjusting each spectral value with a correction factor that depends on the building’s properties and on how waves propagate through the local geology.

6Earthquake Engineering & Structural Dynamics. Response spectrum analysis of structures subjected to spatially varying motions

When engineers need the full picture, including how nonlinear behavior like cracking and yielding affects the outcome, they turn to time-history analysis, which steps through an actual earthquake record second by second. This is more computationally expensive but captures details that spectrum methods miss. For base-isolated buildings, where rubber or sliding bearings decouple the structure from the ground, time-history analysis has revealed that the standard assumption of 5% viscous damping significantly overestimates the forces reaching the superstructure. A damping ratio closer to 1% produces much more realistic results for isolated reinforced concrete buildings.

7Earthquake Engineering & Structural Dynamics. Appropriate viscous damping for nonlinear time‐history analysis of base‐isolated reinforced concrete buildings

When the Ground Changes the Equation

A building does not sit passively on the earth; the structure and the soil interact. Soft soil amplifies shaking, and the building’s own weight and motion feed energy back into the ground. Ignoring this dynamic soil-structure interaction can lead to dangerously unconservative designs. Research comparing fixed-base models with models that account for soil flexibility has found that lateral story drift can increase by roughly 50 to 95% depending on the seismic zone when soil-structure interaction is included.

8Civil Engineering Journal. Influence of Soil-Structure Interaction on the Seismic Response of the Structure on Mat Foundation

The severity of these effects depends on the building type and the foundation. Structures on shallow foundations tend to experience the most pronounced changes, while increasing foundation size for tall buildings helps reduce lateral deflection and rocking. Interestingly, the structural system matters too: wall-frame systems, which combine shear walls with moment frames, tend to be more sensitive to soil-structure interaction effects than pure moment-resisting frames.

9Journal of Physics: Conference Series. Evaluation of Soil-Structure Interaction for Different RC Structural Systems and Foundation Sizes

Seconds That Save Lives

Earthquake early warning systems represent one of the most time-critical applications of seismic analysis. The idea is straightforward: the faster-traveling P wave arrives before the more destructive S wave and surface waves. If you can detect the P wave and quickly estimate how big the earthquake is, you can issue an alert in the seconds or tens of seconds before the strong shaking hits. Even a few seconds of warning can be enough to stop trains, open fire station doors, or trigger automated shutoffs at industrial facilities.

One approach developed for the Mexican early warning system uses just the first three seconds of the P-wave signal on a single vertical-component sensor. Two parameters extracted from the acceleration waveform are fed into a learning algorithm trained on records from deep in-slab earthquakes beneath the subducted Cocos plate, and the system outputs a magnitude estimate fast enough to beat the destructive waves to populated areas.

10Bulletin of the Seismological Society of America. A Fast Earthquake Early Warning Algorithm Based on the First 3 s of the P‐Wave Coda

Japanese systems, working with arguably the densest seismic network on the planet, have taken a complementary approach: measuring peak amplitude parameters in real time from the P wave to predict the peak ground velocity that will follow.

11Geophysical Research Letters. A P wave‐based, on‐site method for earthquake early warning

A persistent challenge is that for large earthquakes, the rupture itself keeps growing after the first P waves have left the source. A technique that detects late-arriving P waves from expanding rupture zones has been shown to significantly increase warning lead times for large events without sacrificing much prediction accuracy. Testing on 23 large earthquakes in Japan confirmed the improvement.

12Geophysical Research Letters. Real‐Time Detection of Rupture Development: Earthquake Early Warning Using P Waves From Growing Ruptures

Deep Learning for Earthquake Detection

Seismologists have traditionally picked out earthquake arrivals by hand or with simple automated algorithms that trigger on sudden changes in signal energy. Both approaches struggle when signals are weak, noise is high, or thousands of stations need processing simultaneously. Deep learning has changed the game. EQTransformer, a neural network model based on an attention mechanism, can simultaneously detect earthquake signals and pick first P- and S-wave arrival times on individual station records. Applied to five weeks of continuous data from Japan, the model generalized well to a region it was not trained on, demonstrating that these networks can transfer across geographies.

13Nature Communications. Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking

Other architectures have pushed this further. DeepPhasePick uses a two-stage approach: a convolutional neural network first detects whether a phase is present, then recurrent neural networks pick the precise arrival times for both P and S waves.

