LNG storage tank status from thermal anomaly patterns
Above-ground LNG tanks leave measurable cold signatures in thermal infrared imagery, and their boil-off flares leave bright SWIR traces at night. Together, these signals let analysts infer fill status and operational tempo without setting foot on a terminal.
Sensors
- ECOSTRESS (ISS-mounted): Thermal infrared radiometer at roughly 70 m ground sampling distance, five TIR bands centred near 8–12 µm. Provides skin-temperature retrievals with a stated precision of approximately 0.5 K, which is sufficient to resolve the 2–6 K surface depression that a full cryogenic tank can impart relative to an empty neighbour. Irregular revisit (ISS precession gives variable overpass times, typically several passes per week at mid-latitudes) is a genuine limitation for daily monitoring.
- Landsat 9 TIRS: Two thermal bands (Band 10 at 10.9 µm, Band 11 at 12.0 µm) at 100 m native resolution, resampled to 30 m in the standard product. Sixteen-day repeat at any given path/row, though adjacent-path overlap improves effective revisit to roughly eight days at mid-latitudes. Noise-equivalent temperature difference (NEDT) is approximately 0.1 K for Band 10, making it the most precise free thermal archive available. The 100 m native footprint means small tanks (diameter under 60 m) may straddle pixel boundaries and produce mixed-pixel artefacts.
- Sentinel-2 MSI (SWIR bands): Bands 11 (1.61 µm) and 12 (2.19 µm) at 20 m resolution, 10-day revisit with a single satellite and roughly five days with both Sentinel-2A and 2B in constellation. SWIR bands are sensitive to high-temperature combustion events such as flare stacks: a flare burning at several hundred degrees Celsius saturates these bands even at small spatial extent, producing a detectable signal well below the 20 m pixel size. Cloud cover is the primary operational constraint.
- VIIRS Day/Night Band (DNB) and VIIRS I4/I5: DNB at approximately 750 m resolution detects nocturnal radiance from flares and terminal lighting. VIIRS I4 (3.74 µm) and I5 (11.45 µm) bands at 375 m provide twice-daily thermal coverage globally, useful for detecting large thermal anomalies and persistent flaring. Resolution is too coarse to distinguish individual tanks at most terminals, but the twice-daily cadence makes VIIRS the best available system for flagging sudden changes in flare activity.
What a cold tank actually looks like from orbit
A full LNG tank holds liquid methane at approximately minus 162 degrees Celsius. The tank wall and roof are heavily insulated, but no insulation is perfect. Heat flow from the environment into the tank is continuous and unavoidable; the cryogenic contents act as a thermal sink that suppresses the apparent surface temperature of the tank roof relative to an identical but empty tank warmed only by ambient air and solar gain.
The magnitude of the contrast depends on insulation quality, ambient temperature, wind speed and how recently the tank was filled or emptied. Published studies using Landsat TIRS data at major LNG terminals in Qatar and Australia have reported roof temperature differences of two to six Kelvin between full and near-empty tanks under favourable night-time conditions. Daytime acquisitions are noisier because solar heating of the roof partially masks the cryogenic signal; night-time passes are strongly preferred. ECOSTRESS, with its finer 70 m sampling, can resolve this contrast even at terminals where tanks are clustered closely together, though its irregular ISS-driven revisit schedule means you cannot guarantee a useful acquisition on any given night.
Boil-off flaring as a corroborating signal
Boil-off gas (BOG) is the vapour that forms continuously inside a full or partially full LNG tank as residual heat infiltrates the insulation. Terminals manage BOG by re-liquefaction, compression for send-out, or controlled flaring. Flaring intensity is not a direct proxy for fill level, but it correlates with operational events: ship loading, tank topping-up, compressor trips and send-out surges all alter BOG volumes and therefore flare behaviour.
Sentinel-2 SWIR bands detect active flares reliably because combustion at several hundred degrees Celsius produces strong mid-infrared emission that saturates Band 12 pixels even when the physical flare is far smaller than the 20 m pixel footprint. The detection is binary rather than quantitative at this resolution: you can confirm a flare is burning, and you can track whether it was burning across consecutive acquisitions, but you cannot derive a precise flare volume from Sentinel-2 SWIR alone. VIIRS I4 at 375 m offers twice-daily coverage and is more useful for continuous flare-state monitoring, while VIIRS Nightfire (a published algorithm from the Colorado School of Mines applied to VIIRS data) provides temperature and radiant power estimates for persistent flares that have been validated against ground measurements at oil and gas facilities.
Combining signals to infer fill status
No single sensor gives a definitive fill-level reading. The operational approach is triangulation across three independent evidence streams: the thermal depression signal from TIRS or ECOSTRESS, the flare-state record from SWIR and VIIRS, and vessel activity at the terminal berths (covered separately in the LNG terminal throughput page). A tank that shows a strong cold anomaly and coincides with a recent ship arrival is almost certainly full or filling. A tank that has warmed to ambient temperature over several consecutive acquisitions, with no vessel calls and no elevated flaring, is likely empty or in maintenance.
The honest limit of this approach is that thermal signals are probabilistic, not metrological. Insulation condition varies between tanks and degrades over time. A poorly insulated empty tank can occasionally appear cooler than a well-insulated full one. Analysts should treat the thermal signal as a relative indicator within a terminal, comparing tanks against each other and against their own historical baseline, rather than applying a universal temperature threshold. An archive of at least twelve months of Landsat TIRS data per terminal is generally needed to establish a reliable per-tank baseline.
