School-proximity NO2 exposure mapping for child respiratory burden

Children breathe the air outside their classrooms for hours each day, yet most cities lack the monitoring density to quantify that exposure school by school. TROPOMI, GEMS and road-proximity downscaling change that calculus.

Sensors

Sentinel-5P TROPOMI: Measures tropospheric NO2 columns in the 405–465 nm window at a native pixel size of 3.5 × 5.5 km (upgraded from 3.5 × 7 km in August 2019). Daily global coverage. Precision on a single overpass is roughly 0.7 × 10¹⁵ molecules/cm²; multi-year averaging reduces uncertainty substantially. The native resolution is far coarser than a school catchment, so statistical downscaling is required.
GEMS (Geostationary Environment Monitoring Spectrometer): Operated by the Korea Meteorological Administration from GEO orbit over Asia. Retrieves NO2 columns at approximately 3.5 × 8 km with hourly daytime cadence across East and South-East Asia. The high temporal frequency captures morning school-run peaks that a once-daily polar overpass misses entirely.
Sentinel-2 MSI: 10 m multispectral imagery used to derive built-up density, impervious surface fraction and road-network proximity. These spatial covariates anchor the statistical downscaling of coarse NO2 columns to sub-kilometre grids. Not an NO2 sensor itself; its role is purely as a land-cover and morphology input.
Copernicus Atmosphere Monitoring Service (CAMS) urban NO2 reanalysis: ECMWF's CAMS produces global atmospheric reanalysis fields at roughly 40 km resolution, blending satellite retrievals with chemistry-transport model output. Used as a bias-correction layer and to extend records back before TROPOMI's 2017 launch. Not a direct observation; model uncertainty should be carried through to final exposure estimates.

Why a 3.5 km pixel is still useful for a 50 m school

TROPOMI's native footprint covers several city blocks at once. That sounds disqualifying for school-level analysis, but the pixel is not the product. Multi-year NO2 column composites from TROPOMI carry spatial signal that correlates strongly with ground-level concentrations when road density, traffic volume proxies and building morphology are used as downscaling covariates. Published land-use regression studies, including work validated against dense sensor networks in European cities, consistently show that satellite-derived NO2 columns combined with road-proximity variables explain 60–80% of the variance in ground-level NO2 at scales of 100–500 m.

The practical workflow is: aggregate three to five years of TROPOMI overpass retrievals into a mean column map, apply a land-use regression or random-forest downscaling model trained on whatever ground monitors exist in the study city, then intersect the resulting sub-kilometre NO2 surface with a georeferenced school register. The output is a ranked list of facilities by estimated chronic NO2 exposure. That ranking, not an absolute concentration figure, is what drives intervention priority. Honest caveat: the absolute values carry uncertainty of 20–40% relative to co-located monitors in cities where validation has been done, so the method is most defensible as a relative screening tool.

What GEMS adds that a polar orbit cannot

Morning school-run traffic produces NO2 spikes that a single daily TROPOMI overpass, timed to local mid-morning, captures only partially. GEMS, covering Asia from geostationary orbit, retrieves NO2 roughly every hour during daylight. That temporal resolution makes it possible to reconstruct a diurnal exposure profile rather than a snapshot, which matters because a child arriving at school at 07:30 and leaving at 15:30 experiences a very different NO2 trajectory than the overpass moment implies.

Outside GEMS coverage, which currently spans roughly 5°S to 45°N and 75°E to 145°E, hourly variation must be inferred from CAMS chemistry-transport model output or from traffic-count data where available. This is a real geographic constraint. Cities in sub-Saharan Africa, Latin America and South Asia west of the GEMS swath rely on TROPOMI alone for satellite NO2, and the single-overpass limitation should be stated plainly in any exposure assessment delivered to health authorities.

Building the school-level exposure surface

The processing chain has four steps. First, quality-filter TROPOMI retrievals to remove cloud-contaminated pixels (the standard qa_value threshold of 0.75 removes scenes with cloud radiance fraction above roughly 50%). Second, compute a multi-year mean column map and apply the downscaling model using Sentinel-2-derived impervious surface fraction and road-network distance as covariates. Third, extract the estimated NO2 concentration at each school location, applying a buffer of 100–300 m to represent the outdoor play area rather than a single point. Fourth, rank schools by exposure percentile within the city.

School location data is not a satellite product. It must come from national or municipal education registries, OpenStreetMap, or field survey. Data quality here is often the binding constraint, particularly in rapidly growing cities where school registers lag physical construction. A missing or mislocated school is a worse error than a 20% uncertainty in the NO2 estimate above it.

What the health literature actually supports

The epidemiological link between traffic-related NO2 and childhood respiratory outcomes is well established in the peer-reviewed literature: reduced lung function growth, increased asthma incidence and exacerbation, and higher rates of respiratory infection. The WHO 2021 air quality guidelines set an annual mean NO2 standard of 10 µg/m³, down from the previous 40 µg/m³, reflecting accumulated evidence of harm at lower concentrations than previously recognised.

