Seismometers are instruments that detect and record ground motion, converting the vibrations passing through rock, soil, and water into signals that scientists can analyze. At their core, they all exploit the same physical trick: an inertial mass that resists movement while the ground shifts beneath it, creating a measurable difference between the two. What makes the field fascinating is how far that simple idea has been pushed, from a bronze vessel in second-century China to sensors sitting on the surface of Mars, and from detecting earthquakes to tracking ocean waves, calving glaciers, and even the absence of human traffic during pandemic lockdowns.
The Pendulum Principle
Every seismometer relies on some version of a mass-on-a-spring. When the ground shakes, the instrument’s frame moves with it, but the mass, suspended by a spring and damper, lags behind. The relative motion between the mass and the frame is what gets recorded. A damper prevents the mass from bouncing endlessly after a shake, so the instrument settles quickly and stays ready for the next event. This setup acts as a kind of physical filter, letting the sensor respond to rapid ground acceleration while ignoring very slow drift.
Modern instruments refine that concept in different ways. A velocimeter records how fast the ground moves (velocity), while an accelerometer records the rate of change in that speed (acceleration). Some broadband designs combine both approaches. One recent design converted a linear force-balance accelerometer into a broadband seismometer by adding an integrator circuit, achieving a flat velocity response from periods of 120 seconds all the way up to 75 hertz, with sensitivity of 1,200 volts per meter-per-second and self-noise well below standard low-noise benchmarks across most of that range.1SpringerLink (Journal of Seismology). Turning a linear geometry force-balance accelerometer to a broadband seismometer: design, modeling, and evaluation That kind of performance matters because the signals scientists care about can range from sharp jolts lasting fractions of a second to deep, slow undulations with periods of minutes.
From Ancient China to the Cold War
The earliest known device for detecting earthquakes was Zhang Heng’s seismoscope, built in 132 AD during the Han Dynasty. The original was lost long ago, but engineering reconstructions suggest its interior mechanism was a planar linkage of at least five bars and six joints, designed so that ground shaking in a particular direction would trigger a lever arm to release a bronze ball from one of eight dragon-head ornaments on the outside of the vessel.2Mechanism and Machine Theory. Reconstruction design of the lost seismoscope of ancient China Separate engineering analyses have explored cam-linkage alternatives that also satisfy the mechanical knowledge available in Zhang Heng’s era, producing multiple feasible interior designs.3Journal of Advanced Mechanical Design, Systems, and Manufacturing. Structural Synthesis of Zhang Heng’s Seismoscope with Cam-linkage Mechanisms The device could not record a waveform or measure magnitude. It simply indicated the direction of a distant quake by dropping a ball into the mouth of a waiting bronze toad. Still, the underlying insight, that a suspended mass responds differently from the ground it sits on, is the same principle used today.
The leap from direction indicator to waveform recorder took almost two millennia. The modern era of seismograph development really accelerated after the Second World War, when electrodynamic seismographs and their amplifiers, recorders, and timing systems were adapted for worldwide deployment under varying field conditions. Between 1948 and 1976, instruments shrank enough to fit inside cased boreholes, which helped them avoid surface noise from wind, traffic, and temperature swings.4Eos, Transactions American Geophysical Union. Earthquake seismograph development: A modern history—Part 1 That push toward smaller, tougher sensors had a political driver too: the Cold War created enormous demand for detecting underground nuclear tests from thousands of kilometers away.
Keeping Noise Out
A seismometer is only as useful as its signal-to-noise ratio. Wind, traffic, temperature fluctuations, and even ocean waves create background vibrations that can swamp the faint signals scientists are trying to read. Several strategies exist for taming that noise.
