Volcanic thermal anomaly detection for eruption onset warning
Shortwave and thermal infrared radiance from MODIS, VIIRS, Sentinel-2 and Landsat-9 reveal volcanic heating days to weeks before an eruption breaks the surface. MIROVA and MODVOLC translate raw radiance into alert thresholds that civil-protection agencies can act on.
Sensors
- MODIS Terra / Aqua: 1 km thermal infrared (bands 21, 22, 31, 32) and 500 m SWIR (band 7, 2.1 µm). Four overpasses per day per volcano. Basis of the MODVOLC and MIROVA alert systems. Saturates above roughly 500 K in standard TIR channels; band 21 extends to ~500 W m⁻² sr⁻¹ µm⁻¹ before saturation, making it the workhorse for high-temperature lava detection.
- VIIRS Suomi-NPP / NOAA-20: 375 m I-band resolution for the M13 (4.05 µm) and I4 (3.74 µm) mid-infrared channels, plus 750 m M-band thermal. VIIRS Nightfire and the FIRMS active-fire product both flag volcanic radiance anomalies. The finer pixel footprint compared with MODIS reduces the fractional-area ambiguity for small lava lakes and vent openings.
- Sentinel-2 MSI: 10–20 m optical and SWIR bands including B11 (1.61 µm) and B12 (2.19 µm) at 20 m. These SWIR channels are not saturated by lava temperatures in the 800–1200 K range, allowing sub-pixel lava area and radiative power estimates at spatial detail that MODIS cannot resolve. Five-day revisit at the equator (2–3 days with both satellites). Useless through thick cloud or volcanic ash.
- Landsat-9 OLI-2 / TIRS-2: OLI-2 SWIR bands at 30 m (1.61 µm and 2.20 µm) combined with TIRS-2 thermal at 100 m (resampled to 30 m). Sixteen-day repeat. Quantitative radiative flux estimation at 30 m resolution is a significant step up from MODIS, though the infrequent revisit makes it a confirmatory rather than monitoring sensor for fast-moving eruptions.
What a floating roof gives away
Every object emits thermal radiation in proportion to the fourth power of its surface temperature. A basaltic lava lake at 1100 K radiates roughly 82 000 W m⁻² from its surface. Even when that lake occupies only a small fraction of a 1 km MODIS pixel, the radiance contrast against cool surrounding rock is detectable. This is the physical foundation of every satellite-based volcanic alert system: you do not need to see the lava directly, only the anomalous radiance it injects into the sensor's field of view.
The practical consequence is that SWIR channels (roughly 1.6–2.2 µm) are far more sensitive to high-temperature volcanic features than standard thermal infrared. At 1000 K, the Wien peak of emitted radiation sits near 2.9 µm, so SWIR channels sit on the rising limb of the Planck curve for lava and are orders of magnitude brighter than for background terrain at 300 K. Standard TIR channels (10–12 µm) saturate quickly on exposed lava but remain useful for detecting lower-temperature features such as cooling lava fields and fumarolic ground heated to 350–500 K.
MIROVA and MODVOLC: what the alert systems actually do
MODVOLC, developed at the University of Hawaii, applies a Normalised Thermal Index threshold to MODIS band 21 and band 32 data. Any pixel exceeding the threshold triggers an alert, typically within hours of the satellite overpass. The system has been running continuously since 2000, giving a 24-year archive of volcanic thermal anomalies across every active volcano on Earth.
MIROVA (Middle InfraRed Observation of Volcanic Activity), developed jointly by the universities of Turin and Florence, takes a different approach. It calculates Volcanic Radiative Power (VRP) in megawatts from MODIS band 21 radiance, correcting for background thermal emission and atmospheric path. VRP correlates empirically with lava effusion rate, which means MIROVA outputs are not just binary alerts but a proxy for eruption intensity. Published studies on Etna, Kilauea and Stromboli show that rising VRP trends in the days before a major eruptive episode often precede visible surface activity, giving civil-protection agencies a quantitative early-warning signal rather than a simple on/off flag.
Both systems are openly accessible and run on near-real-time MODIS data. FIRMS (Fire Information for Resource Management System) at NASA also ingests VIIRS and MODIS detections and is freely queryable. The practical latency from satellite overpass to published alert is typically two to four hours.
Cloud cover: the persistent adversary
Persistently cloudy volcanoes, particularly in tropical island arcs such as Vanuatu, the Philippines or the Lesser Antilles, can remain obscured for days at a time. Cloud opacity in TIR and SWIR bands is essentially total: a thick cumulus deck at 5 km altitude blocks all surface radiance regardless of how intense the thermal anomaly below it is. This is not a limitation that better algorithms overcome; it is a physical constraint.
The practical mitigation is multi-sensor fusion. With MODIS Terra, MODIS Aqua, Suomi-NPP VIIRS and NOAA-20 VIIRS, a given location receives eight or more overpasses per day. The probability of at least one cloud-free acquisition over a 24-hour window is substantially higher than any single overpass would suggest, particularly at volcanoes where convective cloud has a diurnal cycle. Sentinel-2 and Landsat-9 add spatial detail on the days they do acquire clear imagery. Synthetic aperture radar (SAR) can detect topographic change from lava flow emplacement through cloud, though that is addressed separately in the volcanic lahar and edifice-deformation pages in this library.
Sentinel-2 SWIR: resolution where it matters
The spatial resolution gap between MODIS (500–1000 m) and the volcanic features of interest is significant. A lava lake 50 m across, a vent opening 20 m wide, a small hornito field: all of these are sub-pixel at MODIS resolution. The thermal anomaly is still detectable as a radiance excess, but the geometry is lost. Sentinel-2 MSI at 20 m SWIR resolution recovers that geometry.
