Temperature inversion detection for trapped urban pollution health alerts

Thermal inversions cap vertical mixing and compress pollutants into a shallow surface layer that satellite column measurements routinely understate. Combining MODIS/VIIRS land surface temperature, ERA5 boundary-layer height, and TROPOMI NO2 columns makes the trap visible and correctable.

Sensors

MODIS MOD11A1 (Terra/Aqua): Land surface temperature at 1 km resolution, twice-daily overpass (Terra ~10:30 local, Aqua ~13:30 local). Night-time retrievals are particularly useful for detecting the cold surface pools that anchor nocturnal inversions. Cloud cover invalidates individual pixels; clear-sky bias is a known limitation.
VIIRS LST (Suomi-NPP / NOAA-20): Land surface temperature at 750 m resolution with early-morning (~01:30 local) and early-afternoon (~13:30 local) overpasses. Finer spatial detail than MODIS makes it better suited to resolving intra-urban temperature gradients that indicate where cold-air pooling is strongest.
ERA5 atmospheric reanalysis (ECMWF): Hourly global atmospheric profiles at ~31 km horizontal resolution, including boundary-layer height (BLH), temperature at multiple pressure levels, and wind. BLH is the primary mixing-height input for column-to-surface conversion. ERA5 is reanalysis, not a direct measurement; BLH estimates carry uncertainty of ±200–500 m in stable nocturnal conditions.
Sentinel-5P TROPOMI NO2: Tropospheric NO2 vertical column density at 3.5 × 5.5 km (post-August 2019 upgrade) with daily global coverage. Provides the pollutant load that needs correcting. Column measurements integrate the full tropospheric depth; surface concentrations can be several times higher than the column implies when BLH collapses below 200 m.

Why a shallow sky is a public health emergency

A temperature inversion is simply warmer air sitting above cooler air, reversing the normal lapse rate and suppressing the buoyancy that drives vertical mixing. The result is a lid. Emissions from traffic, industry and heating accumulate in a layer that can be as shallow as 50–200 metres during strong nocturnal or winter inversions. The same mass of NO2 that would be diluted through a 1,500-metre well-mixed boundary layer is instead compressed into a fraction of that depth, multiplying surface concentrations by a factor of five to fifteen.

Satellite instruments measuring column-integrated NO2 do not see this compression directly. A 3.5 km pixel integrating the full tropospheric column will record a modestly elevated value while the street below exceeds hourly health thresholds by a wide margin. Without a mixing-height correction, column data alone will consistently understate acute exposure events. That understatement is not a minor technical quibble; it is the difference between issuing a health alert and not issuing one.

Reading the inversion from orbit

The thermal signature of an inversion is detectable before the pollution peak arrives. MODIS and VIIRS land surface temperature retrievals capture the nocturnal radiative cooling that creates surface cold pools in valleys, basins and street canyons. When LST drops sharply relative to surrounding elevated terrain, and ERA5 confirms a low boundary-layer height, the physical preconditions for pollutant trapping are in place. This two-variable check, cold surface plus collapsed BLH, is more reliable than either signal alone.

The operational workflow runs in three steps. First, VIIRS or MODIS night-time LST is compared against a climatological baseline for the same calendar period to flag anomalous surface cooling. Second, the ERA5 BLH field for the same hour is extracted; values below roughly 300 m in urban areas are treated as inversion-probable. Third, the morning TROPOMI overpass NO2 column is divided by the ERA5 BLH to produce a first-order surface concentration estimate in units of µg/m³, which can then be compared against national air quality thresholds. The method follows the general approach described in published literature on boundary-layer height corrections to satellite NO2, including work using the DOMINO and QA4ECV retrieval products.

Honest caveats apply throughout. ERA5 BLH in stable nocturnal conditions is one of the weaker fields in the reanalysis; the model tends to overestimate mixing height during strong inversions, which means the correction can underestimate surface concentrations rather than overestimate them. Cloud cover, which is itself common during inversion-prone synoptic patterns in some climates, blocks both LST and TROPOMI retrievals. On heavily overcast days the system must fall back to ERA5 alone, reducing confidence.

What a floating roof gives away

One underused signal is the spatial pattern of LST within a city. During a strong inversion, the urban heat island partially counteracts surface cooling in dense built-up cores, while peripheral valleys and parks cool most aggressively. VIIRS at 750 m resolves this intra-urban gradient well enough to identify which neighbourhoods are sitting deepest inside the cold pool and therefore experiencing the worst pollutant compression. This spatial differentiation matters for targeted health alerts: a city-wide advisory is less actionable than a ward-level one.

The aerosol optical depth (AOD) signal from MODIS MAIAC or VIIRS can corroborate the diagnosis. During inversion episodes, AOD in the boundary layer rises sharply even when the total column is not extreme, because the aerosol is confined to a thin layer with high optical density per unit depth. A joint LST-BLH-AOD flag therefore provides a more defensible trigger for alert issuance than any single variable.

