Urban heat island intensification and population vulnerability during heatwaves

Landsat-9 TIRS-2 and ECOSTRESS resolve intra-urban land surface temperature to 70–100 m, exposing the thermal inequality that determines who bears the greatest physiological risk during a heatwave. Knowing the gradient is the first step toward targeted intervention.

Sensors

Landsat-9 TIRS-2: Two thermal infrared bands centred at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in distributed products), 16-day repeat at the equator. The split-window pair allows atmospheric correction for water-vapour effects, improving land surface temperature retrieval accuracy to roughly ±1–2 K under clear skies.
ECOSTRESS (ISS-mounted): Five-band thermal radiometer covering 8.28–12.13 µm, approximately 70 m × 70 m ground sample distance. Because it rides the International Space Station rather than a sun-synchronous orbit, overpass times vary across a roughly 51.6° inclination precession cycle, giving acquisitions at different times of day over the same city across successive passes. This is its key advantage for diurnal heat-stress characterisation, though coverage of any single city on a given day is not guaranteed.
MODIS Terra/Aqua: Thermal bands at 1 km resolution, with Terra overpass near 10:30 local time and Aqua near 13:30. Daily revisit makes MODIS the workhorse for multi-year climatological urban heat island baselines and anomaly detection during prolonged heat events, despite resolution too coarse to distinguish individual city blocks.
Sentinel-3 SLSTR: Dual-view thermal radiometer at 1 km resolution, revisit under two days for most mid-latitude cities. Useful for regional heat-dome extent and sea-surface temperature context, but shares the resolution limitation of MODIS for fine-grained intra-urban work.

What the surface energy balance reveals about a city

Urban materials absorb more solar radiation than vegetated land and release it slowly as longwave radiation through the evening. Low albedo surfaces such as asphalt and dark roofing absorb a greater fraction of incoming shortwave energy. Impervious cover eliminates evapotranspiration, which would otherwise consume latent heat and cool the surface. The result is a land surface temperature that can run 5–12 K above a nearby vegetated fringe on a calm, sunny afternoon, a differential well within the detection capability of TIRS-2 and ECOSTRESS.

Land surface temperature (LST) is not the same as air temperature at two metres, the quantity most people associate with heat stress. LST is the radiometric temperature of the surface itself, which can exceed ambient air temperature by 20–40 K on exposed asphalt. The relationship between LST and the physiological heat load experienced by a person standing on that surface is indirect but consistent: high LST drives upward sensible heat flux, elevating the local air temperature and mean radiant temperature that together determine the wet-bulb globe temperature relevant to human health. LST from satellite is therefore a reliable spatial proxy for differential exposure, not a direct health metric.

The diurnal problem: when a polar orbit looks down

Landsat-9 crosses the equator at approximately 10:00 local solar time, descending. For most cities this means TIRS-2 captures mid-morning conditions, before peak surface heating. That is useful for establishing a baseline and detecting areas that are already anomalously warm by mid-morning, but it misses the afternoon maximum and the dangerous overnight retention of heat that prevents physiological recovery. A single Landsat overpass during a heatwave event is informative; it is not a complete picture of the thermal day.

ECOSTRESS partially addresses this by sampling at varying local times as the ISS orbit precesses. Over a multi-week heatwave, successive ECOSTRESS acquisitions of the same city may span early morning, afternoon and evening, allowing analysts to reconstruct something closer to a diurnal profile. The caveat is that acquisition on any specific day at a specific time cannot be scheduled; it is a function of ISS ground track geometry. Cloud cover compounds the constraint: thermal infrared cannot penetrate cloud, so a week of convective afternoon cloud during a heatwave can blank out the most critical sampling window entirely.

Mapping who is exposed, not just where it is hot

A temperature map becomes a vulnerability map when it is intersected with population data. The analytical step is straightforward in principle: derive LST from atmospherically corrected thermal imagery, classify the urban area into temperature percentile bands, then overlay gridded population estimates such as WorldPop or LandScan and available socioeconomic indicators. Districts in the upper temperature quartile with high population density and low green-space fraction represent the highest-priority intervention zones.

The socioeconomic correlation is not accidental. Lower-income urban areas in many cities are characterised by higher building density, less tree canopy, older building stock with poor insulation, and fewer air-conditioned spaces. These are also the populations least able to self-evacuate to cooler environments. Satellite-derived canopy cover from multispectral imagery (NDVI from Landsat OLI or Sentinel-2 MSI) combined with LST produces a paired diagnostic: where canopy is low and LST is high, the intervention case for tree planting or cool-roof programmes is spatially explicit and quantified.

One honest limit: satellite LST captures the outdoor thermal environment. It does not capture indoor temperatures, which are determined by building construction, orientation and mechanical cooling access. Indoor heat mortality risk requires ground-truth data that satellite imagery cannot supply.

Building a multi-year heat climatology before the next event

Landsat's archive extends to 1972, with consistent thermal data from Landsat 5 TM onwards (1984). MODIS provides daily thermal observations from 2000. This depth allows analysts to establish a city's historical LST distribution, identify which districts have consistently been hottest across multiple summers, and quantify whether the urban heat island has intensified over time as the city has grown and green space has contracted.

