Unlicensed sawmill and timber-processing site activity monitoring
Unlicensed sawmills leave distinctive footprints in satellite imagery: log-yard stockpiles, sawdust waste, and vehicle tracks. Cross-referencing detected sites against national concession registers separates the detection problem from the legality question.
Sensors
- PlanetScope SuperDove: 3-metre nominal GSD, 8 spectral bands including red-edge and near-infrared, daily revisit globally. Sufficient to resolve individual log-pile geometry and detect sawdust waste areas by their distinctive pale spectral signature. Archive from 2016.
- Sentinel-2 MSI: 10-metre resolution in visible and NIR bands, 20-metre in red-edge and SWIR. Five-day revisit at the equator (2–3 days with both satellites). Free and open; useful for change detection across large forest concession areas and for time-series of log-yard extent. Cloud cover in tropical regions limits usable acquisitions to roughly 30–50% of passes.
- Maxar WorldView-3: 31-cm panchromatic, 1.24-metre multispectral, 8 VNIR bands plus 8 SWIR bands. Resolves individual logs and vehicle types. Tasked on demand; cost and revisit (1–4.5 days depending on latitude and off-nadir tolerance) make it appropriate for confirmation and legal-quality documentation rather than wide-area screening.
- Sentinel-1 SAR: C-band SAR, IW mode at 10-metre resolution, 6-day repeat (12-day per satellite). Cloud-independent. Rough-surface log piles produce elevated backscatter relative to bare ground, and temporal coherence loss over active yards can flag ongoing disturbance. Useful during prolonged cloud seasons when optical data is unavailable.
What a log yard looks like from 400 kilometres up
Sawmill compounds have a physical grammar that persists across geographies. A cleared rectangular or irregular pad, typically 0.5 to several hectares, sits adjacent to a forest edge or a road junction. Log piles appear as elongated high-reflectance objects casting short, directional shadows. Sawdust and wood-chip waste accumulates in mounds with a characteristic pale yellow-white spectral signature, distinct from bare soil and from the surrounding canopy. Corrugated metal roofing produces strong specular returns. Access tracks show compaction and repeated vehicle disturbance.
At PlanetScope's 3-metre resolution these features are individually resolvable. At Sentinel-2's 10 metres, individual logs merge but the cleared compound, waste areas and spectral contrast with surrounding vegetation remain detectable. Object-based image analysis (OBIA) segments the scene into spectrally and geometrically coherent objects, then classifies them against these signatures. Published work on industrial facility detection in tropical forest contexts, including studies indexed in Remote Sensing (MDPI), has demonstrated that OBIA outperforms pixel-based classifiers for irregular anthropogenic clearings precisely because shape and texture matter as much as spectral value.
Stockpile change as a proxy for throughput
A static sawmill compound tells you a site exists. A time-series of log-yard extent tells you whether it is operating. When the log pile shrinks between two acquisitions and then re-accumulates, timber is moving through. When a yard that was empty three months ago now holds a substantial stockpile, harvesting is active upstream.
PlanetScope's daily cadence makes it possible to track these cycles at roughly weekly effective resolution in cloud-affected tropical regions, depending on cloud persistence. Sentinel-2 provides the longer archive (back to 2015 for Sentinel-2A) at lower cost, which matters when investigators need to establish when a site first appeared and whether it predates or postdates a concession boundary change. The two datasets are complementary: Sentinel-2 for historical baseline and trend, PlanetScope for near-real-time stockpile monitoring.
Honest caveat: persistent cloud cover over the Congo Basin and parts of the Amazon can produce gaps of weeks in optical time-series. Sentinel-1 SAR partially fills these gaps, but log-pile backscatter interpretation requires careful calibration against known targets, and the method is less mature than optical OBIA for this specific application.
Detection is not conviction: the concession-register problem
Identifying an active timber-processing compound from imagery is a tractable technical problem. Determining whether that compound is licensed is a data-integration problem, and the two must not be conflated.
National forestry concession registries, where they exist and are accessible, define the legal boundary. In Myanmar, Global Witness documented in published reports how timber was processed through facilities whose declared output bore no plausible relationship to their visible log-yard activity. In Brazil, the Environmental Investigation Agency has published casework showing how fraudulent DOFS (Documento de Origem Florestal) credits were used to launder timber from unlicensed sources. In the Congo Basin, concession boundaries are sometimes poorly georeferenced in public databases, creating genuine ambiguity even when imagery is clear.
The practical workflow is: detect and geolocate the site from imagery; query it against the available concession layer; flag the discrepancy class (inside a valid concession, outside all concessions, inside an expired or suspended concession, or location ambiguous relative to boundary uncertainty). Each class carries a different enforcement implication and a different evidential burden. Satellite data handles the first step. The rest requires legal and administrative data that varies enormously by country.
