Iron and steel plant emission fingerprinting via co-emitted NO2 and CO columns
Blast furnace and basic oxygen furnace operations produce a co-emission ratio of NO2 to CO that is physically distinct from power stations or traffic. TROPOMI and MOPITT column retrievals, cross-checked against Sentinel-2 facility imagery, can fingerprint integrated steelworks and separate them from electric arc furnace plants.
Sensors
- Sentinel-5P TROPOMI: Provides daily global NO2 and CO tropospheric column retrievals at approximately 3.5 × 5.5 km (NO2, post-August 2019 reprocessing) and 7 × 7 km (CO). Single overpass per day near local solar noon. Minimum detectable NO2 column enhancement above background is roughly 0.5–1 × 10¹⁵ molecules/cm² for large point sources; CO detection limit is around 5–10 ppbv column enhancement in clear conditions.
- MOPITT (Terra): Measures CO vertical profiles and total columns at 22 km nadir resolution with a 3-day revisit at mid-latitudes. Lower spatial resolution than TROPOMI but provides multi-level profile information useful for separating boundary-layer CO from free-tropospheric transport. Archive extends to March 2000, enabling long-term trend analysis.
- Sentinel-2 MSI: 10 m multispectral imagery (VNIR/SWIR) used for facility identification and structural classification: distinguishing blast furnace stacks, basic oxygen furnace shops and electric arc furnace bays by their physical footprints and thermal signatures. 5-day revisit at mid-latitudes with two satellites. Does not measure trace gas columns directly.
- VIIRS (Suomi-NPP / NOAA-20): Day/Night Band and thermal infrared bands confirm furnace thermal activity. VIIRS Nightfire product (Colorado School of Mines / NOAA) detects high-temperature combustion events at sub-pixel scale, confirming whether a blast furnace was in active operation during the TROPOMI overpass window.
Why the NO2-to-CO ratio is a process identity card
Blast furnaces running on coke produce large volumes of blast furnace gas, a mixture dominated by CO (roughly 20–28% by volume) with smaller but measurable nitrogen oxide fractions from combustion of that gas in stoves and ancillary burners. Basic oxygen furnaces add a further CO pulse when the oxygen lance converts molten iron to steel. The result is an atmospheric column signature in which CO is elevated by hundreds of ppbv above local background while NO2 rises more modestly, giving a characteristic high-CO, moderate-NO2 fingerprint.
Electric arc furnaces tell a different story. They melt scrap using electrical energy, so the CO source term is far smaller. Their NO2 emissions from electrode arcing and auxiliary gas burners are not negligible, but the NO2/CO ratio is substantially higher than for integrated blast-furnace routes. This ratio difference is large enough to be detectable in TROPOMI retrievals when wind conditions are favourable and background signals are properly subtracted. It is not a perfect discriminant at the single-pixel level, but across multiple overpasses and wind directions it becomes statistically reliable.
Separating steel plumes from the surrounding airshed
The practical difficulty is that large steelworks rarely sit in isolation. They share airsheds with power stations, coking plants, road networks and port operations, all of which contribute to the observed NO2 and CO columns. Traffic NO2 is spatially diffuse and peaks during morning and evening rush hours, well away from TROPOMI's early-afternoon overpass. Power station plumes are high in NOx but low in CO unless combustion is inefficient. Coking plants are often co-located with steelworks, which complicates attribution but also reinforces the CO signal.
The standard approach is a wind-rotation or plume-dispersion method: TROPOMI pixels downwind of the facility are compared with upwind background columns on the same overpass, controlling for the prevailing wind direction from reanalysis data such as ERA5. Averaging across many overpasses with varied wind directions allows a facility-level emission rate to be estimated. Published studies using this method have attributed NO2 and CO enhancements to individual industrial sites with spatial separations of 20–30 km, though co-located sources within that radius remain difficult to disentangle without additional constraints.
VIIRS Nightfire data provides an independent check: if a blast furnace was banked or in a planned outage during a given overpass, the thermal signal will be absent and that day's TROPOMI observation should be excluded from the active-operation average. This cross-referencing substantially reduces noise in multi-month emission estimates.
What a floating roof gives away (and what TROPOMI cannot resolve)
Sentinel-2 imagery at 10 m resolution can identify individual furnace stacks, slag pits, hot metal transfer ladles and the characteristic rectangular footprint of a basic oxygen furnace shop. This structural intelligence anchors the atmospheric retrieval to a specific process unit rather than a vague industrial zone. When a new electric arc furnace is installed alongside an existing blast furnace complex, Sentinel-2 can detect the new building footprint and the change in stack configuration before any regulatory filing appears in public databases.
