Illegal logging road detection in tropical forest

Unpermitted access roads are the first and most persistent signature of illegal logging. SAR change detection cuts through the cloud cover that shields most tropical cutting fronts, while high-resolution optical imagery confirms surface type and direction of advance.

Sensors

Sentinel-1 C-band SAR (ESA): 5x20 m ground range detected resolution in IW mode; 6-day repeat at the equator with both satellites active. C-band backscatter drops sharply when closed canopy is replaced by bare compacted laterite, making new road cuts detectable regardless of cloud or rain. The primary sensor for systematic tropical-forest road monitoring.
Planet SuperDove (PlanetScope): 3 m native resolution, 8 spectral bands including red-edge and NIR; near-daily revisit globally. Confirms road surface material, width, and direction of advance when cloud clears. At 3 m, roads as narrow as 4-6 m can be resolved as a linear feature, though sub-3 m tracks remain ambiguous.
Landsat 8/9 OLI (USGS/NASA): 30 m multispectral, 16-day repeat per satellite, archive back to 1972 (TM/ETM+/OLI). Provides the long historical baseline for distinguishing newly cut roads from older permitted infrastructure. NDVI and bare-soil indices on OLI bands 4-6 flag canopy loss with high confidence, though roads narrower than roughly 30 m are sub-pixel and require length-based detection methods.
Sentinel-2 MSI (ESA): 10 m in visible and NIR bands, 20 m in SWIR; 5-day revisit with both satellites. Sits between Landsat and Planet in resolution and revisit. SWIR bands (band 11 at 1610 nm, band 12 at 2190 nm) are particularly sensitive to bare laterite soil and logging-track compaction signatures.

Why roads matter more than the trees they remove

Illegal logging does not begin with a chainsaw. It begins with a road. Without access, high-value timber cannot be extracted at commercial scale, and the cutting front cannot advance. A road is also permanent in a way that a felled area is not: secondary vegetation can re-establish canopy within two to five years, but a compacted laterite track remains detectable from orbit for decades. Monitoring the road network is therefore a more durable enforcement signal than monitoring canopy loss alone.

The spectral contrast is stark. Closed tropical canopy reflects strongly in NIR (NDVI typically above 0.7-0.8) and absorbs in SWIR. Bare compacted laterite does the opposite: low NIR, elevated SWIR reflectance, and NDVI values below 0.1. That contrast is what every detection method in this page exploits.

SAR backscatter: the sensor that does not care about cloud

The operational problem with optical monitoring of tropical logging fronts is persistent cloud cover. In the Congo Basin and Amazonia, cloud-free optical observations can be separated by weeks or months during the wet season, which is precisely when many operators push roads forward under cover. Sentinel-1 C-band SAR has no such constraint.

When a road is cut, the volume scattering from the forest canopy is removed and replaced by a specularly smooth or slightly rough bare surface. In VV polarisation, this typically produces a backscatter reduction of 3-6 dB relative to surrounding forest. In VH polarisation, which is more sensitive to volume scattering from canopy structure, the drop can exceed 8 dB. Coherent change detection on repeat-pass Sentinel-1 pairs can flag disturbances covering as little as one to two hectares, though linear features as narrow as a single road are at the edge of IW-mode resolution and require morphological line-detection algorithms rather than simple threshold classification.

The six-day revisit of the dual-satellite Sentinel-1 constellation means that, in principle, a new road segment cut in week one is detectable by week two. In practice, latency from acquisition to processed alert adds two to five days depending on processing pipeline design.

High-resolution optical change pairs: confirming what SAR flags

SAR detects change. It does not always distinguish a new logging road from a riverbank collapse, a landslide scar, or a permitted agricultural track. High-resolution optical imagery resolves that ambiguity. A Planet SuperDove image at 3 m can show road width, surface colour (red laterite is spectrally distinctive), directionality, and the presence of log-drag scars or skid trails branching off the main track.

The standard workflow pairs a SAR-derived change mask with the next cloud-free Planet or Sentinel-2 acquisition over the same area. The SAR mask constrains the optical search to flagged pixels, reducing analyst time and false-positive review burden. Where Planet tasking is available, the latency between SAR alert and optical confirmation can be as short as 24-48 hours. Where the analyst must wait for a cloud-free Sentinel-2 pass, confirmation may take one to three weeks during heavy wet-season cloud.

Landsat's 30 m resolution is too coarse to confirm individual road tracks, but its archive depth makes it the right tool for reconstructing when a road network first appeared, which is critical for establishing whether a permit pre-dates or post-dates observed infrastructure.

What the method cannot see, and where it fails

Honest limits matter here. Sentinel-1 IW mode at 5x20 m ground resolution can detect the clearing associated with a road but will miss tracks narrower than roughly 5-8 m if they are not accompanied by a wider cleared margin. Selective logging under partial canopy, where individual trees are removed without full clearing, produces a backscatter change too small to distinguish reliably from seasonal moisture variation.

Optical sensors face the cloud problem already described, but they also face a canopy-closure problem. A road that has been cut but is beginning to re-vegetate, perhaps after a temporary halt in operations, will show rising NDVI and may drop below detection thresholds within 12-18 months. The road is still physically present and still usable, but it becomes spectrally ambiguous.

