Trace gas attribution to biomass burning versus industrial combustion
CO, NO2, and aerosols pour from both burning forests and industrial stacks, but their ratios, spatial patterns, and co-emitted tracers differ enough to separate the sources when you read the data carefully.
Sensors
- Sentinel-5P TROPOMI: Retrieves tropospheric CO columns (2.3 µm band) and NO2 columns (405–465 nm) at 5.5 × 3.5 km nadir resolution (7 × 3.5 km before August 2019), daily global coverage. The CO/NO2 column ratio is the primary combustion-type discriminant.
- VIIRS (Suomi-NPP and NOAA-20): 375 m active fire detections using the 3.74 µm mid-infrared band, with a minimum detectable fire radiative power around 5–10 MW under clear-sky conditions. Provides the spatial anchor that tells you whether a TROPOMI CO anomaly sits above a real fire or an industrial cluster.
- MOPITT (Terra): Multi-spectral thermal-infrared and near-infrared CO retrievals with sensitivity to both surface and mid-troposphere layers, at roughly 22 km nadir resolution. Useful for tracking aged plumes and validating TROPOMI CO in biomass-burning outflow regions. Archive runs from 2000.
- Sentinel-2 MSI: 10 m optical imagery in visible and near-infrared bands. Identifies burned-area scars, active fire perimeters, and industrial facility footprints that give spatial context when TROPOMI plumes are ambiguous. No cloud penetration; useless during heavy smoke.
Why the same molecule can mean two completely different things
Carbon monoxide is carbon monoxide. A TROPOMI pixel showing an elevated CO column of, say, 3 × 10¹⁸ molecules per cm² does not come labelled with its origin. Biomass burning produces it. Steel mills produce it. Vehicle exhausts produce it. The column measurement alone is silent on the question of attribution.
What breaks the ambiguity is the ratio of co-emitted species and the spatial relationship between the column anomaly and independently observed emission sources. Biomass burning is characterised by high CO relative to NO2, because incomplete combustion dominates and combustion temperatures are lower than in industrial furnaces. Industrial combustion, particularly in high-temperature processes such as power generation and smelting, runs hotter and more completely, producing proportionally more NO2 and less CO per unit of fuel energy. Published emission ratio studies, including work using TROPOMI and its predecessor OMI, place the pyrogenic CO/NO2 enhancement ratio typically above 10 by molar column, while industrial point sources tend toward ratios below 5, often well below. Those are population averages with real scatter, not clean thresholds.
VIIRS fire detections as the spatial anchor
TROPOMI's 5.5 km pixel is large enough to contain both a burning agricultural field and a small industrial facility. The attribution question is therefore partly a co-location problem. VIIRS active fire detections at 375 m resolution, available through NASA FIRMS with latency typically under three hours, provide the spatial anchor. When a TROPOMI CO anomaly sits directly downwind of a cluster of VIIRS fire pixels and no significant industrial infrastructure appears in Sentinel-2 imagery of the upwind area, the pyrogenic attribution is strong.
The reverse is also useful. A persistent NO2 hotspot with no coincident VIIRS detections over multiple days, in a location that Sentinel-2 shows to be an industrial zone, is almost certainly industrial. The method becomes genuinely difficult when fires occur within or immediately adjacent to industrial areas, a situation common in parts of sub-Saharan Africa, Siberia, and South-East Asia where agricultural burning happens near petrochemical or mining infrastructure. In those cases, even careful ratio analysis leaves residual ambiguity that no single sensor resolves.
Emission ratios as combustion-type fingerprints
The enhancement ratio approach compares the slope of co-emitted species in a plume rather than their absolute values. For a fresh pyrogenic plume, the ΔCO/ΔNO2 ratio (the change in CO column versus the change in NO2 column across the plume boundary) reflects the emission factor ratio of the fire itself. Published emission factors for savanna fires, tropical deforestation fires, and extratropical forest fires differ from each other, which means biomass burning is not a single fingerprint. Savanna fires burn more completely and produce lower CO/NO2 ratios than smouldering peat or tropical deforestation fires. This internal variability within the pyrogenic category is an honest complication.
Industrial sources have their own spread. A gas-fired power plant and a blast furnace both qualify as industrial combustion but emit very different CO/NO2 ratios. Where the industrial source type is known from facility registries or prior Sentinel-2 characterisation, the expected ratio range can be constrained and compared against the observed plume ratio. Where the facility type is unknown, the ratio alone cannot distinguish a clean-burning gas turbine from a coke oven.
Aerosol optical depth from TROPOMI's UV aerosol index adds a third dimension. Biomass burning produces absorbing organic aerosols with a characteristic spectral dependence. Industrial combustion, depending on process type, may produce less absorbing or differently sized particles. The combination of CO column, NO2 column, and aerosol index narrows the attribution space even when no single variable is conclusive.