14Geophysical Journal International. DeepPhasePick: a method for detecting and picking seismic phases from local earthquakes based on highly optimized convolutional and recurrent deep neural networks

The practical significance is enormous. These tools let seismologists detect far smaller earthquakes than before, building richer catalogs that improve our understanding of fault behavior and seismic hazard. They also free human analysts to focus on interpretation rather than the tedious work of scanning waveforms.

Listening to the Earth’s Background Noise

You do not always need an earthquake or an explosion to do seismic analysis. The earth is constantly vibrating from ocean waves, wind, traffic, and industrial activity. Seismic interferometry exploits this ambient noise by cross-correlating long recordings from pairs of seismometers. The result approximates the signal you would see if you set off a source at one station and listened at the other, effectively turning ambient noise into a free, always-on seismic source.

15Proceedings of the Geologists’ Association. Seismic interferometry and ambient noise tomography in the British Isles

Beyond imaging static structure, the same technique enables continuous monitoring of changes in the subsurface. By tracking small shifts in wave speed over time, researchers can detect the effects of earthquakes, volcanic activity, groundwater changes, or seasonal temperature cycles on the crust. This opened a completely new type of seismological observation: passive, continuous, and sensitive to subtle changes that would be invisible in traditional surveys.

16Comptes Rendus. Géoscience. Passive image interferometry for monitoring crustal changes with ambient seismic noise

Fiber Optics as Seismic Sensors

One of the most striking recent developments is the use of ordinary fiber optic cables as seismic sensors. Distributed acoustic sensing, or DAS, works by sending laser pulses down a fiber and analyzing the tiny backscattered reflections. Any vibration along the cable changes the backscatter pattern slightly, turning potentially tens of kilometers of cable into an ultra-dense seismic network. A single fiber can act as thousands of sensors spaced just meters apart, offering long-term seismic observation capability and high-fidelity waveform recording.

17PubMed. Single-shot multi-sensitivity distributed acoustic sensing with hardware envelope retrieving acceleration

The appeal is obvious: cities, oil fields, and ocean floors are already laced with telecom fiber. Much of it can be repurposed for seismic monitoring without installing any traditional seismometers. DAS arrays have already been deployed in settings ranging from active volcanoes to urban environments, picking up traffic vibrations and small earthquakes alike. The tradeoff is data volume. With thousands of channels recording continuously, the storage and processing demands are staggering, which is one reason deep learning has become so important for making sense of DAS output.

Earthquakes Caused by Human Activity

Seismic analysis is not just about natural earthquakes. Human activities, especially the injection of wastewater from oil and gas operations into deep wells, can trigger seismicity on faults that would otherwise have remained quiet. The mechanism involves pore pressure: injected fluids slowly raise the pressure in rock surrounding the well, and if that pressure reaches a fault already close to failure, it can push it over the edge. The interaction between fluid flow and fault mechanics is governed by the coupling between pressure diffusion and rock deformation during the nucleation phase of a rupture.

18Geomechanics for Energy and the Environment. The nucleation of injection-induced earthquakes on low-permeability strike-slip faults

The Fort Worth Basin in Texas offers a well-documented case study. Earthquakes there were induced from 2008 through 2020 by wastewater injection. Faults near the largest injection operations became unstable relatively early, when pore pressure had risen by about 0.31 megapascals. Faults farther away destabilized later but at much smaller pressure changes. Once a fault sequence had already been reactivated, it took only about 0.05 megapascals of additional pressure to trigger fresh activity. The response was highly variable: many faults remained stable even at higher pressures, while a few became seismogenic at very small changes.

19Geophysical Research Letters. Pore Pressure Threshold and Fault Slip Potential for Induced Earthquakes in the Dallas‐Fort Worth Area of North Central Texas

This variability makes managing induced seismicity tricky. A traffic-light protocol, where injection is reduced or halted when local seismicity crosses a magnitude threshold, is the most common regulatory approach, but the lag between pressure change and fault response means seismicity can continue or even escalate after injection stops.

Watching Reservoirs Change Over Time

In the oil and gas industry, seismic analysis does not end once a field is discovered. Time-lapse seismic monitoring, known as 4D seismic, involves shooting repeated 3D surveys over the life of a reservoir and comparing them. The concept requires at least two surveys: a baseline before production begins and one or more follow-ups after fluid has been extracted or injected. By subtracting one survey from the other, engineers can spot differences caused by fluid movement, pressure changes, saturation shifts, and even physical compaction of the reservoir rock.

