Post-catastrophe building damage grading for insurance claims triage
SAR coherence loss and very-high-resolution optical change detection let insurers grade building damage and estimate aggregate loss within hours of a catastrophe, before ground teams can enter the affected area.
Sensors
- Maxar WorldView-3: 30 cm panchromatic, 1.24 m multispectral at nadir; 16 spectral bands including SWIR. Revisit roughly 1 day at mid-latitudes with off-nadir tasking. Resolves individual wall collapses, roof displacement and debris extent at the parcel level.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral; four-satellite constellation enables same-day revisit in many regions. Stereo and tri-stereo acquisitions support 3-D surface models that can quantify building height loss against pre-event DSMs.
- Sentinel-1 SAR (C-band): IW mode delivers 5 × 20 m resolution with a 6-day exact repeat (12-day single-satellite). Free, open archive back to 2014. C-band coherence is sensitive to structural change but less discriminating than X-band for individual buildings; useful for city-scale triage and rural spread.
- ICEYE X-band SAR: Spotlight mode reaches approximately 0.5 m resolution; tasking can return imagery within hours of a request. X-band coherence loss is a sharper indicator of structural displacement than C-band because the shorter wavelength is more sensitive to small surface changes. Tasking is commercial and priced per acquisition.
What a coherence map is actually measuring
SAR coherence compares the phase relationship between two acquisitions of the same scene. When a building stands undamaged between a pre-event pass and a post-event pass, the backscatter geometry is stable and coherence stays high. When a structure collapses or shifts, the scattering geometry changes and coherence drops toward zero. That physical relationship is what makes SAR change detection useful here: it does not depend on cloud cover, and it does not require daylight.
The complication is that coherence loss is not unique to structural damage. Vegetation, wet soil after heavy rain, and even parked vehicles that have moved between acquisitions all reduce coherence. In a post-earthquake scene, liquefaction-induced ground deformation and landslides produce coherence loss that looks identical to building collapse at city scale. Analysts must cross-reference with building footprint data and optical imagery to separate structural damage from terrain effects. This is not a limitation that better processing eliminates; it is a physical ambiguity that requires human judgement or supervised classification trained on labelled ground truth.
Grading damage: the Copernicus EMS classification as a common language
The Copernicus Emergency Management Service grades building damage on a five-level scale running from no visible damage through slight, moderate and severe damage to destruction. The grading criteria are defined in terms of observable optical indicators: roof integrity, wall collapse, debris distribution, and floor pancaking. Because the scale is documented and used operationally across European civil-protection responses, it provides insurers with a classification framework that adjusters, reinsurers and catastrophe modellers already understand.
Applying the scale from satellite imagery requires very-high-resolution optical data. At 30 cm, WorldView-3 or Pléiades Neo imagery allows an experienced analyst to distinguish a partially displaced roof from a full structural collapse at the individual-building level. At Sentinel-2's 10 m resolution, only the largest damage clusters are visible; individual parcels are not classifiable. The resolution floor for parcel-level grading in dense urban fabric is approximately 0.5 to 1 m. Below that threshold, classification collapses to neighbourhood-level aggregates, which may be sufficient for portfolio-level loss estimation but not for individual-policy triage.
Revisit latency: the number that actually governs usefulness
Post-event imagery is only useful if it arrives before the insurer's decisions must be made. For a major earthquake, insurers typically need a preliminary loss estimate within 24 to 72 hours to activate reinsurance treaties and mobilise adjusters. The practical constraint is not sensor capability but cloud cover, tasking queue depth, and the time between the event and the first clear acquisition.
SAR is cloud-independent, which is why it dominates the first 24 hours of catastrophe response. ICEYE can task and deliver X-band imagery within a few hours of a request in many regions. Sentinel-1's 6-day repeat means the post-event SAR acquisition may not arrive for several days unless a neighbouring orbit provides partial coverage earlier. Optical tasking from WorldView-3 or Pléiades Neo is theoretically same-day, but cloud cover over a post-hurricane or post-wildfire scene can delay usable optical acquisition by days. A production workflow that relies solely on optical imagery for initial triage is fragile. The practical answer is to use SAR for the first damage extent map and optical for the subsequent parcel-level grading pass once skies clear.
From damage map to aggregate loss estimate
A damage grade layer becomes financially useful when it is joined to a building exposure database. Insurers maintain policy geocodes; catastrophe model vendors maintain replacement-cost databases by construction type and occupancy. Overlaying a damage-grade raster on a geocoded policy portfolio produces a preliminary probable loss by grade band. Grade 4 and 5 structures (severe damage and destruction in Copernicus EMS terms) drive the majority of claim value in most portfolios, so the accuracy of the high-severity tail matters more than the accuracy of the slight-damage category.
