Construction site thermal activity and curing heat signature analysis
Spaceborne thermal infrared sensors detect the heat released by large-scale concrete curing, asphalt laying and plant commissioning, offering an activity signal that works through haze and at night. Resolution limits are real: individual pours are invisible, but block-scale thermal anomalies are not.
Sensors
- Landsat 9 TIRS-2: Two thermal bands centred at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in the data product), 16-day repeat at the equator. The split-window pair enables atmospheric correction, improving land surface temperature retrievals to roughly ±1 K under clear-sky conditions. Archive extends from September 2021.
- ECOSTRESS (ISS-mounted): Five thermal bands from 8.3 to 12.5 µm, approximately 70 × 70 m pixel size, sub-daily revisit at mid-latitudes because the ISS precesses through local times. Captures diurnal temperature cycles that a sun-synchronous sensor misses entirely, which is directly useful for distinguishing curing heat from solar loading. Publicly available through NASA Earthdata.
- Sentinel-3 SLSTR: Dual-view thermal imager with bands at 10.85 µm and 12.0 µm, 1 km resolution in thermal mode. Too coarse for single-site analysis, but useful for regional baseline temperature fields and detecting very large industrial commissioning events such as power plant first-fire or kiln warm-up across an entire industrial zone.
- ASTER TIR: Five thermal bands from 8.1 to 11.7 µm at 90 m resolution, on-demand tasking (not continuous). ASTER's multi-band thermal coverage allows emissivity separation, which helps distinguish concrete, asphalt and bare soil by material rather than just temperature. The sensor is past its design life and scheduling is unreliable; treat it as an opportunistic archive resource rather than a primary monitoring tool.
What the heat actually tells you
Cement hydration is exothermic. A large mass pour, a raft foundation or a concrete dam lift releases heat over hours to days as calcium silicate hydrate forms. Asphalt laid at 140–160 °C cools detectably over several hours. Industrial plant commissioning, from boiler warm-up to kiln first-fire, produces sustained thermal signatures that persist for days. None of these processes can be hidden from a thermal infrared sensor simply by covering the site or scheduling work at inconvenient times for optical tasking.
The practical signal is a land surface temperature (LST) anomaly relative to background. A freshly poured concrete slab covering several thousand square metres can raise the apparent surface temperature of a 100 m pixel by 3–8 K above the surrounding undisturbed ground, depending on pour thickness, ambient temperature and wind. That is well above the retrieval noise floor of Landsat 9 TIRS-2 and ECOSTRESS. Smaller pours, individual columns or isolated slab sections do not produce anomalies large enough to dominate a full pixel at this resolution.
The resolution problem, stated plainly
Landsat 9 TIRS-2 pixels cover 100 m on a side before resampling. ECOSTRESS pixels are roughly 70 m. ASTER reaches 90 m. No current freely available spaceborne thermal sensor resolves a single residential foundation pour, a column cap or a bridge deck section. If the question is 'did they pour the north-east corner column today?', satellite thermal data cannot answer it.
What it can answer is coarser but still commercially useful: is a site thermally active at all? Has a previously dormant zone of a large industrial facility started generating heat consistent with commissioning? Has a major road corridor shown sustained asphalt-temperature anomalies along a new alignment? These block-level questions, across sites measured in hectares rather than square metres, are where the method earns its place. Project-finance monitors, sovereign infrastructure auditors and intelligence-grade activity verification are the natural buyers, not quality-control engineers.
Diurnal timing matters more than most buyers expect
Landsat 9 crosses the equator at approximately 10:00 local solar time. At that hour, solar loading on bare concrete or dark asphalt is already substantial, which compresses the contrast between a thermally active pour and a sun-warmed inactive surface. ECOSTRESS, mounted on the International Space Station, acquires at varying local times across its roughly 90-minute orbital period. This means it can capture a site at 02:00 or 04:00 local time, when solar loading is zero and any residual curing heat stands out against a uniformly cool background.
For asphalt monitoring specifically, a night-time or pre-dawn pass is close to essential. Freshly laid asphalt at 02:00 retains enough heat to produce a clear anomaly; the same surface at 10:00 may be indistinguishable from adjacent bare ground heated by the sun. Analysts who rely solely on Landsat 9 for asphalt detection are working with the least favourable acquisition geometry. Pairing Landsat with ECOSTRESS night passes substantially improves detection confidence.
Cloud cover and the limits of thermal advantage
Thermal infrared is often described as cloud-penetrating. It is not. Clouds are opaque in the thermal infrared; what a sensor sees is the cloud-top temperature, not the ground. The advantage thermal holds over shortwave optical is narrower than the marketing usually admits: thermal sensors do work at night, and they are unaffected by the haze and thin smoke that can degrade optical imagery. Optically thin cirrus that barely affects a Sentinel-2 scene will still degrade a thermal retrieval.
In practice, cloud-affected thermal pixels are flagged in standard products (the Landsat Collection 2 quality assessment band and the ECOSTRESS cloud mask both do this). Analysts should expect 20–60% of acquisitions to be cloud-contaminated depending on the site's climate zone. In persistently cloudy environments, such as equatorial construction sites or monsoon-season projects, thermal satellite monitoring should be treated as a probabilistic signal accumulated over multiple passes rather than a reliable single-date observation.
