Aluminium smelter potline operating status from thermal infrared

Each electrolytic reduction line in an aluminium smelter radiates a distinct thermal signature detectable from orbit. Landsat, ECOSTRESS and VIIRS night passes let analysts count active potlines, distinguish partial curtailments from full shutdowns, and track output capacity changes weeks before official data.

Sensors

Landsat 8/9 TIRS: Two thermal infrared bands centred at 10.9 µm and 12.0 µm, 100 m native resolution (resampled to 30 m in products), 16-day revisit per satellite. The dual-band split-window allows surface temperature retrieval to roughly ±0.5 K under clear sky, sufficient to resolve individual potline buildings at large smelters.
ECOSTRESS (ISS-mounted): Five LWIR channels from 8.3 to 12.5 µm, 70 m spatial resolution, non-sun-synchronous due to ISS precession. Night-time passes suppress solar reflected contamination and improve thermal contrast for hot industrial surfaces. Revisit is irregular, typically 1–5 days at mid-latitudes, but the ISS orbit provides useful off-nadir geometry over many smelter locations.
Sentinel-3 SLSTR: Dual-view radiometer with thermal channels at 10.85 µm and 12.0 µm, 1 km resolution in thermal mode. Too coarse to resolve individual potlines at most smelters, but useful for detecting facility-level heat flux and confirming gross curtailment or restart events. Daily revisit at mid-latitudes.
VIIRS M-band thermal (Suomi NPP / NOAA-20): M15 (10.76 µm) and M16 (12.01 µm) at 750 m nadir resolution, twice-daily overpass. Comparable to SLSTR in spatial terms but provides a consistent long archive from 2012 and is operationally processed through NOAA STAR. Useful for trend analysis and anomaly flagging at facility scale.

What a potline actually looks like from orbit

An aluminium smelter's electrolytic reduction hall, the potline, is a row of Hall-Héroult cells each holding a bath of molten cryolite and alumina at roughly 950 °C. A single potline building is typically 500 to 1,500 metres long and 20 to 30 metres wide. The roof and walls leak heat continuously, producing a linear thermal anomaly that stands out sharply against the surrounding terrain in LWIR imagery, particularly at night when solar heating of adjacent surfaces has dissipated.

Landsat TIRS at 100 m resolution can distinguish individual potline buildings at most large smelters. A facility with four potlines in two parallel halls produces a recognisable dual-stripe signature. When one hall is curtailed, the stripe disappears or dims to background within one to two days as the cells cool. This is not subtle: the brightness temperature difference between an active potline roof and an idle one is typically 10–30 K in night-time imagery, well above the instrument noise floor.

Partial curtailment is the harder problem, and the more valuable one

Full shutdowns are eventually announced. The commercially interesting signal is the partial curtailment: a smelter running three of four potlines at 75% capacity, or restarting a single line after a power-price spike subsides. Aluminium producers rarely issue press releases for these events. The thermal record does.

Published work in the remote-sensing literature has linked potline count to nameplate capacity by combining facility engineering data with satellite-derived thermal anomaly geometry. A smelter's total installed capacity is a matter of public record from industry databases; what is not public is how many lines are running on any given week. Counting lit potlines in a Landsat or ECOSTRESS scene, cross-referenced against the facility's known configuration, produces a capacity utilisation estimate with an uncertainty that is honest to state: roughly ±one potline, or ±5–15% of capacity depending on facility size and image geometry.

Cloud cover is the principal operational limit. Landsat's 16-day revisit collapses to usable clear-sky observations perhaps six to ten times per year at persistently cloudy locations such as smelters in southern China or Norway's fjords. ECOSTRESS and VIIRS improve temporal sampling but do not eliminate the problem. A multi-sensor fusion approach, combining all three where available, reduces the expected gap between cloud-free observations to roughly three to seven days at most smelter latitudes.

The physics that makes night passes indispensable

Daytime thermal imagery of industrial sites is complicated by differential solar heating. A south-facing metal roof absorbs more solar radiation than an adjacent concrete apron, creating brightness-temperature gradients that can mimic or mask industrial heat sources. Night-time passes remove this noise almost entirely. ECOSTRESS, because the ISS orbit is not sun-synchronous, frequently acquires night-time passes over smelter latitudes. VIIRS acquires a night-time overpass at roughly 01:30 local solar time for most facilities.

Landsat 8 and 9 are sun-synchronous with a 10:00 local overpass time, meaning all Landsat thermal acquisitions are daytime. This is a genuine limitation for potline detection at smelters in low-contrast terrain or hot climates. Analysts compensate by using the split-window brightness temperature difference between TIRS Band 10 and Band 11 to isolate emissivity anomalies from solar-heated surfaces, but this correction is imperfect and adds uncertainty. For high-confidence potline counting, night-time ECOSTRESS or VIIRS scenes are preferred where available.

