Ethylene cracker furnace operating status from thermal infrared signatures
Steam cracker furnaces produce thermal anomalies strong enough to saturate standard TIR detectors. Landsat TIRS, ECOSTRESS and Sentinel-3 SLSTR let analysts track operating state, decoking cycles and unplanned shutdowns from orbit.
Sensors
- Landsat 9 TIRS: Two thermal bands centred at 10.9 µm and 12.0 µm, 100-metre native resolution (resampled to 30 m in product). 16-day single-satellite revisit, extendable to roughly 8 days when combined with Landsat 8. Radiance saturation can occur directly over the hottest furnace pixels; surrounding pixels still carry usable brightness-temperature gradients. Archive extends to 2013 for Landsat 8 TIRS.
- ECOSTRESS (ISS-mounted): Five TIR bands from 8.3 to 12.5 µm at approximately 70-metre ground sampling distance. Irregular revisit driven by ISS orbital precession, roughly 1 to 5 days at mid-latitudes. Finer spatial detail than Landsat TIRS and multi-band capability supports emissivity separation, which matters when distinguishing furnace casing temperature from process-gas flue emissions.
- Sentinel-3 SLSTR: Dual-view TIR at 1-km resolution (nadir and oblique), daily global revisit when both Sentinel-3A and 3B are active. Too coarse to resolve individual furnace clusters in a dense complex, but useful for detecting facility-level thermal anomaly persistence and for flagging acquisition windows for higher-resolution sensors.
- Sentinel-5P TROPOMI: Measures tropospheric columns of ethylene precursors, principally formaldehyde (HCHO) and acetaldehyde, at 3.5 × 5.5 km pixel resolution (post-2019 upgrade). Direct ethylene detection is below TROPOMI sensitivity, but elevated HCHO columns downwind of a complex correlate with high cracker throughput. Revisit is daily.
Why a cracker furnace is visible from 700 kilometres
A steam cracker furnace operates its radiant coils at tube-wall temperatures between 850 °C and 1,050 °C. The furnace casing itself, though insulated, radiates at temperatures well above ambient, and the flue-gas stack exit can exceed 200 °C. At 10.9 µm, the Planck function is strongly nonlinear: a surface at 120 °C emits roughly six times the spectral radiance of a surface at 20 °C. A cluster of four to eight furnaces, each with a footprint of roughly 15 × 60 metres, creates a thermal anomaly that is easily distinguished from background even at 100-metre pixel resolution, provided the overpass is cloud-free.
The practical limit is saturation. Landsat TIRS Band 10 saturates at a brightness temperature of approximately 380 K (107 °C). Pixels directly over active furnace casings frequently saturate, which sounds like a problem but is itself diagnostic: a pixel that saturates on one date and reads 310 K on the next has almost certainly dropped to partial or zero load. Analysts therefore use the spatial extent of saturated or near-saturated pixels, not raw radiance values, as the primary operating-state indicator.
Reading a decoking cycle in a time series
Cracker furnaces accumulate coke deposits on tube walls over run lengths that typically range from 30 to 90 days, depending on feedstock and severity. Decoking involves taking individual furnaces offline, passing steam and air through the coils to burn off coke, then returning the furnace to service. The process takes 24 to 48 hours per furnace. During decoking, the affected furnace cools detectably while adjacent furnaces may be pushed harder to maintain plant throughput.
In a Landsat or ECOSTRESS time series, this appears as a rotating pattern of reduced thermal intensity across individual furnace positions within a cluster. The 8- to 16-day revisit of Landsat means a single decoking event may be missed entirely if it falls between overpasses. ECOSTRESS's irregular but sometimes daily revisit improves detection probability. The analytic value is not in catching every individual decoking event but in tracking the aggregate cycle frequency, which is a proxy for run-length and, indirectly, for feedstock quality and cracking severity.
Unplanned shutdowns look different from scheduled ones
A planned turnaround at a large ethylene plant typically takes two to six weeks and affects the entire furnace cluster simultaneously. The thermal signature collapses across all positions in a single revisit interval. An unplanned shutdown driven by a tube rupture, compressor trip or utility failure tends to produce a faster and less orderly thermal decay, sometimes with an initial spike if emergency flaring accompanies the trip. Sentinel-3 SLSTR, despite its coarse 1-km resolution, can detect the facility-level thermal collapse within the same day if the overpass timing is favourable.
Cross-referencing the thermal signal with Sentinel-5P TROPOMI formaldehyde columns adds a second line of evidence. HCHO columns over a major ethylene complex during full operation are measurably elevated relative to regional background, particularly in low-wind conditions. A sustained drop in HCHO column density downwind of a complex, coinciding with a thermal anomaly reduction, strengthens the case for a genuine production outage rather than a sensor artefact or cloud contamination.
Furnace clusters versus other fired heaters: the disambiguation problem
A large petrochemical complex contains many sources of elevated surface temperature: crude distillation furnaces, hydrotreater fired heaters, reformer furnaces, steam superheaters and flare stacks. All appear warm in TIR imagery. Cracker furnace clusters are distinguished by their characteristic rectangular geometry, their grouping in parallel rows of four to twelve units, and their position within the plant relative to the cold-box and compression sections visible in optical imagery.
