Volcanologists draw on a layered toolkit of technologies, from underground seismometers to orbiting radar satellites, to detect the warning signs that a volcano may erupt. No single instrument does the job alone. Eruptions are preceded by a mix of signals: the ground swells, earthquakes cluster, gases shift in composition, and surface temperatures rise. Modern forecasting stitches together data from all of these streams, and increasingly relies on machine learning to spot patterns humans miss. The picture that emerges is impressive but imperfect, and some types of eruptions remain stubbornly difficult to anticipate.
Seismic Monitoring
Seismometers remain the oldest and most fundamental tool in volcano surveillance. Magma forcing its way toward the surface cracks rock, generates tremor, and produces distinctive low-frequency and high-frequency earthquake signals. By tracking how often these quakes occur, where they cluster, and how their character changes over time, scientists can gauge whether a volcano is becoming restless. Networks of seismometers are typically installed on and around the volcano’s flanks so that the location of underground activity can be triangulated in three dimensions.
The challenge is volume: an active volcano can produce thousands of small seismic events per day, and distinguishing a volcanic earthquake from a regional tectonic quake or a rockfall requires skill and speed. At Anak Krakatau in Indonesia, researchers deployed artificial neural networks trained to automatically detect and classify the different signal types, including long-period events, volcano-tectonic earthquakes, and tremor, so that activity levels could be estimated continuously without a human reviewing every trace.
Ground Deformation
When magma accumulates underground, the ground surface above it deforms. It inflates like the skin of a balloon being filled, sometimes by centimeters, sometimes by mere millimeters spread over months. Measuring that deformation with high precision is one of the most reliable ways to track magma movement.
Ground-based instruments include GPS receivers (now more broadly called GNSS), tiltmeters that detect changes in slope angle, and strainmeters that measure how rock stretches or compresses. At Sakurajima volcano in Japan, researchers combined data from all three sensor types during an eruptive episode between 2011 and 2012 and identified three separate pressure sources beneath the volcano, at depths of roughly 0.7 km, 3.3 km, and 9.6 km, each inflating or deflating at different rates.1ScienceDirect. Multiple-pressure-source model for ground inflation during the period of high explosivity at Sakurajima volcano, Japan That kind of three-dimensional picture helps scientists understand not just that magma is moving, but where it is sitting and how it connects between deep reservoirs and the shallow vent.
Satellite Radar and InSAR
You cannot bolt a GPS receiver to every volcano on Earth. Many of the world’s roughly 1,500 potentially active volcanoes sit in remote jungles, on uninhabited islands, or under ice caps, with no ground instruments at all. Satellite radar fills that gap. A technique called Interferometric Synthetic Aperture Radar, or InSAR, compares radar images of the same patch of ground taken on different passes to detect surface displacement down to millimeter scale.
At Asama volcano in Japan, researchers combined 120 radar images from the European Sentinel-1 satellite with 20 from the Japanese ALOS-2 satellite, acquired between 2014 and 2018, and detected asymmetric deformation on two flanks of the volcano that would have been invisible from ground stations alone.2Earth, Planets and Space. Surface deformation of Asama volcano, Japan, detected by time series InSAR combining persistent and distributed scatterers, 2014‒2018 In northern Ecuador, the same technique was applied to four volcanoes between 2017 and 2024, revealing low-magnitude inflation at the Pichincha Volcanic Complex and Pululahua, while a more complicated pattern at the Cayambe Volcanic Complex appeared to be tangled up with tectonic and human-caused ground movement.3Environmental Challenges. Deformation monitoring in northern Ecuadorian volcanoes using the InSAR technique
InSAR’s strength is its ability to survey volcanoes globally with no equipment on the ground. Its weakness is that cloud cover, vegetation, and snow can degrade the signal, and repeat visits from the satellite are spaced days to weeks apart, so rapidly evolving crises can slip between observations.
Volcanic Gas Measurements
Magma carries dissolved gases, primarily water vapor, carbon dioxide, and sulfur dioxide. As magma rises and pressure drops, those gases escape, much like bubbles forming in an opened soda bottle. Shifts in the composition of the gas plume, or in the ratio of one gas to another, can signal that fresh magma is approaching the surface.