From thermal anomaly to commodity intelligence

Raw brightness temperature maps are not the deliverable. The analytic chain runs from scene acquisition through atmospheric correction, land surface temperature retrieval, anomaly extraction against a seasonal background, and finally potline count inference using the facility's known geometry. The output is a time series of estimated operating potlines per facility, updated with each cloud-free observation.

Aggregated across a portfolio of smelters in a given region, this produces a regional capacity utilisation index. Aluminium production is tightly coupled to electricity prices: smelters curtail when power costs exceed the metal price margin. Tracking the thermal record across, say, all major smelters in a grid region during a power-price event allows an analyst to estimate the production response before any official statistics are published. The lag between satellite observation and official production data from national statistics bureaux is typically one to three months. The thermal signal is available within days of acquisition.

Satellize runs this analytic pipeline on open Landsat and VIIRS archives, with ECOSTRESS tasking added on client licence. The methodology is the same class of published industrial thermal remote sensing used in academic literature on aluminium sector monitoring.

Honest limits and what they mean for users

Three limits deserve plain statement. First, spatial resolution: at 100 m, Landsat TIRS cannot resolve individual pots within a potline, only the building as a whole. A potline with a small number of cells offline cannot be distinguished from a fully active one. The method detects building-level status, not cell-level status.

Second, temperature retrieval accuracy degrades with water vapour. Humid tropical locations introduce errors of 1–3 K in surface temperature retrieval even after atmospheric correction, which narrows the margin above background for cooler or partially curtailed lines. This is manageable for large modern smelters with high thermal contrast but can be a problem for smaller or older facilities.

Third, the method requires a known facility configuration to convert thermal anomaly count to capacity utilisation. For smelters where the potline layout is not documented in open sources, the analyst must infer it from the imagery itself, introducing an additional uncertainty. New facilities or recently expanded ones may not yet have published capacity figures that match the observed thermal footprint.

Typical figures

Spatial resolution (primary sensor)	100 m native, 30 m resampled (Landsat 8/9 TIRS); 70 m (ECOSTRESS); 750 m (VIIRS M-band); 1 km (Sentinel-3 SLSTR)
Revisit frequency	16 days per Landsat satellite (8 days combined 8+9); 1–5 days irregular (ECOSTRESS/ISS); ~12 hours (VIIRS, twice daily)
Spectral bands used	LWIR 10.6–12.5 µm (split-window pairs on all sensors); ECOSTRESS adds 8.3–8.6 µm and 8.8–9.2 µm channels
Surface temperature sensitivity	±0.5 K (Landsat TIRS, clear sky); ±0.3 K (ECOSTRESS); ~0.5 K (VIIRS)
Minimum detectable thermal anomaly	Potline building >500 m length reliably detectable; partial curtailment (≥1 potline difference) detectable at most large smelters
Cloud limitation	Thermal LWIR does not penetrate cloud; effective clear-sky observation rate 6–10 passes/year at persistently cloudy sites, higher in arid regions
Archive depth	Landsat from 2013 (TIRS); VIIRS from 2012; ECOSTRESS from 2018
Latency from acquisition to processed output	Landsat Level-2 products available within 12–24 hours of acquisition; VIIRS daily composites within ~6 hours; ECOSTRESS products within 2–5 days
Delivery format	GeoTIFF temperature anomaly maps, CSV time-series per facility, GeoJSON facility status layers

Analytics Satellize can run

Per-facility potline count time series	LWIR brightness temperature retrieval and anomaly extraction against seasonal baseline; potline count inferred from thermal stripe geometry against known facility layout	Monthly CSV or JSON feed of estimated active potlines per monitored smelter
Capacity utilisation estimate	Active potline count divided by known installed potline count from open industry sources; uncertainty bounds stated per facility	Weekly utilisation index (% of nameplate capacity) per smelter, with confidence interval
Curtailment and restart event alerts	Change detection against rolling 30-day thermal baseline; alert triggered when potline count drops or rises by ≥1 line across two consecutive cloud-free observations	Email or API alert within 48 hours of confirmed observation
Regional production capacity index	Aggregation of per-facility utilisation estimates across a defined smelter portfolio (e.g. all facilities in a national grid region or producer group)	Bi-weekly regional index report with facility-level breakdown, PDF and structured data
Historical curtailment reconstruction	Retrospective processing of Landsat and VIIRS archive from 2012 onwards against facility configuration database	Multi-year time series GIS layer and summary table for due-diligence or market-research purposes
Cloud-gap interpolation and confidence flagging	Multi-sensor fusion (Landsat + VIIRS + ECOSTRESS where available) with explicit cloud-gap flagging and interpolation uncertainty stated per observation window	Annotated time series with data-quality flags distinguishing confirmed observations from interpolated estimates

Sources

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.