Sensors
- Planet SuperDove (PlanetScope): 3 m native resolution, near-daily revisit over most of Africa. Eight spectral bands including red-edge and NIR. Used to flag new clearings in time series and detect SWIR-elevated pixels during active firing, though the SWIR band (at roughly 1.37 µm on SuperDove) is narrow and thermal sensitivity is limited compared with dedicated thermal sensors.
- Planet SkySat: 50 cm pan-sharpened resolution, tasked on demand. Confirms kiln morphology: the characteristic circular mound, spoke-pattern felling and bare-earth ring are individually resolvable. Revisit is tasking-dependent rather than systematic, so it is used for confirmation rather than screening.
- Sentinel-2 MSI: 10 m visible and NIR, 20 m SWIR bands (1.61 µm and 2.19 µm), 5-day revisit at the equator. SWIR bands respond to active combustion and freshly disturbed soil. Free and archived from 2015, making it the backbone of change-detection time series at national scale. No thermal band.
- Landsat 8/9 TIRS: Thermal Infrared Sensor provides Band 10 at 100 m (resampled to 30 m in products), sensitive to surface temperature anomalies from active kilns. 16-day single-satellite revisit, 8-day combined. Spatial resolution means a single kiln rarely produces a detectable pixel, but clusters of kilns or large earthmound fires can exceed the detection threshold. Free archive from 2013 (Landsat 8) and 2021 (Landsat 9).
Why charcoal is the deforestation signal that coarse systems miss
Most satellite deforestation alerts are calibrated for large-scale clearing: mechanised agriculture, plantation conversion, logging roads. Charcoal production is different. A single traditional earth kiln occupies a clearing of roughly 10 to 30 metres in diameter. The firing cycle lasts one to three weeks. The site may be abandoned and partially revegetated within a season. At 250 m resolution, the MODIS-era systems that defined the field simply cannot see it.
The scale of the problem is not small. Charcoal is the primary cooking fuel for several hundred million people across sub-Saharan Africa, and the supply chain is largely informal. In countries such as the Democratic Republic of Congo, Tanzania and Zambia, charcoal production has been identified in peer-reviewed literature as a leading proximate cause of forest loss in specific landscapes, even where headline deforestation figures look modest. A monitoring system that cannot detect kilns is, in effect, blind to a substantial fraction of actual forest loss.
Two signatures, two sensor families
Active charcoal kilns produce heat. Traditional earth mounds reach internal temperatures of 300 to 500 °C during carbonisation, though the earthen cover suppresses surface emission considerably. The surface temperature anomaly above an active kiln is detectable in Landsat TIRS Band 10 when kilns cluster, but a single isolated kiln at 100 m thermal resolution is often below the noise floor. SWIR bands are more useful here: the 1.6 µm and 2.2 µm channels on Sentinel-2 and the SWIR bands on SuperDove respond to sub-pixel thermal emitters, a principle well-established in active fire remote sensing. A smouldering kiln that would not trigger a standard fire alert can still elevate SWIR reflectance measurably above background forest.
The second signature is geometric and persistent. Once a kiln is built, the surrounding trees are felled in a rough circle to provide feedstock. This clearing, typically 10 to 30 m across, remains visible for months after firing ends. At 3 m resolution (Planet SuperDove) the clearing is clearly non-forest. At 50 cm (SkySat or WorldView) the conical mound, the radial felling pattern and the charcoal-blackened soil are individually identifiable. The combination of a SWIR anomaly in time series followed by a persistent circular clearing in high-resolution imagery is the detection logic: one sensor class screens, the other confirms.
What a time series actually reveals, and what it cannot
Planet SuperDove's near-daily cadence over Africa means that, in cloud-free conditions, the window between a clearing appearing and being detected can be days rather than months. Change-detection algorithms applied to the red-edge and NIR bands (normalised difference vegetation index differencing, or band-ratio anomaly scoring) flag pixels that transition from vegetated to bare. A new circular clearing of 15 m diameter covers roughly 175 square metres, which at 3 m resolution occupies fewer than 20 pixels. That is detectable, but the signal is not unambiguous: small agricultural plots, termite mounds and natural tree falls produce similar footprints.
Cloud cover is the honest limiting factor. Across the Congo Basin and parts of East Africa, persistent cloud can suppress optical observations for weeks during the wet season. SAR (Sentinel-1 C-band) can detect clearings through cloud, but the spatial resolution and the difficulty of distinguishing small bare patches from other low-backscatter surfaces make it a secondary rather than primary tool for kiln-scale detection. Thermal detection from Landsat is similarly cloud-blocked. No optical or thermal system solves the cloud problem; it is a physical constraint, not a processing one.
Archive depth matters for enforcement. Sentinel-2 data is available from 2015, Landsat from 1972 (though at 30 m). A retrospective time series can establish when a forest parcel was intact, when the first clearing appeared and how many firing cycles have occurred. That historical record is often more useful to a forestry authority than a real-time alert alone.
