Refinery throughput proxies from stack SO₂ and NOₓ column density
TROPOMI measures tropospheric SO₂ and NOₓ columns at daily global coverage. Because distillation furnace firing rates scale with crude throughput, those columns are a working proxy for refinery feed rates, independent of company disclosures.
Sensors
- Sentinel-5P TROPOMI: Daily global coverage; SO₂ and NO₂ tropospheric column retrievals at 3.5 × 5.5 km pixel size (reprocessed to 3.5 × 5.5 km from August 2019 onward, earlier data at 3.5 × 7 km). Vertical column uncertainty for SO₂ is roughly 0.5–1 Dobson Unit per pixel under clear-sky conditions. Operational since November 2017.
- Sentinel-5 (pre-operational): ESA's follow-on to TROPOMI, targeting sub-7 km pixel resolution for trace gases including SO₂ and NO₂. Launch planned for mid-2020s. Will improve point-source isolation at smaller facilities once operational.
- ERA5 wind reanalysis (ECMWF): Hourly atmospheric wind fields at 0.25° horizontal resolution and 37 pressure levels. Used to model plume advection, convert observed column enhancements into emission-rate estimates, and filter days where wind direction sweeps adjacent industrial sources into the refinery pixel.
- OMI (Aura): NASA's Ozone Monitoring Instrument provides a historical SO₂ and NO₂ record from 2004 onward at 13 × 24 km nadir pixel. Coarser than TROPOMI but useful for establishing multi-year baseline trends at large refinery clusters.
Why stack chemistry correlates with crude throughput
Crude distillation requires sustained heat. The atmospheric distillation unit and its associated vacuum tower are fed by fired heaters burning refinery fuel gas or residual fuel oil. Firing intensity scales with the volume of crude being processed. That combustion produces sulphur dioxide in proportion to the sulphur content of the fuel, and nitrogen oxides in proportion to combustion temperature and volume. Both relationships are documented in EPA AP-42 emission factors and the EU's EMEP/EEA Air Pollutant Emission Inventory Guidebook, which provide fuel-specific emission factors that allow a column enhancement to be back-calculated to an approximate firing rate.
The inference is not a direct meter reading. Sulphur content of refinery fuel gas varies by crude slate, and NOₓ formation depends on burner design and excess-air settings. What the satellite observes is a column density integrated through the atmosphere, not a stack concentration. The relationship between column density and throughput is therefore statistical rather than deterministic, and it tightens considerably when multi-day averages are used and when the facility is geographically isolated enough to attribute the signal unambiguously.
The pixel problem: when 3.5 km is both enough and not enough
TROPOMI's 3.5 × 5.5 km pixel is genuinely useful for large, isolated refineries. A facility processing 300,000 barrels per day typically occupies one to three square kilometres and sits within a single pixel if it is separated from other significant SO₂ sources by at least five kilometres. Many coastal export refineries meet that criterion. The signal-to-noise ratio improves further because large refineries emit continuously, whereas meteorological variability is partly averaged out over the pixel.
Dense industrial zones are a different story. A refinery sharing a petrochemical corridor with a power station, a fertiliser plant and a coking facility cannot be cleanly isolated in a single TROPOMI pixel. Attribution becomes ambiguous, and the column enhancement reflects the aggregate of all sources. The honest answer in those situations is that TROPOMI-derived throughput proxies carry material uncertainty and should be treated as zone-level indicators rather than facility-level estimates. Sentinel-5's improved resolution will help, but it is not operational yet.
Multi-day averaging and the meteorological noise problem
A single overpass on a single day is rarely interpretable on its own. Cloud cover blocks the retrieval entirely. Wind direction determines whether the plume falls within the source pixel or is advected into an adjacent one. Boundary-layer height controls how much the column is diluted. Published studies using TROPOMI SO₂ over industrial point sources typically use 30-day or 90-day accumulations to produce stable emission-rate estimates, applying ERA5 wind fields to weight each clear-sky overpass by its plume-transport geometry.
For commodity intelligence purposes, a 30-day rolling average is generally the minimum useful window. That cadence suits month-on-month throughput comparisons, which align naturally with the reporting cycles that commodity analysts and credit analysts actually use. Shorter windows are possible at facilities where TROPOMI achieves a high clear-sky overpass rate, which in practice means latitudes below about 60° and away from persistently cloudy maritime climates. Operators in the Arabian Gulf, the US Gulf Coast and parts of East Asia are well served; Northwest European refineries less so.
