Earthquake building-damage proxies via SAR coherence loss
Interferometric SAR coherence drops sharply where structures have collapsed or shifted, giving emergency managers a damage proxy within hours of an earthquake. The signal is real but indirect: coherence loss is not a body count or a building count, and field validation remains essential.
Sensors
- Sentinel-1 IW (C-band, ESA/Copernicus): 5 m × 20 m ground range detected resolution in Interferometric Wide swath mode, 250 km swath, 6-day repeat at the equator with both satellites, down to 1-2 days over seismically active regions due to orbit geometry. Free and open archive from 2014. Primary operational source for rapid coherence mapping.
- ALOS-2 PALSAR-2 (L-band, JAXA): L-band (1.27 GHz) penetrates vegetation canopy and interacts more deeply with building fabric than C-band, improving coherence discrimination in forested or mixed urban-rural settings. Stripmap mode delivers 3 m resolution; standard repeat is 14 days, though JAXA can task emergency acquisitions within 24-48 hours under disaster protocols.
- ICEYE X-band SAR constellation: Sub-metre resolution (spotlight mode: ~0.25 m × 0.5 m) with commercial tasking latency under 24 hours and revisit potentially same-day. X-band is sensitive to fine structural surface changes, but shorter wavelengths also increase sensitivity to atmospheric moisture, requiring careful baseline selection for coherence analysis.
- Capella Space X-band SAR: Spotlight imagery at 0.35 m resolution, on-demand tasking. Useful for corroborating coherence anomalies at individual-building scale in high-priority urban zones where Sentinel-1's 20 m pixel conflates multiple structures.
What coherence loss actually measures
Interferometric SAR coherence quantifies how consistent the phase relationship is between two radar acquisitions of the same patch of ground. When a building stands undisturbed, the scattering geometry is stable and coherence is high, typically above 0.6 in urban C-band pairs. When a building collapses, the scatterers shift, rotate, or disappear entirely. The phase relationship randomises. Coherence drops toward zero.
The physics is real and well-documented, but the interpretation requires care. Coherence also drops over vegetation (temporal decorrelation from wind and growth), over water, and wherever soil moisture changes sharply between acquisitions. An earthquake that ruptures irrigation channels or saturates soils with liquefaction can produce coherence loss that looks, pixel for pixel, identical to building collapse. This is not a flaw to be engineered away; it is a fundamental property of the measurement that analysts must account for every time.
The pre/post pair: what makes a usable acquisition
A coherence change map requires at least one pre-event image and one post-event image from the same orbital geometry, the same look angle, and ideally the same season. Sentinel-1's 6-day exact repeat (or 12-day for a single satellite) means a pre-event image is almost always in the archive. The post-event image must be acquired after the mainshock, and every hour of delay translates into uncertainty about aftershock-driven secondary damage.
Temporal baseline matters enormously. A 6-day pre/post pair over a dry urban area will show coherence loss almost exclusively where something physically changed. A 24-day pair will accumulate vegetation decorrelation that muddies the signal. In practice, Copernicus EMS rapid-mapping analysts prefer the shortest available temporal baseline that brackets the event. Perpendicular baseline (the spatial separation between orbits) should ideally be below 150 m for Sentinel-1 to avoid topographic phase contamination, though the constellation's tight orbital tube keeps this manageable.
L-band from ALOS-2 tolerates longer temporal baselines before decorrelating over vegetation, which is why JAXA's emergency observation requests are particularly valuable in tropical cities or mountainous terrain where C-band coherence degrades quickly even without an earthquake.
From coherence map to damage proxy: the classification step
Raw coherence values are continuous, not categorical. Turning them into a damage proxy requires thresholding or, more commonly, computing the coherence change between a reference pair (two pre-event images) and the co-seismic pair (pre- and post-event). The reference pair establishes the background decorrelation rate for that landscape. Pixels where co-seismic coherence drops significantly below the reference rate are flagged as anomalous.
The ARIA project at NASA JPL, which has produced rapid damage proxy maps for events including the 2023 Turkey-Syria earthquake sequence, uses exactly this approach, publishing damage proxy maps (DPMs) as GeoTIFF rasters with three damage likelihood classes. Copernicus EMS Rapid Mapping activations similarly produce grading maps that combine coherence change with optical change detection where cloud cover permits. Both services are explicit that the output is a proxy, not a verified damage assessment, and both recommend field-survey validation before using the maps for resource allocation.
Accuracy studies on past events suggest that high-coherence-loss pixels correlate with heavily damaged or collapsed structures at rates between roughly 70 and 85 percent, depending on building typology and the quality of the image pair. Reinforced concrete frame buildings produce stronger coherence loss signals than timber or adobe, which can deform significantly without fully randomising the radar return. That range of uncertainty is wide enough to matter in life-safety decisions.
