Thermal infrared urban heat island quantification
Thermal infrared sensors retrieve land surface temperature across entire cities in a single pass, resolving the heat penalty paid by dense impervious surfaces versus parks and water bodies. The physics is well understood; the constraints are resolution, revisit and cloud.
Sensors
- Landsat 8/9 TIRS: Two thermal bands (Band 10 at 10.6–11.19 µm, Band 11 at 11.50–12.51 µm) at 100-metre native resolution, resampled to 30 m in products. 16-day revisit per satellite; combined Landsat 8 and 9 fleet halves that to roughly 8 days at mid-latitudes. Enables split-window atmospheric correction and resolves individual city blocks.
- ECOSTRESS (ISS): Five thermal infrared bands centred near 8.3–12.5 µm, 70-metre spatial resolution. Mounted on the International Space Station, so its non-sun-synchronous orbit produces overpass times that drift across the full diurnal cycle over weeks, capturing pre-dawn minima and afternoon maxima that fixed-overpass sensors miss entirely.
- ASTER (Terra, heritage): Five thermal infrared bands (8.125–11.65 µm) at 90-metre resolution with on-demand tasking. No longer acquiring new data routinely since 2023, but its 23-year archive remains a reference for long-term urban heat studies and emissivity libraries.
- Sentinel-3 SLSTR: Two thermal channels (10.85 µm and 12.0 µm) at 1-kilometre resolution, daily global coverage. Useful for city-scale trend monitoring and split-window LST retrieval across metropolitan regions, but cannot resolve intra-urban variation below roughly one square kilometre.
What a floating roof gives away
Land surface temperature (LST) is not air temperature. It is the radiometric temperature of whatever the sensor sees: asphalt, a membrane roof, a tree canopy, bare soil. In urban areas the gap between LST and the air temperature measured at a weather station can exceed 10 °C on a calm summer evening, because impervious surfaces absorb shortwave radiation during the day and release it slowly through the night rather than evaporating it away as a vegetated surface would.
Thermal infrared sensors measure upwelling radiance in atmospheric window bands, typically 8–14 µm, where the atmosphere is relatively transparent. Converting that radiance to a surface temperature requires correcting for atmospheric water vapour absorption and for the emissivity of the surface itself. Concrete and asphalt have emissivities close to 0.95, which makes the correction tractable. The split-window algorithm, standard in Landsat TIRS and Sentinel-3 SLSTR processing, uses the differential absorption between two adjacent thermal bands to estimate and remove the atmospheric contribution without needing a concurrent radiosonde.
Why 100 metres matters and 1 kilometre does not
MODIS thermal bands operate at 1-kilometre resolution. A pixel that size in a typical European or Asian city contains a mixture of roads, rooftops, courtyards and small parks. The averaged LST from that pixel is real but analytically weak: it cannot tell you whether the heat source is the industrial estate on the western edge or the residential blocks on the east. Planning decisions require that distinction.
Landsat TIRS at 100 metres (native) resolves a footprint roughly the size of a city block. That is sufficient to separate a large park from the surrounding built fabric, to identify a cool-roof intervention that covers a single large building, or to map the temperature gradient along a river corridor. It is not sufficient to resolve individual rooftops in dense urban fabric, where commercial very-high-resolution thermal sensors in the 3–5 metre class would be needed. Landsat sits in a useful middle ground: free, consistent since 1982 in the thermal bands, and fine enough for neighbourhood-scale policy.
ECOSTRESS at 70 metres offers marginally better spatial detail, but its principal advantage is temporal. A sun-synchronous satellite like Landsat always passes at roughly the same local time, around 10:00 local solar time for the descending node. That captures a mid-morning snapshot, not the peak afternoon heat or the nocturnal retention that most concerns urban planners. ECOSTRESS, precessing across local times as the ISS orbit evolves, can return to the same city at 14:00 on one pass and 03:00 on another, making it the only freely available sensor that directly measures the nocturnal urban heat island from orbit.
Impervious surface fraction as the explanatory variable
The relationship between LST and impervious surface fraction (ISF) is one of the most replicated findings in urban remote sensing. As ISF rises from 0 (fully vegetated) to 1 (fully sealed), daytime LST increases and nocturnal cooling slows. The mechanism is straightforward: sealed surfaces have low albedo, high heat capacity and near-zero latent heat flux. The practical consequence is that LST maps derived from Landsat TIRS, when regressed against ISF derived from multispectral classification of the same or concurrent imagery, produce a quantitative heat-exposure model that can be used to prioritise greening or cool-roof interventions.
ISF itself is derived from optical bands, not thermal ones. The combination of a 30-metre multispectral Landsat or Sentinel-2 classification with the co-registered Landsat TIRS thermal layer is a well-established analytical pairing. The thermal layer anchors the heat signal; the optical layer explains it.
