Industrial rooftop condition assessment from thermal and optical data
Membrane failures, ponding water and insulation loss leave distinct thermal and spectral fingerprints detectable from orbit. Combining Landsat TIRS, ECOSTRESS and WorldView-3 SWIR turns roof condition into a quantified underwriting input.
Sensors
- Landsat 8/9 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 revisit at the equator. Adequate for detecting large-area insulation anomalies and systematic ponding across logistics parks; too coarse to resolve individual roof sections on buildings smaller than roughly 50 × 50 m.
- NASA ECOSTRESS: ISS-mounted thermal radiometer with five bands spanning 8–12.5 µm, ~70 m ground sampling distance, non-sun-synchronous orbit giving variable overpass times. The variable timing is genuinely useful: acquiring scenes at different times of day reveals whether thermal anomalies are diurnal moisture signatures or persistent structural defects.
- Maxar WorldView-3 SWIR: Eight SWIR bands from 1195 nm to 2365 nm at 7.5 m resolution. Distinguishes roofing membrane chemistry (modified bitumen, TPO, EPDM, built-up felt) by reflectance signature, and flags oxidised or moisture-saturated surfaces that appear spectrally similar in visible bands but diverge sharply in SWIR. Commercial tasking; not free.
- Airbus Pléiades Neo: 30 cm panchromatic, 1.2 m multispectral, same-day stereo possible. Provides the geometric detail to map individual roof sections, drainage outlets, HVAC penetrations and visible membrane blistering or seam separation. No thermal capability; works as the geometric anchor for co-registered thermal layers.
What a flat roof actually emits at night
A well-insulated flat roof stores very little daytime heat. A roof with degraded or saturated insulation stores a great deal, and releases it slowly after sunset. That differential cooling rate is the core physical signal. Landsat TIRS acquires a single nighttime pass per 16-day cycle at most latitudes; ECOSTRESS, because it rides the ISS at variable inclination, can capture the same site at 2 a.m. on one pass and 11 p.m. on another, giving analysts two points on the cooling curve rather than one snapshot.
The practical detection floor is roughly 0.5 K of surface temperature difference between a healthy and a compromised roof section, under calm, clear-sky conditions. Cloud cover eliminates thermal signal entirely, which is the method's most significant operational constraint. In climates with frequent overcast, a single usable thermal acquisition per month is realistic. Scheduling tasking around forecast clear windows is not optional; it is the job.
Shortwave infrared tells you what thermal cannot
Thermal emissivity shows that something is wrong. SWIR reflectance can suggest what. Modified bitumen membranes, thermoplastic polyolefin (TPO) and EPDM rubber each have distinct SWIR spectral profiles, particularly in the 1600 nm and 2200 nm bands that WorldView-3 captures. Oxidised bitumen shifts its 2200 nm reflectance measurably compared with fresh material. Moisture ingress into a built-up felt system suppresses SWIR reflectance differently from surface ponding on an intact TPO sheet.
This spectral discrimination matters for underwriting because the reinstatement cost of a 10,000 m² bitumen roof and a TPO single-ply system of the same area differ by a factor that insurers care about. Getting material type wrong at survey stage propagates into rebuild estimates. SWIR from orbit cannot replace a physical core sample, but it can flag which sections warrant one, reducing the number of invasive surveys needed across a large portfolio.
Ponding water: the slow failure mode
Standing water on a flat roof is a compounding problem. It adds structural load, accelerates membrane degradation through freeze-thaw cycling, and eventually finds any seam or penetration. From orbit, ponding appears as a spectrally dark, thermally cool patch in daytime imagery and a thermally warm patch at night, because water has high heat capacity and cools more slowly than dry membrane. The coincidence of both signatures in co-registered daytime optical and nighttime thermal data is a strong indicator of persistent ponding rather than recent rainfall.
Pléiades Neo at 30 cm resolution resolves ponding areas down to a few square metres in favourable lighting conditions. Landsat TIRS at 100 m native resolution will miss anything smaller than a section of roof that fills several pixels. For a 20,000 m² distribution centre, Landsat is a screening tool; WorldView-3 and Pléiades are the diagnostic instruments. Using them in sequence, rather than defaulting immediately to the expensive commercial tasking, is the sensible workflow.
The limits you should know before commissioning a survey
Resolution is the first constraint. No current open-data thermal sensor resolves individual roof sections on a standard 5,000 m² unit. ECOSTRESS at 70 m and Landsat at 100 m are portfolio-level screeners. They identify which buildings in a 200-asset logistics estate are anomalous; they do not map which quadrant of a specific roof has failed.
