NO2 column mapping over industrial facilities
Tropospheric NO2 columns retrieved from TROPOMI and OMI can fingerprint persistent emission enhancements above named power stations, smelters, and industrial clusters. Multi-year composites resolve facility-level signals at 3.5 × 5.5 km, with honest caveats on cloud screening and air-mass factor uncertainty.
Sensors
- Sentinel-5P TROPOMI: Primary workhorse. UV-visible DOAS retrieval of tropospheric NO2 at 3.5 × 5.5 km (reprocessed from original 7 × 3.5 km after August 2019 upgrade). Daily global coverage, latency roughly 3 hours for near-real-time product and 5 days for the offline reprocessed product. Archive from October 2017.
- Aura OMI: 13 × 24 km nadir pixel, daily global revisit. Lower spatial resolution than TROPOMI but an archive extending to 2004, making it the reference dataset for long-term trend analysis at large facilities. Row-anomaly data gaps must be flagged.
- GEMS (GEO-KOMPSAT-2B): Geostationary UV-visible spectrometer covering East Asia from 60°E to 145°E. Hourly daytime observations at roughly 3.5 × 8 km. Captures diurnal NO2 cycles above industrial corridors in China, South Korea and India that polar orbiters cannot resolve.
- Sentinel-4 (forthcoming): ESA geostationary instrument planned for Meteosat Third Generation, covering Europe and North Africa. Hourly UV-visible-NIR observations at sub-8 km resolution. Will complement TROPOMI for European regulatory enforcement once operational.
What DOAS actually measures, and what it does not
Differential optical absorption spectroscopy isolates NO2 by fitting its characteristic absorption fingerprint in the 405 to 465 nm window against the solar backscatter spectrum. The retrieval yields a slant column density, the integrated concentration along the photon path through the atmosphere. Converting that to a vertical column density requires an air-mass factor, which depends on the assumed vertical profile shape of NO2, surface albedo, and the viewing and solar geometry at the moment of overpass.
The profile shape is taken from a chemical transport model, typically TM5-MP in the TROPOMI operational chain. Where the real atmosphere differs sharply from the model, for instance above a facility with a very tall stack injecting NO2 high into the free troposphere, the air-mass factor can be wrong by 20 to 50 percent. That is not a flaw to hide; it is a calibration problem that facility-specific profile corrections can partially address, and any credible analysis should report it explicitly.
From pixels to facilities: the attribution problem
A single TROPOMI overpass over a 500 MW coal plant typically shows a column enhancement of 1 to 5 × 10¹⁵ molecules cm⁻² above local background, depending on wind speed, mixing-layer depth, and load factor. That signal sits near the retrieval noise floor on any given day. The technique becomes powerful when you stack two to four years of cloud-screened overpasses and compute a time-averaged composite. Persistent enhancements that survive the averaging are almost certainly real; transient industrial accidents or agricultural burning events wash out.
Attribution to a named facility rather than a general area requires cross-referencing with wind fields. Rotating the composite into a wind-aligned coordinate system, a method published in the peer-reviewed literature for TROPOMI and OMI data, concentrates the downwind plume signal and allows an emission rate estimate via mass-balance integration. The approach works best for large isolated sources. A cluster of smelters and chemical plants within a 20 km radius will produce overlapping plumes that are genuinely difficult to disaggregate without additional information, such as stack-height data or co-located SO2 signatures.
Cloud fraction and surface albedo: where retrievals go wrong
The TROPOMI NO2 product applies a cloud radiance fraction threshold, typically 0.5 in the offline product, to exclude heavily clouded scenes. In practice, analysts often tighten this to 0.2 or 0.3 to reduce cloud contamination at the cost of losing a larger fraction of observations. In persistently cloudy regions, such as equatorial industrial zones or high-latitude smelter sites in winter, this can reduce the usable observation count by 60 to 80 percent, stretching the time needed to build a statistically stable composite.
Surface albedo matters because bright surfaces, snow, salt flats, and certain industrial rooftops, increase the effective light path in ways that inflate the retrieved column. The operational retrieval uses a monthly climatological albedo database. Facilities that sit on anomalously bright ground, such as open-pit mines with white tailings ponds, can carry a systematic positive bias that must be corrected before any compliance comparison is meaningful. This is a known, documented limitation, not a reason to distrust the method, but it must be stated upfront to any regulatory client.
What multi-year composites reveal that inspectors miss
Ground-based inspectors sample a facility on a handful of days per year, often announced in advance. A satellite composite built from 500 or more cloud-free overpasses over three years reflects the facility's typical operating state across all seasons and all hours of the overpass window, which for TROPOMI is around 13:30 local solar time. Facilities that throttle emissions during announced inspections but run hard overnight or on weekends will show up in the composite at their true average load.