Converting columns to concentrations: the arithmetic and its limits

The surface concentration estimate used in health-alert systems is typically expressed as: C_surface ≈ Ω / H, where Ω is the tropospheric column density (molecules/cm²) from TROPOMI and H is the mixing height (m) from ERA5, with appropriate unit conversions to µg/m³. This is a zeroth-order approximation. It assumes the pollutant is uniformly distributed within the boundary layer, which is physically reasonable during well-mixed daytime conditions but less so immediately after an overnight inversion begins to break.

More sophisticated implementations apply a vertical shape factor derived from chemical transport model profiles, effectively weighting the column by the fraction estimated to lie below the BLH. Studies using WRF-Chem or GEOS-Chem profiles have shown that shape-factor corrections improve the correlation between satellite-derived and ground-monitored NO2 by 10–30 percentage points in urban settings, though the improvement is dataset- and city-specific. Ground-truth validation against a dense surface monitor network remains the only reliable way to calibrate the correction for a specific city's topography and emissions mix.

Revisit frequency is a real operational constraint. TROPOMI provides one overpass per day, typically mid-morning local time. Inversions that form overnight and break by mid-morning may be missed entirely; those that persist into the afternoon are captured but only once. Combining TROPOMI with geostationary NO2 data, such as that from GEMS over East Asia or Sentinel-4 when operational, will eventually close this gap, but neither is yet globally available.

Putting the alert system together

A practical near-real-time alert pipeline ingests ERA5 operational analysis (available with roughly a five-hour lag), VIIRS LST from the NOAA-20 morning pass, and the TROPOMI NO2 Level-2 product from the Copernicus Data Space. The three inputs are co-registered to a common grid, the BLH correction is applied, and grid cells exceeding a configurable threshold are flagged. Alert polygons are generated, attributed with estimated surface NO2 in µg/m³, and pushed to a public health operations centre via GeoJSON or WMS feed.

Satellize applies this pipeline as part of its satellite-data analytics service, running on open constellations and validated against ground monitor records where clients operate surface networks. For cities without dense ground networks, which is the majority of cities in lower-income countries, the satellite-derived estimate is often the only timely sub-regional signal available, making its honest uncertainty characterisation especially important.

Typical figures

LST spatial resolution	MODIS MOD11A1: 1 km. VIIRS (NPP/NOAA-20): 750 m.
LST revisit	Up to 4 overpasses/day combining Terra, Aqua, NPP and NOAA-20; cloud cover reduces usable frequency significantly.
TROPOMI NO2 spatial resolution	3.5 × 5.5 km (post-August 2019 reprocessing); daily global coverage.
TROPOMI NO2 latency	Near-real-time product available within ~3 hours of overpass; offline product within 5 days.
ERA5 boundary-layer height resolution	~31 km horizontal, hourly temporal; BLH uncertainty ±200–500 m in stable nocturnal conditions.
ERA5 operational analysis lag	Approximately 5 hours behind real time for operational product; full reanalysis available to 1940.
Minimum detectable inversion signal	BLH collapse to <300 m detectable in ERA5; LST anomaly threshold typically set at ≥3 K below climatological mean for the calendar date.
Archive depth	MODIS LST from 2000; VIIRS from 2012; TROPOMI from May 2018; ERA5 from 1940.
Output formats	GeoTIFF (gridded fields), GeoJSON (alert polygons), WMS/WMTS tile feed, CSV (time-series per district).

Analytics Satellize can run

Inversion probability map	Joint threshold classification on VIIRS LST anomaly and ERA5 BLH; rule-based with configurable thresholds per city climatology.	Daily GeoTIFF and GeoJSON polygon layer, gridded at 750 m, flagging inversion-probable cells.
BLH-corrected surface NO2 estimate	TROPOMI tropospheric column divided by ERA5 BLH with unit conversion to µg/m³; optional vertical shape-factor weighting from published CTM climatologies.	Daily gridded GeoTIFF at TROPOMI native resolution (~3.5 km), attributed with estimated surface concentration and uncertainty band.
Acute health-alert polygons	Thresholding of BLH-corrected NO2 against national or WHO guideline values (e.g. WHO 2021 interim target: 25 µg/m³ 24-hour mean); spatial union of exceedance cells into alert zones.	GeoJSON alert polygons with severity tier, estimated population exposed (from WorldPop grid), and timestamp; push-compatible with WMS or API feed.
Inversion episode climatology	Multi-year ERA5 BLH and LST archive analysis to derive frequency, duration and seasonal distribution of inversion events per urban area.	Annual summary report with monthly exceedance statistics and mapped hotspot districts; suitable for city air quality management plans.
Ground-monitor calibration report	Regression of BLH-corrected satellite NO2 against client-supplied surface monitor time series; bias and RMSE quantification by season and inversion class.	Calibration coefficients and residual diagnostics delivered as a PDF technical report and updated lookup table for operational pipeline.
Inversion-onset early warning	ERA5 operational forecast BLH combined with ECMWF medium-range temperature profile forecast to predict inversion onset 12–24 hours ahead; alert triggered when forecast BLH falls below threshold.	Twice-daily GeoJSON forecast alert with lead time, confidence tier and recommended public health action level.

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.