Traffic-related NO2 hotspot mapping in urban corridors
TROPOMI's tropospheric NO2 columns are too coarse to name a junction, but multi-year temporal averaging and land-use regression can push actionable hotspot maps below one kilometre. Here is what the method can and cannot promise.
Sensors
- Sentinel-5P TROPOMI: Primary workhorse. Tropospheric NO2 vertical column density at a native footprint of 3.5 × 5.5 km (reprocessed to 3.5 × 5.5 km from 2019; earlier data at 7 × 3.5 km). Daily global revisit. Detection limit roughly 0.1–0.3 × 10¹⁵ molecules cm⁻², depending on scene albedo and solar angle. Free, open archive from May 2018.
- Aura OMI: Predecessor instrument providing NO2 columns at ~13 × 24 km from 2004 onward. Coarser than TROPOMI but offers a longer record useful for trend baselines. Row-anomaly artefacts affect a portion of the swath and must be masked.
- GEMS (GEO-KOMPSAT-2B): Geostationary sensor covering East Asia. Retrieves NO2 columns roughly eight times per day at approximately 3.5 × 8 km, enabling diurnal cycle characterisation. Launched February 2020. Spatial footprint is still street-network-blind without downscaling.
- Sentinel-4 (forthcoming): Planned geostationary sensor for Europe on MTG-S1, targeting hourly NO2 retrievals at roughly 8 km nadir resolution. Launch expected mid-2020s. Will improve temporal sampling over European urban corridors but will not close the spatial gap to individual roads.
What a 3.5 km pixel actually contains
A single TROPOMI pixel over a city centre encompasses hundreds of streets, several land-use classes, and a mixture of industrial, domestic and traffic emission sources. The sensor measures the total tropospheric NO2 column above that footprint, integrated through the atmosphere. It cannot, on its own, tell you whether the signal comes from a bus depot on the eastern edge or a motorway junction in the middle.
That ambiguity is not a flaw in the sensor; it is physics. NO2 retrieved in the ultraviolet-visible range (around 405–465 nm for the DOAS fitting window) reflects column abundance, not surface concentration at a specific address. Buyers who expect TROPOMI to replace a roadside monitor will be disappointed. Buyers who understand it as a city-scale constraint that can be disaggregated are working with the right mental model.
Temporal averaging: turning noise into signal
The key statistical trick is simple in principle and demanding in practice. Average TROPOMI overpasses over a single city for one to three years, masking cloudy scenes (cloud radiance fraction threshold typically 0.5 or lower), and the random retrieval noise averages down while persistent emission patterns accumulate. Studies published in Remote Sensing and Atmospheric Chemistry and Physics have demonstrated that annual-mean TROPOMI maps over European cities begin to show spatial gradients that correlate with road-network density and traffic counts, even though no single overpass resolves a street.
The honest caveat: averaging suppresses episodic events. A corridor that is heavily congested only during school-run hours on weekdays will appear weaker than its peak contribution suggests. GEMS and, eventually, Sentinel-4 help here by providing multiple overpasses per day, allowing diurnal profiles to be built for the same coarse footprint. That temporal richness is genuinely new information, even if the spatial resolution is unchanged.
Downscaling with street-network priors
Land-use regression (LUR) has been applied to satellite NO2 columns for over a decade. The method regresses observed column values against spatial predictors, including road-length density by category (motorway, arterial, local), traffic count data where available, population density, and land-cover fractions. The resulting model redistributes the coarse satellite signal to a finer grid, typically 100–500 m, guided by the assumption that roads explain a known fraction of the variance.
A more recent class of methods uses emission inventories or traffic models as spatial disaggregation weights. The satellite column acts as a top-down constraint: the disaggregated surface concentrations must, when convolved with an atmospheric transport model and the averaging kernel, reproduce the observed TROPOMI value. This approach is more physically grounded than pure LUR but requires a credible prior emission map and a working chemistry-transport model. The output is not a measurement; it is a constrained estimate. Uncertainty at the sub-kilometre level is typically 30–60 % and should be communicated explicitly in any regulatory context.
Ground-monitor networks remain the validation anchor. Where a city operates a dense network of NO2 monitors (London, for example, runs over a hundred roadside and background monitors), satellite-derived maps can be cross-validated and bias-corrected. Where monitoring is sparse, which is most of the world, the satellite product is the only available spatial picture, and its uncertainty bands widen accordingly.
What the maps are genuinely useful for
City authorities and transport ministries have used satellite-derived NO2 maps for several practical purposes. Identifying which districts persistently exceed WHO guideline levels (10 µg m⁻³ annual mean for NO2, revised in 2021) without relying on interpolation between sparse monitors is one. Evaluating the before-and-after effect of low-emission zones is another: TROPOMI's archive from 2018 onward is long enough to detect statistically significant column reductions following zone introduction, provided the signal is not swamped by interannual meteorological variability.
