Underground mine ventilation shaft thermal signature monitoring
Ventilation shafts betray underground mine activity through thermal infrared anomalies detectable by Landsat-9 TIRS and ECOSTRESS. The contrast is sharpest at high-latitude or high-altitude sites where ambient swings are large, but the method has real resolution and revisit limits buyers should understand before commissioning it.
Sensors
- Landsat-9 TIRS: Two thermal infrared bands centred at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in products), 16-day repeat at the equator and better at high latitudes where orbits converge. The primary workhorse for this use case given its free archive back to 2021 and Landsat-8 continuity to 2013.
- ECOSTRESS (ISS-mounted): Five TIR bands spanning 8.3 to 12.5 µm at approximately 70 m resolution, with irregular revisit driven by ISS orbital precession (typically 1 to 5 days at mid-latitudes). Captures diurnal variation that Landsat cannot, which helps separate geothermal background from ventilation-driven anomalies.
- ASTER TIR: Five bands from 8.125 to 11.65 µm at 90 m resolution, 16-day repeat. The archive extends to 1999, making it the deepest thermal record for mine-state comparison. On-demand tasking ended in 2019 but the historical archive remains accessible via NASA Earthdata.
- Airbus Pléiades Neo (daytime proxy): No dedicated TIR band, but sub-metre VNIR imagery at 30 cm resolution can detect snow-melt halos and condensation plumes around active exhaust shafts as a visible-band proxy for thermal output. Useful for corroborating shaft locations identified in coarser thermal data.
What a ventilation shaft actually does to the ground around it
Underground mines require continuous airflow to dilute blasting fumes, control methane, and keep workers alive. Exhaust shafts carry warm, humid air from depth to surface. In winter at high-latitude or high-altitude sites, that air is substantially warmer than the frozen ground around it. The shaft collar and the immediate surface downwind of the outlet show a persistent thermal anomaly. In summer the relationship inverts: deep rock temperatures are relatively stable (typically between 10°C and 25°C depending on depth and geothermal gradient), so exhaust air can appear cooler than sun-baked surface ground. The inversion is less dramatic and harder to detect, but it is real.
The physics are straightforward. The diagnostic signal is the temperature contrast between the shaft outlet area and a reference patch of undisturbed ground at the same elevation and aspect, sampled in the same image. Absolute temperature matters less than relative anomaly. A shaft pushing 5,000 cubic metres per minute of air at 18°C into a landscape sitting at minus 15°C produces a detectable warm patch of several hundred square metres around the collar, well within the footprint of a single Landsat-9 TIRS pixel at 100 m.
Where the method works and where it fails
The technique is most reliable at mines above roughly 55 degrees latitude or above 2,000 metres elevation, where seasonal ambient temperature swings exceed 30°C and the ventilation contrast is large relative to sensor noise. Landsat-9 TIRS has a noise-equivalent temperature difference (NEDT) of approximately 0.4 K, which sets a practical floor on detectable anomalies. A marginal shaft at a temperate-climate mine in spring or autumn, when ambient and exhaust temperatures are close, may produce a contrast below that threshold.
Cloud cover is a hard blocker. Thermal infrared does not penetrate cloud. High-latitude mines can experience weeks of persistent overcast, and a 16-day Landsat repeat offers no guarantee of a clear acquisition in any given window. ECOSTRESS partially compensates through its irregular revisit, but it is also blocked by cloud. Analysts must work with whatever clear-sky acquisitions the archive provides, which can mean multi-week gaps in the time series.
Spatial resolution is the other constraint. At 100 m native pixel size, TIRS cannot resolve individual shaft collars at a complex multi-shaft mine. It detects the aggregate thermal footprint of an area. If two shafts are 80 metres apart, they appear as one warm blob. Pléiades Neo imagery can pin the collar location to within half a metre, but that is a visible-band proxy, not a direct temperature measurement. The combination of coarse thermal confirmation and high-resolution location is more useful than either alone.
Reading operational status from the thermal time series
A single thermal image tells you whether a shaft was warm or cool at the moment of acquisition. A time series tells you something more interesting: whether the mine is running, idling, or shut down. Active ventilation produces a persistent anomaly across multiple acquisitions in winter. A care-and-maintenance shutdown typically collapses the anomaly within days as the airflow stops and the shaft collar equilibrates to ambient. Seasonal ramp-ups and shutdowns produce characteristic signatures in a multi-year Landsat archive.
Estimating airflow intensity from thermal data is harder. The surface temperature anomaly depends on airflow volume, exhaust air temperature, wind speed at the surface, and ground thermal properties. None of those are directly observable from orbit. What satellite data can do is rank shafts by relative anomaly magnitude and flag step-changes that suggest significant operational shifts. Quantitative airflow figures require ground calibration against known ventilation data, which is rarely available for third-party monitoring.