From pixel to prosecution: what the analytic chain produces
Detection is not enforcement. The gap between a flagged pixel and a usable evidence package is where most remote-sensing programmes fail in practice. A credible analytic product for a forestry authority needs: a georeferenced location accurate enough to navigate to in the field (Planet 3 m positional accuracy is typically within one to two pixels after orthorectification); a temporal record showing when the clearing appeared; a high-resolution confirmation image showing kiln morphology; and an estimate of the area of forest removed.
Area estimation at kiln scale carries genuine uncertainty. A 20 m circular clearing represents roughly 0.03 hectares. Multiply that by hundreds of sites across a landscape and the aggregate is significant, but individual-site area figures should carry error bars of 20 to 40 percent at 3 m resolution, wider if the clearing boundary is irregular or partially shaded by adjacent canopy. Honest reporting of that uncertainty is part of what makes the evidence defensible.
Practical constraints a buyer should price in
SkySat and WorldView tasking over Africa is available but competes with demand from other customers and is subject to cloud risk on any given pass. A confirmation image may take days to acquire in a high-cloud season. Building a workflow that automatically triggers a tasking request when a SuperDove anomaly is detected reduces latency, but does not eliminate it.
Kiln sites in dense forest are sometimes partially canopy-covered, particularly early in the firing cycle before full clearing. A sensor looking straight down may see only a gap in the canopy rather than the mound itself. Oblique viewing geometry from SkySat (which can collect off-nadir) can improve mound visibility, but introduces geometric distortion that complicates area measurement.
Satellize's analytics stack runs on open constellations (Sentinel-2, Landsat) for screening and adds commercial tasking on client licence for confirmation. The workflow is similar in structure to the crop-estimation approach used in the Kingdom of Tonga programme, adapted for change detection rather than phenological modelling. Forestry authorities wanting to pilot the approach should start with a defined landscape of known charcoal pressure, run a 12-month retrospective to establish baseline detection rates, and calibrate against field-verified kiln locations before scaling nationally.
Typical figures
| Screening spatial resolution | 3 m (Planet SuperDove); 10–20 m (Sentinel-2 MSI) |
| Confirmation spatial resolution | 50 cm (Planet SkySat); 30–50 cm (WorldView tasking) |
| Thermal band resolution | 100 m native / 30 m resampled (Landsat 8/9 TIRS Band 10) |
| Screening revisit | Near-daily (Planet SuperDove, cloud-permitting); 5 days (Sentinel-2 at equator) |
| Confirmation revisit | Tasking-dependent; typically 1–5 days from order to collect (SkySat) |
| Minimum detectable clearing | ~10 m diameter at 3 m resolution; ~30 m diameter reliably at 10 m resolution |
| SWIR bands used | Sentinel-2 Band 11 (1.61 µm) and Band 12 (2.19 µm); Landsat 8/9 Band 6 (1.57 µm) and Band 7 (2.11 µm) |
| Archive depth | Sentinel-2 from 2015; Landsat from 2013 (L8) / 2021 (L9); Planet from ~2016 for most African coverage |
| Cloud limitation | Optical and thermal blocked by cloud; Congo Basin and coastal West Africa can see 10+ consecutive cloudy days in wet season |
| Typical alert latency | 1–3 days from acquisition to screened alert; 3–7 days to confirmed high-resolution image in good conditions |
Analytics Satellize can run
| New clearing alert layer | NDVI differencing and band-ratio anomaly scoring on Planet SuperDove time series; threshold-based change flagging | GeoJSON alert feed with location, date of first detection and thumbnail chip |
| SWIR thermal anomaly map | Sub-pixel fire / thermal emitter detection adapted from MODIS active fire methodology, applied to Sentinel-2 Band 11/12 and Landsat TIRS; anomaly scoring against local background | Raster layer of SWIR anomaly magnitude, updated per available acquisition |
| Kiln morphology confirmation report | Visual and automated shape analysis of SkySat or WorldView 50 cm imagery; circular clearing detection using Hough transform or template matching | PDF site report per confirmed kiln cluster: coordinates, area estimate, acquisition date, annotated image chip |
| Retrospective site history | Dense time-series stack from Sentinel-2 archive (2015–present); per-pixel trajectory analysis to date first clearing and count recurrence | GIS polygon layer with first-detected date, number of firing cycles inferred, cumulative cleared area |
| Landscape-scale kiln density map | Spatial clustering of confirmed and candidate sites; kernel density estimation over a defined forest management unit or protected area buffer | Heatmap raster and summary statistics table by administrative unit, suitable for enforcement prioritisation |
| Forest area loss attribution | Intersection of kiln-associated clearings with forest baseline mask (Hansen/UMD or national forest inventory); area calculation with stated uncertainty bounds | Tabular report of hectares lost by site class, period and administrative boundary, with confidence intervals |
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.