20Journal of Natural Gas Science and Engineering. The Role of Time Lapse(4D) Seismic Technology as Reservoir Monitoring and Surveillance Tool: A Comprehensive Review

The same principle is being applied to carbon dioxide sequestration projects. After CO₂ is injected underground, 4D seismic can track the plume as it spreads, verifying that the gas is staying where it should and not migrating toward unintended pathways. The key assumption underpinning 4D work is that the acquisition parameters and processing steps stay as consistent as possible between surveys, so any differences in the seismic images can be attributed to genuine subsurface changes rather than survey artifacts.

Probabilistic Hazard Assessment

Seismic hazard analysis aims to answer a deceptively simple question: how hard is the ground likely to shake at a given location over a given time period? The standard approach, probabilistic seismic hazard analysis, combines information about fault locations, earthquake recurrence rates, and ground motion attenuation to produce hazard curves and maps that feed directly into building codes. A recent advance merges physics-based earthquake simulations for nearby faults with statistical ground motion prediction equations for more distant sources, accounting for the uncertainties inherent in both methods.

21International Journal of Disaster Risk Science. A Probabilistic Seismic Hazard Analysis Method Incorporating Physics-Based Simulation and Ground Motion Prediction Equation

Part of what makes hazard assessment difficult is the limited historical record. Instrumental seismology covers only the last century or so, and written records extend a few thousand years at best in some regions. Paleoseismology fills this gap by digging trenches across active faults and reading the geological record of past ruptures. At the Salt Lake site along the Haiyuan Fault in China, a high-resolution stratigraphic sequence preserved evidence of three and possibly four surface-rupturing events since about 1500 AD. The study showed that moderate-magnitude earthquakes, not just the largest events, can be preserved in sediment layers if deposition is fast enough and continuous.

22Journal of Geophysical Research: Solid Earth. Variability in magnitude of paleoearthquakes revealed by trenching and historical records, along the Haiyuan Fault, China

Seismic Analysis on Mars

Perhaps the most dramatic expansion of seismic analysis in recent years has nothing to do with Earth. NASA’s InSight lander placed a seismometer on Mars in late 2018, and the data it returned transformed our understanding of another planet’s interior. Using direct and surface-reflected body waves from eight low-frequency marsquakes, researchers mapped the Martian upper mantle down to about 800 kilometers. They found a thermal lithosphere much thicker than Earth’s, underlain by a low-velocity zone, and estimated that the Martian crust is enriched in heat-producing elements by a factor of 13 to 20 relative to the planet’s primitive mantle composition.

23PubMed. Upper mantle structure of Mars from InSight seismic data

The crust itself turned out to be surprisingly variable in its possible thickness. Analysis of reflected and converted seismic phases beneath the InSight landing site yielded two candidate models: one with a crustal thickness of about 20 kilometers and another around 39 kilometers, depending on which subsurface boundary is interpreted as the crust-mantle transition. Extrapolating from this single measurement using gravity and topography data, the average crustal thickness for the entire planet falls somewhere between 24 and 72 kilometers.

24PubMed. Thickness and structure of the martian crust from InSight seismic data

Imaging the Deep Mantle

Back on Earth, seismic analysis has been the primary tool for understanding the planet’s deep interior since long before anyone dreamed of putting a seismometer on Mars. Global seismic tomography, which works on a principle loosely analogous to a medical CT scan, uses the travel times of seismic waves from earthquakes around the world to build three-dimensional maps of velocity variations throughout the mantle. P-wave tomography has provided evidence for mantle-wide convective flow, revealing long, narrow high-velocity structures in the lower mantle that can be traced from their deep locations all the way up to sites where tectonic plates currently converge at the surface.

25Nature. Evidence for deep mantle circulation from global tomography

These structures are interpreted as slabs of ancient ocean floor that sank into the mantle at subduction zones hundreds of millions of years ago and have been slowly descending ever since. Their presence at great depth is one of the strongest pieces of evidence that convection in Earth’s mantle extends from the surface all the way to the core-mantle boundary, rather than being confined to separate upper and lower circulation cells. It is a finding that reshaped plate tectonics from a theory about the surface into a theory about the entire planet, and it came almost entirely from the patient recording and analysis of seismic waves.