Cloud cover and revisit: the real operational constraints
Thermal infrared is not immune to cloud. Optically thick cloud completely blocks the surface signal in all TIR bands. At terminals in persistently cloudy regions, such as parts of Southeast Asia or the US Gulf Coast in summer, usable clear-sky acquisitions may occur only once every two to four weeks even with Landsat and ECOSTRESS combined. This is a genuine limitation that any honest monitoring programme must account for.
The mitigation is multi-sensor fusion. Landsat 9 and Landsat 8 together provide an eight-day revisit on the same path; adding ECOSTRESS passes and any available commercial thermal data (for clients with the licence) increases the probability of a cloud-free acquisition in any given fortnight. VIIRS, with its coarser resolution but twice-daily cadence, provides a useful continuity layer for flare monitoring even when optical and thermal imaging is blocked. SAR coherence, covered in the pipeline construction page, does not contribute directly to fill-level estimation but can confirm whether a terminal is physically active.
What the analytics actually deliver
A well-constructed monitoring programme produces a per-tank status score updated on each cloud-free acquisition: a normalised temperature anomaly relative to the terminal's own historical baseline, a flare-state flag from SWIR, and a confidence rating that reflects cloud cover, acquisition geometry and the number of independent observations in the preceding window. Delivered as a time-series feed or a GIS layer, this lets a commodity analyst or terminal operator track inventory trends over weeks and months without needing access to proprietary metering data.
Satellize runs this kind of multi-sensor fusion on open constellations for clients who need independent verification of publicly reported storage levels. The Tonga crop-estimation programme demonstrated the same underlying principle at a different scale: extracting quantitative physical inferences from open satellite data where no ground truth is freely available. For LNG terminals, the analytic output is not a certified inventory figure. It is an independent, physics-based signal that complements, challenges or confirms what operators and traders report through other channels.
Typical figures
| Thermal spatial resolution | ECOSTRESS ~70 m; Landsat 9 TIRS 100 m native (30 m resampled); VIIRS I5 375 m |
| Temperature sensitivity (NEDT) | Landsat 9 TIRS Band 10: ~0.1 K; ECOSTRESS: ~0.5 K stated precision |
| Revisit (thermal) | Landsat 8+9 combined: ~8 days at mid-latitudes; ECOSTRESS: variable, typically several passes per week; VIIRS: twice daily |
| SWIR flare detection resolution | Sentinel-2 Band 12 at 20 m; sub-pixel flares detectable when combustion temperature exceeds ~500 °C |
| Revisit (SWIR) | Sentinel-2A+2B combined: ~5 days at most latitudes |
| Minimum detectable thermal contrast | ~2 K difference between full and empty tank roofs under clear-sky night-time conditions (published literature range: 2–6 K) |
| Cloud impact | Optically thick cloud fully blocks TIR and SWIR; clear-sky frequency at cloudy terminals may fall to one usable pass per 2–4 weeks |
| Archive depth | Landsat: back to 1972 (TIR from Landsat 4, 1982); Sentinel-2: 2015–present; ECOSTRESS: 2018–present; VIIRS: 2012–present |
| Spectral bands used | TIR: 10.9 µm, 12.0 µm (Landsat); 8–12 µm (ECOSTRESS). SWIR: 1.61 µm, 2.19 µm (Sentinel-2). DNB: 0.5–0.9 µm panchromatic (VIIRS) |
| Delivery formats | GeoTIFF thermal anomaly layers, CSV time-series per tank, GIS-compatible vector status layer, PDF monitoring report |
Analytics Satellize can run
| Per-tank thermal anomaly score | Land surface temperature retrieval (split-window or single-channel algorithm) applied to Landsat TIRS and ECOSTRESS; per-tank baseline normalisation using 12-month rolling archive | GeoTIFF layer and CSV time-series updated on each cloud-free acquisition, with confidence flag |
| Flare-state detection and history | SWIR band thresholding on Sentinel-2 Band 12; VIIRS Nightfire radiant power estimates for persistent flares | Binary flare-on/off log per stack per acquisition, exportable as CSV or GIS point layer with radiance values |
| Tank fill-status classification | Fusion of thermal anomaly score and flare-state flag; Bayesian update against vessel-arrival priors where AIS data is available | Weekly status report classifying each monitored tank as likely full, partially filled, likely empty or unknown (cloud-obscured) |
| Terminal-level inventory trend index | Aggregation of per-tank scores into a normalised terminal index; trend decomposition to separate seasonal thermal effects from operational changes | Monthly trend chart and index value, suitable for commodity desk integration |
| Anomaly alert on sudden thermal or flare change | Change detection against rolling 30-day baseline; alert triggered when anomaly exceeds two standard deviations from baseline or flare state changes on consecutive passes | Email or API alert within 24 hours of a qualifying acquisition being processed |
| Cloud-gap analysis and monitoring confidence report | Per-terminal cloud-cover frequency statistics derived from Sentinel-2 scene classification layer; used to characterise expected data gaps | Quarterly monitoring-confidence summary showing percentage of days with usable thermal coverage per terminal |
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.