The most direct is to go underground. A study using a vertical array of seismometers installed in a 285-meter-deep borehole in northern Italy found that noise levels drop following a logarithmic trend with depth, and the signal-to-noise ratio for recorded earthquakes consistently increased at deeper stations.5Seismological Research Letters. Seismic Noise Reduction as a Function of Depth Recorded by a Vertical Array Installed in a 285‐m‐Deep Borehole at a Gas Storage Field in Northern Italy Even events with negative magnitudes became detectable at depth. For stations that must remain on the surface, insulating enclosures help. Research on broadband seismometer housings showed that surface-based installations benefited more from thermal and wind shielding than stations already in vaults or caves, likely because underground environments are naturally more thermally stable to begin with.6PubMed Central. Noise reduction characteristics of broadband seismometer enclosures
On the ocean floor, the challenges are different. Ocean-bottom seismometers deal with noise from tidal currents and permanent deep-sea flows. Short-period noise and long-period tilt noise are both driven by bottom currents, and in some cases the currents interact with the mechanical housing of the instrument to create harmonic tremors that have no geological origin at all. A newer broadband ocean-bottom design built by researchers at the University of Lisbon addressed this by hiding non-essential external components from the current flow, which eliminated the worst harmonic contamination.7Natural Hazards and Earth System Sciences. The effect of deep ocean currents on ocean-bottom seismometers records
Earthquake Early Warning and Machine Learning
When an earthquake starts, it sends out two main types of waves. The P-wave (primary) travels fastest and arrives first but usually causes less damage. The S-wave (secondary) and surface waves follow and do the real shaking. Early warning systems try to detect the P-wave, estimate the earthquake’s size and location, and push an alert to people in the path of the slower, more destructive waves. Even a few seconds of warning can be enough to duck under a desk, stop a train, or shut a valve on a gas line.
The bottleneck has always been speed and accuracy of that initial P-wave detection, especially in regions where seismic stations are far apart. Deep learning models are transforming this step. A convolutional neural network tested in Indonesia achieved P-wave detection F1 scores above 99% on local data and 80% on 12 hours of continuous waveforms across the country’s sparse network.8Applied Computing and Geosciences. Deep learning for real-time P-wave detection: A case study in Indonesia’s earthquake early warning system In New Zealand, a lightweight model with roughly 38,000 trainable parameters achieved 97% overall accuracy and could identify 98% of P-wave segments, all while running on a Raspberry Pi with sub-seven-millisecond inference times.9PubMed Central. Lightweight convolutional neural network for real-time earthquake P-wave detection on edge devices in New Zealand That matters because it means the detection algorithm can run directly at the sensor, without waiting for data to travel to a central server and back.
False alarms are another persistent headache. A slamming door, a passing truck, or a quarry blast can produce impulsive signals that look enough like a P-wave to fool simple detectors. One approach trained a generative adversarial network on 300,000 waveforms from southern California and Japan to learn the specific character of genuine P-wave arrivals, giving the system a way to distinguish earthquakes from local noise before an alert is issued.10Geophysical Research Letters. Machine Learning Seismic Wave Discrimination: Application to Earthquake Early Warning
Nuclear Test Monitoring
One of the most consequential jobs seismometers perform has nothing to do with natural hazards. The Comprehensive Nuclear-Test-Ban Treaty relies on a global network of monitoring stations, and seismic detection is the backbone of that network. The challenge is sorting the signal of an underground explosion from the thousands of natural earthquakes that occur worldwide each year. Forensic seismologists use several discrimination methods, including the ratio of body-wave to surface-wave magnitudes, the ratio of high-frequency P-wave to S-wave energy, model-based comparisons, and estimates of source depth.11Annual Review of Earth and Planetary Sciences. Forensic Seismology and the Comprehensive Nuclear-Test-Ban Treaty Explosions tend to produce stronger body waves relative to surface waves than natural earthquakes of the same size, and they originate at shallow depths.
Even very low-yield or deliberately hidden explosions have proven difficult to conceal entirely. An analysis of high-frequency seismic data up to 30 or 40 hertz found that even fully decoupled underground nuclear tests (detonations set off inside large cavities to muffle the signal) could be detected and identified at regional distances, where the monitoring network picks up signals with high signal-to-noise ratios.12Reviews of Geophysics. An evaluation of seismic decoupling and underground nuclear test monitoring using high‐frequency seismic data The science of telling earthquakes from explosions has become remarkably refined, and it gives the treaty real enforcement teeth.
Beyond Earthquakes
Seismometers have become versatile environmental sensors. Their ability to detect any ground vibration has been applied to a surprisingly wide set of problems.
