Gas compressor station operating state from thermal infrared signatures
Gas compressor stations betray their operating state through waste-heat signatures detectable in longwave infrared imagery. Landsat 8/9 TIRS and ECOSTRESS resolve the thermal anomalies, but spatial resolution and co-located infrastructure demand careful multi-temporal analysis.
Sensors
- Landsat 8/9 TIRS: Two longwave infrared bands (Band 10 at ~10.9 µm, Band 11 at ~12.0 µm) at 100 m native resolution (resampled to 30 m in products). 16-day single-satellite revisit, 8-day combined revisit for Landsat 8 and 9 together. Adequate for large compressor stations; marginal for smaller sites below roughly 2–3 ha of heated surface.
- ECOSTRESS (ISS-mounted): Five LWIR channels spanning 8.0–12.0 µm at approximately 70 m resolution. Non-sun-synchronous ISS orbit provides variable overpass times including night acquisitions, which improve signal-to-noise by eliminating solar heating. Revisit is irregular, typically 3–5 days at mid-latitudes, but not guaranteed.
- Sentinel-3 SLSTR: Dual-view thermal instrument with 1 km resolution in LWIR channels. Too coarse to isolate individual compressor stations but useful for cross-checking anomaly persistence at regional scale and for cloud-gap filling in time series.
- Planet SuperDove (contextual optical): 8-band multispectral at 3–4 m resolution with near-daily revisit. Carries no thermal capability but provides site-layout context: turbine enclosures, exhaust stacks, cooling systems and vehicle presence can be mapped to disambiguate which thermal anomaly belongs to which facility component.
What a gas turbine driver actually radiates
A gas compressor station driven by aeroderivative or industrial gas turbines operates at turbine inlet temperatures of 1000–1300 °C, but what escapes to the atmosphere is far cooler and far more diffuse: exhaust gases typically leave stacks at 400–600 °C, and the station footprint as a whole warms by only a few degrees Celsius above ambient. That modest surface temperature excess is the signal. At 100 m TIRS resolution, a station must present a thermally anomalous area of at least several pixels to register reliably above background noise, which is why only larger stations, typically those with installed driver power above roughly 10–15 MW, are consistently detectable. Smaller metering and regulation stations, which share the same fenced footprint and similar building stock, rarely generate enough waste heat to distinguish themselves.
The physics that make detection possible are the same that impose its limits. Longwave infrared (8–14 µm) measures emitted radiance from surfaces, not transmitted heat from underground pipes. The signal is therefore dominated by whatever is hottest at the surface: exhaust stacks, engine casings, compressor cooler banks and warm concrete aprons. Cloud cover blocks LWIR entirely, which is a hard constraint in high-latitude or tropical pipeline corridors. A single overpass tells you the state at one moment; it does not tell you whether the station has been running continuously or just started up an hour before the satellite passed.
The discrimination problem: compressors versus metering buildings
Pipeline operators typically co-locate compressor stations with metering and regulation (M&R) facilities, control buildings and sometimes pig-launching infrastructure. From a 100 m pixel, all of these can appear as a single warm cluster. A metering station without active compression will still show mild thermal elevation from heated control-room buildings, trace-heating on instrumentation, and solar loading on metal roofs. The temperature excess is usually 1–3 K above background; an active compressor station running multiple turbine drivers can show 5–15 K or more depending on ambient conditions, wind speed and the number of units on-line.
Separating these requires two things. First, contextual optical imagery at high resolution to map the site layout and identify the presence of turbine enclosures, exhaust stacks and cooler banks before thermal analysis begins. Planet SuperDove or archived commercial imagery at 0.5–3 m serves this purpose. Second, a temperature-excess threshold calibrated against ambient land-surface temperature derived from the same TIRS acquisition, not a fixed absolute threshold, because a station in Siberia in January and the same station in Texas in August will show very different absolute brightness temperatures even at identical operating states.
Multi-temporal stacking to read operational cycling
A single thermal image is ambiguous. A time series is not. Compressor stations on transmission pipelines cycle with gas demand: they run hard in winter heating seasons, throttle back in summer, and may shut individual units for maintenance. Permanent shutdown following decommissioning produces a characteristic pattern where the thermal anomaly collapses to background and stays there across multiple seasons, including winter peaks when a live station would be expected to show elevated output.
Building a stack of 20–40 Landsat acquisitions over one to two years, after filtering for cloud cover and applying land-surface temperature normalisation, produces a seasonal thermal profile for each station. Active stations show a repeating annual cycle correlated with heating-degree days or regional demand signals. Stations with partial unit outages show reduced anomaly magnitude without full suppression. Permanently shut stations show a flat, background-level signature. The Landsat archive extends back to 1982 for TM/ETM+ thermal bands and to 2013 for TIRS, giving substantial historical depth for baseline construction. ECOSTRESS adds night-time passes that eliminate solar contamination and can sharpen the detection of low-level anomalies that daytime acquisitions obscure.
