Thermal anomaly detection for military activity inference
Spaceborne thermal and shortwave infrared sensors detect heat signatures from aircraft engines, vehicle exhausts, and burning materiel. Open data from Landsat and ASTER provides the baseline; commercial SWIR adds resolution where it matters.
Sensors
- Landsat 8/9 TIRS: Two thermal infrared bands centred at 10.9 µm and 12.0 µm, native 100 m resolution resampled to 30 m in distributed products. Revisit 16 days per satellite, 8 days combined for the pair. Sufficient for detecting large industrial thermal anomalies and open burns; cannot resolve individual vehicles or aircraft.
- ASTER TIR: Five thermal bands from 8.125 µm to 11.65 µm at 90 m resolution. On-demand tasking historically available through NASA/USGS; instrument has experienced intermittent operational issues since 2008. Useful for multi-band emissivity separation at dual-use facilities.
- MODIS/VIIRS Active Fire: MODIS detects fires at 1 km resolution; VIIRS I-band improves this to 375 m. Near-real-time global coverage (twice daily per platform). Effective for open burning of materiel or fuel dumps but far too coarse for facility-level attribution.
- Planet SkySat SWIR: Shortwave infrared channels sensitive to hot targets in the 1.6 µm and 2.2 µm atmospheric windows. Sub-metre panchromatic context with SWIR overlay; commercial tasking with same-day revisit in some orbits. Detects engine plumes and exhaust heat invisible to standard optical bands.
- Maxar WorldView-3 SWIR: Eight SWIR bands from 1.195 µm to 2.365 µm at 3.7 m resolution, paired with 0.31 m panchromatic. Best commercially available spatial resolution for SWIR thermal inference. Can distinguish warm engine nacelles from ambient apron surfaces where Landsat TIRS cannot.
What thermal emission actually tells you
Every object above absolute zero radiates energy, and the wavelength of peak emission shifts with temperature according to Planck's law. At ambient ground temperatures (roughly 280–310 K), peak emission falls in the longwave infrared around 10 µm. Aircraft turbofan exhaust at several hundred degrees Celsius shifts the peak toward mid-wave infrared (3–5 µm) and produces strong contrast in shortwave IR bands (1.6–2.5 µm) that is essentially invisible in standard optical imagery. That physics is the foundation of the whole discipline.
The practical consequence is that a running jet engine on a cold apron is a bright anomaly in SWIR even when the aircraft itself is camouflaged. An armoured vehicle that has been driven for an hour retains elevated exhaust and engine-deck temperatures for 30–60 minutes after stopping. Open burning of fuel, ammunition, or destroyed vehicles produces intense mid-wave signatures detectable by MODIS and VIIRS at continental scale within minutes of ignition.
The resolution problem no one should pretend away
Landsat TIRS is distributed at 30 m pixels but the native thermal measurement is 100 m. A main battle tank is roughly 3 m wide and 7 m long. It occupies a fraction of a single Landsat thermal pixel. You cannot detect an individual vehicle with Landsat TIRS; you can detect a cluster of vehicles, a heated compound, or a burning fuel store. This is not a flaw in the analysis pipeline, it is a physical constraint that shapes what questions the data can honestly answer.
ASTER's 90 m thermal resolution is only marginally better for vehicle-level detection. The step change comes with commercial SWIR. WorldView-3's 3.7 m SWIR bands can resolve the thermal footprint of a single large aircraft engine nacelle under favourable contrast conditions. Planet SkySat SWIR sits in between. Neither replaces dedicated airborne or spaceborne mid-wave IR sensors operated by national intelligence agencies, but they are the best available in the open and commercial market.
Dual-use facilities and the baseline problem
A steel mill, a cement plant, and a military propellant factory can produce similar thermal signatures from space. Attribution requires a baseline. The standard method is to build a time-series of thermal observations at a facility during a period when its declared function is known, then flag deviations: a site that ran cold for six months and then showed sustained elevated temperatures across multiple TIRS acquisitions warrants investigation. This is the approach documented in open-source conflict monitoring work on sites in North Korea and Iran, where Landsat thermal archives stretching back to 1982 (Landsat 4 TM) provide the long baseline needed to identify anomalous operational periods.
The 16-day Landsat revisit is a real constraint for fast-moving situations. Cloud cover compounds it: a single acquisition gap in a cloudy season can be 30–60 days. VIIRS and MODIS fill some of that gap for large signatures but not for facility-level detail. Commercial tasking of WorldView-3 or SkySat can compress the revisit to days or hours for a priority site, at cost.
