Tropospheric NO2 exposure mapping from orbit
Satellite spectrometers measure nitrogen dioxide columns daily at sub-city resolution, exposing gaps between reported emissions and actual population exposure. TROPOMI, GEMS and TEMPO together now offer both the spatial detail and the hourly diurnal cycle needed for credible public-health burden estimates.
Sensors
- Sentinel-5P TROPOMI: UV-visible-NIR-SWIR push-broom spectrometer; NO2 tropospheric column at 3.5 × 5.5 km pixel (upgraded from 7 × 3.5 km in August 2019); daily global coverage; ~15:00 local overpass; Level-2 products freely available within 3 hours of sensing via Copernicus dataspace.
- GEMS (GEO-KOMPSAT-2B): Geostationary UV-visible spectrometer covering East and Southeast Asia; NO2 columns at roughly 3.5 × 8 km nadir; scans the full disk every hour during daylight, enabling diurnal pollution cycles that polar orbiters cannot resolve. Operational since 2020.
- TEMPO (Tropospheric Emissions: Monitoring of Pollution): NASA geostationary instrument aboard Intelsat 40E covering North America; sub-hourly NO2 columns at approximately 2 × 4.5 km; operational since 2023. First geostationary pollution spectrometer over the Americas, resolving morning rush-hour peaks invisible to TROPOMI.
- Ozone Monitoring Instrument (OMI, Aura): Predecessor to TROPOMI; 13 × 24 km nadir pixel; daily global coverage since 2004. Archive depth is its primary value now: nearly two decades of consistent NO2 retrievals underpin long-term trend analysis and epidemiological cohort studies.
What the spectrometer actually measures, and what it does not
Every DOAS retrieval begins with sunlight. The instrument records solar radiation backscattered from the Earth's surface and atmosphere across the UV-visible window, roughly 405 to 465 nm for NO2. Nitrogen dioxide absorbs at characteristic wavelengths with a known cross-section. DOAS fits the measured slant-column absorption against reference spectra, corrects for the photon path length using an air-mass factor derived from radiative-transfer modelling, and separates the tropospheric column from the stratospheric background using assimilated stratospheric fields.
The result is a tropospheric vertical column density, expressed in molecules per square centimetre, typically in the range of 1 to 30 × 10¹⁵ mol/cm² over polluted urban areas. What it is not is a surface concentration. The column integrates NO2 through the entire tropospheric depth. On a day with strong vertical mixing, boundary-layer NO2 dominates. Under a temperature inversion, the same column value could represent a far more dangerous near-surface accumulation. Separating these regimes requires either coincident meteorological profiles or chemical-transport model output, and that introduces its own uncertainty, typically 20 to 50 percent in surface-layer attribution depending on boundary-layer height assumptions.
Why one overpass a day is the wrong tool for exposure assessment
TROPOMI crosses the equator at roughly 13:30 local solar time. That single daily snapshot misses the morning traffic peak, the midday photochemical minimum, and the evening secondary maximum that together define a city's actual NO2 burden. Studies using GEMS data over Seoul have shown that the midday overpass can underestimate the daily mean column by 15 to 30 percent relative to a full diurnal integration, depending on season and meteorology.
GEMS and TEMPO resolve this by scanning their fixed continental footprints every hour through daylight. The diurnal cycle matters clinically: epidemiological evidence links peak morning exposure, when commuters are outdoors, to acute cardiovascular events more strongly than it links daily-mean exposure. Geostationary data can be matched to time-activity patterns in a way that a single polar-orbiter snapshot cannot. The practical constraint is that both GEMS and TEMPO cover limited regions. Outside East Asia and North America, TROPOMI remains the only operational source, and analysts must account for the overpass-time bias explicitly.
The pixel-size problem and what can be done about it
TROPOMI's 3.5 × 5.5 km pixel is large enough to smooth over the sharp NO2 gradients that exist within cities. A street canyon beside a motorway can carry NO2 concentrations three to five times higher than a park 500 metres away, yet both fall inside a single TROPOMI pixel. This matters enormously for exposure justice analysis, where the question is which communities, often low-income and near freight corridors, bear disproportionate burdens.
Statistical downscaling methods address this by combining the satellite column with high-resolution proxies: road-network density, traffic count data, land-use regression outputs, and in some cases commercial NO2 ground monitors. The approach, sometimes called land-use regression fusion, can push effective resolution to 100 to 500 metres, but the uncertainty widens considerably and the method is only as good as its proxy inputs. Honest practice requires communicating both the downscaled estimate and its confidence interval, not just the headline map. Where ground-truth monitors are sparse, validation is genuinely difficult.
Population-weighted exposure: turning columns into health numbers
A column map becomes an exposure estimate when it is intersected with gridded population data, typically LandScan or WorldPop at 100 m to 1 km resolution. The population-weighted mean column gives a single number summarising how much NO2 the average resident breathes, weighted by where people actually live rather than by land area. That distinction is not trivial: industrial zones with high columns but few residents contribute less to population burden than moderately polluted dense residential areas.
