Aluminium smelter operational status estimation as electricity-demand and commodity proxy
Aluminium electrolytic cells run at continuous high current and emit a stable thermal signature. Curtailment or shutdown drops that radiance measurably, giving commodity traders and lenders a near-real-time production proxy from VIIRS and Landsat thermal time series.
Sensors
- VIIRS M-band thermal (Suomi NPP / NOAA-20 / NOAA-21): 750 m pixel resolution in M13 (4.05 µm) and M15/M16 (11–12 µm) bands; twice-daily global overpass. The 4 µm band is highly sensitive to high-temperature industrial point sources, making it the primary channel for detecting active potlines and flagging curtailment events within a single diurnal cycle.
- Landsat 8/9 TIRS: 100 m native thermal resolution (resampled to 30 m in data products) in two bands centred at 10.9 µm and 12.0 µm; 16-day revisit per satellite, 8-day combined. Provides higher spatial detail than VIIRS, allowing individual potline buildings within a smelter to be distinguished. Useful for confirming partial curtailment affecting one potroom while others remain active.
- ECOSTRESS (ISS-mounted): Five thermal infrared bands from 8.3 to 12.5 µm at approximately 70 m resolution. Non-sun-synchronous orbit from the ISS gives irregular but occasionally useful off-hour overpasses. Useful as a cross-check against Landsat TIRS readings; archive extends from 2018. Revisit is irregular, typically 1–5 days depending on latitude.
- Sentinel-3 SLSTR: 1 km resolution in thermal channels S7–S9 (3.7–12 µm); twice-daily global coverage. Coarser than VIIRS M-band but dual-view geometry improves atmospheric correction. Best suited to detecting large smelters or clusters of facilities rather than resolving individual potlines.
Why a smelter's thermal signature is unusually legible
Hall-Héroult electrolytic cells operate at bath temperatures of roughly 940–970 °C and draw continuous direct current at densities typically between 300 and 400 kA per potline. That sustained thermal load radiates through the roof and ventilation of potline buildings in a pattern that is stable for months or years during normal operation. Unlike a steel furnace or cement kiln, which cycle through charge-and-tap rhythms, aluminium potlines run without interruption. The signature is therefore not a pulse but a persistent baseline.
That stability is precisely what makes curtailment visible. When grid operators instruct a smelter to shed load, or when a smelter operator voluntarily curtails in response to power price spikes, cells are taken offline in sequence. Each potline removed drops the facility's radiance contribution measurably. A full potroom shutdown at a mid-sized smelter (say, 250,000 tonnes per annum capacity) can reduce thermal radiance in the VIIRS M13 band by 30–60 percent relative to the baseline, depending on facility geometry and atmospheric conditions on the day.
Reading the time series: baseline, ramp and curtailment signatures
The analytical workflow starts with constructing a per-facility radiance baseline from at least 90 days of cloud-free thermal observations. VIIRS provides the temporal density: two overpasses per day means a typical smelter accumulates 50–80 usable night-time observations per quarter at mid-latitudes, cloud permitting. Landsat adds spatial resolution to disambiguate which potrooms are active. Together they allow analysts to separate three states: full operation, partial curtailment (one or more potrooms offline), and full shutdown.
Ramp signatures matter as much as the curtailment itself. Restarting a Hall-Héroult cell after a cold shutdown takes 24–72 hours to reach operating temperature, during which radiance climbs in a characteristic curve. That ramp is visible in the VIIRS time series and can be used to estimate restart timing independently of company announcements. The same logic applies in reverse: a gradual radiance decline over several days suggests planned maintenance rather than an emergency curtailment, which tends to produce a sharper step.
What the sensor cannot tell you, and why that matters
Thermal sensors measure radiance, not cause. A radiance drop consistent with 20 percent curtailment could reflect a power-price-driven load shed, a delayed alumina shipment, a potline rebuild, or a grid fault. The satellite cannot distinguish between them. Analysts who present curtailment estimates without that caveat are overselling the method.
Cloud cover is the other hard constraint. Thermal infrared at 10–12 µm is blocked by thick cloud, though the 4 µm VIIRS M13 band is somewhat less affected by thin cirrus. At high-latitude smelters in Norway, Iceland, Canada or Russia, winter cloud cover can create gaps of 5–10 consecutive days in usable observations. Sentinel-3's dual-view geometry helps with atmospheric correction but does not solve cloud opacity. Analysts should quote observation-gap statistics alongside curtailment estimates so clients understand the confidence interval on any given reading.
Spatial resolution also sets a floor on what is resolvable. VIIRS at 750 m will conflate a smelter with adjacent industrial heat sources if facilities are co-located on an industrial estate. Landsat TIRS at 100 m (effective) is better for separating potlines, but its 8-day combined revisit means a curtailment event that lasts only 48 hours may be missed entirely between passes.
From radiance to production: the commodity and power-market link
Primary aluminium output is tightly coupled to electricity consumption. The theoretical minimum energy requirement for electrolytic reduction is approximately 6.3 kWh per kilogram of aluminium, and modern smelters operate at 13–15 kWh per kilogram in practice. That means a 500,000-tonne-per-annum smelter draws roughly 7–8 TWh per year, comparable to the residential demand of a mid-sized European city. Any detectable curtailment therefore has a direct electricity-market implication, which is why power utilities and grid operators pay attention to smelter status independently of commodity traders.
