Wet-bulb globe temperature estimation for outdoor worker heat stress

Wet-bulb globe temperature integrates heat, humidity, wind and solar load into the occupational standard for outdoor work limits. Satellite inputs from MODIS, VIIRS, CERES and ERA5 can estimate gridded WBGT where no ground station exists, but a ±2–3°C uncertainty is real and must be declared to any user making work-rest decisions.

Sensors

MODIS Terra / Aqua (Land Surface Temperature): Retrieves LST at 1 km spatial resolution using thermal infrared bands 31 and 32 (10.78–12.27 µm). Terra overpass near 10:30 local, Aqua near 13:30 local, giving two daytime snapshots. Instantaneous skin temperature, not air temperature; a conversion step is required. Cloud contamination is the dominant gap: MODIS reports quality flags but affected pixels must be excluded or gap-filled.
VIIRS SNPP (Land Surface Temperature): Comparable thermal retrieval to MODIS at 375 m resolution in the I-band thermal channel (I5, 10.5–12.4 µm), with a single daily overpass near 13:30 local. The finer pixel size improves urban and field-edge discrimination but does not resolve the fundamental LST-to-air-temperature ambiguity.
CERES (Clouds and the Earth's Radiant Energy System) on Terra / Aqua: Measures top-of-atmosphere and surface shortwave and longwave fluxes. Surface downwelling shortwave irradiance from CERES SYN1deg products is provided at 1° spatial resolution and hourly temporal resolution. This is the most defensible satellite source for the solar radiation component of WBGT, though 1° is coarse for field-level work.
ERA5 Reanalysis (ECMWF): Not a satellite sensor but an essential complement: ERA5 provides 2 m air temperature, 2 m dewpoint (from which vapour pressure and relative humidity are derived), and 10 m wind speed at ~31 km horizontal resolution and hourly frequency. It assimilates satellite radiances alongside surface observations. Its humidity fields are generally more reliable than any single-satellite humidity retrieval at the spatial scales relevant to occupational health.

What WBGT actually measures, and why that matters for satellite estimation

The wet-bulb globe temperature is not a single physical quantity. It is a weighted composite: 0.7 × natural wet-bulb temperature (capturing humidity and air movement), 0.2 × black-globe temperature (capturing radiant heat load), and 0.1 × dry-bulb air temperature. The ISO 7243 standard and the American Conference of Governmental Industrial Hygienists both use it to set work-rest schedules for outdoor labour. At 28°C WBGT under moderate physical work, rest ratios begin to change. At 32°C, continuous heavy work is contraindicated.

Each component of WBGT maps imperfectly onto what satellites observe. Satellites measure radiance from the surface or from atmospheric columns. They do not measure the temperature of a wetted wick in the shade, nor the temperature of a 150 mm black globe suspended in ambient air. Every satellite-derived WBGT estimate therefore involves a chain of physical assumptions and empirical conversions, and each step introduces error. Declaring that chain honestly is not a weakness in the analysis; it is the analysis.

How the inputs are assembled from orbit

The standard approach in the published literature converts MODIS or VIIRS LST to near-surface air temperature using statistical or physical models, often trained against weather-station records. A commonly cited relationship uses a linear regression of LST against station-measured air temperature, with corrections for land cover and viewing angle. Errors at the pixel level typically run 1.5–2.5°C RMSE under clear-sky conditions, rising sharply near cloud edges.

The natural wet-bulb temperature is then estimated from air temperature and vapour pressure, the latter drawn from ERA5 dewpoint fields. CERES SYN1deg surface downwelling shortwave irradiance feeds the globe temperature estimate, sometimes supplemented with ERA5 surface solar radiation. Wind speed comes almost entirely from ERA5, since no operational satellite currently retrieves 10 m wind over land with the spatial resolution needed for field-level heat stress assessment. Scatterometers such as ASCAT retrieve ocean surface wind but are not applicable here.

The resulting WBGT grid is typically available at 1 km or coarser spatial resolution, with a latency of several hours to one day depending on how quickly MODIS LST granules, CERES daily composites and ERA5 analysis fields are ingested. Near-real-time applications require accepting ERA5 short-range forecasts rather than final reanalysis, which adds further uncertainty.

The ±2–3°C problem, and when it becomes a safety issue

Published validation studies comparing satellite-derived WBGT to collocated ground measurements consistently report RMSE in the range of 2–3°C under clear skies, with larger errors under partial cloud and in complex terrain. A 2020 study published in the International Journal of Environmental Research and Public Health (Zhao et al., using MODIS LST over China) reported RMSE of approximately 2.4°C against station-measured WBGT. Similar figures appear in work using ERA5 alone, suggesting the reanalysis humidity and wind fields are often the binding constraint, not the LST retrieval.

Two or three degrees sounds modest. At the WBGT thresholds where ISO 7243 mandates a shift from continuous work to a 75/25 work-rest cycle, it is not. A site reading 26°C on the satellite product could genuinely be 28–29°C on the ground. Conversely, a product showing 30°C might reflect 27°C in reality. Neither error is acceptable if the product is used as the sole basis for stopping work. The correct use of satellite WBGT is as a spatial screening tool: identifying high-risk zones and times where ground measurement or worker monitoring should be prioritised, not as a replacement for a calibrated instrument at the worksite.

