Volcano inflation and deflation monitoring by InSAR
Interferometric SAR detects ground swelling and subsidence above magmatic and hydrothermal systems with millimetre-scale sensitivity, giving volcanologists a continuous geodetic record that precedes eruptions by days to years.
Sensors
- Sentinel-1 A/B (C-band SAR): 5.6 cm wavelength; Interferometric Wide Swath mode delivers 5 × 20 m resolution across a 250 km swath. Six-day repeat at mid-latitudes with both satellites; now single-satellite after Sentinel-1B failure, giving 12-day repeat over most volcanoes. Free and open archive from 2014. Workhorse for Campi Flegrei and Fagradalsfjall operational monitoring.
- COSMO-SkyMed 1–4 / Second Generation (X-band SAR): 9.6 GHz; Stripmap mode gives 3 m resolution; Spotlight reaches ~1 m. One-day revisit possible by tasking multiple satellites. X-band is more sensitive to subtle surface change but loses coherence faster on vegetated or actively degassing terrain. Used operationally by INGV for Campi Flegrei and Etna.
- ALOS-2 PALSAR-2 (L-band SAR): 23.6 cm wavelength; maintains interferometric coherence through dense tropical vegetation and on rough lava flows where C- and X-band decorrelate entirely. 3–10 m resolution in Stripmap; 14-day nominal repeat, though tasking can shorten this. Critical for Hawaiian and Indonesian volcanic monitoring.
- RADARSAT-2 (C-band SAR): 5.6 cm wavelength; flexible beam modes from 3 m Ultra-Fine to 100 m ScanSAR. Commercially tasked, so revisit and geometry can be optimised for a specific volcano. Useful for ascending/descending pair acquisition to decompose line-of-sight displacement into vertical and horizontal components.
What the fringes are actually measuring
An InSAR interferogram encodes the difference in radar travel time between two passes over the same ground. Each full colour cycle, or fringe, represents half the sensor wavelength of displacement along the satellite's line of sight: roughly 2.8 cm for C-band Sentinel-1, about 1.2 cm for X-band COSMO-SkyMed. A Mogi-type inflation source, the idealised point-pressure model for a spherical magma body, produces a roughly circular bullseye pattern of fringes centred above the source. Count the fringes, fit the pattern, and you recover the source depth and volume change simultaneously.
The line-of-sight geometry matters enormously. Sentinel-1 ascending and descending passes look at the ground from different azimuths and incidence angles, typically 30–46 degrees from vertical. Combining both geometries allows separation of vertical uplift from horizontal outward motion, which is essential for distinguishing a shallow sill intrusion from a deep spherical source. Without at least two look directions, source geometry is underdetermined and model outputs carry large uncertainty. This is not a limitation of the physics; it is a geometry problem that careful acquisition planning can largely resolve.
Three volcanoes, three monitoring lessons
Campi Flegrei, the restless caldera beneath Naples, has been monitored continuously by InSAR since the ERS-1 era. Sentinel-1 and COSMO-SkyMed together now resolve uplift episodes at the centimetre level with near-weekly cadence. The caldera has risen more than 1.2 metres since 2005 in the current bradyseismic crisis, a figure well documented in published INGV bulletins. The deformation field is consistent with a shallow hydrothermal source at roughly 3 km depth, though the relative contributions of magmatic versus hydrothermal pressure remain actively debated. InSAR alone cannot resolve that ambiguity; it needs tiltmeter and GNSS corroboration.
Kīlauea's 2018 eruption and summit collapse produced dramatic subsidence, with the Halema'uma'u crater floor dropping more than 500 metres. InSAR coherence on fresh pahoehoe lava is poor at C-band because the surface changes between passes, but ALOS-2's longer L-band wavelength preserved coherence well enough to track the collapse geometry. Fagradalsfjall on Iceland's Reykjanes Peninsula, erupting intermittently from 2021, sits in a relatively low-vegetation environment where Sentinel-1 coherence is good, and the Icelandic Meteorological Office integrated Sentinel-1 time series with a dense GNSS network to track dyke propagation in near real time before each fissure opened.
The consistent lesson across all three is that InSAR provides the spatial context that point sensors cannot: a GNSS station tells you what the ground is doing at one location; an interferogram shows you the full deformation footprint and lets you distinguish a localised shallow source from a broad deep one.
Where coherence breaks down
Coherence, the statistical similarity between two SAR acquisitions that makes interferometry possible, degrades whenever the ground surface changes between passes. Active lava flows are the worst case: fresh surfaces, thermal emission, and rapid morphological change destroy C-band coherence within hours. Dense tropical rainforest is nearly as bad at C-band; vegetation canopy motion between passes introduces phase noise that masks real deformation signals. This is why ALOS-2 PALSAR-2 is indispensable for Indonesian and Papua New Guinean volcanoes, where C-band interferograms are often unusable.
Atmospheric delay is the other major error source. Water vapour in the troposphere adds a phase delay that mimics ground deformation, particularly in volcanic environments where hydrothermal activity creates strong local humidity gradients. Corrections using ERA5 reanalysis data or GACOS (Generic Atmospheric Correction Online Service) reduce but do not eliminate this noise. For slow deformation signals below roughly 1 cm per year, atmospheric artefacts can dominate the interferogram unless time-series methods such as SBAS or PS-InSAR are applied across many acquisitions to average them down.
Source modelling: what the inversion can and cannot tell you
Fitting a deformation pattern to a source model is an inverse problem with non-unique solutions. The Mogi model assumes a point pressure source in an elastic half-space; it is analytically simple and works surprisingly well for many calderas. Sill models (horizontal penny-shaped cracks) better fit flat, wide deformation patterns such as those seen at Yellowstone and some Andean volcanoes. Prolate spheroid models capture elongated magma bodies. All three assume a homogeneous elastic crust, which is demonstrably wrong beneath active volcanic systems where rock properties vary with temperature and alteration.