At Mount Etna, researchers made continuous real-time measurements of CO₂ concentration and the CO₂-to-SO₂ ratio in the volcanic plume over five consecutive days, and observed both tracers peaking on a day when an eruption was already underway, consistent with early degassing of CO₂ from newly ascending magma.4ScienceDirect. Real-time measurements of δ13C, CO2 concentration, and CO2/SO2 in volcanic plume gases at Mount Etna, Italy, over 5 consecutive days CO₂ tends to bubble out of magma at greater depth than SO₂, so a rising CO₂-to-SO₂ ratio can be an early indicator that a new batch of magma has entered the system from below. Ground-based spectrometers, ultraviolet cameras, and even sensors mounted on drones now allow gas flux to be measured in near real time at many volcanoes.
Thermal Imaging From Space
Satellites carrying thermal infrared sensors can detect hotspots on and around a volcano’s summit. A rise in surface temperature may indicate that hot magma or hydrothermal fluids have migrated closer to the surface. At Merapi volcano in Indonesia, satellite thermal imagery identified positive thermal anomaly areas that matched heat sources independently revealed by seismic imaging and resistivity surveys, demonstrating that the remote-sensing approach can complement ground-based monitoring by providing a broader spatial view and more frequent snapshots.5Journal of Volcanology and Geothermal Research. Spatio-temporal surface temperature variations detected by satellite thermal infrared images at Merapi volcano, Indonesia
Thermal monitoring is especially useful at volcanoes with lava domes. A dome that is quietly cooling shows one temperature profile; one being fed by new magma from below shows another. Tracking those changes over weeks or months helps forecasters judge whether an eruption is building.
Infrasound
Explosive volcanic eruptions send powerful low-frequency sound waves through the atmosphere, well below the range of human hearing. Infrasound stations, originally built to monitor nuclear test-ban compliance, can pick up these signals from enormous distances. In fact, infrasound is the only ground-based monitoring technique capable of detecting explosive eruptions from thousands of kilometers away.6Scientific Reports. Long range infrasound monitoring of Etna volcano That makes it invaluable for remote or unmonitored volcanoes: even if there is no seismometer within hundreds of kilometers, a distant infrasound array may capture evidence of an eruption within minutes.
The global International Monitoring System maintains dozens of infrasound stations, and their data are increasingly being combined with satellite imagery and seismic records to build a near-real-time picture of volcanic activity worldwide.
Machine Learning and Eruption Forecasting
The sheer volume of data generated by modern monitoring networks has created an opening for machine learning. Algorithms can sift through continuous seismic records, satellite imagery, and gas measurements to identify subtle patterns that precede eruptions, patterns that a human analyst might overlook or recognize only in hindsight.
A striking example comes from Axial Seamount, an underwater volcano on the Juan de Fuca Ridge. Researchers applied unsupervised machine learning to the spectral features of volcanic earthquakes recorded there and discovered a class of mixed-frequency signals that rapidly increased in number about 15 hours before an eruption began. These signals had not been flagged by conventional analysis.7Geophysical Research Letters. Volcanic Precursor Revealed by Machine Learning Offers New Eruption Forecasting Capability Machine learning has also been applied to classify seismic events at Krakatau and to detect volcanic ash clouds from space, where a convolutional neural network architecture achieved roughly 90% accuracy in identifying ash plumes in satellite imagery, outperforming older threshold-based methods.8ScienceDirect. Enhancing detection of volcanic ash clouds from space with convolutional neural networks
The promise of these tools is not that they replace volcanologists, but that they compress the time between a precursory signal appearing in raw data and someone recognizing it. At volcanoes with short warning windows, those extra hours matter.
