Sensors
- Landsat 8 TIRS: Two thermal infrared bands (Band 10 at 10.6–11.19 µm, Band 11 at 11.50–12.51 µm), 100-metre native resolution resampled to 30 m in data products, 16-day exact repeat. Primary workhorse for train-level cold anomaly detection at large, physically separated facilities.
- Landsat 9 TIRS-2: Identical band configuration to Landsat 8 TIRS with improved radiometric calibration and lower noise-equivalent temperature difference (NEdT ~0.2 K). Operates in the same WRS-2 orbit offset by 8 days, giving a combined 8-day revisit with Landsat 8.
- ECOSTRESS: ISS-mounted thermal radiometer at approximately 70 × 38 m pixel size across five TIR bands (8–12.5 µm). Non-sun-synchronous orbit yields irregular overpass times including night acquisitions, which suppress solar heating artefacts and improve cold-anomaly contrast. Revisit is variable, typically 1–5 days at mid-latitudes.
- Sentinel-3 SLSTR: Dual-view thermal scanner at 1-km resolution in TIR channels. Too coarse to resolve individual trains but useful for facility-level temperature anomaly trending and cross-calibration of Landsat retrievals over time.
What a cold surface gives away
An LNG liquefaction train does not radiate heat in the way a refinery furnace or a power-plant condenser does. It does the opposite. The refrigeration cycles used in propane pre-cooling and mixed-refrigerant main cryogenic stages pull heat out of the process stream and, in doing so, chill the surrounding steel structures, pipe racks, cold boxes and heat exchangers to well below ambient. Published land-surface temperature retrievals from Landsat TIRS routinely resolve temperature contrasts of 2–4 K against background ground at the 100-metre pixel scale. An operating train at a facility like those in Qatar, Australia or the United States Gulf Coast presents a spatially coherent cold patch that is physically distinct from a warm standby train sitting idle in ambient air.
The physical mechanism is straightforward. Cryogenic insulation is imperfect. Piping and structural steel in contact with process streams at temperatures as low as minus 162 °C conduct cold outward. On a clear night, when solar heating is absent and the atmospheric correction problem is at its simplest, this cold signature is most legible. Daytime acquisitions are usable but require careful removal of differential solar heating across the facility's varied surface materials.
Why 100 metres is both the method's power and its ceiling
Landsat TIRS pixels are 100 metres on a side before resampling. That is a genuine constraint. At a facility like Sabine Pass or Gorgon, individual trains are physically separated by 150–400 metres, making them resolvable as distinct thermal objects. At a compact facility where trains share a common pipe rack and are separated by 40–60 metres, the cold signals blend into a single mixed pixel and train-level discrimination collapses. Buyers should assess facility layout geometry before commissioning this analysis.
ECOSTRESS partially relaxes this constraint. Its approximately 70 × 38 m ground sampling distance can separate trains that Landsat cannot, and its variable overpass timing includes pre-dawn passes where the thermal scene is cleanest. The trade-off is that ECOSTRESS coverage is not guaranteed on any given date; the ISS orbit precesses and does not follow a fixed repeat ground track in the way Landsat does. For consistent time-series monitoring, Landsat remains the backbone. ECOSTRESS serves best as a high-resolution confirmation layer when a Landsat anomaly needs disambiguation.
Building a status time series: method in practice
The analytical workflow begins with surface temperature retrieval. For Landsat TIRS, the standard USGS Collection 2 Level-2 product delivers surface temperature in Kelvin with an absolute accuracy of approximately ±2 K under clear-sky conditions, using the ASTER GED emissivity database for surface emissivity correction. Scenes with cloud cover over the facility are discarded; at most tropical and subtropical LNG export locations, clear-sky acquisition frequency on a 16-day Landsat cycle is adequate for monthly status assessment, though not for day-to-day tracking.
Once a clean temperature image is in hand, each train's footprint is defined using a facility layout derived from high-resolution optical imagery. Mean and minimum pixel temperatures within each footprint are extracted. The key metric is the temperature anomaly relative to a stable background reference, typically the facility's paved apron or a nearby area of similar surface type. A train showing a persistent cold anomaly of more than approximately 2 K below background across multiple acquisitions is flagged as operating. A train whose temperature converges toward background is flagged as standby or in maintenance.
Temporal consistency matters more than any single scene. A maintenance outage typically produces a warming trend over several days as cryogenic inventory dissipates and the cold box warms toward ambient. Restart produces the reverse: a gradual cooling signature over the first hours to days of operation. Landsat's 8-day combined revisit can capture this arc across two or three scenes if cloud allows.
Honest limits: what the method cannot tell you
Thermal infrared cannot see through cloud. At facilities in tropical coastal locations, cloud cover can suppress clear-sky acquisitions for weeks during monsoon periods. Sentinel-3 SLSTR at 1-km resolution can maintain some facility-level signal through broken cloud conditions, but individual train discrimination at that resolution is not possible.