Satellite-derived exposure estimates have been used in burden-of-disease studies, but the causal inference chain from column retrieval to health outcome involves several modelling steps, each adding uncertainty. A satellite-derived school exposure ranking is a triage instrument. It tells a health authority which schools to prioritise for ground-level monitoring or filtration assessment. It does not replace an epidemiological cohort study, and presenting it as such would overstate what the data can support.

Limits that belong in every client report

Cloud cover is the most immediate operational limit. In tropical cities during monsoon season, TROPOMI retrieval rates can fall below 30% of days in a month. Multi-year compositing mitigates this but cannot recover information that was never observed. Cities with persistent cloud cover require longer averaging periods, which in turn means the exposure estimate reflects a historical mean rather than current conditions.

Vertical profile assumptions matter too. TROPOMI retrieves total tropospheric NO2 columns; converting those to surface concentrations requires an assumed vertical profile shape from a chemistry-transport model. In cities with strong local sources and complex topography, that assumed profile can introduce systematic bias. The CAMS reanalysis provides one correction, but it operates at 40 km resolution and will not capture street-canyon effects. Downscaling models trained on cities with dense ground networks should not be transferred without revalidation to cities with different building morphology or emission source mixes.

Satellize runs this downscaling pipeline on open TROPOMI and Sentinel-2 archives, with CAMS bias correction, and can deliver ranked school exposure layers as GIS outputs for city or national health authorities. The Tonga crop-estimation programme demonstrated the same principle of extracting actionable sub-pixel signals from open constellations in data-sparse settings.

From ranked list to intervention decision

The output of this analysis is most useful when paired with a decision rule. A reasonable one: schools in the top exposure decile, within 150 m of a road carrying more than a defined traffic threshold, and serving populations without existing monitoring, become candidates for portable NO2 sensor deployment or HEPA filtration assessment. The satellite layer identifies the candidates; ground instruments confirm the diagnosis.

Siting intervention is harder. Recommending that a school be relocated is a multi-decade decision involving land tenure, community ties and capital budgets that no satellite product can resolve. What the data can do is inform new school siting policy so that the next generation of facilities is not built into the same exposure trap. Several cities, including some in the United Kingdom, have begun incorporating air quality mapping into school planning applications. Satellite-derived NO2 surfaces are a cost-effective input to that process, particularly in cities where the ground monitoring network is too sparse to produce reliable interpolated maps on its own.

Typical figures

Native NO2 pixel size (TROPOMI)	3.5 × 5.5 km (post-August 2019 upgrade)
Native NO2 pixel size (GEMS)	Approx. 3.5 × 8 km; hourly daytime cadence over Asia
Downscaled output resolution	100–500 m, depending on covariate density and ground-monitor availability for training
TROPOMI revisit	Daily global coverage; single overpass per day at fixed local time (~13:30 equatorial)
Spectral bands used	TROPOMI: 405–465 nm (NO2 absorption); Sentinel-2 MSI: 10–60 m bands for built-up index
Cloud contamination threshold	Pixels with qa_value < 0.75 excluded; retrieval rates can fall below 30%/month in tropical monsoon seasons
Minimum meaningful averaging period	12 months for temperate cities; 24–36 months recommended in high-cloud-fraction regions
Absolute concentration uncertainty	20–40% relative to co-located monitors in validated urban studies; relative ranking more reliable than absolute values
TROPOMI archive depth	From May 2018 (science data); CAMS reanalysis extends back to 2003
Delivery formats	GeoTIFF exposure surface, GeoPackage school-ranked attribute table, PDF summary report

Analytics Satellize can run

City-wide downscaled NO2 exposure surface	Land-use regression or random-forest downscaling of multi-year TROPOMI composites using Sentinel-2 impervious surface fraction and road-distance covariates	GeoTIFF raster at 100–500 m resolution, with uncertainty band layer
School-level exposure ranking table	Zonal statistics extraction over 100–300 m school buffers from downscaled NO2 surface, percentile ranking within city	GeoPackage attribute table with school ID, estimated mean NO2, exposure percentile and road-proximity flag
Diurnal NO2 profile at school locations (Asia only)	Hourly GEMS column retrieval time series extracted at school coordinates, smoothed with LOESS filter	CSV time series and chart showing morning peak, midday and afternoon exposure windows
Priority intervention shortlist	Rule-based filter: top exposure decile, within 150 m of classified road, no existing ground monitor within 500 m	Ranked PDF report with map, facility metadata and recommended next step (sensor deployment or filtration assessment)
Historical exposure trend (2018 to present)	Annual TROPOMI composite sequence with CAMS bias correction, differenced to show improvement or deterioration	Multi-year GeoTIFF stack and trend summary chart per school
New-siting suitability layer	Inverse NO2 exposure surface combined with land-use and population density layers to identify low-exposure candidate zones for future school construction	GIS polygon layer with suitability score and planning-authority-ready PDF

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.