Cement plant clinker production monitoring by kiln thermal signature

Rotary cement kilns produce a persistent thermal plume readable by spaceborne infrared sensors. Time-series radiance anomalies reveal kiln-on/off states and partial-load operation, giving lenders and commodity desks an independent production signal.

Sensors

Landsat 8/9 TIRS: Two thermal infrared bands (Band 10: 10.6–11.2 µm, Band 11: 11.5–12.5 µm) at 100 m native resolution (resampled to 30 m). 16-day exact repeat per satellite; combined Landsat 8 and 9 gives roughly 8-day revisit at mid-latitudes. Radiometric accuracy better than 0.5 K, sufficient to resolve kiln-on radiance against background. Free archive to 2013.
ECOSTRESS (ISS-mounted): Five thermal infrared bands (8–12.5 µm) at approximately 70 m resolution. Non-sun-synchronous orbit from the ISS gives variable overpass times, occasionally sampling night-time kiln states that Landsat misses. Revisit is irregular, roughly 3–5 days at most latitudes, with gaps. Useful for cross-validating Landsat-derived anomalies.
VIIRS M-band thermal (M14/M15/M16): 375 m resolution thermal bands at 3.74 µm (M13, optimised for high-temperature anomalies) and 11/12 µm. Daily global coverage from Suomi-NPP and NOAA-20. At 375 m a single large kiln is a sub-pixel source, so radiance is mixed; useful for detecting clear on/off states and tracking multi-kiln complexes rather than individual units. The VIIRS Nightfire algorithm (Colorado School of Mines / NOAA) is the published method for sub-pixel high-temperature source detection.
Planet Fusion thermal composites: Planet's Fusion product blends high-cadence PlanetScope optical data with Landsat thermal to produce near-daily thermal-consistent composites. Spatial resolution follows Landsat TIRS at 100 m. Reduces cloud-gap problem at the cost of temporal smoothing; not a direct radiance measurement but useful for filling the 8-day Landsat revisit window.

What a rotary kiln actually radiates, and why it is readable from orbit

A cement rotary kiln is a steel cylinder up to 6 m in diameter and 90 m long, inclined slightly and rotating at roughly one to three revolutions per minute. The burning zone inside reaches 1400–1500 °C. Even after the clinker exits into a grate cooler, the discharge hood and the first metres of the cooler vent air at several hundred degrees Celsius. That heat load is not subtle. The clinker cooler exhaust alone can represent 30–40 percent of total kiln heat output, and it exits through a stack or louvred vents that are spatially compact and thermally distinct from the surrounding plant.

Long-wave infrared sensors in the 10–12 µm range detect surface-emitted radiance. At 1450 °C a kiln shell and discharge structure radiate far above the background terrain temperature of, say, 25–40 °C. Even accounting for atmospheric attenuation, the contrast is detectable at Landsat TIRS resolution. Mid-wave infrared (around 3.7–4 µm, the VIIRS M13 band) is more sensitive to high-temperature anomalies because the Planck function peaks shift dramatically with temperature: a 1400 °C source emits orders of magnitude more mid-wave radiance than ambient background. This is the physical basis that makes kiln monitoring tractable even from sensors not designed for industrial surveillance.

From radiance anomaly to kiln state: the inference chain

The analytic workflow begins with per-pixel radiance time series extracted at known kiln locations. Plant coordinates come from public sources: satellite base imagery, OpenStreetMap industrial tagging, and national environmental permit registries where available. For each cloud-free acquisition, the at-sensor radiance in the thermal band is compared against a baseline derived from the same pixel in confirmed kiln-off periods (scheduled maintenance shutdowns, public holidays in cement-producing markets, or periods corroborated by AIS-derived clinker shipment gaps).

A radiance value exceeding the baseline by a threshold determined from the sensor's noise-equivalent temperature difference (NEdT) indicates kiln-on. Landsat TIRS NEdT is approximately 0.4 K; a running kiln typically produces anomalies of 5–15 K above background at 100 m resolution, well above the noise floor. Partial-load operation is harder. Below roughly 30 percent of rated throughput, the thermal signature compresses toward the detection margin, and the method cannot reliably distinguish a kiln running at 25 percent capacity from one in a warm-standby state. That limit is honest and should be built into any credit model that uses this signal.

Cloud cover is the principal operational constraint. Thermal infrared cannot penetrate cloud. In monsoon-affected markets (South and Southeast Asia, parts of West Africa) cloud-free acquisitions may be separated by three to six weeks during peak season. Fusion composites and VIIRS daily passes partially compensate, but VIIRS at 375 m cannot resolve individual kilns at smaller plants. A practical approach combines Landsat for spatial precision with VIIRS for temporal density, accepting that VIIRS detections are plant-level rather than kiln-level.

Translating kiln-on time to clinker volume: calibration and its limits

Kiln-on hours are not directly clinker tonnes. The conversion requires a plant-specific calibration that links operating time to throughput. Published engineering ranges for modern dry-process kilns run from roughly 1,500 to 10,000 tonnes of clinker per day depending on kiln diameter and length. That is a wide range. Without ground-truth data (plant capacity filings, clinker export manifests, or energy-consumption disclosures) the satellite-derived signal gives a relative utilisation index rather than an absolute volume.

