Wildfire perimeter dynamics and evacuation route accessibility
Near-real-time perimeter extraction from VIIRS, Sentinel-3 SLSTR and GOES-16/17 ABI, overlaid on road networks, tells emergency managers which evacuation corridors are already compromised and which will be within the hour.
Sensors
- VIIRS (Suomi-NPP / NOAA-20): 375 m active fire product (VNP14IMG / VJ114IMG) using the I4 band at 3.74 µm and I5 at 11.45 µm. Detects sub-pixel fires; nominal overpass every 12 hours per satellite, roughly 6-hour combined revisit at mid-latitudes. NASA FIRMS distributes detections within ~3 hours of overpass.
- GOES-16 / GOES-17 ABI: Geostationary imager covering the Americas. Full-disk scan every 10 minutes, CONUS sector every 5 minutes. Fire detection via the 3.9 µm band (Band 7) at 2 km nadir resolution. Coarser than VIIRS but the sub-hourly cadence captures rapid perimeter advance that polar orbiters miss between passes.
- Sentinel-3 SLSTR: Dual-view thermal imager at 1 km resolution in the S7 (3.74 µm) and S8 (10.85 µm) bands. Two Sentinel-3 satellites together provide roughly twice-daily global coverage. ESA's Fire Radiative Power product is available within a few hours of acquisition via the Copernicus Data Space.
- Sentinel-2 MSI: 10 m optical and 20 m shortwave-infrared (SWIR, bands B11 and B12 at 1.61 µm and 2.19 µm). Not a thermal detector, but SWIR saturates over active fire pixels and clearly delineates the burned-area boundary. Five-day revisit at the equator, better at higher latitudes; cloud and smoke are genuine blockers.
Why perimeter velocity matters more than perimeter position
Emergency managers are not primarily asking where the fire is. They are asking where it will be in 90 minutes, and whether the road their residents are driving out on will still be passable. A static perimeter snapshot answers neither question. What is needed is a rate-of-spread estimate derived from successive perimeter positions, combined with observed wind speed and direction, so that a projected spread zone can be drawn around each road segment.
This is operationally harder than it sounds. A single VIIRS pass gives a position. Two passes, separated by roughly six hours at mid-latitudes when both Suomi-NPP and NOAA-20 are used together, give a displacement vector. That vector, divided by elapsed time and corrected for terrain slope using a digital elevation model, produces a spread rate in metres per minute. Errors compound quickly: cloud contamination in one pass, or a lull in wind between passes, can make a fast fire look slow. The honest operational floor is that spread-rate estimates from polar orbiters carry meaningful uncertainty and should be treated as order-of-magnitude guidance rather than precise forecasts.
What the geostationary layer adds, and what it costs
GOES-16 ABI changes the calculus for the Americas. A five-minute CONUS scan cadence means that perimeter displacement can be computed over intervals short enough to catch a fire that crowns and runs ahead of the main front. The NOAA GOES-R Fire Detection and Characterisation Algorithm (FDCA) flags active fire pixels in the 3.9 µm band and estimates Fire Radiative Power. Latency from scan to public product is typically under 15 minutes via NOAA's CLASS and STAR portals.
The cost is spatial resolution. At 2 km nadir, a GOES fire pixel is roughly 28 times the area of a VIIRS 375 m pixel. A road corridor 10 metres wide is entirely sub-pixel. This means GOES data can tell you that a fire is approaching a road corridor, but it cannot confirm whether the fire has crossed it. Sentinel-2 SWIR, when a cloud-free pass is available, provides the confirmation at 20 m resolution. The operational workflow is therefore layered: GOES for rate, VIIRS for position, Sentinel-2 for boundary precision when smoke permits.
Road-network overlay: the geometry of a closing corridor
Once a spread-rate vector is estimated, a projected spread zone is constructed as a buffered polygon around the leading fire edge, scaled to a chosen time horizon, typically 60 to 120 minutes for evacuation planning. That polygon is then intersected with the road network, drawn from OpenStreetMap or a national authoritative dataset. Any road segment whose centreline falls inside the polygon, or within a configurable safety margin, is flagged as potentially compromised.
The safety margin matters. A 100-metre buffer around a road segment is meaningless if spotting embers are landing 500 metres ahead of the main front, which is documented behaviour in chaparral and eucalyptus fuels under Foehn-type winds. The buffer should be informed by fuel type, derived from land-cover classification, and by observed spotting distance if aerial observation data are available. Without that, the geometry is precise but the risk estimate is not.
Terrain is the other complication. Fire spread accelerates upslope at roughly double the rate for every 10-degree increase in slope angle, a relationship formalised in Rothermel's fire spread model and its successors. A road that sits in a valley below an advancing upslope fire may be cut off far faster than a flat-terrain spread rate would suggest. Slope correction using a 30 m SRTM or Copernicus DEM is a minimum requirement for any credible route-risk product.
Smoke, cloud and the gaps you cannot ignore
Thermal infrared penetrates smoke well enough for active fire detection. VIIRS I4 and I5 bands, and GOES Band 7, are not meaningfully attenuated by smoke. But Sentinel-2 optical and SWIR bands are. A dense smoke column over the fire perimeter will corrupt or blank the Sentinel-2 boundary delineation precisely when the fire is most active. Analysts must flag this explicitly in any delivered product rather than presenting a smoke-obscured perimeter as if it were confirmed.
