Sensors
- Landsat 8/9 TIRS: Two thermal bands (Band 10 at 10.6–11.2 µm, Band 11 at 11.5–12.5 µm) at 100 m native resolution, resampled to 30 m in products. 16-day exact repeat; split-window LST retrieval requires both bands and an emissivity input layer. Landsat 9 TIRS-2 has lower stray-light noise than the original TIRS-1.
- ECOSTRESS (ISS): Five thermal bands from 8.3 to 12.5 µm at approximately 70 m resolution. Non-sun-synchronous ISS orbit gives variable overpass times, which is scientifically valuable for capturing diurnal heat cycles but makes time-series comparison harder. Typical revisit is irregular, roughly 3–5 days for mid-latitudes.
- MODIS MOD11 LST (Terra/Aqua): Generalist workhorse for city-scale or regional heat mapping. 1 km spatial resolution, twice-daily overpass per satellite. Adequate for large metropolitan comparisons but cannot resolve intra-urban variation at neighbourhood scale.
- ASTER TIR (Terra): Five thermal bands at 90 m resolution, with on-board emissivity separation via the Temperature and Emissivity Separation (TES) algorithm. Revisit is not fixed (pointable instrument), so archive depth over any given city is sparse. Useful for high-quality single-date calibration scenes.
What a thermal band actually measures, and why that matters
Thermal infrared sensors do not measure air temperature. They measure the radiance emitted by the surface itself, which a retrieval algorithm converts to land surface temperature (LST). The distinction is critical for urban heat island work: a rooftop covered in dark bitumen at 55 °C and a tree canopy at 28 °C can sit within the same 100 m pixel, and the sensor records a mixture of both. What you get is a spatially weighted, emissivity-modulated composite, not a thermometer reading.
The split-window algorithm, used by both Landsat TIRS and MODIS MOD11, exploits differential atmospheric absorption between two adjacent thermal channels to correct for water vapour in the column above. The approach is well-validated and computationally tractable, but it propagates emissivity uncertainty directly into the LST estimate. Published error budgets for Landsat TIRS split-window retrievals typically cite ±1–2 K under clear-sky conditions, rising toward ±3 K where emissivity assignments are poorly constrained.
The emissivity problem over heterogeneous cities
Emissivity is the ratio of actual thermal emission to that of a perfect blackbody. Vegetation emissivity in the thermal infrared sits around 0.97–0.99. Dry concrete is closer to 0.92–0.95. Polished metal roofing can drop below 0.80. A city block mixes all three in proportions that change with season, rainfall, and redevelopment. Assign the wrong emissivity and the retrieved temperature shifts by 1–4 K, which is large enough to misclassify a neighbourhood as a heat island when it is not, or to understate intensity where it is real.
The standard mitigation is to derive emissivity from a land-cover classification or from a vegetation index (NDVI) using the NDVI-threshold method or the vegetation fraction method. Both introduce their own errors over impervious surfaces where NDVI is near zero regardless of actual thermal properties. ASTER's TES algorithm is more physically rigorous but requires the full five-band ASTER TIR stack, which limits operational use. For routine Landsat-based monitoring, emissivity maps derived from urban land-cover products are the pragmatic choice, with explicit uncertainty flagging over industrial and commercial zones.
Resolution floors and what they hide
The 100 m native resolution of Landsat TIRS is sufficient to distinguish a large park from adjacent streets, or a dense commercial district from a residential suburb. It is not sufficient to resolve individual street canyons, single buildings, or the thermal plume of a rooftop HVAC unit. Planners who want to site cooling interventions at the block level need to understand this limit before commissioning an analysis.
MODIS at 1 km is useful for city-to-city comparisons or for tracking heat island intensity across an entire metropolitan region over time. It cannot answer questions about specific neighbourhoods. ECOSTRESS at roughly 70 m is the best currently operational open-access option for sub-100 m work, but its irregular revisit means you cannot build a consistent monthly or seasonal time series without careful gap-filling. The practical answer for most urban planning clients is Landsat for time series and trend analysis, with ECOSTRESS scenes pulled for specific high-interest dates such as heatwave events.
Cloud cover is the other hard constraint. Thermal retrievals require clear sky. In humid tropical or monsoon-affected cities, cloud-free Landsat scenes can be rare during the hottest months, which are precisely the months of greatest interest. A cloud-compositing strategy using all available clear pixels within a season is the standard workaround, but it blurs the temporal signal.
Isolating the urban heat island signal from surface moisture
A wet surface cools by evaporation. A dry urban surface does not. This means that a city photographed one day after rain can appear cooler than surrounding dry farmland, temporarily inverting the expected heat island pattern. Surface moisture confounds LST-based heat island intensity (UHI-I) estimates unless the analysis explicitly accounts for it.
