Petroleum refinery throughput monitoring by thermal infrared
Thermal infrared and shortwave infrared satellites detect heat from distillation units, flare stacks and cooling towers, providing an independent proxy for refinery operating rate that does not depend on operator disclosure.
Sensors
- VIIRS DNB / M-band (Suomi-NPP, NOAA-20, NOAA-21): The Day/Night Band detects visible and near-infrared radiance from flares at 750 m nadir resolution. The M-band thermal channels (M12–M16, 3.7–12 µm) support radiative power retrieval. VIIRS Nightfire (Colorado School of Mines / NOAA) uses multi-band fitting to estimate flare temperature and radiant heat output. Daily global coverage, but 750 m pixels mean a single pixel can contain multiple process units at a large refinery.
- Landsat 8 / 9 TIRS: Two thermal infrared bands centred at 10.9 µm and 12.0 µm at 100 m native resolution (resampled to 30 m in distributed products). Provides quantitative brightness temperature that can be converted to surface radiance. Revisit is 8 days per satellite, 16 days per instrument, which limits the ability to track short-duration operational changes. Archive extends to 2013 (Landsat 8) and 2021 (Landsat 9), enabling multi-year baseline construction.
- Sentinel-3 SLSTR: Sea and Land Surface Temperature Radiometer carries nine spectral channels from 0.55 µm to 12 µm. Nadir pixel size is 500 m for visible/SWIR and 1 km for thermal. Revisit is approximately one day globally at mid-latitudes when both Sentinel-3A and 3B are active. Useful for tracking large refinery complexes and flare events but too coarse for unit-level attribution.
- Landsat 8 / 9 OLI SWIR (bands 6–7): Shortwave infrared bands at 1.61 µm and 2.20 µm, 30 m resolution, saturate over active flares but can detect thermal anomalies from hot process equipment during daytime passes. Useful for identifying which part of a facility is thermally active when combined with TIRS data from the same overpass.
- Planet SuperDove: Eight spectral bands from 431 nm to 885 nm at approximately 3–4 m resolution with daily revisit over most land areas. No thermal capability, but visible and NIR imagery can detect flare plumes, steam from cooling towers and site activity (vehicle counts, visible smoke) that corroborate thermal findings. Acts as a high-resolution context layer rather than a primary thermal sensor.
What a flare stack gives away
A petroleum refinery cannot operate its distillation columns, hydrocrackers or cokers without generating heat, and it cannot suppress that heat entirely. Flare stacks burn associated gas that cannot be recovered or routed elsewhere. Cooling towers shed waste heat into the atmosphere as warm vapour plumes. Distillation units maintain column temperatures that, at scale, are detectable as elevated surface radiance in the thermal infrared. None of these signatures require the operator to file a report.
The physics is straightforward. A blackbody at flare temperatures (roughly 1,200–1,800 K for a well-combusted gas flare) emits strongly in the shortwave infrared around 1.6 and 2.2 µm, which is why VIIRS M-band and Landsat OLI SWIR channels saturate over active flares. VIIRS Nightfire exploits multi-band radiance ratios to fit a Planck curve and retrieve both flare temperature and radiant heat output in megawatts. Published studies using this method have estimated associated gas flaring volumes at country and facility level, and the results have been used by the World Bank's Global Gas Flaring Reduction Partnership to cross-check national reporting.
From radiant power to operating rate: the inference chain
Flare radiative power does not map directly to crude throughput. The relationship is indirect and requires calibration against known operational states. When a refinery is running at full capacity, its flaring reflects steady-state gas management. When throughput drops, associated gas volumes fall and flare radiance typically decreases. When a unit shuts for unplanned reasons, flaring may spike briefly as pressure is relieved before dropping to near zero. The analyst must read the temporal signature, not just the instantaneous value.
Cooling tower thermal plumes are a more stable proxy. A tower serving an active distillation unit maintains a consistent thermal footprint in Landsat TIRS brightness temperature data. Comparing a facility's thermal signature across multiple Landsat overpasses against a multi-year baseline reveals whether it is operating near historical norms or significantly below them. The 8-day single-satellite revisit limits responsiveness: a two-week shutdown might be captured by only one or two cloud-free acquisitions.
VIIRS, with daily coverage, fills the temporal gap but at the cost of spatial detail. The practical workflow combines VIIRS for daily flare-event detection and Landsat for quantitative thermal mapping of the facility footprint. Sentinel-3 SLSTR adds a second daily thermal observation for large complexes where its 1 km pixel is not a binding constraint.
The maintenance flaring problem
The most persistent source of analytical error is the confusion between planned maintenance and genuine throughput reduction. A refinery undergoing a scheduled turnaround will show dramatically reduced thermal signatures for weeks, which looks identical from orbit to an unplanned shutdown or a sanctions-driven curtailment. Conversely, a facility can maintain high flare radiance during a partial shutdown if it is burning off stored gas rather than processing new crude.
Disentangling these requires context. Published maintenance schedules, when available, can be cross-referenced with the satellite record. AIS-derived crude tanker arrival data (covered separately in the crude tanker cargo estimation page) provides an independent check: if crude deliveries to a facility's marine terminal have stopped, the satellite thermal evidence of reduced activity becomes more credible. The combination is more informative than either dataset alone, but it still cannot definitively attribute the cause of a shutdown without corroborating ground intelligence.
Resolution limits and the small-refinery problem
Large export refineries, those processing 200,000 barrels per day or more, have footprints of several square kilometres and multiple independently detectable heat sources. For these facilities, Landsat TIRS at 100 m native resolution can distinguish cooling tower clusters from distillation areas, and VIIRS can reliably detect flare events. The inference chain is relatively well-constrained.
