Urban heat island intensity mapping for outdoor telecoms equipment thermal management
Landsat 8/9 TIRS and ECOSTRESS resolve land surface temperature at 30–70 m, identifying urban micro-zones where persistent heat stress pushes outdoor telecoms equipment beyond rated thermal envelopes. LST is not air temperature; engineering use requires an aerodynamic resistance correction.
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 product; 16-day revisit per satellite, ~8 days combined; daytime and night-time passes available; free archive from 2013 (Landsat 8) and 2021 (Landsat 9) via USGS.
- ECOSTRESS (NASA/JPL, aboard ISS): Five thermal infrared bands, 70 m resolution, variable revisit of 1–5 days depending on ISS orbital precession; captures diurnal temperature variation that polar-orbiting sensors miss; particularly useful for identifying peak afternoon heat stress windows.
- Sentinel-3 SLSTR (ESA): Dual-view thermal sensor at 1 km resolution; too coarse for individual cabinet siting but useful for city-scale UHI intensity climatology and cloud-gap filling across multi-year time series; daily revisit.
- ASTER (NASA/METI, Terra): Five thermal infrared bands at 90 m resolution; on-demand tasking model limits systematic revisit, but the global emissivity database derived from ASTER is widely used to correct Landsat LST retrievals; archive extends to 2000.
What a hot roof actually tells an engineer
Land surface temperature (LST) retrieved from thermal infrared imagery is the radiative skin temperature of whatever material the sensor sees: bitumen membrane, concrete parapet, metal cabinet lid. It is not the air temperature at the height of a remote radio unit or the internal temperature of an outdoor cabinet. The gap matters. On a calm, sunny afternoon in a dense city, LST on a dark roof can run 20–30 °C above the ambient air temperature measured at screen height, and 10–15 °C above the air temperature at one metre above the surface, where convective exchange with equipment actually occurs.
The correction that bridges these quantities is grounded in the surface energy balance. Sensible heat flux from a surface to the air above it depends on the temperature difference divided by aerodynamic resistance, which is itself a function of wind speed, surface roughness and atmospheric stability. For engineering purposes, a simplified bulk aerodynamic formula using local wind climatology and a roughness length appropriate to rooftop or street-level installations gives a defensible estimate of the air temperature envelope an equipment enclosure will experience. Satellite LST provides the spatial input; the physics provides the translation.
Why impervious fraction and albedo are the two numbers that matter most
Urban heat island intensity correlates strongly with two surface properties that satellite imagery can quantify independently of thermal bands. Impervious surface fraction, derived from multispectral classification of Sentinel-2 or Landsat OLI imagery at 10–30 m, determines how much incoming solar radiation is absorbed and stored rather than evaporated. A surface that is 90 % impervious and dark has almost no latent heat pathway; virtually all absorbed shortwave radiation converts to sensible heat and longwave emission.
Albedo, the broadband shortwave reflectance, controls how much radiation is absorbed in the first place. A fresh white roof coating with albedo of 0.65–0.75 absorbs roughly a third of what a weathered bitumen surface at albedo 0.05–0.10 absorbs. Mapping albedo from Landsat OLI or Sentinel-2 using published narrowband-to-broadband conversion coefficients allows a network planner to rank candidate base-station sites by their expected thermal load before any ground survey. Sites with high impervious fraction and low albedo in dense urban canyons are the ones that will stress cooling systems hardest.
Sensors, resolution floors and the cloud problem
Thermal infrared imagery is useless through cloud. This is not a minor caveat in tropical and temperate cities. Landsat's 16-day revisit means that in monsoon climates or maritime cities with persistent summer overcast, acquiring a clear-sky image during the seasonal peak heat period can require compositing across several years of archive. ECOSTRESS partially addresses the revisit gap with its variable 1–5 day return, but the ISS orbit means coverage is not guaranteed on any specific date.
Resolution floors matter for equipment siting. At 100 m native TIRS resolution, a single Landsat pixel can straddle a rooftop edge and mix the thermal signature of the roof surface with the cooler street below, biasing the LST estimate downward. The 30 m resampled product helps presentation but does not recover information lost at the detector. ECOSTRESS at 70 m is marginally better for individual rooftops in cities with large building footprints. For sites where the relevant surface is a narrow street-level cabinet in a dense canyon, no current free-access thermal sensor resolves the geometry adequately; the satellite product then serves as a neighbourhood-level thermal index rather than a site-specific temperature estimate.
Emissivity uncertainty is a second source of error. LST retrieval requires knowing the surface emissivity, which varies between about 0.92 for concrete and 0.97 for vegetation. The ASTER Global Emissivity Database provides per-pixel emissivity estimates that Landsat Collection 2 LST products use by default, but urban surfaces change faster than the database updates. A freshly painted roof or a new metal cladding can shift emissivity enough to introduce 1–3 °C error in the LST retrieval.
Building the thermal risk layer for network planning
The practical workflow starts with a multi-year LST time series, selecting only clear-sky acquisitions during the hottest part of the diurnal cycle (Landsat overpass is typically around 10:00–10:30 local solar time, ECOSTRESS varies). From this stack, a 90th-percentile LST map identifies persistent thermal hotspots rather than single-day anomalies. A site that appears at the 90th percentile of LST across five years of summer acquisitions is genuinely exposed; a single-image spike may reflect a transient event.