Where satellite coverage genuinely adds value

Ground weather stations are sparse in precisely the regions where outdoor heat stress is most acute. Large agricultural zones in sub-Saharan Africa, South and Southeast Asia, and parts of Latin America have station densities that make interpolated WBGT maps unreliable at the district level. Satellite-derived estimates, despite their uncertainty, can reveal spatial gradients that no ground network would capture: the difference in heat load between an irrigated paddy field and an adjacent dry-land plot, or between a coastal construction site and an inland one twenty kilometres away.

For military planners, agricultural employers and public health agencies assessing risk across large areas, a gridded WBGT product at 1 km resolution and daily frequency is qualitatively different from a handful of point observations. It allows prioritisation: which sites need portable wet-bulb instruments deployed, which shifts need rescheduling, which regions are approaching threshold conditions. That is a legitimate and valuable use. It is not a substitute for ground truth at the point of decision.

Satellize has applied similar multi-source satellite analytics to agricultural conditions in the Pacific, including the Tonga crop-estimation programme, and the same data fusion principles that underpin crop stress mapping transfer directly to heat stress spatial screening.

Practical limits the buyer should know before commissioning

Cloud cover is the hardest constraint. MODIS and VIIRS LST retrievals fail under cloud, and tropical agricultural and construction sites often experience afternoon convective cloud precisely when solar heat load peaks. Gap-filling with temporal compositing (using the most recent clear observation) introduces temporal error that can be substantial during multi-day cloud episodes. ERA5 can fill the gap for humidity and wind, but LST-derived air temperature becomes unavailable.

Spatial resolution is the second constraint. At 1 km, a pixel averages over multiple land cover types. A small construction site surrounded by vegetated land will have its LST pulled toward the cooler surroundings. VIIRS at 375 m improves this marginally, but fine-scale heterogeneity within a worksite remains unresolved. Landsat 8/9 thermal band (100 m resampled to 30 m) could in principle sharpen LST estimates, but its 16-day revisit makes it unsuitable for operational daily heat stress monitoring.

Finally, the black-globe temperature component of WBGT is sensitive to direct beam solar radiation and reflected radiation from surrounding surfaces. Satellite irradiance products capture the downwelling component but not the complex reflected and re-emitted environment that a globe thermometer integrates in practice. This is a structural limitation of the satellite approach, not a data quality issue that better sensors will fully resolve.

Typical figures

LST spatial resolution	1 km (MODIS), 375 m (VIIRS I-band)
LST revisit (daytime)	2 passes per day (Terra ~10:30, Aqua ~13:30 local); VIIRS SNPP adds a third near 13:30
Solar irradiance resolution	CERES SYN1deg: 1° (~111 km); hourly temporal resolution
Humidity and wind resolution	ERA5: ~31 km horizontal, hourly; final reanalysis latency ~5 days; forecast fields available in near-real-time
Derived WBGT grid resolution	Typically 1 km (LST-limited); coarser where CERES or ERA5 dominate
WBGT uncertainty vs ground instrument	±2–3°C RMSE under clear sky; higher under partial cloud or complex terrain
Cloud-affected data loss	LST retrieval fails under cloud; temporal gap-filling required; tropical wet-season coverage substantially degraded
Archive depth	MODIS from 2000; VIIRS SNPP from 2012; ERA5 from 1940; CERES from 2000
Latency (operational product)	4–24 hours depending on ERA5 forecast vs reanalysis and MODIS granule availability
Output formats	GeoTIFF gridded raster, NetCDF time series, tabular CSV per administrative or worksite polygon

Analytics Satellize can run

Daily gridded WBGT estimate	LST-to-air-temperature regression (published split-window retrieval), ERA5 humidity and wind, CERES surface irradiance; ISO 7243 WBGT formula	GeoTIFF raster at 1 km, delivered daily with embedded uncertainty flag per pixel
Threshold exceedance alert	Pixel-level comparison of derived WBGT against ISO 7243 action levels (28°C, 30°C, 32°C for moderate work); spatial clustering to suppress isolated noisy pixels	Daily alert shapefile or GeoJSON identifying zones at or above each threshold, with cloud-mask caveat layer
Seasonal heat stress calendar	Multi-year MODIS/ERA5 climatology; percentile ranking of daily WBGT by location and calendar week	PDF or interactive chart showing historical frequency of threshold exceedance by week for a defined worksite or district
Spatial risk screening report	Comparison of derived WBGT across a portfolio of sites (agricultural blocks, construction zones, military training areas) to rank by heat exposure frequency	Ranked site table with WBGT statistics, ground-instrument deployment recommendations and confidence classification
Cloud-gap-filled WBGT composite	Temporal compositing using ERA5 air temperature as a fallback when LST is cloud-contaminated; composite flagged to distinguish satellite-derived from reanalysis-filled pixels	NetCDF time series with provenance flag per pixel per day
Work-rest schedule input layer	WBGT grid clipped to user-defined worksite polygons; ISO 7243 work-rest table applied to derive recommended continuous work minutes per hour at four metabolic rate categories	Tabular CSV per site per day, formatted for integration into workforce scheduling systems

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.