Smaller refineries, particularly the simple topping plants common in parts of Africa, Central Asia and Southeast Asia, may have footprints of a few hundred metres. At this scale, a single Landsat TIRS pixel covers a substantial fraction of the facility, and unit-level attribution is not possible. VIIRS is essentially useless for spatial attribution at these sites. Planet SuperDove can confirm whether the facility appears active in visible wavelengths, but without a thermal band it cannot quantify heat output. For small refineries, the honest answer is that satellite thermal methods provide a binary active/inactive indicator rather than a throughput estimate.
Cloud, latency and what the archive cannot see
Thermal infrared penetrates haze and thin smoke but not cloud. Refineries in persistently cloudy regions, the Gulf of Guinea coast, parts of Indonesia, the US Gulf Coast during hurricane season, face significant data gaps. VIIRS and Sentinel-3, with daily revisit, reduce the probability of a multi-week gap compared with Landsat alone, but a sustained cloud event can still obscure a facility for ten or more consecutive days.
Latency depends on sensor and processing pipeline. VIIRS Nightfire products from NOAA and the Colorado School of Mines are typically available within 24 hours of acquisition. Landsat Collection 2 Level-2 products are generally available within 24 hours of downlink. Sentinel-3 SLSTR Level-2 land surface temperature products are distributed through the Copernicus Data Space Ecosystem, usually within a few hours. For financial intelligence applications requiring near-real-time alerts, VIIRS is the primary instrument; Landsat provides the quantitative baseline.
The Landsat archive back to 1972 (with thermal data from Landsat 4 onwards) and the VIIRS archive from 2012 allow analysts to construct long-run operating baselines for individual facilities. This is genuinely useful for distinguishing a facility that has been structurally curtailed from one that is simply in a seasonal low. Satellize applies this kind of multi-year baseline analysis in its analytics work, including the methods developed for the Kingdom of Tonga crop-estimation programme, where temporal baselining against sparse historical data is similarly central to the inference.
What this method can and cannot tell a financial analyst
Satellite thermal monitoring can answer several questions with reasonable confidence: Is this refinery currently operating? Has its thermal signature changed materially over the past month relative to its historical baseline? Has flaring increased in a pattern consistent with an unplanned event? These are useful inputs for commodity traders, credit analysts assessing refinery-backed debt, and insurers pricing business interruption cover.
It cannot reliably answer: What is the precise throughput in barrels per day? Is the reduction due to feedstock shortage, mechanical failure, labour action or deliberate curtailment? Will the facility return to normal operation within a specific timeframe? The method is a signal, not a measurement. Used alongside tanker tracking, night-light radiance data and published operational reports, it narrows the range of plausible interpretations substantially. Used alone, it supports probability-weighted assessments rather than definitive conclusions. Buyers of this intelligence should price that uncertainty into their decisions.
Typical figures
| Thermal spatial resolution (Landsat TIRS) | 100 m native, resampled to 30 m in distributed products |
| Thermal spatial resolution (VIIRS M-band) | 750 m at nadir |
| Thermal spatial resolution (Sentinel-3 SLSTR) | 1,000 m thermal channels; 500 m SWIR channels |
| Revisit frequency | VIIRS: daily global; Sentinel-3A+B: ~1 day; Landsat 8+9 combined: 8 days per site |
| Spectral bands used | SWIR 1.6 µm and 2.2 µm (flare detection); TIR 10–12 µm (brightness temperature); VIIRS DNB 0.5–0.9 µm (visible flare) |
| Minimum detectable flare (VIIRS Nightfire) | Approximately 1–3 MW radiant heat power under clear-sky conditions (published detection limit varies by study) |
| Archive depth | VIIRS from 2012; Landsat 8 TIRS from 2013; Landsat 9 from 2021; Sentinel-3 SLSTR from 2016 |
| Cloud penetration | None: thermal infrared is blocked by cloud cover |
| Typical data latency | VIIRS Nightfire: ~24 hours; Landsat Level-2: ~24 hours; Sentinel-3 Level-2: a few hours |
| Deliverable formats | GeoTIFF thermal anomaly maps, time-series CSV, GeoJSON facility polygons, PDF intelligence reports |
Analytics Satellize can run
| Facility operating-state classification | Thresholded brightness temperature anomaly against multi-year Landsat TIRS baseline; binary active/inactive flag with confidence score | Weekly GeoJSON layer with per-facility status and confidence band |
| Flare radiative power time series | VIIRS Nightfire multi-band Planck fitting (published Colorado School of Mines / NOAA method) applied to M-band and DNB radiance | Daily CSV time series of radiant heat output in MW per facility, with cloud-gap flagging |
| Throughput deviation alert | Statistical control chart (CUSUM or similar) applied to rolling 30-day VIIRS flare radiance and Landsat thermal anomaly index; alert triggered when deviation exceeds two standard deviations from seasonal baseline | Near-real-time alert (email or API push) with supporting imagery thumbnail and anomaly magnitude |
| Maintenance versus curtailment disambiguation report | Thermal signature pattern matched against published turnaround schedule data and cross-referenced with AIS tanker arrival data at associated terminals | Narrative PDF intelligence report with probability-weighted scenario table |
| Multi-facility portfolio dashboard | Aggregated VIIRS and Landsat thermal indices for a client-specified list of refineries, normalised to each facility's historical operating range | Monthly interactive dashboard (web or GIS) showing portfolio-level operating rate index |
| Long-run structural activity baseline | Landsat TIRS archive analysis from 2013 to present, computing seasonal decomposition and trend for each facility's thermal footprint | Facility-level baseline report in PDF and GeoTIFF, suitable for underwriting or credit due diligence |
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.