On glaciers and ice sheets, seismic monitoring reveals processes that are hard to observe directly. Iceberg calving events generate low-frequency seismic signals below 0.1 hertz, dominated by surface waves, that can be recorded hundreds of kilometers away. At Jakobshavn Isbræ in Greenland, broadband stations captured the seismic signature of two full-thickness icebergs capsizing in sequence, and the signals could be correlated with time-lapse photography of the event.13Journal of Geophysical Research. Analysis of low-frequency seismic signals generated during a multiple-iceberg calving event at Jakobshavn Isbræ, Greenland Earlier work at Columbia Glacier in Alaska established a rough empirical relationship between the volume of a calved iceberg and the duration of the seismic signal it produces, allowing researchers to estimate calving rates from seismic data alone without needing to watch the glacier continuously.14Journal of Geophysical Research: Solid Earth. Calving icebergs: A source of low‐frequency seismic signals from Columbia Glacier, Alaska This kind of remote sensing is increasingly valuable as ice-sheet dynamics become central to projections of sea-level rise. The broader field, sometimes called cryoseismology, can detect brittle fracture, basal sliding, transient water flow, and calving, all processes that are otherwise hidden beneath hundreds of meters of ice.15The Cryosphere. Towards the systematic reconnaissance of seismic signals from glaciers and ice sheets – Part 1: Event detection for cryoseismology
Volcanoes are another natural fit. A seismic network at Sakurajima volcano in Japan classifies volcanic tremors and B-type earthquakes, both generated in the conduit beneath the craters, to track subsurface magma movement and assess eruption potential.16Geophysical Journal International. Classification of volcanic tremors and earthquakes based on seismic correlation: application at Sakurajima volcano, Japan In the French Alps, a seismic network installed on the Séchilienne rockslide recorded thousands of events, mostly rockfalls but also local and regional earthquakes, allowing researchers to distinguish and classify mass movements from their signal characteristics.17Journal of Geophysical Research: Earth Surface. Seismic monitoring of Séchilienne rockslide (French Alps): Analysis of seismic signals and their correlation with rainfalls
Listening to the Ocean Through the Ground
Ocean waves constantly pound the seafloor and coastlines, generating a background hum in seismic records known as microseisms. Two distinct mechanisms produce these signals at different period ranges. The primary mechanism, where ocean waves exert direct traction on the shallow seafloor, explains the power and variability of seismic noise at periods from about 13 to 300 seconds. A secondary mechanism, involving the interaction of opposing wave trains, dominates below 13 seconds.18Geophysical Research Letters. How ocean waves rock the Earth: Two mechanisms explain microseisms with periods 3 to 300 s
Because microseisms are generated by ocean waves hitting shallow coastal waters, they serve as a long-term proxy for near-coastal swell activity. A study analyzing the vertical-component primary microseism wavefield recorded by 52 globally distributed seismographs at periods of 14 to 20 seconds found evidence of increasing ocean wave energy from the late 1980s through 2022.19Nature Communications. Increasing ocean wave energy observed in Earth’s seismic wavefield since the late 20th century Seismometers, in other words, are inadvertently climate sensors. They have been recording ocean wave energy for decades in datasets that were originally collected for an entirely different purpose.
What the Pandemic Revealed
In early 2020, COVID-19 lockdowns brought an unplanned experiment in seismic silence. Human activity, from traffic and construction to factory machinery, generates high-frequency seismic waves that propagate into the ground. When much of the world stopped moving, seismometers noticed. Global anthropogenic seismic noise dropped by up to 50% during the lockdowns, making it the longest and most prominent reduction in human-caused seismic noise ever recorded. The quieting was strongest at surface stations in populated areas but extended for kilometers laterally and hundreds of meters into the subsurface.20PubMed. Global quieting of high-frequency seismic noise due to COVID-19 pandemic lockdown measures
Regional studies told a more textured story. In Gujarat, India, urban stations saw daytime noise levels in the 5 to 15 hertz band fall by 27 to 79%, while remote stations showed almost no change.21Scientific Reports. Effect of coronavirus lockdowns on the ambient seismic noise levels in Gujarat, northwest India In Greece, the first lockdown produced a 43% reduction in seismic noise across urban strong-motion stations, with the single quietest day, Easter Sunday 2020, seeing reductions exceeding 80%.22Seismica. The impact of COVID-19 lockdown measures on high-frequency seismic ambient noise in Greece: Utilizing strong-motion seismograph networks for human activity monitoring in urban environments The practical upshot was that faint signals from small earthquakes and subsurface sources that normally would have been buried in city noise became visible during the quiet period. The correlation between seismic noise and independent measures of human mobility was so strong that seismology could serve as a real-time, absolute gauge of how much a population is moving around.
