Refinery flare stack combustion state and upset event detection
VIIRS, Sentinel-3 SLSTR and TROPOMI together reveal flare ignition, intensity shifts and extinction events at refineries, flagging process upsets and emergency depressurisations without a single site visit.
Sensors
- VIIRS Day/Night Band (DNB): 750 m ground sample distance, daily global overpass near 01:30 local solar time. DNB detects low-radiance nocturnal combustion down to roughly 2 nW/cm²/sr, making it sensitive to small or partially extinguished flares that MODIS-era sensors missed. Used for ignition and extinction timing.
- VIIRS SWIR bands (M10, M11, I4, I5): 375 m (I-band) to 750 m (M-band) resolution. Shortwave infrared at 1.6 µm and 2.25 µm saturates on large flares but retains dynamic range on moderate events; mid-infrared I4 (3.74 µm) captures combustion temperatures in the 600–1200 K range consistent with refinery flaring.
- Sentinel-3 SLSTR: 500 m nadir resolution in S5 (1.6 µm) and S6 (2.25 µm) SWIR channels, plus 1 km thermal bands. Dual-view geometry aids aerosol correction. Revisit roughly one day at mid-latitudes with two satellites. SLSTR fire radiative power retrievals are documented in the Copernicus Land Service; the same physics applies to industrial flares.
- Sentinel-5P TROPOMI: Ethane (C₂H₆) column retrievals at 7 × 5.5 km pixel (upgraded to 5.5 × 3.5 km from August 2019). Ethane is a co-product of wet-gas refinery flaring and upstream production flaring alike, but its ratio to simultaneously retrieved formaldehyde or propane can help distinguish source type. Daily global coverage. Published detection limit roughly 0.1–0.5 ppb column enhancement above background.
- Sentinel-5P TROPOMI SO₂ and NO₂: Same 5.5 × 3.5 km pixel. SO₂ spikes accompany sour-gas flaring at refineries processing high-sulphur crude; NO₂ rises with combustion intensity. Neither gas is diagnostic alone, but combined with SWIR radiance they constrain fuel type and combustion completeness.
What a flare stack is actually telling you
A flare stack is a pressure-relief valve made visible. When a refinery unit trips, a compressor surges, or an emergency depressurisation is ordered, the excess hydrocarbon stream is routed to the flare header and burned rather than vented raw. The intensity of combustion, its duration, and the time of day it occurs are all operationally meaningful signals.
Routine, low-rate flaring is normal and nearly constant at many sites. What matters analytically is deviation: a sudden ignition on a stack that was cold, a step-change in radiance on one that was already burning, or an extinction that coincides with a unit shutdown. Those deviations are the events worth flagging. The sensors described here detect them; interpreting them requires cross-referencing with the site's known process configuration and any concurrent AIS, SO₂ or wind data.
Why nocturnal radiance comes first
The VIIRS Day/Night Band was designed for city lights and moonlit cloud. Flares are, by the standards of the DNB, extremely bright point sources. A single overpass near 01:30 local time captures radiance in the panchromatic visible range (0.5–0.9 µm), and the VIIRS Nightfire algorithm, maintained by the Colorado School of Mines Earth Observation Group and published openly, fits a Planck curve to multi-band SWIR and DNB observations to retrieve combustion temperature and fire radiative power. For refinery flares, published Nightfire retrievals typically resolve temperature in the 1400–1900 K range and can detect sources with radiant heat output as low as a few megawatts.
The practical limit is temporal, not radiometric. One overpass per night means an ignition event that starts and extinguishes within a few hours may be missed entirely, or caught only at peak or only at tail. Pairing VIIRS with Sentinel-3 SLSTR, which can provide a second daily observation at many latitudes, narrows that gap. It does not close it. For high-value sites where intra-day timing matters, commercial tasking of high-revisit thermal sensors is the honest answer.
TROPOMI ethane: refinery versus upstream, and the limits of the distinction
Not every flare plume belongs to a refinery. Associated gas flaring at upstream production fields, gas processing plants, and petrochemical complexes all produce SWIR-bright combustion signatures. TROPOMI's ethane column is useful here because refinery flaring, which burns off light hydrocarbon fractions from distillation and cracking operations, tends to produce elevated ethane-to-methane ratios compared with dry-gas production flaring. The distinction is probabilistic, not categorical.
TROPOMI's pixel footprint of 5.5 × 3.5 km means that in dense industrial corridors, a single pixel may contain signal from multiple sources. Attribution to a specific refinery, rather than to a general industrial cluster, requires that the site sits in relative isolation or that wind-field back-trajectories are used to separate contributions. HYSPLIT or equivalent trajectory modelling is standard practice in published source attribution studies. Cloud cover degrades TROPOMI retrievals in the same way it degrades optical sensors; tropical and maritime refineries face this limitation acutely.
One further constraint: TROPOMI does not retrieve ethane columns in real time for operational users. The standard offline product has a latency of roughly one to two days. Near-real-time SO₂ products are available faster, which is why SO₂ is often the first-alert layer for sour-gas upset events.
Distinguishing combustion states: ignition, steady burn, upset peak, extinction
Classifying a flare into discrete combustion states requires a baseline. For any given site, a multi-year archive of VIIRS Nightfire retrievals and SLSTR SWIR observations establishes the site-specific distribution of radiance and fire radiative power under normal operations. Deviations beyond two or three standard deviations from that baseline, sustained across at least two consecutive overpasses, meet a reasonable threshold for flagging as an upset event.