The second constraint is atmospheric correction. Thermal radiance from a roof surface passes through the entire atmosphere before reaching the sensor. Water vapour in particular absorbs in the same spectral region. Accurate surface temperature retrieval requires either a split-window algorithm (using Landsat's two TIRS bands) or coincident atmospheric profile data. Errors of 1–2 K are plausible if correction is done carelessly, which is enough to produce false positives on marginally anomalous roofs.
The third constraint is solar geometry. Daytime thermal imagery of a roof includes a solar-heating component that varies with roof colour, slope, aspect and time of acquisition. A dark bitumen roof on a south-facing slope will appear warmer than a white TPO roof regardless of insulation condition. Pre-dawn or early-morning acquisitions minimise this confound; midday acquisitions maximise it. ECOSTRESS's variable overpass time is an advantage precisely here.
From anomaly map to underwriting input
The analytic chain runs from raw thermal and SWIR data through atmospheric correction, spectral unmixing of roofing materials, and thermal anomaly scoring, to a per-asset condition index. Each roof in a portfolio receives a score reflecting the area-weighted proportion of thermally anomalous surface, the spectral evidence of membrane degradation, and the presence of persistent ponding signatures. That score can be mapped directly onto a reinstatement cost adjustment factor, provided the insurer has defined the cost curve for the asset class.
Satellize structures this as a GIS layer delivered alongside a ranked anomaly report, so underwriting teams can triage physical inspections by severity rather than scheduling them uniformly across a portfolio. The Tonga crop-estimation programme demonstrated that structured satellite-derived scoring can replace or sharply reduce ground-survey frequency even in operationally demanding environments; the same logic applies to industrial roof portfolios, where the cost of physical inspection across hundreds of geographically dispersed assets is itself a significant line item.
One honest caveat: satellite-derived condition assessment is an input to underwriting judgement, not a substitute for it. A roof that scores poorly on thermal anomaly may have a recent repair that is not yet visible in the archive. A roof that scores well may have a localised failure below the detection threshold of any current spaceborne sensor. The value is in portfolio-level prioritisation, not in replacing the surveyor's report on any individual asset.
Typical figures
| Thermal spatial resolution (Landsat 8/9 TIRS) | 100 m native, 30 m resampled product |
| Thermal spatial resolution (ECOSTRESS) | ~70 m ground sampling distance |
| SWIR spatial resolution (WorldView-3) | 7.5 m per band, 8 bands 1195–2365 nm |
| Optical resolution (Pléiades Neo) | 30 cm panchromatic, 1.2 m multispectral |
| Landsat revisit | 16 days at equator; combined Landsat 8+9 reduces to ~8 days |
| ECOSTRESS overpass frequency | Variable; typically 1–5 usable acquisitions per month per site depending on cloud and ISS orbit |
| Minimum detectable thermal anomaly | ~0.5 K under clear-sky, calm conditions; larger uncertainty with residual atmospheric moisture |
| Minimum roof area reliably assessed (thermal) | Approximately 5,000 m² for ECOSTRESS/Landsat screening; no firm floor for WorldView-3 SWIR |
| Cloud impact | Total signal loss under cloud; thermal acquisition must be scheduled on forecast clear windows |
| Landsat archive depth | Landsat 8 from 2013, Landsat 9 from 2021; free via USGS EarthExplorer |
Analytics Satellize can run
| Portfolio thermal anomaly screen | Split-window land surface temperature retrieval (Landsat TIRS bands 10 and 11) with per-asset z-score ranking against local background | Ranked asset list with anomaly score and flagged roof area fraction, delivered as CSV and GIS polygon layer |
| Membrane material classification | Spectral unmixing of WorldView-3 SWIR bands using published endmember libraries for bitumen, TPO and EPDM reflectance | Per-roof material-type map at 7.5 m resolution, GeoTIFF with class confidence scores |
| Ponding water probability layer | Multi-temporal fusion of daytime SWIR reflectance suppression and nighttime ECOSTRESS thermal retention; persistence filter across at least three acquisitions | Ponding probability raster per asset, updated on each clear-sky acquisition cycle |
| Insulation degradation index | Normalised thermal anomaly score derived from pre-dawn ECOSTRESS or Landsat TIRS, corrected for roof colour and solar-loading history using daytime optical reflectance | Per-asset index value (0–1 scale) with confidence band, included in underwriting condition report |
| Change detection: condition deterioration over time | Time-series differencing of seasonal thermal anomaly scores using Landsat archive from 2013 onwards; Mann-Kendall trend test on per-asset score sequence | Trend report flagging assets with statistically significant deterioration, suitable for renewal-cycle review |
| Reinstatement cost adjustment layer | Combination of material classification output and anomaly-scored degraded area fraction, mapped to client-supplied unit reinstatement cost schedules | Adjusted reinstatement cost estimate per asset as tabular report and GIS attribute 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.