A climatological baseline built from 20 years of MODIS and 10 years of Landsat 8 and 9 data gives emergency planners something more useful than a single heatwave snapshot: it tells them which areas are structurally hot, not just hot today. That distinction matters for capital investment decisions such as cool-roof subsidies or park placement, which operate on decade-long timescales.

What the method cannot do

Sixteen-day revisit from a single Landsat satellite means that a short, intense heatwave lasting three to five days may produce zero cloud-free acquisitions. ECOSTRESS improves the odds but does not guarantee coverage. For real-time operational heat-alert systems, satellite LST is a supporting input, not the primary trigger. Ground-based weather station networks and numerical weather prediction models carry that load; satellite data contextualises and spatialises their outputs.

Atmospheric correction of thermal imagery requires accurate water-vapour profiles, typically sourced from radiosonde data or reanalysis products. Errors in water-vapour input propagate into LST retrieval errors that can reach 2–3 K in humid conditions, which is meaningful when the temperature difference between a park and an adjacent street may be only 4–6 K. Analysts should report retrieval uncertainty alongside LST estimates, not present thermal maps as if they were precise thermometer readings.

Satellize runs this analytical workflow on open Landsat and ECOSTRESS archives and can integrate commercial very-high-resolution multispectral data for canopy and albedo inputs where city-scale detail is required.

From thermal map to operational planning product

The deliverable that city planners and public-health agencies actually need is not a false-colour temperature image. It is a ranked list of priority zones with population counts, a quantified green-space deficit, and a change layer showing whether each zone has grown hotter or cooler over the past decade. That product is built from the same satellite inputs described above, processed through a reproducible pipeline that can be rerun each summer to track whether interventions are working.

Cooling centre placement, emergency medical pre-positioning and targeted welfare checks for elderly residents all benefit from a spatially explicit vulnerability index rather than a city-wide average temperature. The satellite data does not make those decisions; it removes the excuse that the geography of heat exposure is unknown.

Typical figures

Best spatial resolution (LST)	~70 m (ECOSTRESS); 100 m native / 30 m resampled (Landsat-9 TIRS-2)
Revisit (Landsat-9)	16 days at equator; ~8 days combined with Landsat-8
Revisit (ECOSTRESS)	Variable; ISS precession cycle of ~3 days average but no guaranteed daily coverage
Revisit (MODIS Terra+Aqua)	Daily, two fixed local overpass times (~10:30 and ~13:30)
Thermal spectral bands	TIRS-2: Band 10 (10.9 µm), Band 11 (12.0 µm); ECOSTRESS: 8.28–12.13 µm (5 bands)
LST retrieval accuracy (clear sky)	±1–2 K (TIRS-2 split-window); ±1.5 K (ECOSTRESS, published mission spec)
Cloud penetration	None; thermal infrared is blocked by cloud. Microwave sensors can see through cloud but at 10–25 km resolution, unsuitable for intra-urban mapping
Archive depth	Landsat thermal: 1984–present (TM); MODIS: 2000–present; ECOSTRESS: 2018–present
Latency (open archive)	Landsat: ~6–12 hours after acquisition to USGS distribution; ECOSTRESS: hours to days via NASA Earthdata
Minimum detectable temperature contrast	~0.3 K (TIRS-2 noise-equivalent temperature difference); practical intra-urban mapping threshold ~1 K

Analytics Satellize can run

Intra-urban LST percentile map	Split-window atmospheric correction (TIRS-2 Bands 10+11) or temperature-emissivity separation (ECOSTRESS); emissivity estimated from NDVI-based vegetation fraction	GeoTIFF LST layer at 30–100 m resolution, classified into decile bands, with metadata on retrieval uncertainty
Population-weighted heat-exposure index	Zonal statistics overlay of LST percentile layer with WorldPop or LandScan gridded population; optional weighting by age-structure or deprivation index	District-level ranked table and choropleth map; CSV for integration into emergency management GIS
Green-space deficit and cool-surface fraction	NDVI from Landsat OLI or Sentinel-2 MSI co-registered to thermal layer; per-pixel classification of vegetated, impervious and water surfaces	GIS polygon layer of priority planting or cool-roof intervention zones with area and estimated LST reduction potential
Decadal heat-island trend analysis	Time-series regression on annual summer-maximum LST composites from MODIS MOD11 and Landsat archive; Mann-Kendall trend test per pixel	Change map showing statistically significant warming or cooling trends per district; PDF summary report
Diurnal LST profile (ECOSTRESS multi-pass)	Collation of all available ECOSTRESS acquisitions over a defined event window; local solar time assigned per pass; interpolated diurnal curve per zone	Time-series chart per priority zone showing morning, afternoon and nocturnal LST; GIS layer of overnight heat-retention hotspots
Cooling-centre placement optimisation	Spatial optimisation (maximum coverage or p-median model) using heat-exposure index and walking-distance accessibility network	Recommended facility locations ranked by population served per unit, with coverage maps

Sources

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.