That percentile map is then intersected with the planned or existing base-station footprint. Sites where the 90th-percentile LST exceeds a threshold derived from the aerodynamic resistance correction (typically 5–8 °C above the equipment's rated maximum ambient) are flagged for enhanced cooling specification review. The output is a ranked list of sites, not a binary pass/fail, because the thermal margin available to a given cabinet design varies by vendor specification. Satellize packages this as a GIS layer with per-site temperature exceedance statistics, delivered alongside the aerodynamic correction parameters so the network engineer can apply their own equipment thermal model.
Albedo and impervious fraction layers are included as explanatory variables. A site flagged for thermal risk because of low albedo may be addressable by roof treatment rather than upgraded cooling hardware, a distinction worth making before procurement.
Archive depth as a design input, not just a data source
Landsat's continuous archive from 1984 means that for most cities, a decade or more of thermal data predates the current network. That history is directly useful: it shows how LST at a given location has trended as the city has densified, and it allows a planner to ask whether a site that is marginal today will be comfortably within envelope or critically overheated in ten years as surrounding impervious fraction increases. The trend is not always upward at every location; urban greening programmes and cool-roof mandates in some cities have produced measurable LST reductions over the past decade in specific districts.
For new network deployments, particularly in cities undergoing rapid densification, the historical trend slope is arguably more useful than the current absolute LST. A site at 38 °C today with a trend of +0.4 °C per year crosses a 45 °C equipment limit within less than two decades, well within the expected service life of the infrastructure.
Typical figures
| Spatial resolution (thermal) | 70 m (ECOSTRESS), 100 m native / 30 m product (Landsat 8/9 TIRS), 90 m (ASTER), 1 km (Sentinel-3 SLSTR) |
| Revisit cadence | ~8 days combined Landsat 8+9; 1–5 days ECOSTRESS (ISS-dependent); daily Sentinel-3; on-demand ASTER |
| LST retrieval accuracy (clear sky) | ±1–2 °C under typical conditions; ±3 °C or more where emissivity is poorly constrained |
| Thermal spectral bands | Landsat TIRS: 10.6–11.2 µm and 11.5–12.5 µm; ECOSTRESS: 8.3–12.5 µm (5 bands); ASTER TIR: 8.1–11.7 µm (5 bands) |
| Archive depth | Landsat: 1984 to present (thermal from 1984 via TM/ETM+, TIRS from 2013); ECOSTRESS: 2018 to present; ASTER: 2000 to present |
| Cloud limitation | All thermal IR sensors blind through cloud; clear-sky compositing over multi-year archive required in high-cloud-frequency climates |
| Albedo / impervious fraction resolution | 10 m (Sentinel-2 OLI-equivalent), 30 m (Landsat OLI) for ancillary surface property layers |
| Data access | Landsat Collection 2 LST: free via USGS EarthExplorer; ECOSTRESS: free via NASA Earthdata; Sentinel-3: free via Copernicus Data Space |
| Delivery formats | GeoTIFF (LST raster, percentile composites), GeoPackage or Shapefile (site-level statistics), CSV (per-site exceedance table) |
Analytics Satellize can run
| 90th-percentile LST hotspot map | Multi-year clear-sky stack compositing using Landsat Collection 2 LST product and ECOSTRESS L2 LST; percentile reduction per pixel | GeoTIFF raster at 30–70 m resolution, city or corridor extent, with per-pixel 90th-percentile LST value |
| Per-site thermal exceedance score | Point extraction from LST percentile raster at planned or existing base-station coordinates; aerodynamic resistance correction applied using ERA5 wind climatology | CSV or GeoPackage table with estimated peak air temperature at equipment height, exceedance margin against user-supplied thermal envelope |
| Albedo and impervious surface fraction layer | Landsat OLI or Sentinel-2 multispectral classification using published narrowband-to-broadband albedo coefficients and spectral unmixing for impervious fraction | GeoTIFF pair (albedo, impervious fraction) at 10–30 m; used as explanatory and mitigation-planning layers alongside LST |
| LST trend analysis (historical densification effect) | Linear regression on annual 90th-percentile LST per pixel across Landsat archive; Mann-Kendall trend test for significance | GeoTIFF of LST trend slope (°C per year) and significance mask; site-level trend statistics in accompanying CSV |
| Thermal risk tier classification for site portfolio | Combination of LST exceedance margin, trend slope and impervious fraction into a ranked risk tier (high / medium / low) using thresholds agreed with client thermal engineers | Ranked site list in GeoPackage and PDF summary report; flagged sites include recommended cooling specification review trigger |
| Cool-roof intervention impact estimate | Albedo sensitivity analysis using surface energy balance parameterisation; modelled LST reduction from albedo increase scenarios (e.g. 0.10 to 0.65) at flagged sites | Per-site table of estimated LST reduction under roof treatment scenarios, formatted for input to equipment cooling system specification review |
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.