The 16-day problem
Landsat 8 and Landsat 9 together achieve a revisit of roughly 8 days at mid-latitudes under clear skies. That qualifier matters. Cloud cover in humid tropical and temperate cities can render a majority of passes unusable for thermal retrieval. Over a European summer, a city might yield four to six cloud-free Landsat thermal acquisitions in three months. That is enough for seasonal characterisation but not for tracking the thermal response to a specific heatwave event.
ECOSTRESS partially fills this gap. Its revisit is irregular and orbit-dependent, but it has achieved multiple cloud-free passes over individual cities within a single week during heatwave periods. Sentinel-3 SLSTR, with daily coverage at 1 kilometre, provides the temporal continuity that neither Landsat nor ECOSTRESS can match, at the cost of spatial specificity. A practical analytical approach combines all three: SLSTR for daily temperature anomaly detection, Landsat TIRS for spatial disaggregation, and ECOSTRESS for diurnal cycle characterisation when passes align.
No current freely available thermal sensor simultaneously offers sub-100-metre resolution, daily revisit and full diurnal sampling. That is an honest statement of where the technology sits.
What the analysis actually produces
A typical urban heat island study outputs a gridded LST map for each valid acquisition, a mean summer LST composite, a temperature anomaly layer showing deviation from a rural reference, and a ranked list of heat-exposed census units or planning zones. These products are directly usable by urban planners, public health departments and infrastructure managers without specialist remote sensing knowledge.
Longer time series, ten years or more of Landsat thermal data, can detect statistically significant LST trends attributable to urban densification, large-scale green infrastructure programmes or changes in industrial activity. The archive depth is a genuine asset. Landsat thermal data extends back to Landsat 4 TM in 1982, though band calibration comparability across generations requires care.
Satellize runs this class of analysis on open Landsat and ECOSTRESS data for government and enterprise clients. The Tonga crop-estimation programme uses a related methodology of pairing spectral retrieval with field-truth calibration, and the same calibration discipline applies to urban thermal work. Ask for a sample LST composite for your city before committing to a full programme.
Typical figures
| Spatial resolution (Landsat TIRS) | 100 m native thermal, 30 m resampled product |
| Spatial resolution (ECOSTRESS) | 70 m |
| Spatial resolution (Sentinel-3 SLSTR) | 1 km (thermal channels) |
| Revisit (Landsat 8 + 9 combined) | ~8 days at mid-latitudes, cloud-free acquisitions typically 4–6 per summer season |
| Overpass time (Landsat) | Fixed ~10:00 local solar time (sun-synchronous, descending node) |
| Overpass time (ECOSTRESS) | Variable across full diurnal cycle; precesses ~3 minutes per day relative to local solar time |
| Thermal bands used for split-window | Landsat TIRS Band 10 (10.6–11.19 µm) and Band 11 (11.50–12.51 µm); SLSTR S8 (10.85 µm) and S9 (12.0 µm) |
| LST retrieval accuracy (split-window, clear sky) | Typically ±1–2 K under good atmospheric conditions; degrades with high water vapour or uncertain emissivity |
| Archive depth | Landsat thermal from 1982 (TM); ECOSTRESS from 2018; ASTER from 2000 to ~2023 |
| Delivery formats | GeoTIFF LST grids, NetCDF time series, vector anomaly zones, PDF summary reports |
Analytics Satellize can run
| Mean summer LST composite | Per-pixel averaging of cloud-masked Landsat TIRS acquisitions over a defined season, split-window atmospheric correction applied per scene | GeoTIFF raster at 30 m, georeferenced to city boundary, with rural reference baseline |
| Urban heat island intensity map | LST anomaly relative to rural buffer zone, computed per acquisition and aggregated; method follows published USGS Landsat LST product documentation | Ranked zone shapefile with mean UHI intensity per administrative or planning unit |
| Diurnal temperature cycle profile | Multi-pass ECOSTRESS LST extraction at fixed urban polygons, assembled into time-of-day profiles across available acquisitions | CSV time series and chart report showing daytime peak and nocturnal retention by land-cover class |
| LST trend analysis (decadal) | Linear regression of annual mean summer LST against year, using full Landsat archive; Mann-Kendall trend test for significance | Trend magnitude raster (°C per decade) and statistical significance layer, with written interpretation |
| Heat-exposure vulnerability index | Combination of LST anomaly layer with population density and impervious surface fraction; weighted composite following published urban heat vulnerability frameworks | GIS polygon layer with ranked priority zones for intervention planning |
| Cool-roof or greening intervention monitoring | Pre/post LST difference maps at 30 m, controlling for meteorological conditions using concurrent SLSTR regional temperature anomaly | Before-and-after report with temperature delta statistics for intervention footprint versus control areas |
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.