Wildfire smoke aerosol optical depth and population health exposure
Satellite aerosol optical depth from MODIS, VIIRS, and Sentinel-5P TROPOMI can estimate how many people are breathing hazardous air during a wildfire event, though cloud cover and coarse resolution introduce real uncertainty that any honest analysis must quantify.
Sensors
- MODIS Terra and Aqua: Retrieves aerosol optical depth at 550 nm via the Dark Target and Deep Blue algorithms at 3 km and 10 km spatial resolution. Two daily overpasses per sensor give up to four looks per day globally, though cloud contamination can eliminate all of them. MAIAC (Multi-Angle Implementation of Atmospheric Correction) retrieval refines AOD to 1 km over land.
- VIIRS Suomi-NPP and NOAA-20: Provides aerosol optical depth at 750 m resolution (aggregated to 6 km for standard products) and supports continuity with the MODIS record. The VIIRS Deep Blue algorithm extends retrievals over bright desert surfaces where Dark Target fails. Revisit is once per day per satellite.
- Sentinel-5P TROPOMI: Measures UV Aerosol Index and absorbing aerosol optical depth at 3.5 km x 5.5 km (improved to 3.5 km x 3.5 km post-August 2019) with daily global coverage. TROPOMI's strength is its wide swath (2600 km) and its co-registered SO2 and CO columns, which help characterise fire-specific aerosol chemistry. Its resolution is too coarse to resolve intra-urban exposure gradients in cities smaller than roughly 20 km across.
- GOES-16 and GOES-17 ABI: Geostationary sensors covering the Americas with 5-15 minute repeat imagery. ABI aerosol optical depth retrievals are less accurate than polar-orbiter products but provide near-real-time plume tracking and can capture rapid smoke advection events that a twice-daily polar pass would miss entirely.
What aerosol optical depth actually measures, and what it does not
Aerosol optical depth is a dimensionless measure of how much sunlight a column of atmosphere scatters or absorbs. A value of 1.0 means roughly 63 percent of direct sunlight is extinguished before reaching the surface. During major wildfire events, AOD values above 2.0 are common downwind; during the 2019-2020 Australian fires, MODIS recorded AOD exceeding 3.0 over parts of the Tasman Sea.
The clinical problem is that health effects depend on surface-level PM2.5 concentration, not on the column integral. Converting AOD to PM2.5 requires knowing the mixing layer height (how deep the atmosphere is actually being breathed) and the aerosol extinction efficiency, which varies with particle size, composition, and relative humidity. Published conversion factors from studies using paired ground monitors and MODIS retrievals show PM2.5-to-AOD ratios that can differ by a factor of three or more depending on meteorological conditions. Any satellite-derived PM2.5 estimate carries that uncertainty, and responsible reporting says so explicitly.
The sensor stack: what each instrument adds and where it fails
MODIS Terra and Aqua together offer up to four polar-orbit passes per day, which sounds generous until a frontal cloud system parks over a burning region for 72 hours. Cloud screening removes contaminated pixels, but it does so non-randomly: cloud and smoke frequently co-occur, so the days when exposure is highest are often the days when retrievals are missing. MAIAC retrieval at 1 km partially addresses this by using multi-temporal information to separate aerosol from surface reflectance, but it cannot see through optically thick cloud.
TROPOMI's UV Aerosol Index is particularly useful for absorbing aerosols such as black carbon from biomass burning, because it is less sensitive to cloud contamination than passive visible retrievals. Its 3.5 km pixel, however, means a city of 300,000 people might fall within two or three pixels. Exposure estimates at the neighbourhood scale are not credible from TROPOMI alone.
GOES fills the temporal gap. Its 10-15 minute cadence over the Americas allows analysts to track plume advection in near-real-time and to identify which population centres are under the plume at any given hour. The trade-off is accuracy: GOES ABI AOD retrievals are noisier than MODIS over land, and the geostationary viewing angle degrades retrieval quality at high latitudes.
From plume extent to exposed population: the method and its honest limits
The standard workflow combines a smoke plume mask (derived from AOD thresholds or the NOAA Hazard Mapping System fire and smoke product) with gridded population data such as WorldPop or LandScan to count the number of people under the plume. A secondary step applies a published AOD-to-PM2.5 conversion, typically a linear or log-linear regression derived from co-located ground monitor networks such as AERONET, to estimate the fraction of the exposed population likely experiencing PM2.5 above WHO guideline thresholds (15 µg/m³ for 24-hour mean) or above national regulatory thresholds.
Several compounding uncertainties deserve explicit acknowledgement. Population grids themselves carry errors, particularly in rapidly growing cities or informal settlements. The AOD-to-PM2.5 relationship is calibrated on days when ground monitors and satellites both have valid data, which may not represent the most extreme smoke events. And the plume boundary is not a hard line: AOD gradients at plume edges mean that people just outside the masked area may still be experiencing elevated PM2.5. Presenting results as exposure probability distributions rather than single-point estimates is more honest and more useful to public health decision-makers.
