Selective logging and forest degradation detection
Selective logging removes individual trees without triggering canopy-loss alerts, yet it degrades carbon stocks measurably. Radar backscatter and canopy-fraction signals reveal the disturbance, but only within a narrow temporal window before regrowth obscures the evidence.
Sensors
- ALOS-2 PALSAR-2 (L-band SAR): L-band at 1.27 GHz penetrates the upper canopy and interacts with woody trunks and branches. Stripmap mode delivers 3 m single-look or 6 m multi-look resolution. Backscatter drops of 1–3 dB at logging gaps and skid trails are detectable even under cloud cover, which is critical in humid tropics. Revisit is 14 days at a given incidence angle.
- Sentinel-1 C-band SAR: C-band (5.4 GHz) is more sensitive to surface roughness and upper-canopy leaves than to woody structure, making it less penetrating than L-band but still useful for detecting skid-trail networks. IW mode gives 10 m ground range resolution and 6-day revisit over most tropical regions with the two-satellite constellation. Free and open archive back to 2014.
- Planet SuperDove: Eight-band multispectral imagery at 3 m resolution with near-daily revisit. The red-edge and NIR bands support NDVI and canopy-fraction differencing that can resolve gaps wider than roughly 10–15 m. Cloud cover limits usable acquisitions in the humid tropics to perhaps 30–50 clear observations per year, so temporal compositing is essential.
- Sentinel-2 MSI: 10 m resolution in visible and NIR bands, 20 m in red-edge and SWIR. Five-day revisit with both satellites. SWIR bands are sensitive to exposed soil and woody debris on skid trails. Free archive from 2015. Cloud contamination in the Congo Basin or Amazon can reduce usable scenes to fewer than 20 per year without compositing.
- WorldView-3: 31 cm panchromatic, 1.24 m multispectral, with SWIR at 3.7 m. Sufficient resolution to map individual felling gaps and count skid-trail intersections. Tasked commercially, so coverage is on-demand rather than systematic. Useful for ground-truth validation of coarser SAR or multispectral detections, not for wall-to-wall monitoring.
Why selective logging hides in plain sight
Clear-felling removes canopy in patches large enough to trigger standard deforestation alerts, which look for binary land-cover change above a threshold, typically 0.5 ha or larger. Selective logging is different. A harvesting crew extracts five to fifteen stems per hectare, leaving the surrounding canopy largely intact. The felling gap for a single large tree is 20–50 m across. The skid trail connecting it to the extraction road is 4–8 m wide. Neither feature crosses the detection threshold of a 30 m pixel if only one or two pixels are disturbed, and both begin to close within six to eighteen months as pioneer species fill the light gap.
This is not a minor accounting footnote. Published estimates suggest that selective logging in the Brazilian Amazon alone degrades tens of millions of tonnes of carbon per year, a figure that does not appear in canopy-loss datasets. Under REDD+ degradation modules and the Verra VM0048 methodology, project developers are required to account for this pool. Satellite evidence is the only practical way to do it at scale.
The physics of detection: what radar sees that optical cannot
L-band SAR is the workhorse sensor for this application because the 23 cm wavelength penetrates the upper leaf canopy and interacts with trunks, large branches and the soil beneath. When a tree is felled, the volume scatterers it contributed disappear and the ground beneath is disturbed. Backscatter in HV polarisation drops by 1–3 dB at the gap, a signal that ALOS-2 PALSAR-2 can resolve in 3–6 m Stripmap acquisitions. The signal persists for roughly six to twenty-four months before regrowth rebuilds the woody volume. That window is the detection opportunity.
C-band Sentinel-1 responds more to surface roughness and leaf moisture than to woody structure, so it is less sensitive to sub-canopy disturbance. It is still useful for detecting the exposed mineral soil on skid trails, which produces a distinctive low-backscatter signature in VV polarisation. Combining L-band and C-band time series reduces false positives from wind damage or phenological variation, which can produce similar single-date anomalies.
The temporal window: act before the canopy closes
Tropical forest recovers quickly. Published studies using ALOS PALSAR time series in the Amazon and Southeast Asia show that L-band backscatter at logged gaps returns to within 1 dB of surrounding forest within two to four years, depending on rainfall and soil type. For optical sensors, NDVI recovery at felling gaps can be faster still, sometimes within one wet season, as fast-growing pioneer species establish.
This sets a hard operational constraint. To catch a logging event with high confidence, imagery must be acquired within roughly twelve months of the disturbance. For a carbon project covering hundreds of thousands of hectares in the Congo Basin or Borneo, that means systematic, scheduled acquisitions rather than opportunistic tasking. A monitoring plan that relies on annual composites may miss events that occurred and healed within the same twelve-month period. Quarterly SAR acquisitions are the minimum credible standard for degradation monitoring under most current MRV frameworks.
