Thermal infrared detection of buried structures and voids
Subsurface masonry and compacted surfaces retain heat differently from surrounding soil, producing thermal anomalies detectable in pre-dawn satellite imagery. ASTER TIR and Landsat TIRS have confirmed this at sites across Egypt and the Near East, though their coarse resolution limits detection to large features.
Sensors
- ASTER TIR: Five thermal bands (8.125–11.65 µm) at 90 m spatial resolution, 16-day revisit on Terra. The multi-band configuration allows surface emissivity separation, which reduces confusion between thermal inertia signals and emissivity contrasts from different soil mineralogy.
- Landsat 8 / 9 TIRS: Two thermal bands (Band 10: 10.6–11.19 µm; Band 11: 11.50–12.51 µm) at 100 m resolution resampled to 30 m pixels. 16-day revisit per satellite, 8-day combined. Radiometric calibration is well-characterised; Band 11 has documented stray-light issues that reduce its quantitative reliability for split-window retrievals.
- ECOSTRESS: Five TIR bands (8.28–12.13 µm) on the International Space Station at roughly 70 m resolution and non-sun-synchronous orbit, giving variable overpass times that can capture pre-dawn and post-sunset windows unavailable to fixed morning-pass sensors. Revisit is irregular, typically 1–5 days depending on latitude.
- Airborne FLIR (reference baseline): Thermal cameras mounted on aircraft or UAVs achieve sub-metre resolution and allow precise timing of pre-dawn or post-sunset acquisition. Not a satellite system, but the published archaeological literature uses airborne FLIR results as ground truth against which spaceborne detections are validated.
Why buried stone behaves like a slow-release battery
Thermal inertia is a material's resistance to temperature change. It is proportional to the square root of the product of thermal conductivity, density and specific heat capacity. Compact limestone, fired brick and compacted earthen floors all have higher thermal inertia than loose, porous, dry desert sand. During the day both surfaces absorb solar radiation. After sunset the sand cools rapidly; the masonry does not. By the time the sun rises again, a buried wall may still be one to five degrees Celsius warmer than the surrounding soil, depending on burial depth, moisture content and the diurnal temperature range at the site.
That differential is the signal. It is not large. A single-degree contrast sits near the noise floor of spaceborne TIR sensors under unfavourable conditions, which is why acquisition timing is everything. The physics also imposes a depth limit: features buried deeper than roughly one to two metres are too well insulated from the surface thermal cycle to produce a detectable signal in a single diurnal observation. Moisture complicates matters further, because wet soil has higher thermal inertia and can mimic buried structure signals or suppress them, depending on geometry.
The two acquisition windows and why one is better than the other
Pre-dawn acquisition, typically one to two hours before local sunrise, captures the maximum thermal contrast between high-inertia buried features and the cooled surrounding soil. This is the preferred window in the published literature. Post-sunset imagery, taken one to three hours after sunset, captures an earlier stage of the same cooling curve and can also show the contrast, though it is generally smaller and more variable because surface heating during the day is uneven across different surface materials.
Landsat 8 and 9 pass in the mid-morning (approximately 10:00–10:30 local solar time), which is the worst possible moment: surface temperatures are rising and thermal contrasts from overnight differential cooling have largely equalised. ASTER on Terra passes at roughly 10:30 local time for the same reason. Both sensors are therefore used primarily for diurnal-difference analysis: subtracting a daytime scene from a night-time scene acquired on the same or a nearby date. ASTER acquired dedicated night-time scenes over selected archaeological sites under its on-demand tasking programme. ECOSTRESS, with its variable overpass times, is the only current spaceborne sensor that routinely captures genuine pre-dawn passes without special tasking.
What the published record actually shows
The most-cited spaceborne thermal work in archaeology comes from Egypt and the Near East. Studies using ASTER night-time data over sites in the Nile Delta and Upper Egypt have identified broad thermal anomalies consistent with buried mudbrick enclosures and compacted surfaces, later partially confirmed by ground survey. Work at sites in the Mesopotamian alluvial plain has used diurnal-difference ASTER composites to distinguish tell mounds from their surroundings and to identify linear features interpreted as ancient field boundaries and canal banks.
The honest summary of this body of work is: spaceborne TIR confirms the principle, but the resolution constraint is severe. At 90–100 m, a buried feature must be tens of metres across to produce a pixel-level anomaly. Individual walls, pits and small chambers are invisible. What the sensors detect are large architectural complexes, broad compacted surfaces and the aggregate thermal signature of dense buried rubble. Airborne FLIR at sub-metre resolution has resolved individual wall lines at sites where the spaceborne data showed only a diffuse warm patch. The two scales are complementary rather than interchangeable.
Confounders that produce false positives and how to screen them
Geological heterogeneity is the primary confounder. Variations in soil texture, mineralogy and moisture across a site produce thermal contrasts that have nothing to do with archaeology. A patch of clay-rich soil surrounded by sandy matrix will appear warm in pre-dawn imagery for purely pedological reasons. Screening requires at minimum a multispectral reflectance scene to map surface soil composition, and ideally a published geological or soil survey of the area.
