Wildfire perimeter mapping for real-time insured-asset exposure aggregation
Active wildfire perimeters mapped from VIIRS, GOES-16/17 and Sentinel-2 let insurers intersect a live burn scar with geocoded policy databases before containment, compressing the gap between ignition and reinsurance treaty notification from days to hours.
Sensors
- VIIRS 375 m Active Fire (Suomi-NPP / NOAA-20 / NOAA-21): 375-metre pixel footprint in the I4 (3.74 µm) and I5 (11.45 µm) bands. Detects flaming combustion with a nominal false-alarm rate below 5% over non-desert land. Two polar-orbiting passes per satellite per day give four to six detections per day globally when all three satellites are counted. NASA FIRMS distributes near-real-time shapefiles with latency typically under three hours of overpass.
- GOES-16 / GOES-17 / GOES-18 ABI: Geostationary imager covering the Americas at five-minute full-disk cadence, one-minute in mesoscale sectors. The 2-km resolution in the 3.9 µm band (Channel 7) enables continuous fire radiative power tracking and perimeter evolution monitoring between polar passes. Latency from observation to NOAA product availability is typically under 15 minutes.
- Sentinel-2 MSI (10–20 m, SWIR bands B11/B12): The 20-metre SWIR bands at 1.61 µm and 2.19 µm saturate over active flame but map fresh burn scars precisely via the Normalised Burn Ratio (NBR). Revisit is five days at the equator with both Sentinel-2A and 2B, closer to two to three days at mid-latitudes. Cloud cover is the principal constraint; smoke can partially obscure imagery even when cloud-free flags are set.
- Landsat 8 / 9 OLI-TIRS: 30-metre OLI multispectral and 100-metre TIRS thermal bands. The eight-day combined revisit (both satellites) is slower than Sentinel-2 but the 100-metre thermal band independently identifies residual hotspots and smouldering areas that SWIR reflectance mapping may undercount. USGS delivers Collection 2 Level-2 products within 24 hours of acquisition.
- MODIS Terra / Aqua (500 m, 1 km): Legacy active-fire and burn-area products (MOD14/MYD14, MCD64A1) provide a consistent archive back to 2000, useful for historical exposure modelling and actuarial baseline construction. Spatial resolution is coarser than VIIRS; the 500-metre burn-area product is best used for trend analysis rather than individual-structure exposure.
What the sensors actually see, and when
A wildfire large enough to threaten insured structures typically produces a thermal signal detectable by VIIRS within one orbital pass of ignition. That means the first machine-readable fire-location point can be in an analyst's hands within roughly three hours of a daytime ignition, or up to twelve hours for a fire that starts between passes and grows slowly. GOES-16 fills the gap: its five-minute full-disk cadence over the Americas means the fire radiative power curve is essentially continuous, and the one-minute mesoscale sector can be targeted at a known fire within minutes of an operator request.
The catch is resolution. A VIIRS pixel is 375 metres across. A single-family home is roughly 15 metres wide. The pixel will register the fire, but it will not tell you which parcel is burning. GOES is coarser still at 2 km in the thermal channels. For structure-level exposure, the perimeter polygon derived from these sensors is a probabilistic envelope, not a cadastral boundary. That distinction matters enormously when an insurer is deciding whether to pre-position loss adjusters or trigger a reinsurance notification.
Burn-scar mapping closes the resolution gap, with a delay
Once a Sentinel-2 acquisition falls over the fire area, the Normalised Burn Ratio (NBR = (NIR - SWIR2) / (NIR + SWIR2)) maps the burn scar at 20-metre resolution with enough fidelity to distinguish burned from unburned parcels in most suburban layouts. The differenced NBR (dNBR), comparing a pre-fire baseline image to the post-fire acquisition, produces a continuous severity gradient: low severity where surface litter burned but canopy survived, high severity where soil is exposed and structures are likely destroyed.