Calibrating the proxy: what the public record allows
Calibration requires at least one period where independent throughput data is available. Publicly reported refinery run rates, published by national energy agencies or disclosed in company earnings materials, can be matched against contemporaneous TROPOMI retrievals to derive a site-specific scaling coefficient. The EPA's Continuous Emissions Monitoring System (CEMS) database provides stack-level SO₂ and NOₓ measurements for US facilities, which are particularly valuable for validating the satellite-to-throughput relationship at well-characterised sites.
Once a coefficient is established for a facility, it can be applied forward in time even when official data is delayed or absent. The coefficient degrades if the refinery changes its crude slate significantly, installs flue-gas desulphurisation equipment, or switches fuel sources. Those changes are themselves detectable: a sudden step-down in SO₂ column density that is not accompanied by a corresponding NOₓ reduction suggests scrubber installation rather than throughput decline. Tracking both species together reduces the risk of misreading operational changes as throughput signals.
Honest limits and what this method cannot do
This approach does not work for refineries that have installed sulphur recovery units or wet-gas scrubbers on their stacks, because those systems decouple SO₂ emissions from crude throughput. It also cannot distinguish between a refinery running at full capacity on low-sulphur crude and one running at reduced capacity on high-sulphur crude: both configurations can produce similar SO₂ columns. NOₓ partially breaks that degeneracy, since it tracks combustion volume more directly than sulphur content, but the ambiguity is not fully resolved.
Cloud cover is the other structural constraint. TROPOMI is a passive UV-visible instrument and cannot retrieve through optically thick cloud. In persistently overcast regions, valid retrievals may occur on fewer than half of all days, stretching the averaging window needed for a stable estimate. The method works best as a complement to other signals, including the thermal-infrared refinery activity indicators and optical activity proxies covered in sibling pages in this library, rather than as a standalone throughput meter.
Satellize runs TROPOMI-based SO₂ and NOₓ analytics on open archive data and can apply site-specific calibration where public throughput records exist.
Typical figures
| Spatial resolution (TROPOMI SO₂/NO₂) | 3.5 × 5.5 km per pixel (post-August 2019 reprocessing) |
| Revisit frequency | Daily global coverage (one overpass per day at equatorial latitudes) |
| Clear-sky data latency | Near-real-time products available within 3 hours of overpass; offline reprocessed products within 2–5 days |
| Spectral bands used | UV (312–326 nm for SO₂ DOAS retrieval); UV-visible (405–465 nm for NO₂ retrieval) |
| SO₂ column detection limit | Approximately 0.5–1 Dobson Unit per pixel under clear-sky conditions; reliable facility attribution typically requires >1 DU above background |
| Archive depth (TROPOMI) | November 2017 to present; OMI extends the SO₂/NO₂ record to 2004 |
| Wind reanalysis resolution (ERA5) | 0.25° horizontal, hourly, 37 pressure levels |
| Minimum facility size for reliable attribution | Refineries processing roughly 100,000+ barrels per day in geographically isolated settings; smaller or clustered facilities carry higher uncertainty |
| Averaging window for stable throughput proxy | 30-day rolling minimum recommended; 90-day preferred for calibration |
| Delivery formats | Time-series CSV, GeoTIFF column-density composites, facility-level summary reports |
Analytics Satellize can run
| Monthly SO₂ column composite per facility | TROPOMI L2 SO₂ offline product aggregation with cloud-fraction filtering (CF < 0.3) and ERA5 wind-direction screening to exclude days where adjacent sources contaminate the target pixel | GeoTIFF monthly composite with per-pixel uncertainty estimate, delivered as a time-series stack |
| Throughput proxy index (normalised) | Site-specific linear scaling from SO₂ column enhancement to relative throughput, calibrated against publicly available refinery run-rate disclosures or EPA CEMS data where the facility is in the US | Monthly index time series in CSV, with confidence interval derived from retrieval count and meteorological variability |
| NOₓ column trend monitor | TROPOMI NO₂ tropospheric column retrieval aggregated at facility level; compared against SO₂ trend to flag crude-slate changes or scrubber installation events | Dual-species time-series chart with anomaly flags, delivered as a PDF report or JSON feed |
| Facility isolation assessment | Spatial analysis of SO₂ and NO₂ background fields within a 20 km radius of target facility to quantify contamination risk from adjacent sources; informed by land-use data and published industrial inventory registers | One-page facility suitability report, produced once at project initiation |
| Operational status change alert | Statistical process control on rolling 30-day SO₂ column mean; alert triggered when the value falls more than two standard deviations below the site-specific baseline, consistent with a shutdown or major throughput reduction | Email or API alert with supporting column-density map and retrieval-quality metadata |
| Multi-facility portfolio dashboard | Parallel processing of TROPOMI retrievals across a client-defined list of refineries; normalised indices presented on a common scale to allow cross-facility comparison | Interactive dashboard or monthly PDF summary covering up to 20 named facilities |
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.