Building a monitoring programme: data choices and cadence
A practical monitoring stack for a high-latitude mine starts with the Landsat-8 and Landsat-9 archive from 2013 onwards. That gives roughly 200 to 300 clear-sky winter acquisitions at most sites, enough to characterise baseline anomaly magnitude and inter-annual variability. ECOSTRESS adds diurnal sampling where ISS passes align. ASTER fills the pre-2013 gap at the cost of no new acquisitions.
For ongoing monitoring, a monthly composite of all clear-sky Landsat-9 acquisitions in the preceding 30 days is a reasonable operational cadence. It smooths noise, handles partial cloud, and still detects shutdowns that last more than a few weeks. If near-real-time status is required, ECOSTRESS irregular acquisitions can be ingested as they become available, typically within 24 to 48 hours of acquisition through NASA Earthdata. Neither sensor offers same-day tasking in the way optical commercial constellations do.
Satellize applies this approach within its analytics platform, combining open thermal archives with commercial visible tasking for shaft-collar geolocation. The Tonga crop-estimation programme demonstrates the same principle of pairing open-constellation temporal depth with targeted commercial imagery for spatial precision.
Honest limits and what to do about them
Three limits are worth stating plainly to any buyer. First, this method cannot determine what is being mined, how much ore is moving, or whether the mine is profitable. It answers one question: is ventilation running? Second, at temperate latitudes in summer, the signal may be undetectable entirely. A monitoring contract that promises year-round detection at a mine in central Europe should be treated with scepticism. Third, geothermal background, surface water, and industrial waste heat from surface processing plants can all produce thermal anomalies that resemble ventilation signatures. Careful selection of reference pixels and cross-checking against mine layout maps is necessary to avoid misclassification.
The countermeasure to cloud is archive depth and patience. The countermeasure to resolution limits is combining sensors. The countermeasure to ambiguity is knowing the site: a thermal anomaly that appears precisely over a mapped shaft collar, persists through winter, and disappears after a reported shutdown is a ventilation signature. One that appears in a field 300 metres from any shaft is probably something else.
Typical figures
| Thermal spatial resolution (Landsat-9 TIRS) | 100 m native, resampled to 30 m in standard products |
| Thermal spatial resolution (ECOSTRESS) | Approximately 70 m |
| Thermal spatial resolution (ASTER TIR) | 90 m |
| Revisit cadence | Landsat-9: 16-day repeat (shorter effective repeat at high latitudes with Landsat-8 combined). ECOSTRESS: irregular, typically 1 to 5 days at mid-latitudes |
| Spectral bands used | Landsat-9 TIRS Band 10 (10.6–11.2 µm), Band 11 (11.5–12.5 µm); ECOSTRESS bands 4 and 5 (10.5–12.5 µm) |
| Noise-equivalent temperature difference (NEDT) | Landsat-9 TIRS: approximately 0.4 K; ECOSTRESS: approximately 0.3 K |
| Minimum detectable anomaly (practical) | Approximately 1.5 to 2 K above reference ground, assuming clear sky and winter conditions at high latitude |
| Archive depth | Landsat: 2013 (Landsat-8) to present; ASTER: 1999 to 2019 (new acquisitions ceased); ECOSTRESS: 2018 to present |
| Cloud penetration | None. Thermal infrared is blocked by cloud. SAR cannot substitute for thermal in this use case. |
| Delivery formats | GeoTIFF anomaly maps, time-series CSV, GIS-ready vector shaft-status polygons, PDF monitoring reports |
Analytics Satellize can run
| Shaft operational status classification | Per-pixel thermal anomaly detection relative to local reference ground, thresholded against sensor NEDT, applied to each clear-sky acquisition | Monthly status report (active / idle / indeterminate) per identified shaft location, delivered as GIS layer and PDF summary |
| Winter anomaly magnitude time series | Standardised anomaly index (shaft pixel brightness temperature minus median reference temperature) computed across Landsat-8/9 archive from 2013 to present | Annual CSV time series with acquisition dates, anomaly values, and cloud-cover flags; chart-ready for trend analysis |
| Shutdown and restart event detection | Change-point detection on the anomaly time series, flagging acquisitions where anomaly drops below detection threshold for two or more consecutive clear-sky passes | Alert notification with date range, confidence level, and supporting image thumbnails |
| Shaft collar geolocation | Sub-metre visible-band mapping from Pléiades Neo or similar commercial tasking, cross-referenced to thermal anomaly centroid from TIRS | Vector point layer of confirmed shaft collar locations with coordinates accurate to 2 to 5 metres |
| Diurnal ventilation behaviour characterisation | Multi-acquisition ECOSTRESS stack sampled at different local times, compared to single daily Landsat pass to separate geothermal background from active ventilation signal | Site characterisation report with diurnal anomaly profiles and confidence assessment of ventilation attribution |
| Multi-year operational history reconstruction | ASTER TIR archive (1999 to 2019) combined with Landsat archive to produce long baseline anomaly record, enabling detection of historical shutdowns and production ramp-ups | Annotated timeline graphic and underlying data table for due-diligence or regulatory reporting purposes |
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.