Operational services that run this workflow today
Two services deliver coherence-based damage proxy products under operational mandates. ARIA (Advanced Rapid Imaging and Analysis), run by NASA JPL and Caltech, publishes DPMs freely within hours to days of major events, using Sentinel-1 data processed through the ISCE InSAR software stack. Products are distributed via the ASF DAAC. Copernicus EMS Rapid Mapping, activated by EU member states or qualifying third countries through the Emergency Management Service portal, produces grading maps that can include SAR coherence layers alongside optical damage assessment; activation typically occurs within hours of a request following a significant event.
JAXA operates its own rapid-response pipeline for ALOS-2, coordinated partly through the International Charter on Space and Major Disasters, which allows charter members to task commercial and agency sensors under a single activation. The Charter's workflow is covered separately in the sibling page on that topic.
Commercial X-band operators, ICEYE in particular, have supplied sub-24-hour coherence products to humanitarian organisations during recent events, though these are procured separately and at cost rather than through open-access mechanisms.
Honest limits before you brief a minister
Cloud cover does not affect SAR, which is one of the technique's genuine advantages over optical post-event imagery. Night acquisition is equally unproblematic. But several limits are non-trivial.
Sentinel-1's 20 m pixel conflates many individual structures in dense urban fabric. A single pixel flagged as high coherence loss might represent one collapsed tower or several damaged low-rise units. Sub-metre commercial SAR reduces this ambiguity but introduces its own complications: very high resolution images decorrelate faster over even minor surface changes, making the coherence threshold harder to set. Atmospheric water vapour, particularly in tropical coastal cities, adds phase noise that degrades coherence independently of damage.
Perhaps the most important operational limit is the absence of a pre-event image in the right geometry. For a sudden-onset event on a fault not previously monitored, the archive may hold only a 24-day-old pre-event acquisition, or one from a different ascending/descending pass. The resulting coherence map is still useful, but the noise floor is higher and the false-positive rate rises.
Satellize integrates Sentinel-1 coherence change analysis into its analytics stack and can scope a rapid-mapping pipeline for governments that want sovereign, in-country processing capacity rather than dependence on external activation queues. The practical question for any client is not whether the method works, it is whether the pre-event archive is deep enough and the post-event tasking fast enough to produce a product before the search-and-rescue window closes.
Typical figures
| Primary sensor frequency | C-band (5.405 GHz, Sentinel-1); L-band (1.27 GHz, ALOS-2 PALSAR-2); X-band (9.65 GHz approx., ICEYE, Capella) |
| Typical coherence map pixel size | 20–100 m (Sentinel-1 IW, depending on multi-looking); 3–10 m (ALOS-2 Stripmap); 1–5 m (ICEYE/Capella Spotlight) |
| Revisit interval (Sentinel-1 dual-satellite) | 6 days global average; 1–3 days over seismically monitored regions with ascending and descending passes combined |
| Post-event product latency (operational services) | ARIA DPM: typically 6–24 hours after post-event image availability; Copernicus EMS: 12–48 hours after activation |
| Minimum detectable damage unit | Approximately one Sentinel-1 resolution cell (~100 m²); individual building discrimination requires commercial sub-metre SAR |
| Temporal baseline (recommended) | Shortest available pair bracketing the event; ideally 6–12 days for C-band urban; up to 28 days tolerable for L-band |
| Perpendicular baseline (Sentinel-1 practical limit) | Below ~150 m preferred to limit topographic phase contribution |
| Archive depth (Sentinel-1) | Global coverage from April 2014 (Sentinel-1A); Sentinel-1B archive 2016–2021 |
| Delivery formats (ARIA/EMS) | GeoTIFF raster (damage proxy classes), KMZ, Shapefile polygon overlays; ARIA products via ASF DAAC |
| Cloud-cover sensitivity | None: SAR penetrates cloud and operates day/night |
Analytics Satellize can run
| Co-seismic coherence change map | Interferometric SAR coherence estimation from pre/post image pair; coherence change relative to reference pair baseline | GeoTIFF raster with continuous coherence change values, clipped to affected urban extent |
| Three-class damage proxy map | Thresholded coherence change classification (low/moderate/high anomaly) following ARIA DPM methodology | Polygon GIS layer with damage likelihood classes, compatible with UN OCHA common operational datasets |
| Building-footprint damage attribution | Zonal statistics of coherence change raster against OpenStreetMap or national cadastral building footprints | Tabular report of flagged buildings per administrative unit, exportable to humanitarian cluster dashboards |
| Multi-temporal damage progression layer | Sequential coherence change maps across aftershock sequence using successive Sentinel-1 acquisitions | Animated GIS layer or time-series chart showing spatial spread of coherence anomalies over 7–30 days post-event |
| False-positive screening layer | Cross-masking coherence anomalies against NDVI change (vegetation decorrelation), surface water extent, and liquefaction susceptibility zones | Revised damage proxy map with non-structural coherence loss pixels flagged or removed |
| Rapid field-validation prioritisation grid | Spatial clustering of high-anomaly pixels weighted by estimated population density (WorldPop or national census grid) | Ranked list of survey grid cells for ground-truth teams, formatted as a printable field map and GeoJSON |
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.