Illegal charcoal production kiln detection
Active earth kilns emit detectable thermal signatures in Landsat 8/9 TIRS and VIIRS bands; spent kilns leave spectrally distinct dark-soil scars visible in Sentinel-2 and Planet imagery. Together, these two signals support both real-time enforcement and cumulative deforestation accounting.
Sensors
- Landsat 8/9 TIRS: Thermal Infrared Sensor bands 10 and 11 at 100 m native resolution (resampled to 30 m in products). Detects sub-pixel thermal anomalies from active kilns during the smouldering phase, which can last several days. 16-day revisit per satellite; combined Landsat 8 and 9 fleet halves this to roughly 8 days at the equator.
- VIIRS (Suomi-NPP / NOAA-20): The 375 m I-band and 750 m M-band thermal channels, together with the Day/Night Band, underpin the VIIRS Active Fire product (VNP14) and VIIRS Nightfire. Twice-daily overpasses (day and night) give the best temporal coverage for catching kilns during peak heat output, though the 375 m pixel means a single kiln is always sub-pixel and detection depends on temperature contrast rather than resolved size.
- Sentinel-2 MSI: 10 m visible and 20 m shortwave infrared (SWIR) bands, 5-day revisit at the equator with both Sentinel-2A and 2B. The SWIR bands (1610 nm and 2190 nm) are well suited to mapping the charcoal-soil spectral signature of spent kilns, which shows strongly suppressed reflectance relative to surrounding soil. No thermal band, so cannot detect active kilns directly.
- Planet SuperDove: 3 m resolution, 8-band multispectral including red-edge and NIR. Daily revisit over most land areas. Useful for confirming kiln geometry (the characteristic 5–15 m circular scar), counting individual sites, and change detection between Sentinel-2 passes. No SWIR or thermal, so spectral discrimination of charcoal soil is less definitive than with Sentinel-2.
- VIIRS Nightfire (EOG/Mines): Detects sub-pixel combustion sources in the near-infrared and shortwave infrared at night, with reported sensitivity to sources as small as a few square metres under good conditions. Published by the Earth Observation Group at Colorado School of Mines. Provides a complementary night-time thermal signal to daytime Landsat TIRS, useful where kilns are lit or stoked after dark.
Why a kiln is hard to hide and easy to miss
An earth kiln is a low-technology object: a mound of wood, covered in soil, set alight and left to smoulder for several days at temperatures typically between 300 °C and 600 °C at the core. That thermal output is large enough to register as a sub-pixel anomaly in Landsat TIRS and VIIRS even though the kiln footprint is usually only 5 to 15 metres across. The physics is straightforward. The operational challenge is that kilns are deliberately sited under canopy or in remote clearings, operators time burns to avoid overflights where possible, and cloud cover in humid forest zones can block optical and thermal sensors for weeks at a time.
The detection window matters. A kiln burns actively for roughly three to ten days. Miss that window and the thermal signal is gone. What remains is the scar: a circular patch of dark, carbon-rich soil with a spectral signature measurably different from surrounding bare soil or regrowth. That scar can persist for months to years, which is why the two-signal approach, thermal for active sites and SWIR spectral for spent ones, is more informative than either method alone.
What the thermal bands actually see
Landsat 8 and 9 TIRS records at-sensor brightness temperature. A smouldering kiln elevates the temperature of its pixel well above the background, producing a detectable anomaly even when the kiln occupies only a fraction of the 100 m pixel. Published studies using Landsat TIRS have demonstrated detection of agricultural fires and industrial heat sources at sub-pixel scales using contextual algorithms that compare each pixel against a local background distribution. The same principle applies to kilns, though the signal is weaker than an open fire and the algorithm must be tuned to avoid false positives from sun-heated bare soil or metal roofing.
VIIRS offers twice-daily coverage, which dramatically improves the probability of catching a kiln during its active phase. The VIIRS Active Fire product (VNP14) uses a contextual algorithm across the 375 m I-band. A single kiln will not reliably trigger the standard fire-detection threshold, which is calibrated for larger burns, but a cluster of kilns or a particularly hot burn can. VIIRS Nightfire, which processes the night-time near-infrared and SWIR channels, is more sensitive to small persistent heat sources and has been used in published work to detect gas flares as small as a few square metres. Kiln detection via Nightfire is plausible but not yet a routinely validated operational product in the published literature; that caveat belongs in any honest methodology note.
Reading the scar: SWIR spectroscopy of charcoal soil
Charcoal has very low reflectance across the visible and near-infrared spectrum, and this suppression extends into the SWIR. A freshly spent kiln site shows a distinctive spectral shape in Sentinel-2 bands 11 and 12 (1610 nm and 2190 nm) that differs from mineral soils, organic litter and regrowth. The circular geometry of the scar, typically 8 to 20 m in diameter for a traditional earth kiln, is resolvable at Sentinel-2's 20 m SWIR resolution, though a small kiln may occupy only one or two pixels. Planet SuperDove at 3 m resolves the shape clearly but lacks SWIR bands, so it is better used to confirm geometry after a spectral candidate has been identified in Sentinel-2.
Mapping historical scars allows cumulative impact assessment independent of whether any kiln was active during a satellite pass. An analyst can build a time series of scar density across a landscape, correlate it with forest-cover loss from products such as the University of Maryland Global Land Analysis and Discovery (GLAD) alerts, and produce a spatial estimate of charcoal production pressure over months or years. This is the kind of evidence base that UNODC and national forest authorities have described needing for enforcement prioritisation in the DRC and Somalia contexts.