Translating column concentrations into health burden requires concentration-response functions from epidemiology, most commonly drawn from the Global Burden of Disease study's integrated exposure-response curves for ischaemic heart disease, stroke, and chronic obstructive pulmonary disease attributable to ambient NO2. These functions carry substantial uncertainty, and they were largely derived from ground-level monitor data, not satellite columns. The chain from satellite pixel to disability-adjusted life year involves at least three conversion steps, each with its own error distribution. Analysts who present a single attributable-mortality figure without uncertainty ranges are overstating what the method supports.
Archive depth and trend detection: where OMI earns its keep
OMI has been returning NO2 columns since October 2004. Nearly two decades of consistent, though coarser, data make it the backbone of long-term trend studies. Analyses published using the OMI record have documented NO2 reductions of 40 to 50 percent over parts of Western Europe and the eastern United States between 2005 and 2020, alongside sharp increases over South and East Asian industrial corridors during the same period. COVID-19 lockdowns produced transient column drops of 20 to 40 percent over major cities in spring 2020, visible clearly in both OMI and TROPOMI records.
Trend detection requires careful handling of instrument drift, row anomalies (OMI has suffered a persistent 'row anomaly' affecting roughly a quarter of its cross-track pixels since 2009), and changes in retrieval algorithm version. Comparing OMI and TROPOMI directly requires reprocessing both to a common algorithm baseline. ESA's Copernicus dataspace and NASA's Earthdata both archive reprocessed Level-2 products, but analysts should check algorithm version metadata before concatenating time series across sensors.
From retrieved column to actionable product
The gap between a satellite column map and a decision a health ministry can act on is where most of the analytical work lives. Satellize runs TROPOMI and GEMS retrievals through boundary-layer attribution models, intersects outputs with population grids, and produces exposure quintile maps that identify which administrative units carry the highest per-capita burden.
Practical outputs for government clients typically include monthly and annual population-weighted column averages by district, trend series going back to the OMI era, and exceedance-day counts referenced against WHO guideline values. The WHO 2021 annual mean NO2 guideline is 10 µg/m³, substantially tighter than the previous 40 µg/m³ standard. Many cities that appeared compliant under the old guideline are visibly non-compliant under the new one, and satellite data make that visible without waiting for a ground-monitor expansion programme that may take years to fund and deploy.
Typical figures
| Best available pixel size (TROPOMI) | 3.5 × 5.5 km (nadir, post-August 2019) |
| Best available pixel size (TEMPO) | ~2 × 4.5 km (nadir, North America) |
| Revisit (TROPOMI) | Daily global coverage; single overpass ~13:30 local solar time |
| Revisit (GEMS / TEMPO) | Hourly scans through daylight hours over fixed continental footprints |
| Spectral range for NO2 retrieval | 405–465 nm (UV-visible DOAS window) |
| Typical Level-2 latency (TROPOMI) | ~3 hours after sensing (near-real-time product); reprocessed offline within days |
| Detection sensitivity (TROPOMI) | ~0.5–1 × 10¹⁵ mol/cm² in clean-air conditions; urban columns typically 5–30 × 10¹⁵ mol/cm² |
| Archive depth | OMI: October 2004–present; TROPOMI: May 2018–present |
| Cloud screening threshold | Standard products filtered at cloud radiance fraction < 0.5; stricter thresholds (< 0.3) reduce coverage but improve accuracy |
| Delivery formats | NetCDF-4 (Level-2/3); GeoTIFF and GIS-ready shapefiles for downscaled products |
Analytics Satellize can run
| Monthly population-weighted NO2 exposure by administrative unit | TROPOMI L2 tropospheric column aggregated to L3 grid; intersected with WorldPop or LandScan population raster; area-weighted zonal statistics per boundary | GIS layer (GeoPackage or shapefile) with per-district exposure quintiles and WHO-guideline exceedance flags |
| Long-term NO2 trend series (2005–present) | OMI reprocessed L3 annual means concatenated with TROPOMI L3; Sen's slope estimator for trend magnitude; Mann-Kendall significance test | Time-series report with trend maps, confidence intervals, and annotated policy or event markers (e.g. lockdowns, regulation changes) |
| Diurnal NO2 cycle analysis (GEMS / TEMPO coverage areas) | Hourly geostationary L2 columns stacked and averaged by hour-of-day and season; comparison against TROPOMI overpass-time snapshot to quantify sampling bias | Hourly heatmap per city or region; overpass-bias correction factor for TROPOMI-based exposure estimates |
| Downscaled intra-urban NO2 map (100–500 m) | Land-use regression fusion: TROPOMI column as response variable; road-network density, traffic intensity, land-use class as predictors; cross-validated against available ground monitors | Raster GeoTIFF at target resolution with per-pixel uncertainty band; validation statistics reported explicitly |
| Exceedance-day count and compliance dashboard | Daily TROPOMI L3 columns converted to estimated surface concentrations via boundary-layer height from ERA5 reanalysis; days exceeding WHO 2021 guideline (10 µg/m³ annual mean threshold) flagged and counted | Monthly alert feed (JSON or CSV) by district; annual compliance summary PDF for regulatory reporting |
| Emission-source attribution (industrial vs. traffic vs. background) | Spatial correlation of column anomalies with facility locations from public emissions registers; wind-direction conditional averaging to identify upwind source signatures | Source-apportionment map with ranked facility list and estimated column contribution ranges |
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.