The operational limitation is cloud and smoke. During an active fire, the smoke column frequently contaminates SWIR reflectance even when the scene passes cloud-masking algorithms. In practice, a clean Sentinel-2 burn-scar map may lag the fire front by one to four days depending on acquisition geometry and atmospheric conditions. Landsat 9, with its independent eight-day revisit, provides a useful cross-check but does not solve the latency problem. For the hours immediately after ignition, VIIRS and GOES remain the primary inputs; Sentinel-2 confirms and refines.
Building the exposure-aggregation pipeline
The analytic workflow has three stages that run in parallel rather than in sequence. First, GOES-derived fire radiative power and VIIRS point detections are ingested continuously to generate a convex-hull or alpha-shape perimeter polygon, updated on each new detection cycle. Second, that polygon is intersected with a geocoded policy database, producing a list of policy IDs, sum-insured values and construction types within the perimeter and within a configurable buffer zone, typically one to five kilometres depending on fire behaviour and wind speed. Third, the intersection output feeds a treaty-notification workflow: aggregate exposed value by reinsurance layer, flagged against retention thresholds.
The buffer zone is not cosmetic. Fire spotting, where embers carried by wind ignite structures well ahead of the main front, is a documented cause of loss in Australian and Californian wildfires. A polygon that stops at the thermal detection boundary will undercount exposure. Conversely, an overly generous buffer inflates the notification figure and creates friction with reinsurers. Calibrating the buffer to local fire-behaviour models and real-time wind data is where the analytic judgement sits.
Post-event, the dNBR map from Sentinel-2 or Landsat replaces the probabilistic perimeter with a measured burn extent. At that point the exposure list can be refined: parcels inside the high-severity burn zone are flagged as probable total losses, parcels in the low-severity zone are flagged for field inspection. This two-pass approach, coarse-and-fast followed by fine-and-confirmed, is now standard practice in catastrophe response and is directly supported by the NASA FIRMS and Copernicus Emergency Management Service product suites.
Honest limits: what satellite data cannot tell you
Satellite sensors measure radiance and reflectance. They do not measure whether a structure is standing. A high dNBR value over a parcel strongly suggests severe fire impact, but a metal-roofed building can survive a high-severity crown fire while a wood-frame house burns in a low-severity surface fire. Structure survival depends on construction material, defensible space, ember intrusion and dozens of other variables that no current sensor resolves from orbit.
Night-time acquisitions complicate matters further. VIIRS has a day-night band, and GOES operates continuously, but Sentinel-2 is a passive optical sensor and acquires no useful data at night. A fire that makes its most significant run between midnight and dawn may not be captured in a clean SWIR image until the following afternoon overpass, by which time the perimeter has moved substantially. SAR sensors such as Sentinel-1 can penetrate smoke and operate at night, and while SAR-based burn mapping is an active research area, it is not yet operationally mature for rapid perimeter delineation at the precision insurers require.
Reinsurance notification timelines: what is achievable
For a fire igniting in the western United States or Australia, a first-cut exposure aggregate based on VIIRS detections and a geocoded policy database is achievable within three to six hours of ignition during daylight hours. A refined estimate using a Sentinel-2 burn-scar map is typically available within 24 to 96 hours, depending on cloud cover and the next scheduled overpass. A final confirmed perimeter with dNBR severity grading generally requires five to ten days after containment, by which point the smoke has cleared and multiple clean acquisitions are available for cross-validation.
Most proportional reinsurance treaties require notification within 72 hours of a loss event exceeding a defined threshold. The three-to-six-hour first-cut aggregate sits comfortably inside that window for fires that grow rapidly. The practical bottleneck is not satellite latency but geocoding quality: a policy database with imprecise coordinates, missing construction-type fields or stale sum-insured values will produce an exposure aggregate that is wrong regardless of how good the perimeter polygon is. Satellize's analytics stack for this use case, built on the same open-constellation infrastructure used in the Tonga crop-estimation programme, can ingest a client's geocoded policy extract and return a time-stamped GeoJSON exposure report on each VIIRS update cycle.