Temporal resolution as a public health variable
Wildfire smoke exposure is not a static event. A plume that covers a city for six hours is a different health burden from one that persists for six days, and the satellite record can distinguish between them. By combining GOES near-real-time imagery with the daily MODIS and VIIRS AOD composites, it is possible to reconstruct hourly plume extent and then integrate exposure over time, producing cumulative person-hours above a given AOD threshold.
This matters for hospital admission forecasting and for retrospective epidemiological studies. The 2020 western United States wildfires, for instance, produced multi-week smoke events across several states, and retrospective satellite-based exposure reconstructions have been used in published health-outcome studies to establish dose-response relationships where ground monitor networks were sparse. The satellite record extends back to 2000 for MODIS, giving two decades of comparable data for trend analysis.
Operational delivery: what a government client actually receives
A working smoke-exposure product for a national civil protection agency typically takes the form of a daily GIS layer showing AOD at the best available resolution, a population-weighted exposure count by administrative district, and a confidence flag indicating whether cloud contamination has made the retrieval unreliable for that day. Latency from satellite overpass to delivered product is typically four to six hours for MODIS and VIIRS standard products, or under two hours if near-real-time processing streams are used.
Satellize applies this methodology on open constellations, adding TROPOMI UV Aerosol Index as a cross-check and GOES imagery for inter-pass tracking where coverage permits. The approach is operationally similar to the crop-estimation work run for the Kingdom of Tonga, in that it combines multiple open data streams into a single decision-ready output rather than asking the client to reconcile conflicting sensor products themselves. Clients with dense ground monitor networks can supply those data to constrain the AOD-to-PM2.5 conversion and substantially reduce the uncertainty range.
Typical figures
| Best AOD spatial resolution | 1 km (MODIS MAIAC); 3.5 km (TROPOMI); 750 m native pixel for VIIRS, aggregated to 6 km in standard AOD product |
| Temporal revisit (polar orbit) | Up to 4 passes/day (MODIS Terra + Aqua combined); 1-2 passes/day per VIIRS satellite |
| Temporal revisit (geostationary) | 10-15 minutes (GOES-16/17 ABI over Americas) |
| Product latency (near-real-time) | ~1-3 hours from overpass for MODIS NRT and VIIRS NRT streams via NASA FIRMS/LANCE |
| AOD spectral band | 550 nm reference wavelength (MODIS Dark Target/Deep Blue, VIIRS); UV Aerosol Index at 340/380 nm (TROPOMI) |
| AOD detection threshold | Approximately 0.05 AOD units under clear-sky conditions; retrieval unreliable below this or under cloud optical depth >1 |
| Global coverage | 1-2 days for full global swath (MODIS, VIIRS, TROPOMI); GOES continuous over Americas only |
| Archive depth | MODIS: from 2000 (Terra) and 2002 (Aqua); VIIRS: from 2012; TROPOMI: from 2018 |
| PM2.5 conversion uncertainty | Factor of 2-3 depending on mixing layer height, aerosol type, and relative humidity; ground monitor calibration reduces this |
| Delivery formats | GeoTIFF, NetCDF, GeoJSON district summaries, PDF situation reports |
Analytics Satellize can run
| Daily AOD composite layer | MODIS MAIAC and VIIRS Deep Blue retrieval, cloud-screened and gap-filled where multi-day compositing is appropriate | GeoTIFF at 1 km resolution with confidence mask; updated within 4 hours of overpass |
| Smoke plume extent polygon | AOD threshold classification (typically AOD > 0.3 or > 1.0 for hazardous-level events) combined with NOAA Hazard Mapping System smoke classification | GeoJSON polygon updated daily; GOES-derived version updated every 30 minutes for Americas events |
| Population exposure count by administrative unit | Spatial intersection of plume extent with WorldPop or LandScan gridded population; stratified by AOD severity band | CSV and GIS layer showing person-count per district at three AOD thresholds; daily update |
| Estimated surface PM2.5 field | Published linear regression of AOD against ground-monitor PM2.5 (calibrated on AERONET network data), with mixing layer height from ERA5 reanalysis as a covariate | Gridded PM2.5 estimate in µg/m³ with explicit uncertainty range; flagged where cloud contamination exceeds 40% of pixels |
| Cumulative exposure time series | Integration of daily population-weighted AOD over event duration; comparable to published epidemiological exposure metrics | Time-series chart and CSV of person-days above WHO 24-hour PM2.5 guideline for the event period |
| Multi-sensor cross-validation report | Comparison of MODIS, VIIRS, and TROPOMI AOD retrievals over the same scene to identify retrieval artefacts and bound uncertainty | PDF situation report with sensor-agreement statistics; issued at 48-hour intervals during active events |
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.