SO2 plume mapping for industrial pollution health burden assessment
UV backscatter retrievals from TROPOMI and OMI can pinpoint large SO2 point sources at 3.5 × 5.5 km resolution, feed dispersion models, and estimate downwind population exposure for health impact assessments.
Sensors
- Sentinel-5P TROPOMI SO2: Primary workhorse. Ground pixel 3.5 × 5.5 km (resampled to 3.5 × 3.5 km from August 2019 processor update), daily global coverage. Retrieves SO2 vertical column density in Dobson Units via DOAS fitting of UV backscatter in the 312–326 nm window. Detects large point sources (power stations, smelters) above roughly 0.5–1 DU in clear-sky conditions; sensitivity degrades sharply under cloud fractions above ~0.3.
- Ozone Monitoring Instrument (OMI) SO2: NASA/KNMI instrument on Aura, operational since 2004 with a 13 × 24 km nadir pixel (degraded by row-anomaly masking post-2009). Longer archive than TROPOMI makes it the reference for multi-year trend analysis of large industrial emitters. Same UV DOAS physics, higher noise floor per pixel.
- GEMS SO2: Geostationary Environment Monitoring Spectrometer on GEO-KOMPSAT-2B. Covers East Asia at hourly cadence with roughly 3.5 × 8 km pixels. Temporal resolution allows intra-day plume tracking and captures emission variability tied to plant operating cycles, which daily-overpass instruments cannot resolve.
- IASI (MetOp) SO2: Infrared Atmospheric Sounding Interferometer on MetOp-A/B/C. Retrieves SO2 in the thermal infrared at roughly 12 km circular footprint, twice-daily coverage. Most sensitive to elevated SO2 (volcanic or high-stack industrial plumes above ~2 km); poor sensitivity to boundary-layer emissions below about 1–2 km, making it complementary rather than primary for low-stack industrial sources.
What UV backscatter actually measures, and why that matters for health work
TROPOMI does not see SO2 directly. It measures the ratio of solar UV photons backscattered from the Earth-atmosphere system against a reference solar spectrum, then fits the residual absorption signature to SO2 cross-sections using Differential Optical Absorption Spectroscopy. The output is a vertical column density: the total SO2 in a vertical air column above each pixel, expressed in Dobson Units (1 DU ≈ 2.69 × 10²⁰ molecules per square centimetre).
For health burden work, vertical column density is not the same as ground-level concentration. A health assessor needs surface SO2 in micrograms per cubic metre. Converting column to surface requires an assumed or modelled vertical profile shape, and this is the retrieval's central vulnerability. TROPOMI's standard SO2 product ships with three profile-shape assumptions: a boundary-layer profile peaking at 0.9 km, a free-troposphere profile peaking at 2.5 km, and a stratospheric profile. If a smelter stack injects SO2 at 300 m and the analyst applies the wrong profile assumption, the column-to-surface conversion can be wrong by a factor of two or more. Plume-height information from co-located lidar, radiosonde data, or a meteorological model is therefore not optional; it is the difference between a defensible health estimate and a confident-looking error.
Point sources TROPOMI can resolve, and the floor it cannot cross
At 3.5 × 5.5 km, TROPOMI can attribute SO2 columns to individual large point sources: coal-fired power stations above roughly 500 MW thermal output, copper and nickel smelters, and large sulphur-recovery units. Published studies using TROPOMI have identified and ranked the world's largest SO2 emitters by facility, cross-referencing pixel clusters with known plant locations. This is genuinely useful for national emissions inventories and for identifying facilities whose reported emissions diverge from satellite-derived estimates.
The detection floor sits at roughly 0.5 to 1 DU under clear-sky conditions, which corresponds to annual SO2 emissions of the order of tens of kilotonnes per year for a single facility, depending on stack height and meteorology. Distributed sources, such as residential coal burning across a city, small industrial boilers, or vehicle fleets burning high-sulphur diesel, collectively matter for urban SO2 exposure but individually fall well below TROPOMI's pixel-level sensitivity. Aggregating many small sources into a single pixel can sometimes push the column above the noise floor, but attributing that signal to specific emitters is not possible. For urban health burden work, TROPOMI is best treated as a regional constraint, not a street-level exposure map.
From column density to population exposure: the dispersion step
The analytical chain connecting a satellite SO2 column to a health impact estimate has several steps, each carrying its own uncertainty. First, a source emission rate is estimated by fitting a Gaussian or more sophisticated dispersion model to the downwind column enhancement, using wind fields from a reanalysis product such as ERA5. Published methods (for example, the integrated mass enhancement approach) can infer emission rates from TROPOMI data with uncertainties of roughly 20 to 40 percent for well-isolated large sources under favourable wind conditions.
Second, the emission rate feeds an atmospheric chemistry and transport model to estimate surface SO2 concentrations and the secondary formation of sulphate PM2.5, which is often the larger health burden because fine sulphate particles penetrate deep into the respiratory tract and persist far downwind. Third, gridded population data is overlaid to compute population-weighted exposure. Finally, a concentration-response function from the epidemiological literature translates exposure to attributable disease burden, typically expressed as disability-adjusted life years. Each link in this chain compounds uncertainty, so health impact estimates derived from satellite data alone should be presented with explicit uncertainty ranges rather than point estimates.
Cloud cover, latency and the multi-year archive
TROPOMI's single daily overpass means that on any given day, cloud cover above roughly 30 percent radiance fraction renders a pixel unusable. In humid tropical regions or during monsoon seasons, cloud-free observations over a given location may occur on fewer than half of all days. Monthly or seasonal compositing recovers spatial coverage but loses the ability to track individual emission events.