Low-Cost Sensors and Smartphone Networks
Traditional broadband seismometers are expensive and require careful installation. Scaling a network to blanket a city or a developing country with them is impractical. Two approaches are filling the gap.
The first is MEMS (micro-electromechanical system) accelerometers. These tiny, inexpensive sensors lack the sensitivity of broadband instruments but can reliably record moderate-to-strong earthquakes and sometimes even weaker ones. A MEMS network in Trentino, northeastern Italy, deployed 73 low-cost accelerometers close to inhabited areas and integrated their peak ground acceleration readings into the existing monitoring system, producing rapid shaking maps within minutes of an event.23Copernicus Publications (Natural Hazards and Earth System Sciences). A dense micro-electromechanical system (MEMS)-based seismic network in populated areas: rapid estimation of exposure maps in Trentino (NE Italy) The data are coarser than what a research-grade broadband station would deliver, but the density more than compensates for individual sensor limitations.
The second approach skips dedicated hardware altogether. Android smartphones contain accelerometers, and Google’s earthquake detection system uses the global Android phone network as a distributed seismometer array. Over three years of operation, the system detected an average of 312 earthquakes per month in Türkiye alone, with magnitudes ranging from 1.9 to 7.8, and delivered alerts in 98 countries for events above magnitude 4.5, corresponding to roughly 60 events and 18 million alerts per month.24PubMed. Global earthquake detection and warning using Android phones The earlier MyShake project at UC Berkeley pioneered this idea, developing an app that could distinguish earthquake shaking from ordinary phone movements and relay the detection to a central system.25PubMed Central. MyShake: A smartphone seismic network for earthquake early warning and beyond These phone-based systems will never replace professional instruments for research, but they democratize early warning in regions that cannot afford dense conventional networks.
Seismometers on Mars
Perhaps the most dramatic deployment of a seismometer happened about 225 million kilometers from Earth. NASA’s InSight lander placed the SEIS instrument directly on the Martian surface in late 2018 using a robotic arm, a major improvement over the Viking-era approach of the 1970s, when seismometers were mounted on the lander body and picked up every vibration from wind buffeting the spacecraft. SEIS was protected by a dedicated thermal and wind shield to isolate it from the harsh Martian environment.26PubMed. SEIS: Insight’s Seismic Experiment for Internal Structure of Mars Over the course of the mission, SEIS detected hundreds of marsquakes, mapping the planet’s crust, mantle, and core for the first time. The data revealed that Mars has a thinner crust than many models had predicted and a surprisingly large liquid core, reshaping our understanding of how rocky planets form and cool.
Earthquakes on Earth also leave traces far above the ground. Large quakes generate acoustic and gravity waves that propagate upward into the ionosphere, causing measurable disturbances in the density of electrons overhead. GPS receivers can detect these co-seismic ionospheric disturbances by tracking subtle changes in satellite signal delay. After a magnitude 7.9 earthquake in Alaska in 2018, ionospheric disturbances appeared 8 to 12 minutes after the mainshock, propagating at horizontal velocities consistent with acoustic waves, with maximum amplitudes near the epicenter that faded with distance.27Atmosphere. Co-Seismic Ionospheric Disturbance with Alaska Strike-Slip Mw7.9 Earthquake on 23 January 2018 Monitored by GPS This is not seismometry in the traditional sense, but it extends the detection chain from the solid earth through the atmosphere to the edge of space, hinting at future systems that combine ground sensors and satellite observations for a more complete picture of seismic events.