The maps are less useful for enforcement against individual vehicles or specific junctions. For that, you need roadside sensors or camera-based systems. The satellite product works at the policy scale: districts, corridors, and city-wide trends over seasons and years.
Limits that belong in the briefing document
Cloud cover is the most operationally significant constraint. At cloud radiance fractions above 0.5, retrievals are discarded. In persistently cloudy climates, the effective number of usable overpasses per year can fall to 60–100, stretching the averaging period needed to achieve stable annual means. Tropical coastal cities are particularly affected.
The averaging kernel, which describes the altitude sensitivity of the retrieval, varies with surface albedo, aerosol loading, and solar zenith angle. Urban areas with heavy aerosol can suppress sensitivity in the boundary layer, precisely where traffic NO2 is concentrated. This is not always corrected for in standard data products and can cause systematic underestimation of surface-level contributions in polluted cities.
Sentinel-5P carries no on-board calibration source for absolute radiometric drift; cross-calibration with OMI and ground-based Pandora spectrometers is ongoing. For trend analyses extending beyond five years, version-to-version reprocessing differences matter and should be checked against the ESA reprocessing changelog. Satellize applies these quality flags and reprocessing corrections as standard when building multi-year urban NO2 baselines for clients.
Building a monitoring programme around imperfect data
The productive framing is not 'is the satellite good enough?' but 'what decisions does this information support at what confidence level?' A transport ministry deciding whether to extend a low-emission zone across a new district does not need sub-metre certainty; it needs a credible ranking of corridors and a defensible trend estimate. TROPOMI-derived maps, properly averaged, downscaled with traffic priors, and validated against available monitors, can support that decision.
For cities with no existing monitor network, a satellite-first approach establishes a spatial baseline that guides where to place the first physical sensors for maximum information gain. That sequencing is often more cost-effective than installing monitors uniformly and hoping the coverage is adequate.
The next meaningful improvement in this field will come from combining GEMS-class geostationary diurnal sampling with TROPOMI-class polar-orbit spatial density, using data fusion methods that are already being tested in the literature. Sentinel-4's arrival over Europe will accelerate that work. Until then, the honest answer to 'can you map my city's traffic NO2 at street level from orbit?' is: approximately, conditionally, and with stated uncertainty.
Typical figures
| Native spatial resolution (TROPOMI) | 3.5 × 5.5 km (from August 2019); 7 × 3.5 km before reprocessing |
| Native spatial resolution (GEMS) | Approximately 3.5 × 8 km over East Asia |
| Revisit (TROPOMI) | Daily global coverage; one overpass per location per day |
| Revisit (GEMS) | Approximately 8 observations per day over East Asia (geostationary) |
| Spectral fitting window (NO2) | 405–465 nm (DOAS differential optical absorption spectroscopy) |
| Detection limit (TROPOMI NO2) | Approximately 0.1–0.3 × 10¹⁵ molecules cm⁻² (scene-dependent) |
| Downscaled product resolution (post-processing) | 100–500 m typical, depending on prior data quality; uncertainty 30–60 % |
| Archive depth (TROPOMI) | May 2018 to present; OMI extends NO2 record to 2004 |
| Cloud screening threshold | Cloud radiance fraction < 0.5 (standard); stricter thresholds reduce usable scenes further |
| Delivery formats | NetCDF (native ESA product); GeoTIFF and GIS-ready vector layers for downscaled outputs |
Analytics Satellize can run
| Annual-mean NO2 column map | Multi-year TROPOMI overpass averaging with cloud and quality masking; bias correction against available Pandora or monitor data | GeoTIFF raster layer, city or corridor extent, with uncertainty band layer |
| Downscaled sub-kilometre hotspot map | Land-use regression or traffic-weighted spatial disaggregation constrained by TROPOMI column totals | 100–500 m resolution GIS layer ranked by estimated NO2 enhancement; methodology note included |
| Diurnal NO2 profile (GEMS coverage areas) | Geostationary multi-overpass compositing to resolve morning, midday and evening peaks | Time-series chart per district polygon; CSV export |
| Low-emission zone impact assessment | Before-and-after column comparison controlling for meteorological variability using reanalysis wind and temperature fields | Statistical report with significance test, percentage change estimate, and spatial difference map |
| Monitor siting recommendation | Satellite-derived spatial gradient analysis to identify locations of maximum information gain for new physical sensors | Ranked candidate-site list with supporting NO2 gradient maps |
| Multi-year trend analysis | Combined OMI (2004 onward) and TROPOMI (2018 onward) time series with version-consistent reprocessing corrections applied | Annual trend report with Mann-Kendall significance, version changelog notes, and caveats on inter-sensor calibration |
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.