This is where the technique earns its place in enforcement rather than just research. Several peer-reviewed studies using OMI and TROPOMI data have identified facilities whose satellite-derived emission estimates exceed their permitted or self-reported values by factors of two or more. The satellite does not prove a violation by itself; it identifies candidates for targeted ground investigation. That distinction matters for how regulators should present findings, and for how data providers should characterise what they are selling.
Honest limits and the cases where the method fails
Small facilities are largely invisible. A gas turbine peaker plant or a mid-sized cement kiln operating below roughly 50 to 100 MW thermal output will produce NO2 columns that sit within retrieval noise even after multi-year compositing. TROPOMI at 3.5 × 5.5 km cannot resolve individual stacks within a dense industrial estate; the pixel integrates everything inside it. Future instruments with higher spatial resolution, including commercial hyperspectral imagers currently under development, may push the detection threshold down, but those systems are not yet in routine operational use for NO2.
Diurnal variation is another gap. A polar overpass at 13:30 local time captures one moment. GEMS covers East Asia hourly, which is a genuine advantage for understanding how NO2 builds up during morning rush or industrial shift changes, but geostationary coverage does not yet extend to most of Africa, South America, or South Asia. For those regions, the single daily TROPOMI snapshot is what is available, and interpreting it as a 24-hour average requires a model assumption that adds uncertainty.
Satellize runs TROPOMI-based NO2 compositing and wind-rotation attribution as part of its emissions analytics stack, drawing on the same open ESA data that underpins published regulatory studies. The Tonga crop-estimation programme sits at the other end of the application spectrum, but the underlying pipeline for ingesting, screening, and compositing large volumes of Copernicus data is the same.
Typical figures
| Spatial resolution (TROPOMI, post-Aug 2019) | 3.5 × 5.5 km at nadir |
| Revisit (TROPOMI) | Daily global, single overpass ~13:30 local solar time |
| Revisit (GEMS, East Asia) | Hourly daytime, geostationary |
| Spectral window for NO2 retrieval | 405 to 465 nm (UV-visible DOAS) |
| Typical retrieval precision (single overpass) | ~0.5 × 10¹⁵ molecules cm⁻² (random); systematic uncertainty 20–50% where profile shape deviates from model |
| Cloud fraction screening threshold (standard) | Radiance fraction ≤ 0.5 (offline); analysts typically tighten to ≤ 0.2–0.3 |
| Minimum detectable facility emission (composite) | Roughly 50–100 MW thermal equivalent for isolated sources; larger for clustered facilities |
| Archive depth | TROPOMI from October 2017; OMI from 2004 |
| Near-real-time latency (TROPOMI NRTI) | ~3 hours after overpass |
| Delivery formats | NetCDF-4 (ESA standard), GeoTIFF composite layers, GIS-ready shapefiles with facility attribution |
Analytics Satellize can run
| Multi-year tropospheric NO2 composite | Cloud-screened temporal averaging of TROPOMI offline NO2 product (L2, QA ≥ 0.75), regridded to 0.01° using area-weighted oversampling | GeoTIFF raster layer with per-pixel mean column and standard error, updated annually or on request |
| Facility-level emission rate estimate | Wind-rotation mass-balance method applied to oversampled NO2 composite, using ERA5 reanalysis wind fields at overpass time | Tabular report per named facility: estimated annual NOx emission in kt yr⁻¹ with uncertainty range, compared against permit or self-reported values where available |
| Trend detection and permit compliance flag | Year-on-year column anomaly analysis using OMI archive (2004 onward) extended with TROPOMI; Mann-Kendall trend test on annual composites | Facility scorecard: trend direction, statistical significance, and flag for persistent exceedance above background threshold |
| Cloud-screening and albedo bias assessment | Per-facility data-quality audit: usable observation count, seasonal distribution, surface albedo from MODIS MCD43 climatology, flagged anomalies | Methodology annex suitable for regulatory submission, documenting known biases and their estimated magnitude |
| Diurnal NO2 profile (East Asia, GEMS) | Hourly GEMS L2 NO2 compositing over target facility, averaged by hour-of-day across cloud-free observations | Diurnal cycle chart and GIS layer showing peak-hour enhancement, useful for correlating with reported operational schedules |
| Industrial cluster disaggregation | Plume-separation analysis combining TROPOMI NO2 with co-located SO2 columns and stack metadata, using source-apportionment regression | Per-stack attribution table with confidence rating; low confidence flagged explicitly where spatial overlap prevents clean separation |
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.