Modern infrastructure is a second problem. Buried pipes, irrigation channels, compacted tracks and building foundations all produce thermal anomalies indistinguishable in shape and magnitude from archaeological features at coarse resolution. Sites near agricultural land or settlements require careful masking. Wind is a third factor: even moderate surface wind accelerates evaporative and convective cooling and can reduce or destroy the thermal contrast before acquisition. Selecting scenes from calm nights, cross-referenced against meteorological records, is standard practice in the published literature.
Resolution floors and what they mean for site selection
The 90 m floor of ASTER TIR and the 100 m floor of Landsat TIRS are not soft limits that clever processing can push through. They are Nyquist constraints on the sensor optics. A feature must subtend at least two to three pixels to be reliably distinguished from noise, which means the practical minimum detectable target is roughly 150–200 m across for ASTER and 200 m for TIRS under good conditions. This restricts spaceborne thermal archaeology to a specific class of site: large palatial or administrative complexes, major temple enclosures, city-scale compacted surfaces and broad agricultural systems.
Smaller sites, which constitute the majority of the archaeological record, require airborne or UAV-mounted thermal sensors. Satellize incorporates open-archive ASTER night-time scenes and Landsat TIRS diurnal-difference products into site-screening workflows, as it does for other spectral detection tasks including the Tonga crop-estimation programme, but is direct about the resolution ceiling. A negative result from spaceborne TIR does not mean a site contains nothing; it means the site contains nothing large enough for these sensors to see.
ECOSTRESS partially addresses the timing problem without addressing the resolution problem: its 70 m ground sampling is marginally better than ASTER but still subject to the same Nyquist constraint. Its value lies in the pre-dawn acquisition window and its multi-band emissivity separation, not in spatial detail.
Building a workable detection workflow
A practical spaceborne thermal survey for buried structure detection combines three inputs. First, a diurnal-difference stack: multiple ASTER or Landsat TIRS night-minus-day difference images from different seasons, averaged to suppress noise and meteorological variability. Second, a surface composition mask derived from multispectral reflectance data (Sentinel-2 or Landsat OLI) to flag geological and agricultural confounders. Third, a morphological filter that retains only anomalies with rectilinear or geometrically regular shapes consistent with built structures, rejecting diffuse or irregular warm patches.
Anomalies that survive all three filters become candidate targets for follow-on investigation: higher-resolution optical imagery, SAR coherence analysis, or, where access permits, ground-penetrating radar. The thermal layer is a first-pass screen, not a final answer. Treating it as anything more produces expensive false leads. Treating it as worthless ignores a genuine and low-cost way to prioritise survey effort across landscapes that are too large to cover on foot.
Typical figures
| Spatial resolution (ASTER TIR) | 90 m (5 thermal bands) |
| Spatial resolution (Landsat 8/9 TIRS) | 100 m native, resampled to 30 m in data products |
| Spatial resolution (ECOSTRESS) | ~70 m |
| Revisit (ASTER on Terra) | 16 days; night-time tasking available on request via LP DAAC |
| Revisit (Landsat 8 + 9 combined) | ~8 days at equator |
| Revisit (ECOSTRESS on ISS) | Irregular, typically 1–5 days depending on latitude and ISS orbit |
| Thermal bands covered | 8.1–12.5 µm across ASTER, TIRS and ECOSTRESS |
| Minimum detectable target (spaceborne) | Approximately 150–200 m across under good conditions |
| Detectable temperature differential | Published studies report 1–5 °C for buried masonry vs. surrounding soil |
| Archive depth | ASTER from 1999; Landsat 8 from 2013; Landsat 9 from 2021; ECOSTRESS from 2018 |
Analytics Satellize can run
| Diurnal thermal difference map | Pixel-wise subtraction of night-time TIR brightness temperature from daytime acquisition, multi-date averaging to reduce noise | GeoTIFF thermal anomaly layer with candidate feature polygons, classified by contrast magnitude and shape regularity |
| Thermal inertia index | Apparent thermal inertia (ATI) derived from albedo and diurnal temperature range following published Price (1977) and Cracknell & Xue (1996) formulations | Raster ATI layer with anomaly zones ranked by departure from local background, exported as GIS layer |
| Confounder-masked candidate site list | Overlay of thermal anomaly layer with Sentinel-2 soil composition classification and modern infrastructure mask; morphological shape filter retaining rectilinear anomalies | Prioritised candidate list with coordinates, anomaly area, thermal contrast value and confounder-risk score, delivered as GeoPackage and PDF report |
| Seasonal thermal variability stack | Multi-date ASTER or Landsat TIRS time series across dry and wet seasons to distinguish persistent structural anomalies from transient moisture or vegetation effects | Temporal consistency score per candidate anomaly; anomalies persistent across seasons flagged as higher-confidence targets |
| ECOSTRESS pre-dawn acquisition analysis | Selection and processing of ECOSTRESS scenes acquired within two hours of local sunrise; emissivity-corrected land surface temperature retrieval using the Temperature Emissivity Separation algorithm | Pre-dawn LST map with anomaly overlay, acquisition time and meteorological quality flag noted per scene |
| Multi-sensor fusion briefing | Co-registration and joint interpretation of thermal anomaly candidates with Sentinel-1 SAR coherence and high-resolution optical imagery to cross-validate buried feature hypotheses | Site-level interpretive report with annotated imagery panels and recommended ground-survey priority ranking |
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.