Aerosol optical depth attribution to industrial sources

Satellite sensors retrieve columnar aerosol loading daily over industrial clusters, but attributing persistent AOD anomalies to specific facilities requires multi-angle retrieval, wind-field analysis, and honest acknowledgement of what AOD alone cannot tell you about particle type.

Sensors

MODIS MAIAC (Terra and Aqua): Multi-Angle Implementation of Atmospheric Correction algorithm retrieves AOD at 1 km resolution, twice daily globally. MAIAC's time-series surface-reflectance model allows retrievals over bright urban and desert surfaces where standard Dark Target and Deep Blue algorithms struggle. Archive runs from 2000 (Terra) and 2002 (Aqua).
MISR (Terra): Nine cameras at fixed angles from 70° forward to 70° aft, 275 m to 1.1 km per-camera resolution depending on band. Multi-angle geometry decouples surface and atmospheric contributions without relying on assumed surface reflectance, enabling aerosol-type retrieval (absorbing fraction, particle size) that single-view sensors cannot match. 16-day exact repeat, so it is a complement to MODIS rather than a substitute.
Sentinel-5P TROPOMI AER AI: Absorbing Aerosol Index at 3.5 × 5.5 km (reprocessed to 3.5 × 3.5 km from August 2019), daily global coverage. AER AI is sensitive to UV-absorbing aerosols (black carbon, mineral dust, smoke) and is useful for detecting strongly absorbing plumes from industrial fires and steel mills, but does not directly yield AOD or particle size.
PACE OCI: NASA's Plankton, Aerosol, Cloud, ocean Ecosystem mission, launched February 2024. OCI covers 340–890 nm continuously plus shortwave infrared, enabling multi-spectral aerosol-type discrimination (absorbing versus scattering) at roughly 1 km. Still in validation phase; operational aerosol products are expected to improve on MODIS aerosol-type retrievals, particularly over coastal industrial zones.
VIIRS (Suomi-NPP and NOAA-20): Deep Blue AOD at 6 km, daily global coverage. Useful for continuity with MODIS-era records and for near-real-time monitoring given NOAA-20's afternoon overpass complementing Terra and Aqua. Not multi-angle, so surface-reflectance assumptions remain a source of uncertainty over bright surfaces.

What a column of aerosol actually measures, and what it does not

Aerosol optical depth is a dimensionless measure of how much sunlight a vertical column of atmosphere scatters or absorbs at a given wavelength. An AOD of 0.5 at 550 nm is a hazy day; values above 1.0 indicate dense industrial or dust loading. The number integrates everything in the column from the surface to the top of the atmosphere. It does not tell you whether particles are at 50 metres or 5 kilometres, whether they are freshly emitted black carbon from a coke oven or aged sulphate drifting from three countries away, or whether they are large enough to constitute a PM2.5 health hazard.

This ambiguity is not a sensor failure. It is a physical constraint. Attribution work must therefore combine AOD retrievals with ancillary information: wind reanalysis fields (ERA5 is the standard), back-trajectory modelling, co-located NO2 or SO2 columns from TROPOMI where available, and the spatial pattern of anomalies relative to known facility locations. None of these steps eliminates uncertainty; they reduce it to a defensible, quantified range.

Why multi-angle geometry changes the problem

Standard single-view passive retrievals estimate surface reflectance from a spectral model or a dark-pixel assumption, then attribute the residual signal to aerosols. Over bright desert surfaces or light-coloured urban rooftops, the surface term dominates and small errors in the reflectance model produce large errors in retrieved AOD. This is precisely where many heavy industrial clusters sit: steel mills in arid northern China, petrochemical complexes on pale coastal flats, open-cast mines in semi-arid basins.

MISR's nine simultaneous camera angles observe the same ground pixel under different illumination and viewing geometries within roughly seven minutes. Because the surface bidirectional reflectance distribution function (BRDF) and the aerosol phase function respond differently to angle, the retrieval can separate them with far fewer assumptions. Published comparisons show MISR retrievals over bright surfaces achieving uncertainty of roughly ±0.05 AOD units, compared with ±0.1 to ±0.2 for single-view algorithms in the same conditions. The cost is revisit: MISR's 16-day repeat means it characterises the aerosol climatology of a facility rather than its day-to-day emissions.

MODIS MAIAC takes a different route. It builds a time-series model of surface reflectance from cloud-free observations accumulated over weeks, then uses that model to isolate the aerosol signal on any given day. At 1 km and twice-daily, MAIAC is the workhorse for persistent-anomaly detection over industrial clusters.

Attributing a persistent AOD anomaly to a specific facility

The statistical workflow starts with a multi-year MAIAC or MISR climatology gridded to the facility's location. Seasonal and meteorological confounders are removed using regression against ERA5 boundary-layer height, wind speed, and relative humidity, all of which affect how aerosols accumulate and disperse. What remains is the facility-correlated residual: the fraction of AOD variance that tracks with the facility's operating pattern rather than with weather.

Spatial fingerprinting adds a second constraint. A persistent AOD enhancement that is centred on a steel mill's coordinates, elongated in the prevailing downwind direction, and absent on documented shutdown days is a qualitatively different finding from a diffuse regional haze. Published work using MAIAC over Chinese industrial clusters has identified facility-scale AOD enhancements of 0.05 to 0.2 AOD units above regional background. Those numbers are real but must be treated as columnar loading estimates, not ground-level concentration figures.