In practice, modellers run Bayesian inversions, often using Monte Carlo sampling, to recover posterior distributions on source parameters: depth, volume change, and geometry. The depth estimate is typically more reliable than the volume change, because volume change depends on assumed elastic moduli. Published uncertainties on Mogi source depths from Sentinel-1 data are commonly ±0.2–0.5 km for well-constrained cases. Volume change uncertainties are harder to quote without knowing the host-rock shear modulus, which is rarely measured directly. Honest reporting of these ranges is part of any credible monitoring product.
Integrating InSAR with tiltmeters and GNSS
InSAR is spatially rich but temporally coarse: a 6- or 12-day revisit misses rapid precursory signals that tiltmeters, which can resolve nano-radian ground tilt at one-second intervals, catch immediately. GNSS provides three-dimensional displacement at continuous stations but only at discrete points. The combination is genuinely more powerful than any single method. Tiltmeters and GNSS flag the onset of an unrest episode; InSAR then provides the spatial extent and source geometry that point sensors cannot.
Operational multi-parameter systems at observatories such as INGV-OV (Campi Flegrei), the Hawaiian Volcano Observatory, and the Icelandic Meteorological Office feed all three data streams into joint inversions. For a government or operator without a resident observatory, replicating that integration requires data-sharing agreements, processing pipelines, and analysts who understand the ambiguities in each method. Satellize's analytics capability spans the open-constellation InSAR processing side of that picture; the tiltmeter and GNSS streams come from the client's ground network or national agency feeds.
What a monitoring programme needs to specify upfront
Before commissioning InSAR-based volcano monitoring, a buyer needs to answer four questions. First, what is the expected deformation rate? Slow inter-eruptive creep at a few millimetres per year demands time-series stacking across many interferograms; rapid pre-eruptive inflation at centimetres per day can be captured in a single interferogram pair. Second, what is the vegetation and surface stability? This determines whether C-band Sentinel-1 suffices or whether L-band commercial tasking of ALOS-2 is required. Third, what is the acceptable latency? Open Sentinel-1 data arrives within 24 hours of acquisition via the Copernicus Dataspace; commercial X-band data can be faster but costs more. Fourth, what source model outputs are required, and for which audience? A civil protection agency needs probability-weighted alert levels; a research team needs full posterior distributions.
Getting those specifications right before data collection begins saves considerable rework. Satellize structures initial engagements around a requirements workshop for exactly this reason, drawing on the same scoping discipline used in the Tonga crop-estimation programme, where sensor choice and revisit cadence were fixed before a single image was ordered.
Typical figures
| Spatial resolution (Sentinel-1 IW mode) | 5 × 20 m ground range × azimuth; multi-looked to ~14 × 14 m for interferometry |
| Spatial resolution (COSMO-SkyMed Spotlight) | ~1 m; Stripmap 3 m |
| Spatial resolution (ALOS-2 PALSAR-2 Stripmap) | 3–10 m depending on mode |
| Revisit (Sentinel-1, single satellite post-1B failure) | 12 days at most latitudes; 6 days in overlap zones |
| Line-of-sight displacement sensitivity | ~1–3 mm per interferogram after atmospheric correction; ~0.5 mm/year in stacked time series |
| Minimum detectable volume change (Mogi model) | Approximately 10^5–10^6 m³ for sources shallower than 5 km, depending on noise floor |
| Archive depth (Sentinel-1) | From April 2014 (Sentinel-1A launch); ERS-1/2 and Envisat extend usable archive to 1991 for some sites |
| Radar frequency / wavelength | C-band 5.405 GHz / 5.6 cm (Sentinel-1, RADARSAT-2); X-band 9.6 GHz / 3.1 cm (COSMO-SkyMed); L-band 1.236 GHz / 23.6 cm (ALOS-2) |
| Delivery formats | GeoTIFF displacement maps, NetCDF time-series stacks, KMZ visualisation layers, PDF bulletin with source model outputs |
| Typical processing latency (Sentinel-1 open data) | 24–72 hours from satellite acquisition to interferogram; 5–10 days for full time-series update |
Analytics Satellize can run
| Wrapped and unwrapped interferogram | Two-pass differential InSAR using TOPSAR burst alignment and Goldstein phase filtering | GeoTIFF interferogram pair with coherence mask and unwrapped line-of-sight displacement map |
| Displacement time series | SBAS (Small Baseline Subset) or PS-InSAR stacking across multi-year Sentinel-1 archive | NetCDF stack of cumulative LOS displacement per pixel; CSV of mean velocity and seasonal signal per benchmark point |
| Source geometry inversion | Bayesian Mogi, sill, and prolate spheroid forward modelling with Markov Chain Monte Carlo sampling | PDF report with posterior distributions on source depth, volume change, and geometry; ranked model comparison table |
| Ascending / descending decomposition | Geometric decomposition of two look-direction LOS vectors into vertical and east-west displacement components | GeoTIFF vertical uplift map and horizontal displacement map with propagated uncertainty |
| Atmospheric delay correction | ERA5 or GACOS tropospheric phase screen subtraction | Corrected interferogram with before/after RMS noise comparison; correction quality flag per scene |
| Unrest alert bulletin | Threshold exceedance on cumulative LOS displacement and source volume change rate, cross-referenced with tiltmeter and GNSS feeds provided by client | Automated email or API alert with displacement map attachment and plain-language summary for civil protection use |
| Coherence loss detection on active lava fields | Temporal coherence mapping across C-band and L-band stacks to delineate zones of active surface change | GIS polygon layer of coherent versus decorrelated zones updated per acquisition cycle |
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.