Metal smelter operating status from sulphur dioxide column density
Copper, nickel, and lead smelters emit SO₂ in direct proportion to throughput. Sentinel-5P TROPOMI resolves sulphur dioxide columns at 3.5 km pixels, enough to attribute emissions to individual large facilities and detect curtailments, strikes, or ramp-ups within days.
Sensors
- Sentinel-5P TROPOMI: Primary sensor for this use case. Measures SO₂ vertical column density at 3.5 × 5.5 km (post-2019 upgrade) with daily global coverage. Detection limit for a point source is roughly 1–2 kt SO₂/year under favourable wind and background conditions. Freely available via Copernicus Dataspace.
- NASA/JAXA Ozone Monitoring Instrument (OMI): Predecessor to TROPOMI aboard Aura, 13 × 24 km nadir pixels, operational since 2004. Coarser than TROPOMI but provides an 18-year archive useful for long-run baseline and trend analysis at major smelter complexes.
- Sentinel-3 OLCI: Optical imager at 300 m resolution with roughly 2-day revisit. Does not detect SO₂ directly, but provides corroborating evidence of smelter activity through visible plume colour, stack smoke, and facility-level change detection when cloud cover permits.
- ECMWF ERA5 / operational NWP wind fields: Not a satellite sensor, but an essential input. Hourly wind speed and direction at multiple pressure levels are required to run the Gaussian plume inversion or box-model back-calculation that converts a measured column density into an emission rate in kilograms per second.
Why SO₂ is a reliable throughput proxy
Pyrometallurgical smelting of sulphide ores, the dominant production route for copper, nickel, and lead, oxidises the sulphur content of the concentrate in flash furnaces or reverberatory furnaces. The SO₂ released is chemically tied to the mass of metal produced. A copper smelter processing 1 million tonnes of concentrate per year at a typical 30 per cent copper grade will emit on the order of hundreds of thousands of tonnes of SO₂ annually if scrubbing is absent or partial. Even with acid-plant capture, fugitive and stack emissions scale with throughput. That proportionality is what makes atmospheric SO₂ a useful signal.
The relationship is not perfectly linear. Scrubber efficiency varies, acid plants go offline for maintenance, and wind dispersal can mask or concentrate the apparent column. But across a fleet of smelters observed over weeks, the SO₂ signal is a credible proxy for operational state: running, curtailed, or shut. The method is most reliable at large smelters in regions with low ambient SO₂ background, such as the arid copper belts of Chile, Zambia, and the Democratic Republic of Congo. It is least reliable in industrialised basins where multiple sources overlap.
Source attribution: separating one smelter from its neighbours
TROPOMI's 3.5 km pixel is small enough to distinguish individual large smelters provided they are separated by at least one pixel and the plume is not shared. In practice, attribution requires more than a single overpass. The standard approach is to stack several days of TROPOMI Level-2 SO₂ product (the offline product, OFFL, has lower noise than the near-real-time product) and apply a wind-rotation technique: each overpass is rotated so that the local wind vector points in a fixed reference direction, then the rotated images are averaged. The resulting composite shows the smelter as a persistent point source with a downwind tail, separating it from diffuse background and from other facilities at different azimuths.
The TROPOMI SO₂ product reports vertical column densities in mol/m². The operational Level-2 product distinguishes three SO₂ layers: a boundary-layer product assuming the plume is below 1 km, a mid-troposphere product, and a stratospheric product for volcanic events. Industrial smelter plumes are correctly retrieved with the boundary-layer or 1 km centre-of-mass assumption. Using the wrong layer assumption introduces a systematic bias of 20–40 per cent in retrieved column, so layer selection is not a trivial choice.
Converting column density to emission rate
A column density measurement tells you how much SO₂ is above a given pixel at the moment of overpass. To infer the emission rate, you need to know how fast the plume is moving and how much has already been chemically removed. The simplest approach is the cross-sectional flux method: integrate the SO₂ column across a transect perpendicular to the wind direction at a known distance downwind, then multiply by the wind speed at plume height. The result is an emission rate in kg/s or kt/year.
The main sources of uncertainty are wind speed error, SO₂ lifetime (typically 1–2 days in the boundary layer, shorter in humid or high-UV conditions), and the assumption that the plume is well-mixed vertically within the retrieval layer. Published studies using this method on large copper smelters report uncertainties of 30–50 per cent for individual overpasses, narrowing to perhaps 20 per cent when averaged over monthly periods with multiple wind conditions. That is sufficient to distinguish a smelter running at full capacity from one running at half capacity, but not to resolve finer throughput gradations.
The detection floor and its consequences
TROPOMI's single-overpass detection limit for an SO₂ point source is approximately 1–2 kt/year under clean-background conditions, based on the instrument's noise-equivalent column density of roughly 0.5 mol/m² in the boundary-layer product. A large primary copper smelter processing 500,000 tonnes of concentrate per year with partial acid-plant capture might emit 50–150 kt SO₂/year, well above this floor. A small secondary smelter or a facility running at 5–10 per cent capacity may fall below it.
