Search-and-rescue priority zone delineation after structural collapse
After an earthquake or explosion, satellite-derived damage proxies combined with building inventory and population data can rank urban blocks by probable casualty concentration, giving search-and-rescue commanders a probabilistic triage map within hours, not days.
Sensors
- ICEYE SAR constellation: X-band synthetic aperture radar, spotlight mode delivers roughly 0.5 m ground resolution; the constellation can revisit a target city multiple times per day. Coherence change between a pre-event and post-event pair reveals where scatterers on building surfaces have shifted or disappeared, a reliable proxy for structural change. Cloud and darkness are irrelevant.
- Capella Space SAR: Also X-band, with spotlight products at approximately 0.5 m resolution. Capella's sliding spotlight and stripmap modes allow rapid area coverage at 1 m resolution when city-wide mapping is needed quickly. Tasking latency from order to first collect can be under six hours in good orbital geometry.
- Pleiades Neo: Airbus optical constellation offering 30 cm panchromatic resolution. In clear sky conditions, pre- and post-event image pairs support pixel-level change detection: collapsed floors, exposed rebar, debris fans spreading into streets. Revisit over a given city is roughly once every 24 hours across the four-satellite constellation, though cloud cover can eliminate a pass entirely.
- WorldView-3: Maxar optical satellite with 31 cm panchromatic and 1.24 m multispectral resolution, plus a short-wave infrared band set that can detect heat signatures through thin smoke. Useful for both visual damage classification and, in the SWIR bands, locating residual thermal anomalies in rubble that may indicate survivors or fire risk.
- Sentinel-1 SAR (ESA/Copernicus): C-band SAR at 5 m to 20 m resolution depending on mode. Free and open; typically 6-day repeat at mid-latitudes but can be tasked to 1-3 day repeat in emergency mode. Resolution is too coarse for individual building assessment but adequate for block-level coherence loss mapping across an entire city, useful for initial triage before commercial tasking is confirmed.
What the satellite actually sees, and what it does not
A SAR coherence map does not show casualties. It shows where the radar signal backscattered from a surface has changed between two acquisitions. A building that has partially pancaked will show low coherence over its footprint; a building that has shifted on its foundations without collapsing may show moderate coherence loss; a building that has burned out may show high coherence loss with no structural collapse at all. The proxy is real, but the ambiguity is also real.
Very-high-resolution optical imagery adds a second, more interpretable signal. A collapsed floor produces a debris fan, a change in roof texture, a loss of shadow geometry. Human analysts and machine-learning classifiers trained on labelled post-earthquake imagery can assign damage grades broadly consistent with the European Macroseismic Scale or the HAZUS framework. Accuracy at the individual building level in published studies typically sits between 70 and 85 percent for the binary collapsed/standing distinction, falling when damage is partial or when building types are heterogeneous. That figure needs to be on the table from the start.
Turning damage proxies into a priority grid
The analytic value is not in the damage map alone. It comes from the spatial join with two pre-existing data layers: a building footprint inventory with structural attributes (construction type, number of storeys, build year where available), and a population density grid. WorldPop and Meta's High Resolution Settlement Layer both provide gridded population estimates at 100 m resolution for most earthquake-prone countries, and both are openly accessible. Multiplying estimated occupancy by collapse probability yields a probabilistic casualty-concentration score per block.
The output is a ranked list of urban blocks, not a map of confirmed casualties. Blocks in the top decile of the score distribution become the first priority for ground teams. This is standard triage logic, the same principle used in hospital mass-casualty protocols, applied spatially. The satellite layer compresses the time needed to generate that ranking from days of ground reconnaissance to a few hours of processing after the first post-event image is acquired.
Population grids carry their own uncertainty. Census data in rapidly urbanising cities can be years out of date. Informal construction on upper floors, sub-divided apartments, and nighttime versus daytime occupancy patterns all shift the true occupancy away from the modelled figure. The priority grid should be presented with explicit confidence bands, not as a deterministic answer.
The 72-hour clock and where satellite data fits inside it
Survival probability after structural collapse drops sharply with time. Published search-and-rescue literature places the median survival time for trapped casualties at roughly 24 to 48 hours, with a long tail extending to 72 hours and beyond. The satellite workflow has to deliver something actionable before ground teams have already committed resources by other means.
In practice, the first usable SAR coherence pair requires a pre-event image in the archive (almost always available for any populated city) and a post-event acquisition. For a major earthquake, ICEYE or Capella can typically deliver a first post-event collect within 6 to 12 hours of tasking, depending on orbital geometry and ground-station access. Processing a coherence map from a single interferometric pair takes under an hour with standard software. The spatial join with building and population data adds another hour if the layers are pre-staged. A first-pass priority grid is therefore achievable within 12 to 18 hours of the event, well inside the critical window.
Optical imagery is faster to interpret visually but dependent on cloud cover. In the immediate aftermath of a major earthquake, dust and smoke frequently obscure the scene for 12 to 24 hours. SAR is therefore the primary sensor for the first pass; optical imagery serves as confirmation and for finer-grained damage classification once conditions clear.
False positives, missed detections, and how to communicate uncertainty to commanders
False positives are the dominant operational risk. A block flagged as high priority that contains no casualties draws search-and-rescue teams away from blocks that do. Published false-positive rates for SAR coherence-based damage detection vary widely with urban morphology, typically ranging from 20 to 40 percent at the building level in peer-reviewed studies on events including the 2010 Haiti earthquake and the 2023 Turkey-Syria sequence. That is not a reason to discard the method; it is a reason to weight the satellite layer appropriately alongside ground reports, social media signals, and local knowledge.
