Industrial zone economic activity proxies from tropospheric nitrogen dioxide columns
Tropospheric NO2 columns from Sentinel-5P TROPOMI change measurably when combustion-intensive industry ramps up or shuts down. Weekly averaging turns a noisy atmospheric signal into a usable relative activity index for petrochemical complexes, steel mills and cement plants.
Sensors
- Sentinel-5P TROPOMI: Primary sensor for this method. Measures tropospheric NO2 vertical column density at 3.5 × 5.5 km ground pixel (upgraded from the original 7 × 3.5 km in August 2019), daily global coverage, data latency typically under 3 hours for near-real-time products and under 3 days for offline reprocessed products. Archive from October 2017.
- Aura OMI: NASA predecessor instrument, 13 × 24 km nadir pixel, daily revisit, archive from 2004. Useful for multi-year trend baselines predating TROPOMI, though coarser resolution makes source attribution harder in dense industrial regions.
- GEMS (Geostationary Environment Monitoring Spectrometer): Korean Meteorological Administration instrument on GEO-KOMPSAT-2B, covering East Asia at roughly 3.5 × 8 km with hourly daytime observations from 2021. Hourly cadence allows intra-day activity profiling not possible with low-Earth-orbit sensors, but geographic coverage is limited to Asia-Pacific.
- Sentinel-5P TROPOMI (offline reprocessed L2): The offline product applies additional quality flags and cloud screening, improving data usability in partly cloudy conditions. Effective cloud-fraction threshold for retained pixels is typically set at 0.3 or 0.5 depending on the application, directly trading data volume for accuracy.
What the atmosphere is actually recording
Nitrogen dioxide in the troposphere has two dominant anthropogenic sources in industrial zones: high-temperature combustion in furnaces, boilers and kilns, and road traffic from the logistics corridors that serve those facilities. TROPOMI cannot separate them by spectral means alone. What it measures is the total tropospheric vertical column density, expressed in molecules per square centimetre, integrated from the surface up to the tropopause. The signal over a large industrial complex is typically in the range of 5 to 20 × 10¹⁵ molecules cm⁻², well above the global background of roughly 1 to 3 × 10¹⁵ molecules cm⁻² over clean continental areas.
The physical basis for using this as an activity proxy is straightforward. Combustion intensity scales with fuel consumption, which scales with throughput. A blast furnace operating at full capacity burns substantially more coke and natural gas than one banked or idled. The NO2 column above the facility rises and falls accordingly, with a lag of hours determined by atmospheric mixing and the photochemical lifetime of NO2, which ranges from a few hours in summer sunlight to a day or more in winter. That photochemical lifetime is the first major source of seasonal bias and must be corrected before any month-on-month comparison is meaningful.
Seasonal corrections and the photochemistry problem
NO2 is destroyed primarily by reaction with the hydroxyl radical (OH), whose concentration scales with solar UV intensity. In mid-latitude summer, the tropospheric NO2 lifetime can fall to 4 to 6 hours, meaning that even a large source may not accumulate a detectable column under clear, sunny, afternoon conditions. In winter, the same emission rate can produce a column two to four times larger simply because OH concentrations are lower and the atmosphere holds the gas longer. Any raw time series of NO2 columns over an industrial zone will therefore show a strong seasonal cycle that has nothing to do with production levels.
Correcting for this requires either a chemistry-transport model to estimate the local OH field, or an empirical approach using nearby reference pixels with stable, known emission rates to derive a seasonal correction factor. Neither approach is perfect. Model-based corrections introduce uncertainty from meteorological inputs; empirical corrections require identifying genuinely stable reference sources, which is not always possible in dense industrial regions. The practical result is that month-on-month comparisons within a season are more reliable than cross-season comparisons, and annual averages should be treated with caution unless a full photochemical correction has been applied and validated.
Deconvolving traffic from industrial combustion
A petrochemical complex sits inside a road network. The NO2 column above it reflects both the facility's stacks and the trucks, trains and cars serving it. Separating these contributions is possible only approximately, and the approximation depends on spatial pattern analysis rather than spectral separation.
The approach rests on the fact that traffic NO2 tends to follow linear corridors and peaks in morning and evening rush hours, while industrial stack NO2 tends to be spatially concentrated over the facility footprint and varies on slower operational timescales. With TROPOMI's 3.5 × 5.5 km pixel, a large complex spanning several kilometres can be distinguished from its access roads if the geometry is favourable. For smaller facilities, or where roads run directly through the complex, the deconvolution is unreliable. GEMS's hourly cadence adds a temporal dimension: a facility that shows elevated NO2 continuously through the working day, rather than in morning and evening spikes, is more likely dominated by industrial combustion. This is a probabilistic inference, not a clean separation, and should be presented as such.
Power-sector contributions present a different problem. A coal-fired power plant adjacent to a steel mill will add NO2 that is operationally independent of the mill's output. Where the two facilities are spatially separable at TROPOMI resolution, they can be tracked independently. Where they overlap within a single pixel, the analyst must use ancillary data, such as thermal infrared signatures or grid dispatch records, to apportion the signal. This page does not cover those thermal methods, which are addressed separately in the Satellize use-case library.