For commodity desks, the link runs through the LME aluminium forward curve. Curtailments at smelters representing a meaningful share of regional or global capacity shift near-term supply expectations. The satellite-derived curtailment signal is most useful when it precedes official production announcements by days or weeks, which it frequently does. Published academic work using VIIRS thermal data to monitor industrial facilities has demonstrated detection latencies of under 24 hours for large thermal anomaly changes, though the translation from radiance change to tonnes-curtailed requires facility-specific capacity data from public sources such as company filings and industry databases.
Building a monitoring programme: data cadence, coverage and delivery
A practical smelter-monitoring programme covers a defined universe of facilities. The International Aluminium Institute publishes a list of operating smelters by country; cross-referencing that with coordinates and nameplate capacity allows analysts to build a priority-ranked watchlist. Facilities above 200,000 tonnes per annum capacity account for the majority of global output and are large enough to be reliably detected by VIIRS.
Daily VIIRS processing, supplemented by Landsat passes when cloud-free, gives a latency of roughly 12–36 hours from satellite overpass to client alert under good atmospheric conditions. Sentinel-3 SLSTR provides a secondary check. For lenders with credit exposure to a specific facility, a bespoke alert triggered when radiance falls more than two standard deviations below the rolling 30-day baseline offers a practical early-warning mechanism. Satellize runs this class of thermal time-series analytics on open constellations, applying the same monitoring logic it uses in its Tonga crop-estimation programme to industrial thermal targets. The output is typically a structured data feed or weekly PDF summary with per-facility status flags, radiance anomaly scores and observation-gap disclosures.
Clients who need higher revisit or finer spatial detail can add commercial thermal tasking on licence. That reduces the gap-risk at high-latitude facilities and allows sub-potroom spatial attribution in some cases, though commercial thermal satellites with sub-100 m resolution remain a small and relatively young market.
Typical figures
| Primary thermal sensor | VIIRS M13 (4.05 µm), M15/M16 (11–12 µm) at 750 m; Landsat 8/9 TIRS at 100 m effective resolution |
| Revisit cadence | VIIRS: ~2 passes/day globally; Landsat 8+9 combined: ~8-day repeat at equator; ECOSTRESS: irregular, 1–5 days; Sentinel-3 SLSTR: ~2 passes/day at 1 km |
| Detection latency (clear sky) | 12–36 hours from overpass to processed alert under good atmospheric conditions |
| Minimum detectable curtailment | Approximately 20–30% potline reduction for facilities above ~200,000 t/a capacity using VIIRS; smaller curtailments may require Landsat confirmation |
| Spectral bands used | Mid-wave IR (3.7–4.1 µm) for high-temperature anomaly; long-wave IR (10–12 µm) for facility-wide thermal flux; split-window pairs for atmospheric correction |
| Cloud limitation | Thick cloud blocks 10–12 µm entirely; 4 µm partially penetrates thin cirrus. High-latitude winter gaps can reach 5–10 consecutive days |
| Archive depth | VIIRS: 2012 to present (Suomi NPP); Landsat TIRS: 2013 to present (Landsat 8); ECOSTRESS: 2018 to present |
| Spatial disambiguation floor | Facilities separated by less than ~750 m may be conflated in VIIRS; Landsat TIRS resolves features to ~100 m, enabling potroom-level attribution at larger smelters |
| Delivery formats | Structured JSON or CSV anomaly feed; GeoTIFF radiance composites; weekly PDF status report with observation-gap disclosure |
Analytics Satellize can run
| Per-facility operational status flag | Radiance anomaly detection against rolling 30-day baseline; z-score thresholding on VIIRS M13 and Landsat TIRS time series | Daily status feed (Active / Partial curtailment / Shutdown / Insufficient data) with confidence score and cloud-gap flag |
| Curtailment onset and restart timing | Change-point detection on multi-sensor radiance time series; characteristic ramp-down and ramp-up curve fitting for Hall-Héroult thermal response | Event log with estimated onset timestamp, duration and restart date; delivered as structured data or PDF summary |
| Partial curtailment potroom attribution | Landsat TIRS 100 m spatial decomposition of within-facility radiance distribution; comparison against facility layout from public imagery | GeoTIFF overlay showing active vs. inactive potroom zones; included in weekly facility report |
| Regional curtailment aggregation | Portfolio-level summation of facility status flags weighted by nameplate capacity from public industry sources; normalised to regional output share | Weekly regional curtailment index (percentage of monitored capacity flagged as curtailed or shutdown), delivered as time-series CSV |
| Electricity-demand proxy signal | Conversion of radiance-inferred operational status to estimated load using published kWh-per-tonne benchmarks (13–15 kWh/kg) and nameplate capacity | Estimated MW load change per facility or region; structured feed for integration into power-market models |
| Credit-exposure early-warning alert | Bespoke threshold alert triggered when facility radiance drops more than two standard deviations below rolling baseline for two or more consecutive passes | Push alert (email or API webhook) with supporting radiance chart and observation-gap disclosure; designed for lender covenant-monitoring workflows |
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.