The correct framing for a field commander is: this block has a one-in-three chance of being a false alarm, but it also has the highest estimated trapped-person count of any block we have not yet searched. That framing preserves the probabilistic nature of the input while still supporting a decision. Presenting a colour-coded map without that context is analytically dishonest and operationally dangerous.
Missed detections are a secondary risk. Buildings that collapse into a compact footprint with little lateral debris spread can show moderate coherence loss rather than the near-zero values associated with clear collapse. Single-storey informal construction is particularly hard to detect. Any priority grid should be accompanied by a note on which building types in the inventory are likely to be underdetected by the method used.
Data inputs that must exist before the earthquake, not after
The quality of the priority grid is bounded by the quality of the pre-event building inventory. A city with a current, georeferenced building footprint dataset, including construction type and storey count, will produce a substantially more reliable output than a city where the footprint layer is derived from OpenStreetMap contributions of uneven completeness. This is the strongest argument for investing in building inventory programmes during peacetime.
Satellize works with governments to pre-stage these layers as part of broader sovereign data infrastructure programmes. The Tonga crop-estimation programme demonstrated the value of maintaining current, locally validated geospatial baselines rather than relying on global products of uncertain vintage. The same principle applies here: the satellite imagery acquired in the first hours after a disaster is only as useful as the pre-event data it is compared against.
Sentinel-1 archive depth extends back to 2014 for most of the world, providing a rich coherence baseline. Commercial SAR operators including ICEYE and Capella maintain growing archives. Pre-staging a coherence baseline stack for high-risk cities, computed during the months before any event, reduces processing time after the event and improves the reliability of the change signal.
Honest limits of the current state of the art
No satellite system can detect a living person under rubble. Thermal infrared from orbit lacks the spatial resolution to distinguish a human heat signature from residual structural heat in a collapsed building. Ground-penetrating acoustic and radar systems operated by search-and-rescue teams remain the only confirmed survivor-detection technology. The satellite layer is a spatial prioritisation tool, not a survivor sensor.
Building-level accuracy degrades in dense urban canyons where layover and shadow effects in SAR imagery obscure individual structures. Cities with highly uniform building stock, common in parts of Central Asia and the Middle East, can produce coherence loss patterns that are difficult to attribute to specific buildings rather than whole blocks. At Pleiades Neo or WorldView-3 resolution, individual building assessment is feasible in many urban morphologies, but processing time scales with area and analyst capacity.
The method works best when the event is discrete and the pre-event baseline is recent. A slow-onset structural failure, a building that has been progressively weakening, will not produce a clean coherence step. Explosions in mixed industrial and residential areas produce debris patterns that challenge automated classifiers trained primarily on earthquake damage.
Typical figures
| SAR spatial resolution (spotlight) | 0.5 m (ICEYE, Capella); 5–20 m (Sentinel-1 IW/SM modes) |
| Optical spatial resolution | 30 cm panchromatic (Pleiades Neo); 31 cm panchromatic (WorldView-3) |
| SAR revisit (commercial) | Multiple times per day over a target city (ICEYE, Capella combined) |
| Optical revisit | Approximately 24 hours (Pleiades Neo 4-satellite constellation); cloud cover can eliminate any single pass |
| First-pass priority grid latency | 12–18 hours from event, assuming pre-staged building and population layers and immediate tasking |
| Population grid resolution | 100 m (WorldPop, Meta HRSL); accuracy degrades in rapidly urbanising or informal areas |
| Damage classification accuracy (binary collapse/standing) | 70–85% at building level in published studies; false-positive rate typically 20–40% |
| SAR archive depth | Sentinel-1 from 2014; ICEYE and Capella commercial archives from 2018–2019 onwards |
| Minimum detectable damage unit | Individual building footprint at VHR optical and commercial SAR spotlight; city block at Sentinel-1 resolution |
| Delivery format | GeoTIFF damage-probability raster; GeoJSON priority-ranked block polygons; PDF operational briefing |
Analytics Satellize can run
| SAR coherence-loss damage proxy map | Interferometric coherence differencing between pre- and post-event SAR pair (standard InSAR processing chain) | GeoTIFF raster of coherence change magnitude at building-to-block scale, delivered within 2 hours of post-event image receipt |
| Optical change-detection damage classification | Pixel- and object-based change detection on VHR pre/post image pairs; damage grade assigned per building footprint using published EMS-98-aligned classifier | GeoJSON building polygons with damage grade (D0–D5) and confidence score |
| Probabilistic casualty-concentration grid | Spatial join of damage proxy with building inventory (storey count, construction type) and WorldPop or HRSL population grid; Monte Carlo uncertainty propagation | Ranked block-level GeoJSON with expected trapped-person range and confidence interval; CSV for import into incident-command software |
| Search-and-rescue priority zone map | Decile ranking of casualty-concentration scores; spatial clustering to form contiguous priority zones respecting road-network accessibility | Georeferenced PDF operational map and GeoPackage for field GIS; updated after each new satellite pass |
| False-positive risk annotation | Building-type-specific false-positive rate lookup from published post-earthquake validation studies; flagging of building classes known to be underdetected | Annotation layer attached to priority zone map; written caveat section in operational briefing |
| Coverage gap and cloud-mask report | Automated QA on image acquisition: cloud fraction (optical), incidence angle and layover mask (SAR), identifying blocks with insufficient data confidence | Supplementary GeoTIFF mask and briefing note identifying areas where satellite evidence is absent or degraded |
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.