Cloud is a harder problem than smoke. A convective cloud deck forming over a fire, which happens with pyrocumulus development, blocks both thermal and optical sensors simultaneously. There is no passive optical or thermal workaround. SAR would penetrate cloud but is not a thermal detector and cannot locate the active fire front directly, only the burned area after the fact. This is a genuine capability gap in the current open-data constellation.
Putting numbers into an operational picture
A worked example from publicly documented sensor performance: Suomi-NPP passes at roughly 01:30 and 13:30 local time; NOAA-20 offsets this by approximately three hours. Combined, the two satellites provide VIIRS coverage of a given point roughly every six hours. GOES-16 fills the interval with five-minute updates. If the fire perimeter moves 500 metres between two VIIRS passes six hours apart, the computed spread rate is roughly 1.4 metres per minute, which is slow. If GOES shows the perimeter advancing several pixels between consecutive five-minute scans, the real rate may be an order of magnitude higher. The discrepancy is the signal: something changed between polar passes, and GOES caught it.
Satellize runs this layered fusion workflow on open constellations, adding commercial Sentinel-2 tasking on client licence for post-smoke boundary confirmation. The analytics pipeline is the same class of method used in the Tonga crop-estimation programme: systematic multi-source ingestion, change detection and structured output, applied here to a very different but equally time-critical problem. Outputs are delivered as GIS-ready polygon feeds and structured alert messages keyed to road-segment identifiers, so that they drop directly into existing emergency management systems without manual reformatting.
What this workflow cannot do
It cannot predict ignition of new spot fires ahead of the main front. Spotting behaviour depends on ember transport, which requires atmospheric dispersion modelling beyond what satellite observation alone provides. It cannot give sub-100-metre perimeter accuracy in real time from thermal sensors. And it cannot substitute for ground truth: a road that appears geometrically clear of the projected spread zone may already be impassable due to smoke density, fallen trees or prior ember damage to the road surface itself.
The appropriate use of this product is as a decision-support layer for trained incident commanders, not as an autonomous evacuation trigger. The satellite data narrows the range of plausible situations rapidly. The final call belongs to people with eyes on the ground.
Typical figures
| Best active fire spatial resolution | 375 m (VIIRS I-band active fire product) |
| Geostationary fire detection resolution | 2 km nadir (GOES-16/17 ABI Band 7) |
| Perimeter boundary resolution (cloud-free) | 20 m SWIR (Sentinel-2 B11/B12) |
| Polar-orbiter revisit (VIIRS combined) | ~6 hours at mid-latitudes (Suomi-NPP + NOAA-20) |
| Geostationary revisit (GOES CONUS sector) | 5 minutes |
| Product latency (VIIRS via NASA FIRMS) | Typically 2–3 hours post-overpass |
| Spectral bands used for fire detection | MWIR ~3.7–3.9 µm; TIR ~10.8–11.5 µm |
| Minimum detectable fire (VIIRS) | Sub-pixel flaming fires; published detection floor ~0.1 MW Fire Radiative Power under favourable conditions |
| Terrain correction DEM | 30 m SRTM or Copernicus DEM-30 |
| VIIRS archive depth | Suomi-NPP from 2012; NOAA-20 from 2018 (NASA FIRMS) |
Analytics Satellize can run
| Near-real-time fire perimeter polygon | Active fire pixel clustering and convex-hull or alpha-shape boundary extraction from VIIRS 375 m detections, updated each overpass | GeoJSON polygon layer, timestamped, with confidence class per pixel cluster |
| Perimeter spread-rate vector | Displacement of perimeter centroid or leading edge between successive VIIRS passes, slope-corrected using Copernicus DEM-30 and Rothermel-class spread adjustment | Vector attribute table (direction degrees, rate m/min, uncertainty flag) attached to perimeter GeoJSON |
| Projected spread zone (60/120 min horizons) | Buffered forward projection of spread vector, wind-adjusted using nearest NWP surface wind field (e.g. NOAA HRRR for North America), fuel-type modulated from ESA CCI land cover | Polygon shapefile or GeoJSON with two time-horizon bands and confidence envelope |
| Road-corridor risk flag | Geometric intersection of projected spread zone (plus configurable safety buffer) with OpenStreetMap or national road network; segment-level status: clear / at risk / likely compromised | Road-segment GIS layer with risk attribute and estimated time-to-impact per segment |
| GOES sub-hourly perimeter advance alert | Consecutive GOES ABI Band 7 fire pixel comparison at 5-minute cadence; threshold-based alert when leading-edge displacement exceeds configurable distance in a single interval | Structured alert message (JSON or CAP format) keyed to road-segment IDs, latency target under 20 minutes from GOES scan |
| Post-smoke Sentinel-2 boundary confirmation | SWIR band ratio (B12/B11) thresholding on cloud- and smoke-free Sentinel-2 acquisitions to delineate burned versus unburned boundary at 20 m | Corrected perimeter polygon replacing thermal-derived estimate, flagged with acquisition timestamp and smoke-obstruction assessment |
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.