Historical perimeters and actuarial baseline construction
The MODIS MCD64A1 burned-area product provides a monthly global record from November 2000 onwards at 500-metre resolution. Combined with the VIIRS VNP64A1 product, which extends the record at improved resolution from 2012, an insurer can construct a 20-plus-year fire-frequency and fire-severity map for any portfolio geography. Overlaying that map with current policy locations identifies concentrations of exposure in historically high-frequency fire corridors before a single claim is filed.
This historical layer is also the foundation for parametric product design. A parametric wildfire trigger based on a defined burn-area threshold within a geocoded zone can be verified objectively against the satellite record, removing the basis risk disputes that plague indemnity products in fast-moving catastrophe events. The archive is public, the methodology is auditable, and the trigger can be back-tested against actual loss experience. That combination is increasingly attractive to both cedants and capacity providers in a market where wildfire loss trends are moving faster than traditional rating models.
Typical figures
| VIIRS active-fire pixel resolution | 375 m (I-band), 750 m (M-band) |
| VIIRS / FIRMS detection latency | Typically under 3 hours from overpass; 4–6 detections per day per location with NPP + NOAA-20 + NOAA-21 |
| GOES-16/17/18 ABI thermal cadence | 5-minute full disk; 1-minute mesoscale sector; 2 km resolution in 3.9 µm channel |
| GOES-derived fire product latency | Under 15 minutes from observation to NOAA STAR product |
| Sentinel-2 burn-scar resolution | 20 m (SWIR bands B11, B12); 2–5 day revisit at mid-latitudes with both satellites |
| Landsat 8/9 thermal resolution | 100 m TIRS; 8-day combined revisit; USGS Level-2 delivery within ~24 hours |
| MODIS burned-area archive depth | November 2000 to present (MCD64A1); 500 m monthly composites |
| Minimum reliably detectable fire size (VIIRS) | Flaming fires of ~0.1 ha under favourable conditions; smouldering fires may be missed at sub-pixel scale |
| Delivery formats | GeoJSON perimeter polygons, GeoTIFF dNBR severity rasters, CSV exposure-aggregation tables, timestamped update feeds |
| Cloud / smoke constraint | Sentinel-2 SWIR mapping degraded under thick smoke; GOES and VIIRS thermal less affected but coarser |
Analytics Satellize can run
| Live fire-perimeter polygon | Alpha-shape or convex-hull fitting to VIIRS 375 m active-fire point detections, updated on each FIRMS ingestion cycle | GeoJSON polygon with timestamp, confidence class and fire radiative power attribute; refreshed every 1–3 hours during active event |
| Exposure-aggregation report | Spatial intersection of perimeter polygon (plus configurable buffer) with client geocoded policy database; sum-insured aggregation by reinsurance layer | CSV and PDF report showing policy count, total sum insured, and per-layer exposed value; updated on each perimeter revision |
| dNBR burn-severity map | Differenced Normalised Burn Ratio from pre/post Sentinel-2 or Landsat imagery; severity classes per USGS standard (low / moderate / high) | GeoTIFF raster and classified polygon layer; delivered within 24–96 hours of first clean post-fire acquisition |
| Structure-level loss-probability layer | Overlay of dNBR severity class with parcel footprints and construction-type attributes; probabilistic total-loss flag for high-severity parcels | GeoJSON parcel layer with loss-probability score; input to loss-adjuster dispatch prioritisation |
| Reinsurance treaty notification package | Automated threshold check of aggregated exposed value against client-supplied treaty retention and notification triggers | Structured notification document with perimeter map, exposure table and timestamp; ready for cedant submission within treaty notification window |
| Historical fire-frequency baseline | MODIS MCD64A1 and VIIRS VNP64A1 burned-area time series analysis over portfolio geography; frequency and severity percentile mapping | Portfolio heat map and tabular fire-return-interval statistics for actuarial and parametric product design |
| Parametric trigger verification | Objective measurement of burned area within defined geographic zone against satellite-derived perimeter; back-testable against MODIS/VIIRS archive | Trigger event report with satellite evidence, burn-area measurement and archive comparison; auditable by cedant and reinsurer |
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.