The method also cannot directly measure throughput or feed-gas volume. A train that is operating at 60 percent capacity and one running at full nameplate capacity will both show cold anomalies; the temperature difference between those two states is likely within the retrieval uncertainty of ±2 K. What the method reliably delivers is binary status, operating versus not operating, and the approximate timing of transitions. Quantitative throughput inference requires combining this thermal signal with complementary data, such as vessel loading activity at the export jetty or flare stack combustion state during start-up and shutdown events.
Finally, planned maintenance turnarounds and unplanned outages produce identical thermal signatures. Context from shipping schedules, operator announcements or gas flow data is needed to distinguish the two.
Archive depth and what history reveals
Landsat 8 has been collecting TIRS data since April 2013. Landsat 9 joined in February 2022. That combined archive spans more than a decade of global thermal coverage, including the commissioning periods of major LNG export facilities built during the 2010s expansion wave. For a buyer assessing a counterparty's historical production reliability, this archive is genuinely useful: it is possible to reconstruct which trains were operating, which were in maintenance, and roughly when outages began and ended, going back to 2013.
Satellize runs this type of retrospective analysis as part of its broader satellite-data analytics work. The Tonga crop-estimation programme sits at the opposite end of the resolution and subject-matter spectrum, but the underlying discipline of extracting quantitative signals from open-constellation data is the same. For LNG applications, the analytics output is typically a per-train status table delivered as a structured data feed or GIS layer, updated on each clear-sky Landsat overpass.
Facility types where this works and where it does not
The method is well suited to large baseload export terminals with multiple physically separated trains: facilities in Australia (Gorgon, Wheatstone, Curtis Island), the United States Gulf Coast (Sabine Pass, Corpus Christi, Freeport), Qatar's North Field expansion trains, and similar large-footprint sites. Train separations at these facilities are generally sufficient for Landsat-scale discrimination.
It is poorly suited to floating LNG vessels, where the entire liquefaction plant is compressed onto a single hull and train footprints are not individually resolvable at any currently operational free satellite thermal resolution. It is also less useful at small-scale or peak-shaving facilities where the cold-surface area is below two or three Landsat pixels. For those cases, commercial thermal satellites with sub-30-metre resolution, where available under tasking arrangements, offer a partial solution, though at higher cost and lower revisit frequency than the open Landsat stack.
Typical figures
| Thermal spatial resolution | 100 m native (Landsat 8/9 TIRS); ~70 × 38 m (ECOSTRESS); 1 km (Sentinel-3 SLSTR) |
| Revisit frequency | 8 days combined (Landsat 8 + 9); 1–5 days variable (ECOSTRESS, ISS-dependent); ~1 day (Sentinel-3 SLSTR, facility-level only) |
| Thermal bands used | Landsat Band 10 (10.6–11.19 µm), Band 11 (11.50–12.51 µm); ECOSTRESS Bands 4–5 (10.26–11.32 µm) |
| Surface temperature accuracy | ±2 K absolute under clear-sky conditions (Landsat Collection 2 Level-2); ECOSTRESS target ±1.5 K |
| Minimum detectable temperature anomaly | ~2 K below background reference (practical threshold at Landsat resolution; smaller anomalies within retrieval uncertainty) |
| Cloud constraint | Clear-sky required for valid retrievals; cloud-flagged scenes discarded |
| Archive depth | Landsat 8 TIRS from April 2013; Landsat 9 TIRS-2 from February 2022 |
| Coverage | Global land surface; all major LNG export terminals covered by Landsat WRS-2 path/row grid |
| Delivery latency | Landsat Collection 2 Level-2 products typically available within 12–24 hours of acquisition |
| Minimum resolvable facility footprint | Train-to-train separation of at least ~150 m required for Landsat discrimination; ~80 m for ECOSTRESS |
Analytics Satellize can run
| Per-train binary operating status | Land-surface temperature retrieval (Landsat Collection 2 Level-2) with cold-anomaly thresholding against stable background reference pixels | Structured data feed (CSV or JSON) updated per clear-sky overpass, one row per train per scene |
| Maintenance outage detection and duration estimate | Temporal change detection on per-train temperature anomaly time series; warming trend onset marks outage start, cooling trend marks restart | Outage event log with estimated start and end dates, delivered as a GIS layer or tabular report |
| Facility-level aggregate operating capacity | Sum of active train count across a facility, expressed as fraction of nameplate train complement, derived from per-train status flags | Monthly capacity-utilisation index per facility, suitable for commodity market analysis |
| ECOSTRESS confirmation layer for ambiguous Landsat scenes | Co-registration and comparison of ECOSTRESS TIR retrievals against Landsat anomaly flags where train separation is near the Landsat resolution limit | Annotated scene report flagging confidence level (high/medium/low) for each train status call |
| Historical operating status reconstruction | Retrospective batch processing of Landsat 8/9 TIRS archive from 2013 to present using consistent retrieval and anomaly-detection pipeline | Multi-year per-train status time series, delivered as GIS-ready raster stack and summary table |
| Start-up and shutdown event characterisation | Rate-of-change analysis on temperature anomaly time series to identify warming and cooling arcs consistent with train state transitions | Event timeline report with approximate transition duration and confidence interval |
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.