Near-real-time TROPOMI SO2 products are available through the Copernicus Data Space with latency of roughly three hours after overpass, which is fast enough for acute pollution episode monitoring if combined with automated alerting. The TROPOMI archive runs from November 2017; OMI extends the combined record back to October 2004, giving nearly two decades of data for long-term trend analysis of industrial SO2 emissions. That archive depth is one of the strongest arguments for satellite-based industrial monitoring: it pre-dates most continuous ground-level monitoring networks in lower-income countries, and it is independent of facility self-reporting.
Practical outputs for a health burden assessment
A well-structured satellite SO2 health assessment typically produces four layers. First, a facility-level emission rate time series, showing seasonal and inter-annual variability and flagging anomalous episodes. Second, a mean annual SO2 column map at TROPOMI resolution, clipped to the study region, suitable for communicating spatial patterns to non-specialist audiences. Third, modelled surface SO2 and secondary sulphate PM2.5 concentration grids at the resolution of the transport model used, typically 10 to 25 km for regional assessments. Fourth, a population exposure summary table, disaggregating exposed population by administrative unit and concentration band.
Satellize runs this analytical chain on open TROPOMI and OMI data, combining it with ERA5 wind fields and gridded population products. The workflow is the same one applied to the Kingdom of Tonga crop-estimation programme in terms of open-data pipeline discipline, adapted here for atmospheric retrievals rather than surface reflectance. Clients receive GIS layers, a structured report with explicit uncertainty ranges, and, where agreed, a monitoring feed that flags exceedance events within hours of the daily overpass.
Honest limits, and where ground data remains irreplaceable
Three limits deserve plain statement. First, TROPOMI cannot see below the cloud deck, which means emission episodes during overcast conditions are invisible. Second, the fixed-profile-shape assumption introduces systematic error for low-stack or valley-trapped emissions, and this error is correlated across pixels in the same meteorological regime, so it does not average away with time. Third, the 3.5 km pixel is large enough that a cluster of moderate emitters within a single pixel is indistinguishable from one large emitter.
Ground-level continuous SO2 monitors, where they exist, remain the gold standard for regulatory compliance and acute health alert systems. The satellite's comparative advantage is spatial coverage, independence from facility cooperation, and historical depth. The most defensible assessments combine both: satellite data to establish the spatial pattern and long-term trend, ground monitors to anchor the absolute concentration scale and validate the dispersion model. In regions with no ground network, satellite data is not a perfect substitute; it is simply the best available evidence, and should be described as such.
Typical figures
| Spatial resolution (TROPOMI SO2) | 3.5 × 5.5 km at nadir (3.5 × 3.5 km from August 2019 processor resampling) |
| Spatial resolution (OMI SO2) | 13 × 24 km nadir (row-anomaly affected pixels masked post-2009) |
| Revisit (TROPOMI / OMI) | Daily global coverage (single overpass, ~13:30 local solar time) |
| Revisit (GEMS) | Hourly over East Asia (08:00–17:00 local time) |
| Spectral bands used | UV 312–326 nm (DOAS SO2 retrieval); IASI thermal infrared ~7.3 µm (supplementary elevated-plume retrieval) |
| Detection floor (TROPOMI, clear sky) | ~0.5–1 Dobson Units per pixel; equivalent to roughly tens of kilotonnes SO2/year for an isolated large point source |
| Cloud fraction limit | Pixels with cloud radiance fraction >0.3 typically excluded; tropical/monsoon regions may have <50% usable days |
| NRT product latency (TROPOMI) | ~3 hours after overpass via Copernicus Data Space |
| Archive depth | TROPOMI from November 2017; OMI from October 2004 |
| Delivery formats | NetCDF (native), GeoTIFF (derived grids), CSV exposure tables, PDF assessment report |
Analytics Satellize can run
| Facility-level SO2 emission rate time series | Integrated mass enhancement (IME) or cross-sectional flux method applied to TROPOMI SO2 columns with ERA5 wind fields | Annual and monthly emission rate estimates per facility, with uncertainty ranges, delivered as CSV and facility report |
| Annual mean SO2 column density map | Cloud-screened TROPOMI pixel compositing with quality flag filtering (qa_value ≥ 0.5) | GeoTIFF raster at 3.5 km resolution, clipped to study region, with colour-scaled PDF map |
| Modelled surface SO2 and secondary sulphate PM2.5 concentration grids | Satellite-constrained emission rates ingested into Gaussian or Lagrangian dispersion model; secondary sulphate formation parameterised from published oxidation rate literature | Gridded concentration fields (GeoTIFF, 10–25 km), seasonal and annual averages |
| Population-weighted exposure summary | Overlay of concentration grids with gridded population (e.g. WorldPop or GPWv4), aggregated by administrative unit and concentration band | Tabular exposure report (CSV/XLSX) disaggregated by district, with exposed population counts per WHO guideline exceedance tier |
| Acute episode alert feed | Automated threshold detection on NRT TROPOMI SO2 product (~3-hour latency), compared against rolling 90th-percentile baseline per facility cluster | Automated alert (email or API push) within 4–6 hours of overpass when column exceeds defined threshold |
| Long-term emission trend analysis | Combined OMI (2004–present) and TROPOMI (2017–present) time series with Mann-Kendall trend test and seasonal decomposition | Trend report with significance levels, visualised as annual anomaly chart, suitable for regulatory or treaty reporting |
| Health impact estimate (attributable burden) | Concentration-response functions from published integrated exposure-response literature applied to modelled population exposure; DALY calculation following GBD methodology | Structured health impact report with point estimates and 95% uncertainty intervals, per facility and per administrative region |
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.