For financial intelligence purposes, the relative index is often sufficient. A lender assessing credit exposure to a cement producer does not need to know whether a plant produced 4,200 or 4,400 tonnes on a given day; they need to know whether the plant ran continuously for six weeks, shut down for ten days, then restarted at reduced intensity. That pattern is readable. For commodity-supply nowcasting, the same logic applies: directional changes in aggregate kiln-on time across a regional cluster of plants are a leading indicator of clinker and cement supply, which feeds into construction-sector activity models.

What the signal cannot tell you

Clinker grade is invisible to thermal sensors. High-grade clinker and standard OPC clinker burn at similar temperatures; the kiln signature does not distinguish them. Fuel type is also ambiguous: a kiln co-firing waste-derived fuel and one burning coal produce comparable thermal plumes, though mid-wave radiance ratios can sometimes hint at combustion temperature differences. Capacity utilisation below about 30 percent is below the reliable detection threshold for most sensor configurations.

Stockpile behaviour adds another layer of uncertainty. A plant can run kilns at full capacity while building inventory rather than shipping, decoupling production from sales. Combining thermal kiln monitoring with SAR-based clinker silo or stockpile volume estimation (covered in the grain-storage and coal-stockpile sibling pages) tightens the inference, but that integration requires additional analytic effort and additional data licences. Neither method alone closes the loop between production and market delivery.

Practical applications for financial and commodity clients

Credit analysts at banks with exposure to cement producers use kiln monitoring to cross-check management guidance against physical activity. If a producer reports sustained high utilisation in a quarterly filing but satellite data shows repeated multi-day shutdowns during that period, the discrepancy is a due-diligence flag, not a conclusion. The satellite record supports the question; it does not replace the audit.

Commodity desks tracking regional cement supply use aggregated kiln-on indices across clusters of plants (the Chinese Yangtze Delta, the Indian Gangetic Plain, the Turkish Aegean coast) as a nowcast of near-term clinker availability. These indices can be updated on an 8-day cadence with Landsat and daily with VIIRS at lower spatial precision. For parametric financial instruments tied to construction-sector activity, a kiln-index time series provides a physically grounded, independently verifiable underlying variable.

Satellize runs this type of thermal anomaly analysis on open Landsat and VIIRS archives, with commercial tasking added where a client needs sub-weekly revisit at specific plants. The methodology is the same published radiance-anomaly approach used in peer-reviewed remote sensing literature; what varies is the financial framing, the alert thresholds, and the delivery format.

Typical figures

Spatial resolution (primary sensor)	100 m native, 30 m resampled (Landsat 8/9 TIRS); ~70 m (ECOSTRESS); 375 m (VIIRS M-band)
Revisit cadence	~8 days combined Landsat 8+9 at mid-latitudes; daily with VIIRS (lower spatial precision); irregular 3–5 days ECOSTRESS
Thermal bands used	LWIR 10.6–12.5 µm (Landsat TIRS, VIIRS M15/M16); MWIR 3.74 µm (VIIRS M13 for high-temperature anomaly); 8–12.5 µm (ECOSTRESS)
Noise-equivalent temperature difference (NEdT)	~0.4 K (Landsat TIRS); ~0.3 K (ECOSTRESS); varies by VIIRS band
Minimum detectable kiln state	On/off reliably detected; partial load below ~30% utilisation is below reliable discrimination threshold
Cloud penetration	None. Thermal infrared is blocked by cloud; cloud-affected acquisitions are flagged and excluded
Archive depth	Landsat TIRS from 2013 (Landsat 8); VIIRS from 2012 (Suomi-NPP); ECOSTRESS from 2018
Latency (open data)	Landsat: typically 12–24 hours after acquisition; VIIRS: same-day to next-day via FIRMS/NOAA
Coverage	Global land; polar regions limited by orbit geometry and darkness for optical cross-reference
Delivery formats	GeoTIFF radiance anomaly layers, CSV kiln-state time series, GeoJSON plant-status feed, PDF periodic report

Analytics Satellize can run

Kiln on/off state log	Per-pixel radiance anomaly detection against plant-specific baseline; threshold set at 3-sigma above NEdT noise floor	CSV or JSON time series per plant, flagged by cloud cover and sensor; updated on each cloud-free acquisition
Partial-load utilisation index	Normalised radiance excess above kiln-on baseline, binned into low/medium/high tiers; sub-30% utilisation reported as indeterminate	Weekly GIS layer with utilisation tier per plant; confidence flag included
Regional kiln-activity aggregate	Spatial aggregation of kiln-on fractions across a defined plant cluster; weighted by published plant capacity where available from permit filings	Monthly index chart and underlying data table, formatted for integration into credit or commodity models
Shutdown and restart event alerts	Change-point detection on radiance time series; alert triggered when kiln transitions from on to off (or reverse) across two consecutive cloud-free acquisitions	Near-real-time alert (email or API push) with acquisition timestamp, coordinates, and prior state duration
Historical production pattern baseline	Multi-year Landsat and VIIRS archive analysis to establish seasonal operating patterns, planned maintenance windows, and anomalous shutdown periods	PDF due-diligence report covering up to 10 years of kiln-state history per plant or portfolio
Cross-sensor consistency check	Comparison of Landsat TIRS and VIIRS M13 detections at the same plant to flag sensor artefacts or cloud-mask failures	Quality-assurance annex appended to time-series deliverables; disagreement events flagged for analyst review

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.