The standard approach is to pair LST with a shortwave-derived moisture index such as NDWI or NDMI from the same Landsat overpass, then either stratify the analysis by moisture class or apply a regression correction. Seasonal compositing reduces the problem by averaging across wet and dry days, but it also smooths out the extreme-heat events that are most operationally relevant. For heatwave response applications, single-date analysis with explicit moisture flagging is more honest than a composite that hides the worst days.
Building an intensity map that planners can actually use
Urban heat island intensity is conventionally defined as the LST difference between an urban pixel and a reference rural pixel under comparable land cover and moisture conditions. The choice of rural reference zone matters enormously. A city surrounded by irrigated agriculture will appear cooler relative to its surroundings than an identical city surrounded by dry scrubland, even if the two cities have identical physical properties. Analyses should document the reference zone selection explicitly.
A well-structured output for a planning client includes: a gridded LST map at the best available resolution for the period of interest; a derived UHI intensity layer showing the delta from the rural reference; a confidence or uncertainty layer flagging pixels where emissivity assignment is ambiguous; and a time-series chart of intensity for defined urban zones across available archive dates. Landsat's archive extends to 1972 for lower-resolution thermal data and to 1984 for continuous TIRS-class coverage, which is long enough to quantify multi-decade intensification trends in most cities. Satellize runs this class of analysis on open Landsat and ECOSTRESS archives, with the same methodological transparency it applies to its crop-estimation work in Tonga.
The output is not a map of where people feel hot. It is a map of surface radiative temperature under specific atmospheric conditions. Communicating that distinction to decision-makers, without losing their attention, is half the analytical job.
Typical figures
| Best spatial resolution (LST product) | 100 m native, 30 m resampled (Landsat 8/9 TIRS); ~70 m (ECOSTRESS); 90 m (ASTER TIR); 1 km (MODIS MOD11) |
| Revisit period | 16 days exact repeat (Landsat 8 or 9 individually); ~8 days with both combined; irregular 3–5 days (ECOSTRESS/ISS); twice daily per satellite (MODIS) |
| Thermal bands used | Landsat TIRS Band 10 (10.6–11.2 µm) and Band 11 (11.5–12.5 µm); ECOSTRESS Bands 3–5 (8.3–12.5 µm); MODIS Bands 31–32 (11 and 12 µm) |
| LST retrieval accuracy (clear sky) | ±1–2 K typical; ±3 K or worse over heterogeneous urban surfaces with uncertain emissivity |
| Cloud tolerance | None. Thermal retrieval requires clear sky; cloud masking is mandatory |
| Emissivity input requirement | NDVI-based or land-cover-derived emissivity map; ASTER TES for highest accuracy single dates |
| Archive depth | Landsat thermal back to 1984 (TM Band 6); MODIS from 2000; ECOSTRESS from 2018; ASTER from 2000 |
| Typical latency (open archive) | Landsat Collection 2 Level-2 LST product available within ~12 hours of acquisition via USGS |
| Minimum detectable UHI signal | Approximately 1–2 K above rural reference under clear-sky conditions, given retrieval uncertainty |
| Delivery formats | GeoTIFF (LST and UHI intensity grids), CSV time series, GIS-ready vector zones, PDF technical report |
Analytics Satellize can run
| Land surface temperature map | Split-window algorithm applied to Landsat TIRS Band 10/11 with NDVI-derived emissivity correction | GeoTIFF at 30 m, single date or seasonal composite, with per-pixel uncertainty layer |
| UHI intensity layer | LST delta between urban pixels and a defined rural reference zone, stratified by land cover and moisture class | GeoTIFF showing signed temperature difference from rural baseline; zonal statistics by administrative boundary |
| Multi-decade UHI trend analysis | Time-series regression on Landsat archive LST composites (1984 to present) for defined urban zones | Chart and CSV of annual or seasonal mean UHI intensity; statistical trend slope with confidence interval |
| Heatwave event surface temperature report | Single-date or multi-date ECOSTRESS or Landsat LST retrieval coincident with meteorological heatwave period | PDF report with mapped peak LST, hotspot identification, and comparison to climatological baseline |
| Green-infrastructure cooling assessment | Paired LST and NDVI analysis to quantify park cool-island radius and magnitude from Landsat archive | GIS layer showing distance-decay cooling effect around urban green spaces; tabular summary per park |
| Emissivity uncertainty flag layer | Variance mapping of NDVI-to-emissivity conversion confidence across urban land-cover classes | Raster mask identifying pixels where LST estimates carry elevated uncertainty, for quality-controlled downstream use |
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.