Building a relative activity index
The output of this method is deliberately described as a relative activity index, not a production volume. There is no published, peer-reviewed transfer function that converts a TROPOMI NO2 column directly into tonnes of steel or barrels of petrochemical output with acceptable uncertainty for commercial use. What the column time series does yield is a dimensionless index: facility X in week W was operating at approximately Y% of its peak observed level over the reference period.
Constructing this index requires a baseline. The standard approach is to compute a multi-year weekly mean column for the facility pixel, using the TROPOMI archive from late 2017 onwards or the OMI archive from 2004, apply seasonal and meteorological corrections, and express each new observation as a deviation from that baseline. A sustained deviation of 20% or more over three or four consecutive weeks is generally considered analytically significant, though the exact threshold depends on the facility size, local meteorology and the cloud-fraction filtering applied. Single-week anomalies are rarely reliable; the atmosphere is too variable.
The method works best for large, spatially distinct facilities: integrated steel mills covering several square kilometres, large refinery and petrochemical complexes, or cement clinker lines with multiple kilns. It is not suitable for isolating individual small boilers, single process units, or facilities smaller than roughly 1 to 2 km² in footprint, because the NO2 signal from such sources is unlikely to dominate a TROPOMI pixel.
Honest limits: what this method cannot do
Cloud cover is the most immediate operational constraint. TROPOMI pixels with cloud radiance fractions above 0.5 are typically excluded from analysis, and in persistently cloudy regions, such as parts of East Asia in monsoon season or northern Europe in winter, usable data may be available for only two or three weeks per month. Weekly averaging mitigates this but does not eliminate it.
The method cannot distinguish between a planned maintenance shutdown and an unplanned outage, or between a voluntary production cut and a regulatory curtailment. It observes the atmospheric consequence of combustion, not the operational cause. Interpreting the index requires contextual knowledge of the facility's normal operating calendar, local regulatory environment and commodity market conditions. Used in isolation, a drop in the NO2 index is ambiguous.
Finally, the spatial resolution of TROPOMI means that in dense industrial corridors, such as the Ruhr valley or parts of the Pearl River Delta, individual facility signals overlap and the index reflects the zone rather than any single plant. Satellize applies this method as part of broader commodity intelligence workflows, including for clients building sovereign economic monitoring capabilities. Its Tonga crop-estimation programme demonstrates the same principle of using open-constellation data to derive economic indicators from physical signals, in a very different context.
Typical figures
| Primary sensor pixel size | 3.5 × 5.5 km (TROPOMI, post-August 2019); 13 × 24 km (OMI nadir) |
| Revisit frequency | Daily global (TROPOMI, LEO); hourly daytime over Asia-Pacific (GEMS, GEO) |
| Practical temporal resolution for activity index | Weekly averaging recommended; single-day observations unreliable without meteorological correction |
| Data latency | TROPOMI near-real-time: under 3 hours; offline reprocessed: under 3 days |
| Spectral measurement | UV-visible absorption in the 405 to 465 nm window; NO2 column retrieved by differential optical absorption spectroscopy (DOAS) |
| Minimum detectable column anomaly (facility scale) | Roughly 1 to 2 × 10¹⁵ molecules cm⁻² above local background, depending on cloud conditions and seasonal OH correction quality |
| Minimum facility footprint for reliable isolation | Approximately 1 to 2 km² or larger; smaller facilities not reliably separable at TROPOMI resolution |
| Archive depth | TROPOMI: October 2017 to present; OMI: August 2004 to present (degraded since ~2019) |
| Cloud-fraction threshold for pixel retention | Typically 0.3 to 0.5 radiance fraction; higher cloud cover excludes the pixel |
| Delivery formats | NetCDF (native L2/L3), GeoTIFF time-series stack, CSV activity-index table, GIS polygon layer with weekly index values |
Analytics Satellize can run
| Weekly relative activity index per facility or industrial zone | TROPOMI L2 NO2 offline product, cloud-screened, seasonally corrected against multi-year OMI/TROPOMI baseline using empirical or CTM-based OH correction | CSV or GIS time-series feed, updated weekly, with uncertainty band per observation |
| Anomaly alert: sustained production change | Three-week rolling deviation from seasonal baseline exceeding analyst-set threshold (typically 20 to 30%) | Automated alert report with facility identifier, deviation magnitude, data-quality flags and cloud-fraction metadata |
| Multi-facility zone composite index | Spatial aggregation of TROPOMI pixels over defined industrial polygon; area-weighted mean column density corrected for pixel overlap | Monthly zone-level activity index report suitable for economic monitoring dashboards |
| Traffic versus industrial contribution decomposition | Temporal pattern analysis using GEMS hourly data where available, supplemented by spatial gradient analysis of TROPOMI columns across facility footprint and adjacent road corridors | Qualitative decomposition memo with confidence rating; not a clean quantitative split |
| Long-term trend extraction (5 to 20 years) | OMI archive from 2004 spliced to TROPOMI from 2017 using inter-calibration offset; Mann-Kendall trend test on seasonally adjusted annual means | Trend report with slope estimate, statistical significance and break-point identification |
| Facility-level shutdown and restart event log | Change-point detection on weekly index series; events flagged when index drops below 30% of baseline for two or more consecutive weeks and recovers | Structured event log with dates, duration and estimated depth of curtailment |
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.