At 20 m, Sentinel-2 B12 (2.19 µm) can distinguish individual lava flow lobes, map the active flow front position to within a few tens of metres, and resolve whether a thermal anomaly originates from a summit crater or a flank vent. These distinctions matter operationally: a flank vent opening at low elevation on a populated volcano is a very different hazard from a summit lava lake that has been stable for years. The tradeoff is revisit. Sentinel-2 offers five-day repeat at best, against four overpasses per day from MODIS. For eruption onset warning, MODIS and VIIRS carry the daily monitoring load; Sentinel-2 provides the high-resolution contextual image when conditions allow.
From radiance to decision: what an operational service looks like
A working volcanic thermal anomaly service ingests FIRMS alerts, computes VRP time series from MODIS and VIIRS radiance, flags trend changes against a volcano-specific baseline, and delivers that information to a civil-protection agency in a form they can act on. The baseline matters: Stromboli is almost always thermally anomalous; a threshold calibrated for a dormant stratovolcano would produce continuous false alarms there. Each volcano needs its own historical distribution of VRP values to define what constitutes a meaningful departure.
Satellize runs this class of analytics on open constellation data, with commercial high-resolution tasking added when a client needs sub-10 m SWIR imagery for flow mapping. For national civil-protection agencies managing multiple volcanoes, the most useful deliverable is usually a daily alert digest with VRP trend plots and the most recent cloud-free Sentinel-2 SWIR composite, delivered before the morning operational briefing.
Honest limits: no satellite system currently provides continuous observation of a single volcano. Gaps of four to six hours between overpasses are normal, and cloud can extend those gaps to days. Satellite thermal data is a monitoring layer, not a substitute for ground-based seismic and geodetic networks. It is most powerful when fused with those networks, providing an independent, spatially explicit confirmation of what the seismometers are suggesting.
Typical figures
| Spatial resolution (thermal monitoring) | MODIS: 1 km TIR, 500 m SWIR (band 7); VIIRS: 375 m (I-band), 750 m (M-band) |
| Spatial resolution (high-res SWIR) | Sentinel-2 MSI: 20 m (B11, B12); Landsat-9 OLI-2: 30 m SWIR; Landsat-9 TIRS-2: 100 m TIR |
| Revisit (MODIS + VIIRS combined) | Up to 8 overpasses per day at mid-latitudes across Terra, Aqua, Suomi-NPP and NOAA-20 |
| Revisit (Sentinel-2 / Landsat-9) | Sentinel-2: 5 days equatorial (2–3 days with both satellites); Landsat-9: 16 days |
| Alert latency (MODVOLC / FIRMS) | Typically 2–4 hours from satellite overpass to published alert |
| Key spectral bands | SWIR: 1.6 µm, 2.1–2.2 µm; Mid-IR: 3.7–4.1 µm; TIR: 10–12 µm |
| Minimum detectable thermal anomaly | MODVOLC detects sub-pixel lava features; published detection limit for MODIS band 21 is approximately 0.5 MW radiative power under clear-sky conditions |
| Archive depth | MODVOLC: 2000 to present; MODIS FIRMS: 2000 to present; Sentinel-2: 2015 to present; Landsat: 1972 to present (TIR from Landsat 4, 1982) |
| Cloud penetration | None in optical/TIR/SWIR bands; SAR required for cloud-penetrating surface change detection |
| Volcanic Radiative Power range | Published MIROVA values range from <1 MW (fumarolic heating) to >10 000 MW (major effusive eruptions at Kilauea, Etna) |
Analytics Satellize can run
| Daily VRP time series per volcano | MIROVA radiative power algorithm applied to MODIS band 21 radiance; volcano-specific background baseline subtraction | Daily chart and CSV feed of Volcanic Radiative Power in MW, with trend alert flag when 7-day rolling mean exceeds baseline by a defined sigma threshold |
| Near-real-time thermal alert digest | MODVOLC NTI threshold and FIRMS VIIRS active detections, aggregated per volcano and filtered against known false-positive sources | Morning briefing report (PDF and API JSON) listing all flagged volcanoes, overpass times, and detection confidence |
| Sentinel-2 SWIR lava flow front mapping | Band ratio and thresholding on B12 and B11 reflectance; comparison with prior clear-sky acquisition to isolate new thermal pixels | GeoTIFF and GeoJSON polygon of active flow extent, updated on each cloud-free acquisition; area and advance rate calculated where two sequential images allow |
| Multi-sensor cloud-gap analysis | Cloud-mask statistics across all available overpasses (MODIS, VIIRS, Sentinel-2) for a defined volcano footprint over a rolling 30-day window | Monitoring reliability score: percentage of 6-hour windows with at least one usable thermal observation, informing whether ground-sensor augmentation is warranted |
| Historical eruption onset chronology | Retrospective MIROVA and MODVOLC archive query; identification of VRP inflection points relative to documented eruption dates from the Smithsonian Global Volcanism Program | Volcano-specific lead-time analysis report showing how many days of precursory thermal anomaly preceded past eruptive episodes |
| Fumarolic field temperature trend | Landsat-9 TIRS-2 land surface temperature retrieval using split-window algorithm; time series of mean LST over defined fumarolic zones | Quarterly LST trend plot per fumarolic zone, with anomaly flags referenced to the 16-day Landsat archive |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.