Cloud cover, plume age, and the limits of column retrievals
TROPOMI CO retrievals degrade under optically thick cloud. The instrument retrieves a column-averaged mixing ratio weighted by the averaging kernel, which loses sensitivity to the boundary layer under cloud cover. A fire burning beneath a convective cloud system, common in tropical regions during the burning season, may produce a CO signal that TROPOMI underestimates or misses entirely. VIIRS fire detection has similar cloud limitations at the 375 m band, though the thermal signal is more cloud-penetrating than optical reflectance.
Plume age matters too. As a plume travels downwind over hours to days, CO is relatively conserved (atmospheric lifetime of weeks to months), while NO2 is chemically processed into nitrate aerosols and other products within hours. A TROPOMI overpass catching a plume one day downwind will see a higher CO/NO2 ratio than a fresh measurement of the same source, regardless of combustion type. This means high CO/NO2 ratios in aged plumes are not reliable pyrogenic indicators without trajectory analysis. Back-trajectory modelling using tools such as HYSPLIT, combined with the TROPOMI time series, is necessary to interpret aged plumes correctly.
Building a mixed-source attribution workflow
A practical attribution workflow fuses four data streams: TROPOMI CO and NO2 columns at daily cadence, VIIRS active fire detections, Sentinel-2 land cover and facility imagery, and wind field data for plume trajectory reconstruction. The first step is spatial masking: pixels with coincident VIIRS fire detections above a minimum fire radiative power threshold are flagged as potentially pyrogenic. Pixels in industrial zones with no fire detections are flagged as potentially industrial. The overlap zone, where both conditions are true, is flagged as ambiguous.
Within each zone, the ΔCO/ΔNO2 enhancement ratio is computed from the TROPOMI columns, and the result is compared against published emission factor ranges for the plausible source types. Where the ratio is consistent with only one source category and the spatial evidence supports it, attribution confidence is rated high. Where the ratio falls in the overlap range and both source types are spatially plausible, confidence is rated low and the pixel is reported as mixed.
Satellize runs this workflow on open constellation data for clients who need defensible, auditable attribution rather than a simple fire-count product. Getting that distinction wrong in either direction carries real legal and reputational cost.
Typical figures
| TROPOMI CO spatial resolution | 5.5 × 3.5 km (nadir, post-August 2019) |
| TROPOMI NO2 spatial resolution | 5.5 × 3.5 km (nadir, post-August 2019) |
| TROPOMI revisit | Daily global coverage, single overpass near local solar noon |
| VIIRS active fire resolution | 375 m (I-band); minimum detectable FRP approximately 5–10 MW under clear sky |
| VIIRS fire detection latency | Typically under 3 hours via NASA FIRMS NRT stream |
| Sentinel-2 MSI resolution | 10 m (visible/NIR bands); 5-day revisit at equator with both satellites |
| MOPITT CO resolution | Approximately 22 km nadir; archive from March 2000 |
| TROPOMI CO detection sensitivity | Single-pixel precision approximately 10–15 ppbv; plume detection requires enhancement above background |
| TROPOMI archive depth | May 2018 to present (Sentinel-5P launch) |
| Delivery formats | NetCDF (TROPOMI L2), HDF5 (VIIRS), GeoTIFF (Sentinel-2), CSV or GeoJSON (derived attribution layers) |
Analytics Satellize can run
| Daily CO/NO2 enhancement ratio maps | TROPOMI L2 column differencing against local background, ΔCO/ΔNO2 slope computation per plume cluster | GeoTIFF raster layer with per-pixel ratio values and confidence class, updated daily |
| Pyrogenic vs. industrial source classification | Spatial co-location of TROPOMI anomalies with VIIRS fire detections and Sentinel-2 facility masks; ratio-based decision tree against published emission factor ranges | Classified GIS polygon layer (pyrogenic / industrial / ambiguous) with attribution confidence score |
| Plume trajectory and source linkage | Back-trajectory modelling (HYSPLIT or equivalent) applied to TROPOMI CO anomaly centroids; source linkage to fire clusters or facility locations | PDF report with trajectory maps, source candidates ranked by spatial and chemical consistency |
| Burned-area and facility footprint inventory | Sentinel-2 MSI spectral change detection for burn scars (NBR index); manual or automated facility delineation from high-resolution optical imagery | Shapefile inventory of active burn perimeters and industrial facility polygons used in attribution workflow |
| Multi-week emission anomaly time series | TROPOMI daily CO and NO2 column extraction over defined regions of interest; anomaly detection against rolling climatological baseline | CSV time series with flagged exceedance dates, suitable for regulatory reporting or litigation support |
| Ambiguity assessment report for mixed-source zones | Structured uncertainty quantification combining ratio variance, plume age correction, cloud-cover fraction, and VIIRS detection confidence | Written assessment document identifying irreducible ambiguities and recommending supplementary ground or aircraft measurements where satellite data cannot resolve attribution |
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.