This matters for commodity analysis. A smelter that curtails to minimal care-and-maintenance mode may become invisible to TROPOMI, making it impossible to distinguish a deep curtailment from a full shutdown. The honest answer is that the method cannot reliably characterise very low throughput states. In those cases, corroborating signals, such as Sentinel-3 optical imagery, AIS vessel activity at the nearest concentrate port, or night-light intensity from VIIRS, become necessary to resolve the ambiguity. Cloud cover compounds the problem: TROPOMI is a passive UV/visible instrument and cannot retrieve SO₂ through thick cloud, so persistent overcast over a smelter can create gaps of several days in the time series.
Building a monitoring time series for market intelligence
For commodity analysts, the most useful output is not a single emission estimate but a consistent weekly or monthly time series that can be compared against reported production figures and price movements. The OMI archive from 2004 onwards, combined with TROPOMI from 2018, provides a record long enough to establish seasonal baselines, identify maintenance shutdown patterns, and detect structural capacity changes at individual facilities.
Operationally, TROPOMI data are available in near-real-time (NRTI) within three hours of overpass and in the higher-quality offline product within five days. A monitoring workflow that ingests NRTI for early alerts and replaces estimates with OFFL retrievals for the final record gives a reasonable balance between timeliness and accuracy. Satellize runs this kind of SO₂ time-series analytics on open Copernicus data, applying wind-field corrections from ECMWF ERA5, and can add commercial optical tasking for corroboration at specific facilities of interest. The Tonga crop-estimation programme is a different domain entirely, but the underlying pipeline, ingest open data, apply physics-based correction, deliver a structured analytic output, is the same architecture.
What this method cannot replace
SO₂ column monitoring is a remote signal, not a direct production measurement. It cannot distinguish between a smelter that is curtailed because of a labour dispute and one that is curtailed because of a concentrate supply shortage. It cannot tell you which furnace line is offline. It does not capture production from electric-arc furnaces processing pre-reduced material, which emits little SO₂. And it is blind to smelters that have invested heavily in SO₂ capture, where fugitive emissions may be a small fraction of throughput.
Used correctly, it is a systematic, near-daily signal that covers every major smelter on the planet simultaneously, at no marginal cost per facility. That is something no network of human sources can match. Used carelessly, it will mislead. The combination of a clear physical mechanism, an honest uncertainty budget, and corroborating optical and AIS signals is what makes the method credible rather than merely convenient.
Typical figures
| Spatial resolution (TROPOMI SO₂) | 3.5 × 5.5 km (nadir, post-August 2019 upgrade from 7 × 3.5 km) |
| Revisit frequency | Daily global coverage (Sentinel-5P sun-synchronous, ~13:30 local overpass) |
| Data latency | Near-real-time product (NRTI): within 3 hours of overpass. Offline product (OFFL): within 5 days |
| Spectral bands used | UV (312–326 nm) and UV-2 (325–335 nm) for SO₂ DOAS retrieval; visible channels for cloud fraction |
| Detection floor (point source) | Approximately 1–2 kt SO₂/year under low-background, clear-sky conditions; higher in polluted or cloudy regions |
| Emission rate uncertainty | 30–50% per overpass; ~20% for monthly averages with multiple wind directions |
| Archive depth | TROPOMI: May 2018 to present. OMI (coarser, 13 × 24 km): October 2004 to present |
| Cloud limitation | Retrievals flagged at cloud radiance fraction >0.5; thick cloud causes data gaps of 1–5 days at high-latitude or tropical sites |
| Corroborating optical sensor | Sentinel-3 OLCI at 300 m, ~2-day revisit, for visible plume and facility-state confirmation |
Analytics Satellize can run
| Weekly smelter operating-state classification | TROPOMI OFFL SO₂ Level-2 column stacking with wind-rotation composite; threshold classification against facility-specific baseline | Weekly status report per named facility (running / curtailed / shutdown / ambiguous), delivered as structured JSON feed or PDF |
| Emission rate time series | Cross-sectional flux method using TROPOMI SO₂ columns and ERA5 hourly wind fields at 850 hPa; uncertainty bounds propagated from wind speed error and SO₂ lifetime assumptions | Monthly kt SO₂/year estimates per facility with confidence intervals, delivered as CSV or GIS-compatible time series |
| Curtailment event detection and alert | NRTI SO₂ anomaly detection against rolling 90-day baseline; alert triggered when 7-day mean falls below 2-sigma of historical operating state | Email or API alert within 24 hours of curtailment signal, with supporting TROPOMI overpass imagery |
| Long-run capacity trend analysis | Annual SO₂ budget reconstruction combining OMI archive (2004–2018) and TROPOMI (2018–present); seasonal decomposition to isolate maintenance cycles from structural change | Multi-year trend chart and narrative summary per smelter complex, suitable for investment-committee use |
| Multi-source corroboration package | SO₂ column signal cross-referenced against Sentinel-3 OLCI visible imagery for plume confirmation and AIS vessel activity at nearest concentrate port for supply-chain context | Integrated facility brief combining atmospheric, optical, and maritime signals; updated on event-driven or monthly cadence |
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.