Drones and Close-Range Surveys
Uncrewed aerial vehicles have become a practical way to get instruments close to volcanic vents that would be too dangerous for a person to approach. At La Fossa cone on the Italian island of Vulcano, researchers used drone-based photogrammetry combined with thermal cameras to map the fumarole field in detail. They identified roughly 70,000 square meters of hydrothermally altered ground and showed that high-temperature fumaroles accounted for less than half of the total thermal energy being released; the rest came from broader zones of diffuse degassing.9Copernicus Publications (Solid Earth). Anatomy of a fumarole field: drone remote-sensing and petrological approaches reveal the degassing and alteration structure at La Fossa cone, Vulcano, Italy Without drones, that kind of spatial resolution would have been extremely difficult to obtain.
Drones can also carry miniaturized gas sensors directly into a volcanic plume, collect rock and ash samples from freshly erupted material, and generate high-resolution 3D terrain models that help scientists detect even small surface changes between visits.
Muon Radiography
One of the more exotic tools in the volcanologist’s kit uses cosmic rays. High-energy particles called muons rain down through the atmosphere constantly. Dense rock absorbs more muons than lighter material, so by placing detectors on the flank of a volcano and measuring how many muons make it through, scientists can create a kind of X-ray image of the interior. At Showa-Shinzan lava dome in Japan, this technique was used to image the volcanic conduit and determine its diameter at roughly 100 meters.10Geophysical Research Letters. Imaging the conduit size of the dome with cosmic‐ray muons: The structure beneath Showa‐Shinzan Lava Dome, Japan At Satsuma-Iwojima volcano, muon imaging successfully mapped the density distribution inside the conduit, revealing details of the degassing process.11Geophysical Research Letters. Cosmic‐ray muon imaging of magma in a conduit: Degassing process of Satsuma‐Iwojima Volcano, Japan
Muon radiography is slow (detectors need to accumulate data for weeks or months) and works best on small, steep-sided volcanoes where the detector can be positioned at a favorable angle. It is not a real-time warning tool, but it provides structural information about conduit geometry that no other method can match, and that information feeds into models of how an eruption might unfold.
Monitoring Submarine Volcanoes
A large fraction of Earth’s volcanic activity takes place on the ocean floor, out of sight and beyond the reach of most conventional monitoring tools. Submarine eruptions cannot be observed visually, but they do produce seismo-acoustic signals, combinations of seismic waves propagating through rock and hydroacoustic waves traveling through the water column, that can be picked up by regional and global sensor networks.12Journal of Geophysical Research: Solid Earth. The Seismo‐Acoustics of Submarine Volcanic Eruptions Cabled ocean-floor observatories, like the one monitoring Axial Seamount, add continuous seismic, pressure, and chemical data that make it possible to track magma intrusion in real time even beneath kilometers of seawater.
Crystal Clocks and Petrological Forensics
Not all prediction tools involve electronic instruments. The crystals that grow inside magma as it cools carry a chemical record of their own history. By measuring how elements like magnesium or iron have diffused across zones within a single crystal, scientists can estimate how long ago that crystal was disturbed by, say, an injection of fresh hot magma. This approach, called diffusion chronometry, essentially turns each crystal into a tiny stopwatch.
At Kīlauea, researchers used crystal and melt inclusion timescales to trace the path of magma migration before eruption, finding that crystals recording the highest pressures (deepest storage) also preserved the longest diffusion times, meaning they started their journey earliest.13Nature Communications. Crystal and melt inclusion timescales reveal the evolution of magma migration before eruption Work on plutonic rocks has extended these methods to longer timescales, with crystal-melt segregation timescales on the order of tens of thousands to hundreds of thousands of years.14PubMed Central. Cooling rates and melt extraction timescales determined by diffusion chronometry on shallow crustal plutonic rocks These techniques do not provide real-time warnings, but they help scientists build a picture of how quickly a magma system can recharge and what the lead time to an eruption might look like in the geological record.
