Illegal logging activity detection
Selective illegal felling leaves faint but readable signatures: coherence loss in SAR time-series, skid trails visible in very-high-resolution optical imagery, and clearance polygons that cross-check against published concession cadastres.
Sensors
- Sentinel-1 SAR (C-band, ESA): 10 m ground range resolution in IW mode, 6-day repeat at the equator (12-day for a single satellite). C-band coherence drops measurably within one or two repeat cycles when canopy is disturbed or removed. Cloud-transparent, so it works in the wet tropics where optical sensors are routinely blinded for months.
- Capella Space SAR (X-band): Spotlight mode delivers 0.5 m resolution, tasked on demand with typical revisit of hours to days depending on orbit geometry. X-band is more sensitive to surface roughness changes and small structural features than C-band, making it better suited to detecting individual skid trail incursions, though archive depth is limited compared with Sentinel-1.
- Planet SkySat: 0.5 m pan-sharpened optical imagery, taskable within roughly 24 hours. Useful for confirming trail geometry and log deck extent identified in SAR, and for documenting evidence-grade imagery. Revisit is demand-driven and per-scene cost is significant at scale.
- Maxar WorldView-3: 0.31 m panchromatic, 1.24 m multispectral, 16 SWIR bands at 3.7 m. The SWIR bands can distinguish bare soil, slash debris and standing vegetation in a single pass, aiding classification of disturbed ground even under partial canopy. Again, expensive to task broadly.
- Sentinel-2 MSI (ESA): 10 m visible and NIR, 20 m SWIR, 5-day revisit at mid-latitudes. Useful for monitoring larger clearances and for NDVI time-series that flag canopy loss, but individual felled trees and narrow trails (under roughly 10 m wide) are below the detection floor. Cloud cover in humid tropical forests frequently degrades the optical record.
What a felled tree actually looks like from orbit
A single felled tree is invisible to any medium-resolution sensor. Sentinel-2 at 10 m and Landsat at 30 m cannot resolve a trunk, a stump, or even a small group of stumps. What they can detect is the aggregate signal of canopy loss across a polygon large enough to register: roughly 0.1 hectares at best in 10 m imagery, and that only under good illumination and low cloud cover.
Selective illegal logging is specifically designed to avoid that threshold. A team felling a few dozen high-value trees per hectare may remove only 10 to 20 percent of canopy cover. The remaining canopy closes the gap in medium-resolution optical data within a single growing season. The honest conclusion is that medium-resolution optical sensors alone are insufficient for detecting selective felling; they are useful for monitoring large-scale clearance and for providing spatial context.
Why SAR coherence changes the calculation
Synthetic aperture radar measures the phase relationship between two passes over the same ground. When the canopy is undisturbed, the scattering geometry is stable and coherence between repeat passes is relatively high. Felling, dragging logs, and operating machinery disrupts that geometry. Coherence drops, and the drop persists for several repeat cycles. In C-band Sentinel-1 data, studies in tropical forests have documented coherence loss detectable within one to two 12-day repeat cycles following disturbance events of moderate extent.
The limitation is spatial. A 10 m coherence-change product cannot resolve a 3 m skid trail. What it can do is flag a zone of disturbance, which then justifies tasking a commercial sensor for confirmation. The workflow is therefore tiered: Sentinel-1 coherence change provides wide-area, cloud-transparent screening; Capella Space or SkySat provides the spatial resolution needed to document trail geometry and log deck dimensions. Running that full workflow over a large concession every few weeks is operationally feasible. Running it everywhere at once is not.
Skid trails as the diagnostic signature
Skid trails are the most reliable indicator of active selective logging. A skid trail is a compacted, vegetation-cleared path, typically 3 to 6 m wide, used to drag felled logs to a collection point. At 0.5 m resolution (SkySat, Capella spotlight, WorldView-3) they are clearly visible as linear bare-soil features branching from a central log deck. The log deck itself, a cleared area of 0.1 to 0.5 hectares where logs are sorted and loaded, is detectable even in 3 m PlanetScope imagery under clear conditions.
The diagnostic value of skid trails is that they are persistent. Even after logging ceases, the compacted soil resists revegetation for months to years, leaving a detectable signature in both optical and SAR data. This means that archive imagery can be used retrospectively to establish when a trail network appeared, which is useful for enforcement proceedings that require a timeline.
Concession cross-referencing: where the legal question enters
Detecting a clearance polygon is not the same as proving illegal activity. The analytic step that gives the detection enforcement value is spatial intersection with published concession boundaries. Several countries publish logging concession cadastres as open geospatial data, including Indonesia through the KLHK portal and Brazil through SICAR and SIGEF. Where national cadastres are available, detected disturbance polygons can be automatically flagged when they fall outside licensed boundaries or within protected-area buffers.
The honest caveat is that cadastral data quality varies considerably. Boundaries may be outdated, digitised at coarse resolution, or contested. A polygon that appears to cross a concession line in the satellite data may be within licence when measured against the legal parcel description. This is why satellite evidence is best treated as an investigative trigger rather than a prosecution exhibit in isolation. The spatial analysis narrows the field; ground inspection or legal review closes it.