Legal versus illegal: what satellites can and cannot resolve
Satellite data can locate a disturbance and date it to within a few weeks. It cannot, by itself, determine whether the extraction was authorised. Separating legal concession harvesting from illegal chainsaw logging requires the satellite evidence to be overlaid against concession boundaries, harvesting permits and declared coupe locations, all of which are administrative records held by national forestry agencies.
The geometry of the operation does offer some clues. Legal selective logging under reduced-impact logging (RIL) guidelines produces a planned skid-trail network with controlled branching angles and pre-planned felling directions. Illegal or unplanned extraction tends to produce more chaotic trail networks with higher trail density per extracted stem. These morphological differences are visible in sub-5 m imagery and in the spatial statistics of gap distributions, though the distinction is probabilistic rather than definitive. Any legal finding still requires ground verification or documentary evidence. Satellite analysis narrows the search area; it does not replace the inspector.
Building a detection pipeline that holds up to audit
A credible MRV pipeline for selective logging combines at least two sensor types across multiple dates. The standard approach uses a dense SAR time series, typically ALOS-2 or Sentinel-1, to flag candidate disturbance pixels via change-detection algorithms such as BFAST or LandTrendr adapted for radar. Optical imagery at 3–10 m resolution then confirms the disturbance morphology and provides a spectral signature, exposed soil in SWIR, reduced NIR reflectance, that distinguishes logging from wind damage or disease.
Skid-trail extraction from sub-5 m imagery adds a second line of evidence. Trail length per unit area is a proxy for extraction intensity and can be used to estimate the number of stems removed, which feeds directly into carbon-stock deduction calculations. WorldView-3 or Planet SuperDove at 3 m are the practical choices for this step. The resulting evidence package, dates, locations, gap areas, trail networks, should be archived in a format that a third-party auditor can reproduce independently. Satellize builds these time-series evidence packages for carbon auditors and project developers, drawing on the same open-constellation archive used in its Tonga crop-estimation work.
One honest limit: in areas where cloud cover exceeds 80% of acquisition opportunities and SAR revisit is irregular, there will be gaps in the temporal record. Those gaps are themselves a finding that should be disclosed in any MRV submission, not papered over with interpolated values.
Typical figures
| Minimum detectable gap width | ~20 m with ALOS-2 PALSAR-2 Stripmap (3–6 m pixel); ~30–40 m with Sentinel-2 10 m bands; ~10 m with Planet SuperDove 3 m |
| SAR spatial resolution (ALOS-2 Stripmap) | 3 m single-look, 6 m multi-look |
| Optical spatial resolution | 3 m (Planet SuperDove), 10 m (Sentinel-2 VIS/NIR), 1.24 m multispectral (WorldView-3) |
| SAR revisit (ALOS-2) | 14 days at fixed incidence angle; shorter with multi-angle acquisitions |
| Optical revisit | Near-daily (Planet SuperDove); 5 days (Sentinel-2 two-satellite); on-demand (WorldView-3) |
| Detection temporal window | Approximately 6–24 months post-disturbance before L-band backscatter recovery; 3–12 months in optical NDVI |
| Spectral bands used | L-band HV/HH SAR; C-band VV/VH SAR; Red-edge, NIR, SWIR (optical); Panchromatic at 31 cm (WorldView-3) |
| Archive depth | Sentinel-1 from 2014; Sentinel-2 from 2015; ALOS-2 from 2014; ALOS PALSAR from 2006 (JAXA archive) |
| Minimum mapping unit (practical) | 0.05–0.1 ha for gap detection; individual skid trails mappable at sub-5 m resolution |
| Delivery formats | GeoTIFF change maps, GeoPackage vector layers, timestamped audit-ready evidence packages |
Analytics Satellize can run
| SAR backscatter change map | Bi-temporal or time-series differencing of ALOS-2 HV and Sentinel-1 VH backscatter; BFAST or cumulative-sum change detection | GeoTIFF raster of disturbance probability per pixel, updated quarterly |
| Canopy-fraction anomaly layer | Fractional cover inversion from Planet SuperDove or Sentinel-2 red-edge/NIR; comparison against pre-logging baseline composite | Vector polygons of confirmed canopy-gap events with area, date and confidence score |
| Skid-trail network extraction | Object-based image analysis or convolutional segmentation on sub-5 m optical imagery; morphological filtering to separate trails from natural gaps | GeoPackage line layer of trail centrelines with length, density per ha and estimated extraction intensity |
| Carbon-stock deduction estimate | Gap area and trail length converted to disturbed volume using published allometric relationships and IPCC Tier 2 emission factors for degradation | Tabular report of estimated CO₂e loss per logging event, formatted for REDD+ or Verra MRV submission |
| Legal/illegal disturbance flag | Spatial overlay of detected disturbances against concession boundaries and declared coupe polygons; trail-network morphology scoring | Flagged alert layer with concession-match status and morphological regularity index, for referral to ground inspection |
| Temporal evidence package | Multi-sensor, multi-date archive compilation with provenance metadata; reproducible processing chain documented to ISO 19115 standard | Audit-ready ZIP archive of imagery, derived layers, processing logs and uncertainty statements |
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.