Spatial co-registration of a high-resolution optical base layer (Sentinel-2 at 10 metres, or commercial imagery at finer resolution) with the TIR data is essential. The optical layer identifies unit type by shape, shadow and equipment layout. The TIR layer then assigns thermal intensity to confirmed unit locations. Without this fusion step, a hot reformer furnace and a cracker furnace are indistinguishable in the thermal image alone. ECOSTRESS's 70-metre resolution helps more than Landsat's 100 metres in dense complexes where furnace clusters are separated by as little as 50 to 80 metres.
Cloud cover is the most persistent operational limit. Tropical and coastal sites, where many major ethylene complexes are located, can experience cloud fractions above 70 percent in monsoon months. Building a reliable time series requires accepting that some revisit windows will be unusable and using interpolation or ensemble methods across the available cloud-free acquisitions.
What the signal can and cannot tell a commodity analyst
A confirmed operating-state classification, updated at each cloud-free overpass, tells an analyst whether a plant's cracker section is running, partially curtailed or fully offline. It does not directly measure ethylene output. Cracker yield depends on feedstock composition, severity settings and furnace efficiency, none of which are observable from orbit. The thermal signal is a necessary but not sufficient proxy for production volume.
The signal is most useful for detecting step changes: a plant that was running at apparent full capacity last month and shows a collapsed thermal signature this month has almost certainly reduced output materially. It is less useful for distinguishing, say, 85 percent utilisation from 95 percent. For that kind of granularity, the thermal method needs to be combined with optical activity proxies, vessel traffic at associated terminals, and where available, stack emissions data from Sentinel-5P.
Satellize runs this combined thermal and precursor-column workflow as part of its energy-analytics service. The Tonga crop-estimation programme uses a structurally similar multi-sensor fusion approach, where no single sensor is sufficient and the value comes from disciplined combination of independent observations.
Typical figures
| Primary TIR spatial resolution | 70 m (ECOSTRESS), 100 m native / 30 m resampled (Landsat 9 TIRS), 1 km (Sentinel-3 SLSTR) |
| Revisit at mid-latitudes | 8 days (Landsat 8+9 combined), 1–5 days irregular (ECOSTRESS), daily (Sentinel-3 A+B) |
| Thermal bands used | 10.9 µm and 12.0 µm (Landsat TIRS); 8.3, 8.8, 9.1, 10.5, 12.0 µm (ECOSTRESS); 10.85 µm and 12.0 µm (SLSTR) |
| Brightness temperature saturation limit | Approximately 380 K (107 °C) for Landsat TIRS Band 10; saturation extent used as operating-state indicator |
| Minimum detectable thermal anomaly | Facility-level anomalies of 5–10 K above background detectable at 100 m; individual furnace-row discrimination requires ECOSTRESS 70 m or finer |
| Precursor-column sensor | Sentinel-5P TROPOMI HCHO at 3.5 × 5.5 km, daily global revisit |
| Optical co-registration layer | Sentinel-2 MSI at 10 m for unit-type identification; commercial imagery at 0.3–0.5 m for detailed layout mapping |
| Thermal archive depth | Landsat TIRS from 2013 (Landsat 8); ECOSTRESS from 2018; Sentinel-3 SLSTR from 2016 |
| Cloud-cover limitation | Tropical and coastal sites may yield fewer than 4 cloud-free TIR acquisitions per month; time-series gaps require interpolation |
| Delivery latency | Landsat and Sentinel products available within 3–6 hours of overpass via Copernicus and USGS systems; ECOSTRESS via NASA Earthdata typically within 12–24 hours |
Analytics Satellize can run
| Furnace cluster operating-state classification | Brightness-temperature anomaly mapping with saturated-pixel extent analysis, co-registered to optical unit-type layer | Per-overpass GIS layer with operating state (full, partial, offline) per confirmed furnace cluster; updated at each cloud-free acquisition |
| Decoking cycle frequency index | Time-series decomposition of thermal intensity per furnace position; rotating partial-cooling pattern detection | Monthly report showing estimated decoking frequency per cluster, with uncertainty bounds reflecting revisit gaps |
| Unplanned shutdown alert | Threshold-based anomaly detection on facility-level thermal collapse rate; cross-checked against Sentinel-3 same-day overpass | Alert feed (JSON or email) triggered when thermal intensity drops more than two standard deviations below trailing 90-day baseline within a single revisit interval |
| HCHO column correlation with operating state | TROPOMI HCHO column extraction over downwind receptor cells; Pearson correlation with concurrent thermal operating-state score | Scatter plot and correlation coefficient in monthly summary; flags where HCHO and thermal signals diverge, indicating possible instrument or meteorological artefact |
| Multi-plant regional operating-rate index | Aggregation of per-facility operating-state scores across a defined region; weighted by nameplate capacity from public industry sources | Weekly regional ethylene-sector operating-rate index, delivered as time-series CSV with confidence interval |
| Historical baseline and event reconstruction | Back-processing of Landsat 8/9 TIRS archive from 2013 to present; event-detection applied retrospectively | Facility-level operating-state history dataset, suitable for regression against published trade or price data |
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.