The resolution floor problem, stated plainly
Any honest assessment of this use case has to name the detection limits. For medium-resolution sensors (10 to 30 m), the minimum detectable disturbance is roughly 0.1 to 1 hectare depending on canopy closure and cloud conditions. Sub-hectare selective felling is routinely missed. Very-high-resolution commercial sensors (0.3 to 0.5 m) can detect individual trails and small log decks, but tasking costs are significant per scene, and systematic coverage of large concession areas at that resolution is expensive.
Cloud cover compounds the problem in the humid tropics. In the Congo Basin and parts of the Amazon, optical sensors can be cloud-obscured for 80 to 90 percent of acquisition opportunities during the wet season. SAR is the only practical all-weather option, and its coherence-change products are coarser than the optical systems. The practical consequence is that a monitoring programme must accept some level of detection latency and some rate of missed events at the sub-hectare scale. Designing the monitoring objective around what the sensors can actually deliver, rather than what enforcement ideally requires, is the starting point for any credible programme design.
Satellize structures analytics workflows around this constraint explicitly. In the Tonga crop-estimation programme, the same tiered logic applies: open-constellation screening establishes where to look, commercial tasking confirms what is there. The same architecture applies to forest monitoring.
Building a monitoring cadence that holds up
A single image is a snapshot. Enforcement-grade monitoring requires a time-series with documented acquisition dates, processing versions and detection thresholds. Sentinel-1's six-day repeat (for the two-satellite constellation) makes it the backbone of any operational logging-detection system: it is free, global, cloud-transparent and archived back to 2014. That archive depth allows a baseline coherence model to be built before monitoring begins, which reduces false positives from seasonal vegetation change.
The practical monitoring cadence for a medium-sized concession (50,000 to 200,000 hectares) is: automated Sentinel-1 coherence-change screening every one to two repeat cycles; optical confirmation via Planet or Sentinel-2 when cloud permits; commercial SAR or optical tasking for high-confidence anomalies; and monthly summary reports with flagged polygons cross-referenced against the concession cadastre. That cadence is achievable at a cost that is a small fraction of the timber value being protected. The question is not whether the technology is sufficient. It is whether the monitoring programme is designed with enough honesty about its limits to be defensible when challenged.
Typical figures
| SAR spatial resolution (Sentinel-1 IW) | 10 m ground range; coherence products typically at 20–40 m after multi-looking |
| SAR spatial resolution (Capella Space spotlight) | 0.5 m; suitable for trail-level feature mapping |
| Optical resolution (WorldView-3 / SkySat) | 0.31–0.5 m panchromatic; 1.24–0.5 m multispectral |
| Revisit (Sentinel-1, two-satellite) | 6 days at equator; 12 days per satellite |
| Revisit (commercial SAR/optical tasking) | Hours to days, demand-driven; cost per scene applies |
| Minimum detectable disturbance (medium-resolution optical) | ~0.1–1 ha depending on canopy closure and cloud; sub-hectare selective felling frequently missed |
| Cloud-penetration capability | SAR only; optical sensors routinely obscured 80–90% of acquisitions in humid tropics wet season |
| Sentinel-1 archive depth | 2014 to present; enables multi-year baseline coherence modelling |
| Spectral bands relevant to disturbance classification | NIR, SWIR (Sentinel-2, WorldView-3); C-band and X-band SAR backscatter and coherence |
| Cadastral cross-reference data | Dependent on national open-data availability; quality and currency vary by jurisdiction |
Analytics Satellize can run
| SAR coherence-change disturbance map | Interferometric coherence differencing on Sentinel-1 IW SLC time-series; change flagged against seasonal baseline | Polygon GIS layer (GeoJSON/Shapefile) of disturbance zones with date of first detection and coherence-loss magnitude |
| Skid trail and log deck mapping | Object-based image analysis on very-high-resolution optical or SAR imagery; linear feature extraction and clearing delineation | Vector layer of trail centrelines and log deck polygons with area and estimated first-appearance date |
| Concession boundary exceedance report | Spatial intersection of detected disturbance polygons against client-supplied or open cadastral boundaries; buffer analysis for protected-area proximity | Tabular and mapped report of flagged events with concession ID, exceedance area and confidence rating |
| Optical confirmation tasking brief | Prioritisation of SAR-flagged anomalies by area, coherence-loss magnitude and proximity to concession boundaries; tasking order specification for commercial sensor | Ranked tasking list with AOI coordinates and acquisition window, ready for submission to commercial operator |
| Multi-year disturbance timeline | Retrospective analysis of Sentinel-1 archive (2014–present) and available commercial archive; change-point detection per polygon | Time-stamped disturbance history per parcel, formatted for use in enforcement documentation |
| Monthly monitoring bulletin | Automated screening pipeline with analyst review of high-confidence flags; integrated SAR and optical inputs | PDF and GIS package summarising new detections, confirmed events and areas with persistent cloud or data gaps |
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.