Wildfire smoke attribution and downwind population exposure
Satellite fire radiative power retrievals and Lagrangian dispersion modelling can trace PM2.5 episodes at receptor cities back to specific fire events, but separating wildfire smoke from co-located industrial aerosol remains genuinely difficult.
Sensors
- VIIRS I-band active fire detection (NOAA-20 / Suomi NPP): 375 m pixel resolution at nadir, roughly twice daily global revisit per satellite. Detects sub-pixel fires through a mid-infrared brightness-temperature algorithm; the NASA FIRMS near-real-time product typically delivers detections within 3 hours of overpass. Best available spatial detail for locating individual fire perimeters.
- MODIS Fire Radiative Power (Terra / Aqua): 1 km active fire pixels, up to four overpasses per day combined. The fire radiative power (FRP) product quantifies instantaneous energy release in megawatts per pixel, which is the standard input for estimating smoke emission rates. Archive extends to 2000, enabling multi-decade trend analysis.
- Sentinel-3 SLSTR fire radiative power: 1 km resolution, roughly daily global revisit per satellite (two satellites in constellation). ESA's FRP product is processed operationally by EUMETSAT. Provides an independent cross-check on MODIS FRP estimates and extends the record into the Copernicus era with consistent calibration.
- GOES-16 / GOES-17 ABI fire detection: 2 km resolution but 5-minute scan cadence over the Americas. Temporal density is the asset here: ABI catches rapid fire growth and diurnal emission cycles that polar orbiters miss. Useful for constraining time-of-day emission profiles fed into dispersion models.
What fire radiative power actually measures, and why it matters for smoke
Fire radiative power is not a direct measure of smoke output. It quantifies the rate at which a fire releases radiative energy, expressed in megawatts. The connection to smoke comes through empirical relationships: higher FRP generally correlates with higher fuel consumption rate, which drives particulate and gas emissions. The widely used Fire Radiative Energy approach, developed from MODIS observations, converts integrated FRP over time into estimates of dry matter burned, from which emission factors for PM2.5, carbon monoxide and other species can be applied.
The conversion is imprecise. Emission factors vary by fuel type, burn condition and fire behaviour in ways that satellite FRP alone cannot resolve. A smouldering peat fire and a flaming grass fire can produce similar FRP values but very different aerosol compositions and particle size distributions. This is an honest limitation of the satellite-only chain: emission estimates carry uncertainties of a factor of two or more in many published evaluations.
The attribution chain: from fire pixel to receptor city
The standard workflow runs in three stages. First, active fire detections from VIIRS or MODIS identify candidate source fires with location and timing. Second, FRP values from those detections drive an emissions estimate, typically using look-up tables of fuel-type-specific emission factors. Third, a Lagrangian particle dispersion model, most commonly NOAA's HYSPLIT, transports the emitted particles forward in time through reanalysis or forecast meteorological fields, predicting where smoke concentrations will peak at ground level and when.
HYSPLIT backward trajectories are the complementary tool for receptor-side analysis. Given a measured PM2.5 spike at a monitoring station, a back-trajectory ensemble traces air parcel paths over the preceding 24 to 72 hours to identify probable source regions. Overlaying those trajectories with active fire maps then identifies candidate fires. The attribution is probabilistic, not deterministic. Two fires burning simultaneously upwind of a city, in slightly different wind corridors, can produce overlapping trajectory clusters that are genuinely difficult to separate.
Aerosol optical depth from MODIS or VIIRS dark-target retrievals adds a column-integrated smoke signal that can corroborate the modelled plume path. But AOD is a vertically integrated quantity, and the satellite cannot tell you whether the smoke is in the boundary layer where people breathe or aloft in the free troposphere. Ground-level PM2.5 impact depends heavily on that vertical structure, which requires either lidar measurements or model assumptions about mixing layer height.
Separating wildfire smoke from what was already in the air
Attribution becomes genuinely hard when fires burn near industrial corridors, agricultural regions or urban areas with their own aerosol sources. PM2.5 at a receptor site is a mixture, and satellite aerosol retrievals measure the column total. Chemical transport models can apportion contributions by source type if emission inventories for non-fire sources are reliable, but in many lower-income regions those inventories are sparse or outdated.
Smoke has some spectroscopic fingerprints. The absorption Ångström exponent, derivable from multi-wavelength AOD retrievals, tends to be elevated for biomass-burning aerosol compared with urban pollution. MODIS and VIIRS retrieve spectral AOD at multiple wavelengths, and the MISR instrument on Terra provides multi-angle information that improves aerosol type discrimination. These are useful indicators, not clean separators. A heavy industrial source burning organic material can mimic some smoke optical properties. Candour is warranted: satellite-only attribution of mixed aerosol episodes has known skill limits, and ground-based chemical speciation data, where it exists, substantially improves confidence.