Honest limits: cloud, revisit and the sub-pixel problem
Cloud cover is the dominant operational constraint in humid forest zones. The Congo Basin and Central American forest regions experience cloud fractions above 70 per cent for extended periods, which can reduce effective Sentinel-2 and Landsat revisit to once every several weeks rather than the nominal 5 or 8 days. Thermal bands penetrate thin cloud better than visible bands but not thick convective cloud. SAR (Sentinel-1 C-band) can see through cloud and has been used for forest disturbance mapping, but it does not provide a thermal or charcoal-spectral signal; it detects the structural change in canopy backscatter after clearing, which is a related but distinct signal.
The sub-pixel nature of individual kilns means that detection confidence is probabilistic rather than certain. A single anomalous pixel in TIRS requires corroboration: temporal persistence across multiple passes, spatial clustering with other anomalies, or confirmation in a higher-resolution optical image. False positives from sun-heated laterite soils, small agricultural fires and metal roofing are real. Any operational system should report detection confidence rather than binary presence/absence, and field verification remains necessary before enforcement action.
From detection to enforcement evidence
The analytic workflow moves through three stages. First, a thermal anomaly catalogue is built from VIIRS and Landsat TIRS, filtered by contextual thresholds and cross-referenced against land-cover to exclude known agricultural burning areas. Second, Sentinel-2 SWIR composites are processed to map scar candidates, using spectral indices that emphasise the charcoal-soil signature relative to background. Third, Planet imagery is used for site-level confirmation and counting, producing a georeferenced point layer with confidence scores, estimated scar age and proximity to road or river access routes.
That output is structured for use by forest-law enforcement agencies and prosecutors. A georeferenced record of kiln locations, with timestamps derived from the earliest thermal detection and the spectral scar persistence window, constitutes spatial evidence of production activity. Satellize has built comparable multi-sensor fusion pipelines for agricultural monitoring, including its crop-estimation work for the Kingdom of Tonga, and applies the same sensor-agnostic architecture to enforcement use cases. The deliverable is not a map for its own sake but a structured evidentiary record that can be handed to a legal team or integrated into a national forest-monitoring system.
What a programme actually costs in data terms
Sentinel-2 and Landsat are open-access. VIIRS Active Fire data are free via NASA FIRMS. VIIRS Nightfire data are freely distributed by the Earth Observation Group. The cost of a kiln-monitoring programme is therefore dominated by processing, algorithm development, cloud-gap-filling strategy and analyst time, not by raw data acquisition. Commercial Planet tasking adds cost but is generally only needed for site-level confirmation of high-priority detections rather than area-wide screening.
A practical programme design screens a target area weekly using free thermal and optical data, flags candidate sites above a confidence threshold, and triggers commercial tasking only for confirmed candidates. This keeps commercial data costs proportional to the number of active detections rather than to the area under surveillance. For a large forest concession or a national monitoring programme covering hundreds of thousands of square kilometres, that distinction is significant.
Typical figures
| Thermal detection resolution (Landsat TIRS) | 100 m native, 30 m resampled; sub-pixel sources detectable via contextual algorithms |
| Thermal detection resolution (VIIRS) | 375 m (I-band), 750 m (M-band); individual kilns are sub-pixel |
| SWIR scar mapping resolution (Sentinel-2) | 20 m (bands 11 and 12); small kilns may occupy 1–2 pixels |
| Scar confirmation resolution (Planet SuperDove) | 3 m; resolves kiln geometry but lacks SWIR bands |
| Revisit (VIIRS, day + night combined) | Twice daily globally |
| Revisit (Landsat 8 + 9 combined) | Approximately 8 days at the equator |
| Revisit (Sentinel-2A + 2B combined) | 5 days at the equator under clear sky |
| Minimum detectable kiln diameter (spectral scar, Sentinel-2) | Approximately 10–20 m; smaller scars require Planet confirmation |
| Archive depth | Landsat from 1972; Sentinel-2 from 2015; VIIRS from 2012; Planet from approximately 2016 |
| Delivery formats | GeoJSON point layers, GeoTIFF anomaly rasters, timestamped PDF site reports, GIS-ready shapefiles |
Analytics Satellize can run
| Active kiln thermal anomaly catalogue | Contextual brightness-temperature anomaly detection on Landsat TIRS and VIIRS I-band, filtered against land-cover and known false-positive classes | Weekly georeferenced point layer with confidence score, detection timestamp and pixel temperature delta |
| Spent kiln scar map | SWIR spectral index analysis on Sentinel-2 band 11/12 composites, combined with circular-feature detection for geometric confirmation | GeoJSON polygon layer of scar candidates with estimated age range derived from time-series first-detection date |
| Site-level confirmation imagery | Planet SuperDove tasking triggered by thermal or spectral candidate flags; visual and automated geometry verification | 3 m true-colour and false-colour chip per candidate site, with annotated kiln count and diameter estimate |
| Cumulative production pressure index | Scar density mapping over rolling 12-month windows, correlated with GLAD forest-cover loss alerts to estimate production volume proxies | Monthly raster grid of kiln scar density per km², exportable to GIS or PDF report |
| Night-time combustion detection | VIIRS Nightfire sub-pixel combustion detection in near-infrared and SWIR channels, cross-referenced with daytime thermal catalogue | Alert feed of night-time anomalies with coordinates, estimated radiant heat and overlap with known kiln zones |
| Enforcement prioritisation layer | Spatial clustering of active and historical detections, overlaid with road/river access buffers and protected-area boundaries | Ranked site list with access-route analysis, suitable for patrol planning or legal case file |
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.