Volatile organic compound plume mapping at petroleum refineries
TROPOMI and OMI HCHO column retrievals expose VOC emission intensity above refinery complexes, but biogenic background and coarse pixels demand careful interpretation. Wind-field integration converts column anomalies into flux estimates regulators can act on.
Sensors
- Sentinel-5P TROPOMI: Primary HCHO column product. Native pixel footprint 3.5 × 5.5 km (upgraded from 7 × 3.5 km in August 2019). Daily global coverage. HCHO precision roughly 0.6–1.0 × 10¹⁵ molecules cm⁻² per scene; individual refinery pixels are near the detection floor, so multi-day compositing is standard practice.
- Aura OMI: Provides HCHO column retrievals since 2004, giving a 20-year archive for trend analysis. Pixel footprint 13 × 24 km at nadir, degraded further by the row-anomaly that has affected part of the swath since 2009. Useful for long-term climatology rather than individual event attribution.
- GEMS (GEO-KOMPSAT-2B): Geostationary UV-visible spectrometer covering East Asia at roughly 3.5 × 8 km pixels with hourly daytime revisit. Enables intra-day plume evolution tracking that polar orbiters cannot provide. Operational since 2020; HCHO product still maturing in validation literature.
- Sentinel-2 MSI: No direct VOC sensitivity, but 10 m resolution multispectral imagery provides facility context: active process units, flare visibility, tank farm configuration, and surrounding land-cover classification needed to partition biogenic from anthropogenic HCHO background.
Why formaldehyde is the proxy, not the target
Petroleum refineries emit a broad cocktail of volatile organic compounds including alkenes, aromatics, and oxygenates. Most of these species have no usable absorption feature in the UV-visible window accessible from orbit. Formaldehyde (HCHO) does. It absorbs strongly near 340 nm, is retrievable via Differential Optical Absorption Spectroscopy (DOAS) from backscattered solar radiation, and is produced rapidly in the atmosphere as VOCs oxidise. It is therefore a secondary tracer: elevated HCHO columns above a refinery reflect the facility's aggregate VOC emission load, not a single compound.
The relationship is not direct. HCHO yield per unit VOC mass depends on the specific compound mix. Alkene-rich streams, such as fluid catalytic cracker off-gases, produce HCHO faster than alkane-dominated streams. This means HCHO columns are a useful relative indicator of emission intensity and temporal change, but converting them to total VOC mass requires assumptions about speciation that ground-based stack data or published emission factors must supply.
The biogenic interference problem is not trivial
Vegetation is a prolific isoprene emitter, and isoprene oxidises to HCHO on timescales of tens of minutes. In temperate or tropical climates, the biogenic HCHO background can reach 4–8 × 10¹⁵ molecules cm⁻², comfortably exceeding the anthropogenic enhancement above a mid-sized refinery. Refineries in forested or agricultural surroundings are genuinely harder to isolate than those in arid or coastal-industrial zones.
Standard practice is to subtract a spatially interpolated biogenic baseline derived from MEGAN (Model of Emissions of Gases and Aerosols from Nature) or from upwind pixels of similar land cover, then attribute residual column enhancement to the facility. Seasonal stratification matters: summer isoprene peaks can swamp the industrial signal entirely, making winter or shoulder-season observations more diagnostic in vegetated regions. Sentinel-2 land-cover classification of the surrounding grid cells is an essential input to this partitioning step.
From column anomaly to emission flux: the wind field is the multiplier
A HCHO column enhancement expressed in molecules per square centimetre is not an emission rate. To convert it to a flux in tonnes per hour, analysts apply a mass-balance or cross-sectional flux method: the column excess is multiplied by the wind speed at the relevant transport height and integrated across a downwind transect. ERA5 reanalysis winds at 10 m and 850 hPa are commonly used; for higher accuracy, radiosonde profiles or mesoscale model output from the analysis day are preferable.
Uncertainty in this step is substantial. Wind speed errors of 1–2 m/s translate directly into proportional flux errors. HCHO has an atmospheric lifetime of only a few hours under daytime photolysis, so the plume detected at overpass time may represent a fraction of the day's total emission. Multi-day averaging reduces random noise but conflates different meteorological regimes. Published studies on refinery HCHO fluxes from TROPOMI typically report uncertainties of 30–50% on individual facility estimates, which is honest and should be stated plainly to any regulatory client.
What the pixel size actually limits
TROPOMI's 3.5 × 5.5 km footprint is large relative to most refinery boundaries. A single pixel typically covers the entire facility and some surrounding area. This means the method cannot distinguish between process units within a complex, attribute emissions to a specific tank farm versus a cracking unit, or detect a short-duration fugitive release lasting less than an hour. It is a facility-scale, time-averaged diagnostic.