The honest limits of TROPOMI must be stated plainly. At 3.5 × 5.5 km, a single pixel covers the entire footprint of most steelworks and several hundred metres of surrounding area. Cloud cover invalidates retrievals entirely, and steel-producing regions in East Asia and Eastern Europe frequently experience cloud fractions above 0.5 for weeks at a time. Aerosol loading from the plant itself can bias NO2 retrievals by 5–15% depending on aerosol optical depth. Single-day retrievals are rarely meaningful; credible attribution requires 30 to 90 overpasses of usable data, which translates to two to six months of calendar time in typical mid-latitude cloud regimes.
Building a facility-level emission time series
Once a baseline NO2/CO ratio and absolute column enhancement are established for a given facility, changes in that ratio over time carry information about process changes. A shift toward lower CO relative to NO2 could indicate partial transition to electric arc capacity or installation of CO recovery systems on the blast furnace gas circuit. A sudden spike in both columns simultaneously is consistent with an unplanned release or a combustion control failure.
Monthly aggregation is the practical temporal unit for this kind of monitoring. TROPOMI's archive runs from May 2018, giving roughly six years of continuous data as of mid-2024. MOPITT extends the CO record back to 2000, allowing pre-TROPOMI baselines to be established for facilities that were operating then, though at coarser spatial resolution. Together, these archives make it possible to assess whether a facility's emission intensity has changed following regulatory interventions or capacity expansions, without relying solely on self-reported stack measurements.
Satellize applies this multi-sensor fusion method operationally, drawing on the same open constellation data used in its Tonga crop-estimation programme and other analytics work, to produce facility-level emission scorecards for government enforcement clients.
The ambiguities that remain, and how to manage them
No satellite-only analysis can replace stack measurements for absolute emission quantification. TROPOMI column retrievals carry retrieval uncertainties of roughly 20–30% for individual CO observations and around 15–25% for NO2, depending on surface albedo, aerosol conditions and viewing geometry. These uncertainties shrink with averaging but never disappear. The method is best understood as a screening and prioritisation tool: it identifies facilities whose atmospheric signature is inconsistent with their reported output or whose emission ratios suggest process changes worth investigating on the ground.
Seasonal variation in photochemistry affects NO2 columns independently of emission rates, because OH radical concentrations that control NO2 lifetime vary with sunlight. Summer retrievals over a given facility will show lower NO2 columns than winter retrievals even at constant emission rates. Any time-series analysis must account for this photochemical seasonality, typically by comparing facility columns against a regional background that experiences the same photochemical regime. Ignoring this correction produces spurious apparent emission trends.
Typical figures
| TROPOMI NO2 pixel size | 3.5 × 5.5 km (post-August 2019 reprocessing) |
| TROPOMI CO pixel size | 7 × 7 km |
| TROPOMI revisit | Daily, single overpass near 13:30 local solar time |
| MOPITT CO resolution | 22 km nadir; 3-day revisit at mid-latitudes |
| Sentinel-2 MSI resolution (facility ID) | 10 m (VNIR); 5-day revisit with two satellites |
| VIIRS Nightfire detection threshold | Sub-pixel high-temperature combustion sources; published detection at flame temperatures above ~1000 K |
| Minimum detectable NO2 enhancement | Approximately 0.5–1 × 10¹⁵ molecules/cm² above background for large point sources (single overpass) |
| Minimum usable overpasses for facility attribution | 30–90 cloud-free overpasses; typically 2–6 months calendar time |
| TROPOMI archive depth | May 2018 to present; MOPITT CO from March 2000 |
| Retrieval uncertainty (single observation) | CO: 20–30%; NO2: 15–25%, depending on aerosol loading and surface albedo |
Analytics Satellize can run
| Facility NO2/CO ratio fingerprint | Wind-rotation plume isolation applied to TROPOMI NO2 and CO column pairs; ratio computed per overpass and averaged across wind-screened subset | Facility scorecard PDF and GIS point layer with confidence intervals, updated quarterly |
| Monthly emission anomaly alert | Rolling 30-day column enhancement compared against 12-month baseline; threshold exceedance flagged when enhancement exceeds 2 standard deviations | Automated alert feed (JSON or email) with supporting TROPOMI imagery and wind metadata |
| Process-type classification (integrated vs. EAF) | NO2/CO ratio discrimination combined with Sentinel-2 structural classification of furnace footprint and stack configuration | Facility classification table with confidence rating and Sentinel-2 image evidence |
| Thermal activity cross-check | VIIRS Nightfire daily product matched to TROPOMI overpass timestamps to flag active versus inactive furnace days | Overpass validity filter applied to all TROPOMI retrievals; activity log included in monthly report |
| Multi-year emission trend analysis | TROPOMI 2018-present combined with MOPITT CO for pre-2018 baseline; seasonal photochemical correction using regional background normalisation | Time-series chart and underlying data table (CSV) covering available archive depth |
| Airshed source apportionment | Spatial decomposition of TROPOMI columns using wind-sector analysis to separate steel-plant contribution from co-located power and traffic sources | Source contribution breakdown map (GeoTIFF) with uncertainty ranges per source type |
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.