The absorbing-versus-scattering ambiguity matters most when the output is intended to inform health assessments. Black carbon from coke ovens absorbs strongly and is more harmful per unit mass than sulphate aerosol, yet both can produce similar AOD values. MISR's retrieved single-scattering albedo (SSA) and Ångström exponent partially resolve this. SSA below roughly 0.9 at 550 nm indicates a significant absorbing fraction. PACE OCI's continuous UV-to-NIR spectral coverage is expected to improve SSA retrievals further, though operational validation is ongoing.

Limits that any buyer should understand before commissioning analysis

Cloud cover is the dominant operational constraint. MAIAC requires cloud-free pixels; in persistently overcast industrial regions such as the Yangtze Delta in winter or equatorial mining zones, monthly cloud-free composites may contain fewer than five valid observations. Temporal gaps of two to three weeks are common and must be reported honestly rather than interpolated away.

Spatial resolution sets a hard floor on attribution confidence. MAIAC's 1 km grid means that a facility smaller than roughly 2 to 3 km across is difficult to separate from its neighbours. Petrochemical complexes and large steel mills typically exceed this threshold; smaller foundries or individual process units within a complex do not. MISR's 275 m per-camera resolution helps at the characterisation stage but not at daily monitoring frequency.

Finally, AOD attribution is not the same as emissions quantification. Converting a retrieved AOD anomaly to a mass flux requires assumptions about mixing-layer height, particle density, and size distribution that introduce multiplicative uncertainties often exceeding 50 percent. For emissions quantification, SO2 and NO2 columns from TROPOMI are generally more defensible. AOD attribution is best framed as a persistent-loading and health-exposure indicator, not a tonnage figure.

What a structured analytics programme looks like in practice

A credible monitoring programme for an industrial region typically runs a MAIAC time series from 2000 to present, establishes a pre-facility or pre-expansion baseline, and then tracks monthly anomalies against that baseline with meteorological debiasing. MISR multi-angle retrievals are used to characterise aerosol type at seasonal resolution. TROPOMI AER AI and, where relevant, SO2 or NO2 columns are brought in to cross-validate attribution and distinguish absorbing from scattering sources.

Satellize runs this class of analysis on open constellations, including MAIAC, MISR, and TROPOMI, with commercial tasking added on client licence where higher-resolution optical or SAR context is needed. Outputs are delivered as GIS layers, monthly summary reports, and, where appropriate, automated anomaly alerts keyed to threshold exceedances. The Overhead column occasionally publishes worked examples of the underlying retrieval logic for readers who want to follow the methodology before commissioning a study.

Typical figures

MAIAC AOD spatial resolution	1 km
MISR per-camera resolution	275 m (red band nadir camera); 1.1 km (off-nadir, multi-spectral)
TROPOMI AER AI resolution	3.5 × 3.5 km (from August 2019); 3.5 × 5.5 km earlier
Revisit (MAIAC / MISR / TROPOMI)	MAIAC: ~twice daily (Terra + Aqua); MISR: 16-day exact repeat; TROPOMI: daily global
Spectral bands used for AOD	MAIAC: 0.47–2.1 µm (7 bands); MISR: 0.44–0.87 µm (4 bands, 9 angles); PACE OCI: 340–890 nm continuous
Typical AOD retrieval uncertainty	MAIAC over dark surfaces: ±(0.05 + 0.15 × AOD); MISR over bright surfaces: ~±0.05 AOD units
Minimum detectable facility-scale anomaly	Approximately 0.05 AOD units above regional background (multi-year composite, cloud-free)
Archive depth	MODIS Terra (MAIAC, MISR): from March 2000; Aqua MAIAC: from July 2002; TROPOMI: from May 2018
Cloud-cover constraint	All passive optical retrievals fail under cloud; monthly valid-pixel counts can fall below 5 in persistently overcast regions
Delivery formats	GeoTIFF monthly composites, NetCDF time-series stacks, GeoJSON anomaly polygons, PDF summary reports

Analytics Satellize can run

Multi-year AOD climatology per facility cluster	MAIAC Level-2 retrieval aggregated to monthly medians; ERA5 meteorological debiasing via multivariate regression	GeoTIFF raster stack with per-pixel AOD anomaly and uncertainty band; annual trend summary report
Aerosol-type characterisation (absorbing vs. scattering fraction)	MISR retrieved single-scattering albedo and Ångström exponent at seasonal resolution; cross-validated with TROPOMI AER AI	Seasonal SSA and Ångström exponent maps as GeoTIFF; narrative interpretation of source-type likelihood
Facility-scale AOD attribution score	Spatial fingerprinting: AOD anomaly centroid tracking relative to facility coordinates; wind-aligned plume geometry analysis using ERA5	Monthly attribution confidence score per facility with documented uncertainty; GIS polygon layer of anomaly extent
Shutdown and curtailment detection	Change-point analysis on MAIAC daily AOD time series correlated with known production calendars or reported outages	Event log with dates, magnitude of AOD reduction, and statistical confidence; delivered as structured CSV and PDF
Persistent-loading exceedance alerts	Rolling 30-day MAIAC composite compared against facility-specific baseline; threshold set at client-defined AOD anomaly level	Automated alert feed (JSON or email) triggered on threshold breach, with supporting map attachment
Cross-sensor consistency report	Comparison of MAIAC, MISR, and TROPOMI AER AI retrievals over the same facility cluster to identify retrieval artefacts and build confidence in anomalies seen across multiple sensors	Quarterly PDF report with scatter plots, bias statistics, and recommended primary sensor for the specific surface type

Sources

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.