One honest caveat: if a station is shut down for a scheduled maintenance window of two to four weeks and the available cloud-free acquisitions happen to fall within that window, the time series will undercount operating days. Revisit frequency is the binding constraint, not analytical sophistication.
What the archive cannot resolve
Resolution is the dominant limitation and it is worth being direct about it. At 100 m TIRS pixels, a station occupying a 200 m × 150 m fenced area is represented by roughly three to six pixels. If the turbine hall and the cooler bank are at opposite ends of the site, they may fall in different pixels with different temperature excesses. Sub-pixel mixing with surrounding cool terrain reduces the apparent anomaly, particularly for stations in forested or wet environments where vegetation keeps background temperatures low and variable. ECOSTRESS at 70 m offers modest improvement but is not a substitute for a dedicated airborne thermal campaign when precise per-unit operating state is needed.
Sentinel-3 SLSTR at 1 km is essentially useless for individual station detection; it is included in the sensor set only because its regional coverage and twice-daily revisit make it useful for flagging gross anomalies across a whole pipeline corridor before targeting finer sensors. Anyone expecting pixel-level attribution of which compressor unit is running should commission airborne LWIR at 1–5 m resolution, not satellite data.
Practical applications for pipeline operators and commodity analysts
For a pipeline operator, the primary value is independent verification of throughput capacity along a corridor. If multiple compressor stations on a trunk line show elevated thermal signatures simultaneously, the line is likely running near capacity. If stations are thermally quiet in a period of reported high demand, that is a signal worth investigating. Commodity analysts use the same logic in reverse: compressor station activity along export corridors can serve as a leading indicator of gas flows to liquefaction terminals or border interconnects, ahead of official flow data which often lags by days or weeks.
Sanctions monitoring is a documented application of thermal remote sensing to energy infrastructure. Tracking whether compressor stations on specific pipeline systems remain operational under export restrictions requires exactly the kind of systematic multi-temporal thermal analysis described here. The method does not require any cooperation from the operator and is not affected by reported figures.
Satellize runs this class of analysis on open Landsat and ECOSTRESS archives, with commercial tasking added where revisit gaps require it. The analytical approach is the same stack-and-normalise method applied in the Tonga crop-estimation programme, adapted from agricultural to industrial thermal signatures.
Typical figures
| Native spatial resolution (TIRS) | 100 m (Band 10 and Band 11); resampled to 30 m in standard products |
| Native spatial resolution (ECOSTRESS) | ~70 m across five LWIR channels |
| Revisit cadence | 8 days combined (Landsat 8 + 9); irregular 3–5 days typical for ECOSTRESS at mid-latitudes |
| Spectral bands used | LWIR 10.6–11.2 µm and 11.5–12.5 µm (TIRS); 8.0–12.0 µm (ECOSTRESS) |
| Minimum detectable thermal excess | Approximately 2–3 K above ambient for a station filling multiple 100 m pixels; smaller anomalies require ECOSTRESS night passes |
| Minimum reliably detectable station size | Stations with driver power broadly above 10–15 MW and footprint above ~2–3 ha; smaller sites are marginal |
| Cloud cover constraint | LWIR blocked by cloud; effective revisit in cloudy climates may fall to once per month or fewer |
| Archive depth | Landsat TIRS from 2013 (Landsat 8 launch); ECOSTRESS from 2018; Landsat TM/ETM+ thermal from 1982 (coarser, 120 m / 60 m) |
| Contextual optical resolution | 3–4 m (Planet SuperDove) for site-layout disambiguation |
| Data latency (open archive) | Landsat standard products typically available within 24 hours of acquisition via USGS EarthExplorer |
Analytics Satellize can run
| Station operating-state classification (active / partial / idle / shutdown) | Land-surface temperature anomaly detection against multi-scene ambient baseline; threshold calibrated per acquisition | Monthly GIS layer with per-station state label and confidence score |
| Seasonal thermal profile per station | Multi-temporal TIRS stack normalised for ambient LST; time-series decomposition to extract trend and seasonal components | Per-station CSV time series and annotated chart; 12-month rolling update |
| Corridor-level capacity utilisation index | Aggregation of per-station anomaly magnitude along a defined pipeline route; weighted by installed driver capacity where public engineering data exists | Weekly index value per pipeline corridor, delivered as structured data feed |
| Permanent shutdown detection alert | Change-point detection on thermal time series; flag triggered when anomaly magnitude falls to background across three or more consecutive cloud-free acquisitions spanning at least 60 days | Alert report with acquisition dates, pre/post thermal imagery thumbnails and confidence assessment |
| Site-layout disambiguation map | High-resolution optical classification (Planet SuperDove or equivalent) to identify turbine enclosures, exhaust stacks and cooler banks; registered to TIRS pixel grid | GIS polygon layer with facility-component labels, used as input mask for thermal analysis |
| Historical baseline reconstruction | Full Landsat TIRS archive processing from 2013 to present; cloud-masked stack with per-pixel median and percentile surfaces | Station-level historical operating-state database; one-off delivery with annual refresh option |
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.