Attribution confidence improves markedly when the facility is isolated, when wind direction is stable and well-characterised, and when a multi-week composite is used rather than a single overpass. Conversely, confidence degrades in urban-industrial clusters where multiple emitters share the same pixel, in coastal zones with complex sea-breeze circulation, and during cloud cover, which invalidates the retrieval entirely. TROPOMI's cloud fraction threshold for valid HCHO retrievals is typically set at less than 0.3, reducing effective clear-sky observations in persistently cloudy regions to fewer than half of all overpasses.
Building a monitoring time series: archive depth and operational cadence
OMI's record from 2004 gives nearly two decades of HCHO columns, sufficient to establish pre-operational baselines before a facility expands, or to document emission trends across regulatory regime changes. TROPOMI, operational since late 2017, provides the higher spatial resolution needed for credible facility-level attribution and is the preferred operational sensor.
A practical monitoring workflow combines a rolling 30-day TROPOMI composite to suppress noise, a biogenic correction applied monthly using MEGAN output, and a wind-flux calculation using ERA5 for each composite period. Anomaly alerts, where the composite column exceeds a threshold set from the facility's own historical distribution, can be generated automatically. Satellize applies this composite-and-correct approach across open TROPOMI data streams; the underlying method is the same framework used in published academic monitoring of refinery districts in the US Gulf Coast and the Middle East. For clients wanting facility-level context, Sentinel-2 imagery is pulled on the same schedule to flag visible operational changes such as new tank construction or flare activity.
Honest limits and what complements the method
Satellite HCHO is a screening and trend tool, not a substitute for continuous emissions monitoring systems or fence-line sensors. It cannot resolve individual stack emissions, cannot operate through cloud, and carries flux uncertainties that most regulatory frameworks would not accept as a standalone compliance instrument. What it does well is independent, wide-area surveillance: it covers every facility in a country simultaneously, requires no site access, and provides a historical record that predates any ground instrument deployment.
The most defensible use case is anomaly detection and prioritisation. A persistent HCHO enhancement in a 30-day composite, after biogenic correction, is a credible basis for directing ground-based inspection resources. Combined with co-located NO2 columns from the same TROPOMI overpass and SO2 data where relevant, the multi-species fingerprint narrows the source attribution further. That combination is where satellite VOC monitoring earns its place in a national enforcement programme.
Typical figures
| Primary sensor pixel footprint | TROPOMI: 3.5 × 5.5 km (post-August 2019); OMI: ~13 × 24 km at nadir |
| Revisit cadence | TROPOMI: daily global; OMI: daily global (partial swath affected by row anomaly); GEMS: hourly daytime over East Asia |
| HCHO retrieval precision (TROPOMI) | ~0.6–1.0 × 10¹⁵ molecules cm⁻² per scene; facility signals often 1–5 × 10¹⁵ above background |
| Spectral retrieval window | UV, approximately 328–356 nm (DOAS HCHO fit window varies by product version) |
| Cloud fraction threshold for valid retrieval | Typically < 0.3; higher cloud fraction scenes flagged as invalid |
| Archive depth | OMI: from October 2004; TROPOMI: from November 2017 |
| Flux estimation uncertainty | Typically 30–50% on individual facility estimates using mass-balance wind integration |
| Contextual optical resolution (Sentinel-2) | 10 m (visible/NIR bands) for land-cover and facility mapping |
| Latency (TROPOMI offline product) | Approximately 3–5 days for the standard offline HCHO product; near-real-time product within hours but lower quality |
Analytics Satellize can run
| Monthly HCHO column anomaly map | 30-day TROPOMI composite with biogenic baseline subtraction using MEGAN-derived isoprene oxidation correction; residual attributed to facility cluster | GeoTIFF layer and PDF report showing column anomaly magnitude per facility grid cell, flagged against facility's own historical percentile distribution |
| VOC emission flux estimate | Cross-sectional mass-balance using HCHO column excess integrated across downwind transect, multiplied by ERA5 transport-layer wind speed; uncertainty bounds reported explicitly | Tabular flux estimate in kg/hr with confidence interval, per composite period, delivered as CSV with methodology annex |
| Biogenic-anthropogenic partition assessment | Sentinel-2 land-cover classification of surrounding 20 km radius combined with seasonal MEGAN output to quantify expected biogenic contribution; residual assigned to industrial sources | Partition report with confidence rating (high/medium/low) based on land-cover homogeneity and season |
| Long-term emission trend analysis | Annual HCHO column composites from OMI (2004 onwards) and TROPOMI (2017 onwards), deseasonalised using harmonic regression to isolate multi-year trend signal | Time-series chart and statistical trend summary (slope, significance) for use in regulatory reporting or facility due diligence |
| Multi-species co-emission fingerprint | Co-registration of HCHO, NO2, and SO2 TROPOMI columns over the same facility footprint to distinguish refinery signature from adjacent industrial or traffic sources | Overlay GIS layer with species ratio annotation; narrative interpretation in accompanying brief |
| Anomaly alert feed | Automated threshold exceedance detection on rolling 30-day composite; threshold set at 90th percentile of facility's own 12-month baseline distribution | Email or API alert with overpass date, column value, wind conditions, and link to supporting imagery within 5 days of composite update |
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.