Building an anomaly baseline and avoiding false positives
The most common analytical failure is flagging a thermally warm pixel without accounting for the background. Industrial rooftops, standing water, dark-coloured spoil heaps and metal cladding all produce elevated LST readings that have nothing to do with construction activity. A credible analysis requires a pre-construction thermal baseline, typically drawn from archive Landsat scenes covering the same site at the same time of year and the same local solar time, so that seasonal and material effects are controlled.
Change detection then compares current LST against the baseline distribution. Anomalies exceeding two or three standard deviations from the baseline, persisting across multiple acquisitions and spatially consistent with the known footprint of active construction zones, are the signal worth reporting. Single-pass anomalies without persistence should be treated as uncertain. Satellize applies this baseline-differencing approach operationally; the same logic underpins the thermal component of its Tonga crop-estimation work, where distinguishing soil moisture signals from genuine crop stress requires the same kind of controlled comparison.
ASTER's emissivity separation capability adds a useful check. If a thermal anomaly coincides with an emissivity signature consistent with freshly laid asphalt or concrete rather than bare soil or metal, confidence in the construction interpretation rises materially.
What a commissioning signature looks like versus a curing signature
Concrete curing produces a spatially diffuse, relatively brief anomaly. A large raft pour might show elevated temperatures for 24–72 hours before the surface cools to ambient. The anomaly is warm but not extreme, typically in the 3–10 K above-background range at the pixel scale.
Industrial plant commissioning is different in character. A gas turbine test, a kiln first-fire or a boiler warm-up produces a more localised, hotter and longer-duration anomaly. Stack exhaust and hot surfaces can push LST anomalies well above 20 K at the pixel scale and sustain them for days or weeks. ECOSTRESS and Landsat 9 both detect these readily. The spatial pattern also differs: commissioning heat tends to cluster around specific plant structures rather than spreading across the whole site footprint. Distinguishing the two signature types is usually possible from the spatial morphology and duration of the anomaly, even at 100 m resolution.
Typical figures
| Typical spatial resolution (thermal) | 70–100 m (ECOSTRESS ~70 m, Landsat 9 TIRS 100 m native, ASTER TIR 90 m, Sentinel-3 SLSTR 1 km) |
| Revisit frequency | Landsat 9: 16 days at equator. ECOSTRESS: sub-daily at mid-latitudes (ISS precession). Sentinel-3: ~1 day globally. ASTER: on-demand, no guaranteed revisit. |
| Spectral bands used | Thermal infrared 8.1–12.5 µm; split-window pairs at ~10.9 µm and ~12.0 µm for atmospheric correction |
| Land surface temperature retrieval accuracy | ±1–2 K under clear-sky conditions (Landsat 9 TIRS-2 published specification); degraded under partial cloud or heavy aerosol |
| Minimum detectable anomaly (practical) | ~3–5 K above background across a pixel-scale area (~0.5–1 ha); sub-pixel anomalies not reliably detected |
| Cloud limitation | Opaque cloud blocks thermal signal entirely; expect 20–60% cloud contamination depending on climate zone and season |
| Archive depth | Landsat 9: from September 2021. ECOSTRESS: from July 2018. ASTER: from 2000 (scheduling unreliable post-2020). Sentinel-3: from 2016. |
| Latency (standard open-data products) | Landsat 9 Collection 2 LST: typically 12–24 hours post-acquisition. ECOSTRESS: 2–5 days via NASA Earthdata. Sentinel-3: ~3 hours (NRT product). |
| Delivery formats | GeoTIFF (LST raster), cloud-masked anomaly layer, time-series CSV, GIS-ready shapefile of flagged anomaly zones |
Analytics Satellize can run
| Thermal anomaly detection report | Baseline-differenced land surface temperature analysis using Landsat 9 TIRS-2 and ECOSTRESS; anomalies flagged at >2 SD from pre-construction baseline | PDF report with annotated thermal imagery, anomaly magnitude table and persistence score per site zone |
| Activity timeline reconstruction | Multi-date LST stack over archive acquisitions; temporal clustering of anomaly events to infer active construction periods | Time-series chart and GIS layer showing onset, peak and decay of thermal anomalies by site sector |
| Commissioning onset alert | Threshold-based detection on ECOSTRESS night passes and Landsat acquisitions; alert triggered when LST anomaly exceeds defined threshold for two consecutive clear-sky passes | Automated alert (email or API push) with coordinates, anomaly magnitude and acquisition timestamp |
| Asphalt-laying corridor mapping | Night-pass ECOSTRESS LST differencing along a defined road alignment; anomaly pixels spatially intersected with corridor boundary to estimate active paving front | GIS polyline showing estimated active paving extent with date and confidence flag |
| Material-type disambiguation layer | ASTER TIR emissivity separation (where archive scenes are available) to distinguish concrete, asphalt and bare soil thermal signatures within anomaly zones | GIS raster layer of dominant surface material class within flagged thermal anomaly footprint |
| Persistent inactivity flag | Absence of thermal anomaly across multiple passes where construction schedule predicts activity; compared against baseline and weather-screened for cloud-affected acquisitions | Tabular report flagging schedule deviations with supporting cloud-cover statistics and acquisition log |
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.