How Satellites Detect Active Volcanoes

Volcanoes rarely erupt without warning. The problem is that those warnings are often subtle, and easy to miss.

A slight increase in temperature. A small area that stays warmer than usual. Changes that happen slowly, sometimes over months or years, often in places where no one is really looking.

Monitoring all of that from the ground is very complicated. Many volcanoes are remote, hard to access, or simply too dangerous to observe continuously. Of roughly 1,350 potentially active volcanoes worldwide, fewer than a quarter have multi-parameter ground networks. That’s why more and more of this work is moving to space.

Looking for heat, not just eruptions

When people picture volcanoes from space, they imagine lava flows glowing in false colour. That’s the least interesting part. What matters is what happens before, and it turns out to be far less dramatic than most people expect.

In 2021, a study in Nature Geoscience looked at five volcanoes that had erupted, and worked backwards through the satellite thermal archive. They found something remarkable. Over the span of multiple years, a broad, faint warming of up to about 1 °C in median radiant temperature had appeared across tens of square kilometres of the surrounding edifice. Reporting on the study put the lead times at roughly two years for Ontake and as much as seven for Calbuco.

Up to one degree. Across an entire mountainside. Years in advance.

That signal is invisible to anyone standing near the volcano. It’s below the threshold of most ground networks. But it is exactly the kind of thing a satellite measuring surface temperature, consistently, over a long enough baseline, can pull out of the noise. Two of the eruptions in that study, Ontake in 2014 and Calbuco in 2015, had not been anticipated by conventional monitoring at all.

The wavelengths doing the work

“Thermal imaging” is not specific to only one imaging band, and volcanic monitoring is a good illustration of why.

Long-wave infrared (roughly 10–12 µm) is where Earth’s surface emits most strongly at ambient temperatures. This is the band for measuring land surface temperature, and the band that picked up that subtle warming surrounding the volcano. It’s the tool for the slow signal.

Mid-wave infrared (3–5 µm) behaves very differently. The Planck curve is much steeper at shorter wavelengths, so when temperature rises, the emitted light at this frequency increases much faster than it does in the long-wave infrared region. A small, very hot feature (e.g. a lava lake, an active dome, an open vent at 200–700 °C) will dominate a mid-infrared pixel even if it occupies a tiny fraction of it. This is the basis of the two long-running global volcano alert systems, MODVOLC and MIROVA, both built originally on MODIS mid-infrared channels. MIROVA currently tracks over 200 volcanoes in near real time.

Short-wave infrared (1.6–2.2 µm) goes further in the same direction. Sentinel-2’s SWIR bands respond only to genuinely hot material, but they deliver it at a resolution of 20 m per pixel rather than MODIS’s 1 km. That resolution gap is important: contextual algorithms on Sentinel-2 SWIR flag small, weak, lower-temperature anomalies that moderate-resolution systems miss entirely. That is why MIROVA has since incorporated Sentinel-2 and VIIRS alongside its original MODIS feed.

That sensor diversification has become critical. MODIS, which underpinned two decades of global volcano thermal monitoring, is reaching end of mission: Terra’s full product generation ceased at the end of 2025 and Aqua follows in 2026. The archive it built remains invaluable as a baseline. The continuity has to come from somewhere else.

So the choice of band is really a matter of which question you need answers for. Is this whole mountain slowly warming? will need a long-wave infrared answer. Is there something hot at the vent right now? will need mid-wave or short-wave infrared ones. Serious monitoring needs both.

One image tells you nothing

A single thermal scene of a volcano is close to useless. Surfaces heat and cool with the time of day, the season, the weather, cloud shadow, recent rainfall, and the angle the sensor happened to be looking from. Any of those can produce a warm pixel that, on its own, means nothing.

The valuable signal only emerges from the sequence.

This is why operational systems don’t report temperature: they report Volcanic Radiative Power (VRP), the radiant energy output of the hot feature, tracked as a time series. A VRP curve tells you whether activity is building, holding steady, or winding down. That trajectory is the actual product. The individual images are just samples of it.

Which puts a hard requirement on the observing system: revisit rate. A time series with a two-week gap can’t resolve a change that develops over ten days. Terra and Aqua between them delivered up to four MODIS overpasses a day at 1 km, good cadence, coarse pixels. Sentinel-2 gives 20 m pixels every five days, fine detail, sparser sampling. Neither is sufficient alone, and with the MODIS record now closing, the gap between them is precisely where dedicated thermal constellations are being built.

Heat is one signal among several

Thermal data is powerful, but it is not the whole picture.

Magma moving underground deforms the surface above it, often by centimetres. InSAR, using radar satellites like Sentinel-1, measures that deformation directly. Unlike thermal sensors, radar also sees through clouds, which matters enormously for volcanoes that generate their own weather.

Rising magma also releases gas. SO₂ plumes are detectable from orbit by atmospheric instruments such as TROPOMI on Sentinel-5P, and a change in degassing rate is one of the more reliable indications that something is shifting at depth.

Systems like MOUNTS now fuse all three (Sentinel-1 deformation, Sentinel-2 thermal, Sentinel-5P gas) alongside ground seismicity. No single one of those signals is decisive. Together, they’re considerably harder to argue with.

Beyond the volcano

Volcanic activity disrupts far more than the ground around it.

In April 2010, Eyjafjallajökull put fine ash into European airspace and effectively closed it for six days. Over 100,000 flights cancelled. Around 10 million passenger journeys disrupted. Roughly 300 airports shut across two dozen countries, the worst air travel disruption since the Second World War, and about €1.3 billion (US$1.7 billion) in lost airline revenue.

The knock-on effects went much further than aviation. Kenyan flower exporters, cut off from European markets, were losing over a million dollars a day as perishable stock spoiled in storage.

While better thermal monitoring would not have prevented that eruption, nor the ash cloud that came with it, it’s in situations like this that it pays off to have the forewarning systems in place. The cost of watching is small and predictable. The cost of being surprised is neither.

What satellite monitoring actually changes

Not exact timing. Nobody can tell you when a volcano will erupt, and any system claiming otherwise should be treated with suspicion.

What changes is the baseline. When you have years of consistent thermal observations of a volcano, you know what normal looks like there, and “normal” is specific to each volcano, which is precisely why a global average is no use to anyone. Anomaly detection requires a long, uninterrupted record of non-anomalous behaviour. That record is the asset satellite monitoring provides.

There’s another increasingly relevant discussion point: who holds the data. Relying entirely on external monitoring services means accepting someone else’s revisit schedule, someone else’s processing choices, and someone else’s priorities about which volcanoes are worth watching. For agencies responsible for populations living on the flanks of an active volcano, independent observing capacity is as important as the observations themselves.

Quiet monitoring

None of this is about capturing spectacular events. The useful work happens years before there’s anything spectacular to capture: a degree of warming spread across a hillside, showing up in a time series that nobody would look at twice on any given day.

Volcanoes will remain unpredictable. But unpredictable and opaque are different problems, and only one of them is unavoidable.