False positives from SAR include riverbank erosion, natural treefall gaps, and agricultural clearings at forest margins. These require optical confirmation or contextual GIS layers (protected area boundaries, concession maps, river network) to filter. No automated pipeline eliminates the need for a trained analyst to review flagged events before enforcement action is taken.

Building a detection system that holds up in court

Enforcement agencies increasingly need evidence chains, not just alerts. A detection workflow that will survive legal scrutiny requires: a documented baseline (Landsat archive imagery pre-dating the alleged road, with acquisition dates); a change event tied to a specific Sentinel-1 acquisition and processing version; and an optical confirmation image with metadata showing sensor, date, and ground resolution. All three elements must be archived with consistent coordinate reference systems and reproducible processing parameters.

The Global Land Analysis and Discovery (GLAD) laboratory at the University of Maryland publishes near-real-time forest disturbance alerts using Landsat and Sentinel-2, which provides a publicly auditable reference layer. A bespoke SAR-based system can run at finer temporal resolution and flag events before GLAD alerts are issued, but the GLAD layer is useful as an independent corroborating source in enforcement documentation.

Satellize structures road-detection analytics to produce GIS-ready outputs with full provenance metadata, so that the acquisition chain from raw satellite data to flagged polygon is documented at every step. The same provenance approach underpins the Tonga crop-estimation programme, where audit trails are required by the client government.

Designing a monitoring programme for a concession boundary or protected area

The practical design question is not whether satellite data can detect illegal roads. It demonstrably can. The question is what revisit frequency and resolution tier are proportionate to the threat level and the enforcement capacity of the client.

A national park authority with ranger capacity to respond within 72 hours needs near-real-time SAR alerts and rapid optical tasking. A ministry of forestry conducting annual concession audits needs a different product: a quarterly change map showing road-network expansion against permitted concession boundaries, delivered as a GIS layer with area statistics. The sensor mix and processing cadence differ substantially between these two use cases, and conflating them wastes budget.

As a starting point for scoping, the key parameters to define are: the total area of interest in square kilometres; the minimum road width the client needs to detect (which sets the resolution floor); the acceptable latency between road creation and alert; and whether the output needs to be court-admissible or is for internal triage. Those four parameters determine the sensor combination, processing frequency, and analyst review load.

Typical figures

SAR spatial resolution (Sentinel-1 IW)	5 m range x 20 m azimuth (ground range detected); 10 m after multi-looking
Optical resolution range	3 m (Planet SuperDove), 10 m (Sentinel-2 VIS/NIR), 30 m (Landsat 8/9 OLI)
SAR revisit (dual Sentinel-1 constellation)	6 days at equator; 3-4 days at higher latitudes
Optical revisit	Near-daily (Planet); 5 days (Sentinel-2 dual); 8 days (Landsat dual)
Minimum detectable linear feature (SAR)	Road clearings wider than ~5-8 m with cleared margins; narrower tracks require optical confirmation
SAR frequency / polarisation	C-band 5.405 GHz; VV+VH dual-pol (Sentinel-1 IW standard)
Key optical bands for road detection	NIR (Sentinel-2 B8, 842 nm), SWIR (B11 1610 nm, B12 2190 nm), Red-edge (B5-B7)
Archive depth	Sentinel-1 from 2014; Landsat from 1972 (TM); Planet from ~2016 at commercial tasking
Alert latency (SAR to processed flag)	Typically 2-5 days from acquisition; optical confirmation adds 1-21 days depending on cloud
Delivery formats	GeoTIFF change rasters, GeoJSON/Shapefile road polygons, tabular area statistics, PDF audit reports

Analytics Satellize can run

SAR backscatter change alert	Bi-temporal log-ratio change detection on Sentinel-1 VV and VH; morphological line enhancement to isolate linear features	GeoJSON alert layer with acquisition date, polarisation delta, and confidence score; refreshed every 6 days
Optical road confirmation layer	NDVI and bare-soil index differencing on Planet SuperDove or Sentinel-2 MSI change pairs, constrained to SAR-flagged pixels	Confirmed road polygon shapefile with surface-type classification (bare laterite, compacted track, partial re-vegetation)
Historical road-network baseline	Landsat OLI time-series NDVI and SWIR analysis from archive; road-network digitisation from earliest detectable clearing	Dated road-network GIS layer showing first-detection year per segment, suitable for permit cross-referencing
Concession-boundary encroachment report	Spatial intersection of confirmed road polygons with permitted concession and protected-area boundary layers; area-change statistics	Quarterly PDF report with encroachment area in hectares, map figures, and coordinate lists for enforcement referral
Road advance rate monitoring	Sequential road-tip position tracking across SAR and optical time series; linear regression on advance distance per month	Time-series chart and GIS layer showing road extension velocity, enabling prediction of where the front will reach within a defined period
Provenance-documented evidence package	Full metadata chain from raw acquisition through processing version to final polygon; coordinate reference system standardisation and checksums	Archived evidence bundle (imagery, processing logs, GIS outputs) formatted for legal or regulatory submission

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.