Stereo DSM differencing adds a useful independent check. A pre-event DSM derived from Pléiades Neo tri-stereo can be compared with a post-event DSM to measure height loss at the building footprint level. A structure that has lost more than 50 % of its pre-event height is almost certainly a total loss regardless of how the optical imagery grades it. This method is less affected by debris obscuring the roof surface, which can cause optical classifiers to underestimate damage severity. The limitation is that DSM differencing requires a pre-event stereo acquisition over the same area, which exists for major cities but is sparse for secondary towns and rural areas.
What the method cannot do honestly
Satellite damage grading cannot see inside a building. A structure that appears intact from above may have suffered severe internal damage, foundation failure, or hazardous material contamination that makes it uninhabitable. Conversely, a building with a collapsed roof section may have a structurally sound frame. Satellite grading is a triage tool that directs field adjusters to the right addresses; it does not replace the adjuster.
In dense urban canyons, building facades are not visible from nadir. Off-nadir optical collection helps but introduces geometric distortion that complicates area calculations. SAR layover and shadow in steep urban topography can mask damage on the far side of tall structures. Wildfire damage presents a specific optical challenge: a standing structure with gutted interior looks largely intact from above until the roof collapses. Thermal infrared can detect residual heat signatures for some hours after a fire, but by the time a satellite pass occurs, the thermal signal may have dissipated. These are known limits, not edge cases.
Putting it into a claims workflow
The practical integration point for an insurer is a damage-grade GIS layer delivered as a georeferenced polygon file, joinable to the policy database by building footprint or geocode. Satellize produces this layer by combining SAR coherence change detection for initial extent mapping with optical classification for parcel-level grading, using the Copernicus EMS schema as the output taxonomy. The workflow is the same one applied in the Satellize Overhead analysis column when public catastrophe events occur.
Reinsurers and ILS investors use aggregate loss estimates from these layers to assess whether a catastrophe bond trigger has been approached. That is a different use case from individual-policy triage, but it runs on the same underlying data product. A single post-event damage map can serve both audiences if the output is structured correctly. The next concrete step for an insurer evaluating this capability is to run a retrospective validation against a historical event where ground-truth claim data already exists, comparing satellite-derived grade distributions against actual paid-loss distributions by postcode.
Typical figures
| Best optical resolution (VHR tasking) | 30 cm panchromatic (WorldView-3, Pléiades Neo) |
| SAR resolution (spotlight mode) | ~0.5 m (ICEYE X-band); 5 × 20 m IW mode (Sentinel-1) |
| Optical revisit (commercial tasking) | Same-day possible; cloud cover is the binding constraint |
| SAR revisit (open archive) | 6-day exact repeat per Sentinel-1 satellite; commercial SAR within hours |
| Minimum detectable damage unit (optical) | Individual building parcel at ≥0.5 m resolution; neighbourhood aggregate at 10 m |
| Damage classification schema | Copernicus EMS 5-grade scale (no damage to destruction) |
| SAR archive depth (Sentinel-1) | 2014 to present (Europe and many global regions) |
| Latency from event to first SAR product | 2–12 hours (commercial X-band); up to 6 days (Sentinel-1 repeat) |
| Delivery format | Georeferenced polygon GeoJSON or GeoPackage, joinable to policy geocodes |
| Height-loss detection (stereo DSM differencing) | Vertical accuracy ~0.5–1 m with Pléiades Neo tri-stereo |
Analytics Satellize can run
| SAR coherence change map | Interferometric coherence differencing between pre- and post-event Sentinel-1 or ICEYE acquisitions; low coherence pixels flagged as probable structural change | City-scale raster layer showing coherence loss, delivered within hours of post-event SAR acquisition |
| Parcel-level damage grade layer | Object-based image analysis on pre/post VHR optical pair; features include roof integrity, debris extent, shadow geometry; classified to Copernicus EMS 5-grade schema | Polygon GIS file with damage grade per building footprint, joinable to insurer policy geocode database |
| Building height-loss layer | DSM differencing between pre-event and post-event Pléiades Neo tri-stereo surface models; height loss expressed as percentage of pre-event building height | Raster and polygon layer flagging structures with >30 % and >50 % height loss |
| Aggregate probable loss estimate by grade band | Spatial join of damage-grade polygons to insurer exposure database; loss calculated by applying grade-conditioned damage ratios to replacement-cost values | Tabular loss summary by postcode, policy segment and grade band; suitable for reinsurance treaty notification |
| Adjuster prioritisation list | Ranked output from damage-grade layer filtered by insured building footprints; grade 4–5 structures sorted by estimated replacement cost | Spreadsheet or API feed of addresses ranked by estimated loss severity, with attached image chips |
| Retrospective validation report | Comparison of satellite-derived grade distributions against historical paid-loss data by postcode for a named past event; precision and recall calculated per grade level | PDF report quantifying model accuracy for a client-nominated historical catastrophe |
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.