Vehicle traffic as a secondary signal
Log trucks are large, slow-moving, and they follow predictable routes between forest, mill and export point. At WorldView-3 resolution, individual vehicles are detectable. At PlanetScope resolution, repeated disturbance of access tracks, and the presence of truck-sized objects in yards, is interpretable. Sentinel-2 cannot resolve individual vehicles but can detect road-surface changes consistent with heavy traffic over time.
Traffic frequency matters because it is correlated with throughput volume. A facility that shows two or three truck-sized objects in a yard in a daily PlanetScope image, consistently over weeks, is processing material at a rate that should appear in declared production statistics. When it does not, the discrepancy is the anomaly that enforcement agencies need. This is the logic that Global Witness applied, using commercial satellite imagery, in its published Myanmar investigations.
Spectral limits and what they cannot tell you
Optical sensors distinguish wood waste from bare soil by reflectance. They do not distinguish species. They cannot determine whether logs are hardwood or softwood, protected species or plantation timber. SWIR bands on WorldView-3 and Sentinel-2 provide some additional discrimination of surface moisture and material texture, but species-level identification from space is not operationally achievable with current systems.
Shadow, seasonal vegetation change and variable sun angle all affect log-pile detection reliability. A pile that is clearly visible in dry-season imagery with low sun angle may be harder to separate from surrounding ground in wet-season imagery with high solar elevation. Detection models trained on one season or one geography require validation before being applied elsewhere. These are not theoretical concerns; they affect the false-positive rate in any operational monitoring programme, and any credible analytics service should report them.
From screening to enforcement-ready documentation
Wide-area screening using Sentinel-2 time-series can flag candidate sites across an entire forest jurisdiction in a single analytical pass. Candidate sites are then triaged by confidence score and proximity to known concession gaps. High-priority sites receive PlanetScope time-series analysis for stockpile change quantification. Sites that pass both filters can be submitted for WorldView-3 tasking to produce sub-metre imagery suitable for inclusion in enforcement documentation.
Satellize runs this kind of multi-tier optical analysis on open and commercial constellations for government clients. The Tonga crop-estimation programme demonstrates the same underlying logic applied to agricultural stockpile change, and the methodology transfers directly to timber-yard monitoring. The practical output for a forestry enforcement agency is a prioritised site list with confidence tiers, georeferenced polygons, and a time-stamped change record, not a single image.
The most useful thing a monitoring programme can do is reduce the investigator's field-verification burden. If satellite analysis can narrow a region of several million hectares to thirty candidate sites with documented stockpile activity and concession-register mismatches, field teams can work efficiently. That is the realistic value proposition, stated plainly.
Typical figures
| Screening spatial resolution | 10 m (Sentinel-2 MSI, visible/NIR bands) |
| Confirmation spatial resolution | 3 m (PlanetScope SuperDove); 31 cm pan / 1.24 m MS (WorldView-3 on tasking) |
| Revisit cadence | Daily (PlanetScope); 2–5 days (Sentinel-2, cloud-free); 6 days per satellite (Sentinel-1 SAR) |
| Spectral bands used | Visible, red-edge, NIR, SWIR (optical); C-band backscatter (SAR) |
| Minimum detectable compound area | Approximately 0.5 ha at Sentinel-2 (10 m); smaller clearings resolvable at PlanetScope 3 m |
| Cloud-cover limitation | Optical unusable under dense cloud; SAR gap-fill available but with reduced interpretability for log piles |
| Archive depth | Sentinel-2 from 2015; PlanetScope from 2016; Landsat back to 1972 for coarser historical baseline |
| Latency (screening alert to delivery) | 24–72 hours from acquisition for automated change flags; 5–10 days for full analytical report |
| Delivery formats | GeoTIFF, GeoJSON, Shapefile, PDF report, GIS-ready vector layers |
Analytics Satellize can run
| Active sawmill compound detection | Object-based image analysis (OBIA) on PlanetScope and Sentinel-2 multispectral imagery; spectral and geometric feature classification | Georeferenced polygon layer of detected compounds with confidence score and date of first detection |
| Log-yard stockpile extent time-series | Multi-date OBIA change detection on PlanetScope daily stack; area calculation per classified log-pile object | Time-series chart and GIS layer showing stockpile area per site per acquisition date |
| Concession-register cross-reference | Spatial join of detected compound polygons against client-supplied or open national forestry concession boundaries; discrepancy classification | Prioritised site list with licensing status flag (inside valid, outside all, expired/suspended, boundary-ambiguous) |
| Vehicle-traffic frequency index | Object detection of truck-scale features in PlanetScope and WorldView-3 imagery; temporal frequency count per site | Per-site traffic index table correlated with stockpile-change periods |
| SAR-based cloud-season gap-fill | Sentinel-1 IW backscatter change detection and temporal coherence analysis over candidate sites during optical-blackout periods | Binary activity flag (change / no change) per site per SAR acquisition, appended to optical time-series |
| Enforcement-documentation image package | WorldView-3 tasking for confirmed high-priority sites; orthorectified sub-metre imagery with metadata chain | Timestamped, georeferenced image set suitable for inclusion in legal or regulatory submissions |
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.