Open burning as a conflict indicator
Deliberate burning of military materiel, fuel depots, or crops as a scorched-earth tactic produces signatures that VIIRS Nightfire and MODIS Active Fire detect reliably. The University of Maryland FIRMS system aggregates these in near-real-time with latency measured in hours. In documented conflict monitoring, clusters of fire detections at or near military logistics sites, combined with optical change detection, have been used to infer depot destruction or deliberate denial.
The ambiguity here is significant. Agricultural burning, industrial accidents, and forest fires produce comparable signatures. Spatial context is everything: a fire detection inside the perimeter of a known military storage site carries different inferential weight than one in open farmland. Combining VIIRS fire data with Sentinel-1 SAR coherence loss and Landsat optical change detection is the published method for reducing false attribution.
What the archive enables that real-time data cannot
The Landsat archive from 1972 onward, and ASTER from 1999, allow retrospective analysis that real-time systems cannot provide. Establishing when a facility first showed anomalous thermal activity, correlating that with known geopolitical timelines, and quantifying the duration of elevated operations are all archive problems. The USGS Landsat archive is publicly accessible and free. ASTER data is available through NASA Earthdata. Both are standard inputs to open-source intelligence workflows.
Satellize runs thermal anomaly time-series on Landsat and VIIRS as part of its analytics stack, applying the same change-detection logic it uses for the Tonga crop-estimation programme to industrial and facility targets. The physics of detecting a thermal deviation from baseline is the same whether the anomaly is a warm paddy field or a warm engine bay. Clients with priority sites can add commercial WorldView-3 or SkySat tasking through their own licence arrangements.
Honest limits of the method
Thermal inference is probabilistic, not conclusive. A warm building at a military facility could mean active machinery, poor insulation, or simply a sunny south-facing wall in winter. Single-pass detections should never drive operational conclusions. The standard of evidence in open-source conflict monitoring requires convergence: thermal anomaly, optical change, SAR coherence shift, and where available, AIS or ADS-B absence consistent with activity.
Revisit gaps, persistent cloud, and the 100 m native resolution floor of the best free thermal data are not solvable problems at the open-data tier. They are features of the landscape that any honest programme design must accommodate. For facilities where a 16-day gap is operationally unacceptable, commercial tasking is the only answer, and even then, cloud is cloud.
Typical figures
| Best open-data thermal resolution | 90 m (ASTER TIR); 100 m native / 30 m resampled (Landsat 8/9 TIRS) |
| Best commercial SWIR resolution | 3.7 m (WorldView-3 SWIR); ~1 m panchromatic context |
| Landsat revisit (8+9 combined) | 8 days at equator; longer at high latitudes |
| VIIRS active-fire latency | 3–6 hours from acquisition to FIRMS alert |
| Spectral bands of interest | SWIR: 1.6 µm, 2.2 µm; MWIR: 3–5 µm; LWIR / TIR: 8–12 µm |
| Minimum detectable thermal anomaly (Landsat TIRS) | ~0.1 K temperature contrast at 100 m scale under clear sky; larger area required for reliable detection |
| VIIRS fire detection threshold | Published detection limit approximately 1 MW fire radiative power at 375 m pixel |
| Open archive depth | Landsat thermal from 1982 (TM band 6); ASTER from 1999; MODIS from 2000; VIIRS from 2012 |
| Cloud impact | Thermal and SWIR bands blocked by cloud; no all-weather thermal capability in open data |
| Typical delivery format | GeoTIFF anomaly layers, time-series CSV, change-alert GIS feed (GeoJSON / Shapefile) |
Analytics Satellize can run
| Facility thermal baseline and deviation alerts | Time-series z-score anomaly detection on Landsat TIRS band 10/11 brightness temperature | Monthly PDF report with anomaly-flagged acquisitions and annotated GeoTIFF layers |
| Open-burn event log | VIIRS Nightfire and MODIS Active Fire clustering filtered by facility perimeter polygons | Near-real-time GeoJSON alert feed with fire radiative power estimates and timestamps |
| SWIR hot-spot mapping at priority sites | Band-ratio analysis on WorldView-3 or SkySat SWIR channels to isolate anomalous radiance | Annotated false-colour SWIR scene with hot-spot polygons and radiance values |
| Multi-sensor convergence report | Fusion of TIRS thermal anomaly, Sentinel-1 SAR coherence loss, and optical change detection | Structured intelligence summary with confidence rating and supporting imagery panels |
| Historical thermal activity timeline | Archive mining of Landsat Collection 2 Level-2 surface temperature product from 1982 to present | Longitudinal chart of mean facility brightness temperature with annotated geopolitical events |
| Dual-use facility operational-state classification | Supervised classification of thermal state (idle / warm / active / burning) across multi-date stack | Quarterly state-transition matrix per monitored site, delivered as spreadsheet and GIS layer |
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.