Population exposure: translating plume position into health numbers
Once a modelled or observed smoke plume is positioned in space and time, exposure estimation requires population data at compatible resolution. Gridded population products such as GPWv4 or WorldPop provide sub-kilometre estimates for most countries. Multiplying modelled PM2.5 concentration by population count in each grid cell, integrated over the duration of a smoke event, yields a person-hours-of-exposure metric that epidemiologists can connect to dose-response functions.
The dose-response relationships for wildfire smoke specifically are an active research area. Wildfire PM2.5 is not identical in composition to urban PM2.5, and some published studies suggest its health effects per unit mass may differ. Applying a single concentration-response function across all aerosol types introduces uncertainty that analysts should flag explicitly when reporting health burden estimates. The satellite data constrains the exposure geography; it does not resolve the toxicological question.
Near-real-time applications are where GOES-16 ABI earns its place. Its five-minute cadence allows smoke plume position to be updated continuously during an active fire event, supporting public health advisories with same-day lead times. Sentinel-3 and VIIRS provide higher spatial detail but with longer gaps between overpasses. Operational systems typically blend both.
What the satellite chain cannot do on its own
Vertical resolution is the persistent gap. Satellites measure column-integrated aerosol well. They do not measure the vertical profile with useful resolution unless a lidar instrument is in the constellation. CALIPSO provided curtain profiles until 2023; its successor capability is not yet fully operational. Without vertical information, the fraction of smoke column that actually reaches the surface breathing zone must be inferred from meteorological model output, introducing model-dependent uncertainty that can be large during temperature inversions or when smoke layers are lofted above 3 km.
Cloud cover interrupts retrievals. A fire burning under persistent cloud, common in tropical regions during the wet season, may be detected by its thermal signature through thin cloud but will produce unreliable AOD retrievals. Smoke transport modelling can continue using the last reliable FRP estimate, but the emission input degrades. Analysts should report data gaps explicitly rather than interpolating silently.
Satellize runs this attribution chain on open constellations, combining VIIRS, MODIS and Sentinel-3 fire products with HYSPLIT ensemble modelling, for clients who need defensible, documented exposure estimates rather than model output that arrives without provenance.
Typical figures
| Active fire spatial resolution | 375 m (VIIRS I-band); 1 km (MODIS, Sentinel-3 SLSTR, GOES-16 ABI) |
| Active fire revisit (polar orbit) | Approximately 12-hour repeat per satellite; NOAA-20 + Suomi NPP combined give roughly 6-hour average global revisit |
| GOES-16 ABI fire scan cadence | 5 minutes (CONUS and full-disk modes); covers Americas only |
| FIRMS near-real-time latency | Typically under 3 hours from satellite overpass for VIIRS and MODIS products |
| Fire radiative power detection floor | Approximately 5–15 MW for MODIS at 1 km; VIIRS I-band detects smaller sub-pixel fires due to finer pixel size, though exact minimum varies with background temperature |
| MODIS FRP archive depth | 2000 to present (Terra); 2002 to present (Aqua) |
| HYSPLIT trajectory ensemble size (typical) | 27-member ensemble standard in NOAA operational runs; user-configurable |
| AOD retrieval cloud contamination | Retrievals masked under cloud; smoke under persistent cloud cover produces data gaps that cannot be filled by optical sensors |
| Vertical profile capability | Not available from VIIRS/MODIS/Sentinel-3; requires separate lidar (CALIPSO archive available; no direct current successor at equivalent coverage) |
| Population exposure output resolution | Typically 0.1° (~10 km) for modelled PM2.5 fields; can be downscaled with land-use regression where ground monitor data permits |
Analytics Satellize can run
| Fire event attribution report | VIIRS/MODIS active fire detection combined with HYSPLIT backward-trajectory ensemble from receptor monitor location | PDF report identifying candidate source fires, trajectory confidence intervals and data-gap flags for a named PM2.5 episode |
| Smoke plume forward-transport forecast | FRP-driven emission estimate fed into HYSPLIT forward dispersion run on GFS meteorological fields | 72-hour animated plume position GeoTIFF series and peak-concentration gridded layer, updated at each VIIRS overpass |
| Population exposure estimate | Modelled PM2.5 concentration field multiplied against WorldPop or GPWv4 gridded population; integrated over event duration | Tabular person-hours-of-exposure by administrative unit, with uncertainty band reflecting FRP-to-emission conversion range |
| Multi-source aerosol apportionment flag layer | Absorption Ångström exponent derived from MODIS/VIIRS multi-wavelength AOD to flag smoke-dominated versus mixed aerosol pixels | GIS polygon layer classifying receptor-area aerosol episodes by probable dominant source type, with confidence rating |
| Seasonal fire emission climatology | MODIS FRP archive aggregated by fire season, fuel type region and interannual variability using the standard Fire Radiative Energy approach | Annual summary dataset and trend chart covering up to 24 years of MODIS record, formatted for integration with health burden models |
| Near-real-time public health advisory input | GOES-16 ABI five-minute fire detections fused with HYSPLIT short-range dispersion for same-day plume position | Hourly updated smoke-impact polygon feed in GeoJSON, suitable for ingestion by public health alert systems in the Americas |
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.