Extinction events are in some ways more informative than ignition events. A flare that goes cold suggests a deliberate unit shutdown, a feed interruption, or a maintenance window. Cross-referencing extinction timing with vessel AIS data at the refinery's crude intake jetty, or with SO₂ column changes in the days prior, can help distinguish planned maintenance from an unplanned outage. That cross-referencing is where the analytical work sits; the satellite observation alone is necessary but not sufficient.
The method cannot resolve which specific process unit is flaring. A large refinery may have three to six separate flare headers serving different process areas. SWIR sensors at 375–750 m resolution see the stack tip, not the header network. Optical daytime imagery at sub-metre resolution can sometimes identify which header is active from plume geometry, but that is a different product requiring separate tasking.
Archive depth and what it enables
VIIRS has operated continuously since the Suomi NPP launch in October 2011, with NOAA-20 adding a second sensor from 2018. The Colorado School of Mines Nightfire archive covers this full period for global point-source combustion. Sentinel-3A SLSTR data runs from 2016. TROPOMI began in May 2018. Together these archives allow construction of multi-year flaring baselines for any refinery that has been operating during this window, which covers the vast majority of global refining capacity.
For commodity and credit analysts, the archive supports retrospective event reconstruction: identifying when a refinery experienced an upset, how long it lasted, and whether the pattern is consistent with a reported force majeure or maintenance claim. Satellize applies this kind of retrospective analysis on open constellations, adding commercial tasking where archive gaps or timing requirements demand it.
Honest limits of the method
Spatial resolution is the binding constraint for process-unit attribution. No current free-access sensor resolves individual flare headers at a complex refinery. VIIRS at 375 m and SLSTR at 500 m see the aggregate radiance of a site, not its internal structure.
Cloud cover is a persistent problem. Many refineries in tropical latitudes, the US Gulf Coast during summer convection season, and northern European sites in winter face cloud fractions that regularly block optical and SWIR observations for multi-day stretches. SAR can confirm site activity through cloud but does not detect combustion directly. Revisit gaps mean that fast-onset, fast-recovery upset events lasting under four hours may not be captured at peak intensity. The method is better suited to events lasting twelve hours or more. Finally, TROPOMI's source attribution ambiguity in dense industrial regions is a real limitation; analysts should be explicit about confidence levels when presenting attribution conclusions to clients.
Typical figures
| VIIRS DNB spatial resolution | 750 m ground sample distance |
| VIIRS I-band (SWIR/MWIR) resolution | 375 m |
| Sentinel-3 SLSTR SWIR resolution | 500 m nadir |
| TROPOMI pixel footprint | 5.5 × 3.5 km (post-August 2019) |
| VIIRS / Sentinel-3 revisit | ~1 overpass per day per sensor; two VIIRS sensors (NPP + NOAA-20) give two nocturnal observations |
| TROPOMI revisit | Daily global coverage; offline product latency ~1–2 days; near-real-time SO₂ faster |
| Spectral bands used | DNB 0.5–0.9 µm; SWIR 1.6 µm, 2.25 µm; MWIR 3.74 µm; TROPOMI ethane ~1.67 µm, SO₂ UV |
| Minimum detectable radiance (DNB) | ~2 nW/cm²/sr (published Nightfire threshold) |
| Archive depth | VIIRS from Oct 2011; SLSTR from 2016; TROPOMI from May 2018 |
| Cloud limitation | Optical and SWIR retrievals blocked by cloud; no combustion proxy available via SAR alone |
Analytics Satellize can run
| Flare ignition and extinction alert | Threshold detection on VIIRS Nightfire fire radiative power time series against site-specific baseline; confirmed on next available SLSTR overpass | Email or API alert within 6 hours of VIIRS overpass processing, with radiance magnitude and confidence flag |
| Upset event intensity classification | Planck-curve fitting to multi-band SWIR and DNB radiance (Nightfire algorithm) to retrieve combustion temperature and fire radiative power; deviation scored against rolling 90-day baseline | Event report with FRP estimate, duration, and comparison to historical site distribution; delivered as PDF and GeoJSON |
| Sour-gas upset flag via TROPOMI SO₂ | TROPOMI near-real-time SO₂ column anomaly detection co-located with SWIR flare event; wind-field back-trajectory used to confirm source attribution | Supplementary flag appended to upset event report; includes SO₂ column value and trajectory plot |
| Refinery versus upstream flare source discrimination | TROPOMI ethane column cross-referenced with SWIR source location and site classification database; probabilistic attribution with stated confidence | Source-type probability score in event report; honest uncertainty range stated where pixel overlap is ambiguous |
| Multi-year flaring baseline and anomaly history | Full VIIRS Nightfire archive ingestion for named site; statistical characterisation of normal operating envelope; retrospective upset event catalogue | Site baseline report with time-series chart and event catalogue in CSV; suitable for force-majeure validation or credit-risk review |
| Monthly flaring state summary | Aggregated VIIRS and SLSTR observations classified into combustion-state categories (cold, routine, elevated, upset) per calendar month | Monthly GIS layer and tabular summary; compatible with commodity analyst workflow |
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.