Crowdsourced Observations
Increasingly, data from non-specialists are filling gaps in official monitoring networks. Citizen science seismometers like those in the Raspberry Shake network, personal weather stations, and even smartphone-based felt-earthquake reports add density to observations. During the 2022 Hunga eruption in Tonga, researchers combined data from official weather stations and seismometers in New Zealand with readings from citizen-science seismo-acoustic stations to map the atmospheric waves generated by the blast across the country.15Communications Earth & Environment. Crowdsourcing human observations expands and enhances volcano monitoring records Crowdsourced data are noisier and less calibrated than professional instruments, but they expand coverage in regions where official stations are sparse, and they sometimes catch transient events that fixed networks miss.
Why Forecasting Still Fails
With all these tools, you might expect eruption forecasting to be highly reliable. It is not. Long-term forecasts, which estimate how likely a volcano is to erupt in the coming decades, rely mainly on geological and historical records of past eruptions.16Journal of Geophysical Research: Solid Earth. Partly Cloudy With a Chance of Lava Flows: Forecasting Volcanic Eruptions in the Twenty‐First Century Short-term forecasts depend on recognizing precursory patterns in monitoring data. The trouble is that many episodes of volcanic unrest, complete with earthquakes, deformation, and gas emissions, never lead to an eruption. A recent analysis across multiple volcanoes found that of 26 instances where monitoring data triggered the highest alert level, only 12 were followed by eruptions, a success rate of about 39%. Five eruptions were missed entirely.17Scientific Reports. Near real-time b-value analysis for volcano traffic light alert systems and eruption forecasting False alarms erode public trust and complicate evacuation decisions, while missed eruptions can cost lives.
The hardest eruptions to anticipate are phreatic explosions, which are driven by the sudden flashing of groundwater to steam rather than by ascending magma. Because these events can occur with few or no conventional precursors, they catch even well-monitored volcanoes off guard.18PubMed Central. Understanding and forecasting phreatic eruptions driven by magmatic degassing The 2014 eruption at Mount Ontake in Japan, which killed 63 hikers, was a phreatic event with minimal warning despite dense instrumentation.
Ash Cloud Detection and Aviation Safety
Volcanic ash is invisible on standard weather radar and can destroy a jet engine. When an eruption sends an ash plume into the flight corridors used by commercial aviation, detecting and tracking that plume becomes an urgent safety problem. Volcanic Ash Advisory Centers around the world use a combination of satellite imagery, atmospheric dispersion models, and pilot reports to issue warnings. The satellite detection side has improved significantly with machine learning: a hybrid neural network architecture trained on satellite data achieved 90% accuracy in distinguishing ash from ordinary cloud, with precision, recall, and related metrics all above 85%.8ScienceDirect. Enhancing detection of volcanic ash clouds from space with convolutional neural networks Older methods that relied on simple temperature-difference thresholds between infrared channels were less accurate, especially when ash mixed with water ice or thin cirrus clouds.
Accurate ash detection does not prevent an eruption, but it prevents the downstream economic and safety consequences from being worse than they need to be. The 2010 Eyjafjallajökull eruption in Iceland closed European airspace for six days, partly because detection and modeling tools at the time could not confidently define where ash concentrations were safe to fly through. Better detection tools reduce the size of the no-fly zone and keep it centered on the actual hazard.
How Alert Systems Translate Data Into Action
All the monitoring data in the world is useless if it does not reach the right people in time. Volcano observatories communicate changing risk through volcano alert level systems, standardized color-coded or numbered scales that summarize a volcano’s current state for civil authorities, emergency managers, and the aviation sector. The U.S. Geological Survey, for instance, uses a four-tier aviation color code (green, yellow, orange, red) alongside a parallel ground-based alert level. These systems are deliberately simple, designed to bridge the gap between scientists who think in probabilities and officials who need a clear recommendation.19PubMed Central. Volcano alert level systems: managing the challenges of effective volcanic crisis communication
The simplicity is also a limitation. Condensing complex, ambiguous monitoring data into a single color forces difficult judgment calls. Raising the alert level prematurely triggers costly evacuations and economic disruption; raising it too late puts lives at risk. Observatories supplement the formal alert level with narrative information statements, hazard maps, and direct conversations with decision-makers, because a single color code cannot capture the nuances of an evolving volcanic crisis.