Illegal timber harvest detection for forest-backed bond collateral verification

SAR coherence loss and optical change detection identify unauthorised clearing within concession boundaries, giving bond trustees and ESG auditors independent, cloud-penetrating evidence of collateral integrity.

Sensors

ALOS-2 PALSAR-2: L-band (1.27 GHz) SAR at 6–25 m resolution depending on mode; penetrates cloud and partial canopy to detect structural change in forest volume; repeat cycle 14 days, with shorter revisit via off-nadir steering. The long wavelength interacts with trunks and large branches, making it sensitive to canopy removal even under persistent cloud cover common in tropical regions.
Sentinel-1 SAR (C-band): C-band (5.4 GHz) at 10 m resolution in Interferometric Wide Swath mode; 6-day revisit at the equator with both satellites. Provides dense temporal stack for coherence-change analysis; less canopy-penetrating than L-band but excellent for detecting full clearing events and monitoring regrowth.
Sentinel-2 MSI: 10–20 m resolution across 13 spectral bands including red-edge and SWIR; 5-day revisit. Used for NDVI-based optical change detection and to confirm and classify disturbance flagged by SAR. Cloud-limited in humid tropics, which is why SAR is the primary detection layer.
Landsat 8/9 OLI: 30 m resolution, 16-day repeat, free archive back to 1972 (Landsat 1). OLI's SWIR bands are particularly useful for distinguishing bare soil from low regrowth. Provides historical baseline against which concession-boundary violations can be dated, which matters when a bond was issued against a specific forest state.

What a forest bond is actually backed by

Forest-backed bonds, including some green bonds and biodiversity-linked instruments, use standing timber volume or intact canopy area as collateral or as the metric against which coupon adjustments are triggered. The implicit assumption is that the forest remains intact. That assumption is verifiable from orbit.

The problem for trustees and auditors is that concession boundaries are legal lines on paper. They do not enforce themselves. Illegal clearing, whether by third-party encroachment or by the concession holder exceeding permitted harvest areas, can remove significant canopy before any ground inspection occurs. In humid tropical regions, cloud cover makes optical monitoring alone unreliable for months at a time. A bond issued in good faith against a forest that has since been partially cleared is a collateral integrity problem that satellite data can surface.

Why coherence loss is the right signal

SAR coherence measures how similar the radar backscatter from a patch of ground is between two acquisition dates. Intact forest canopy, with its stable arrangement of trunks, branches and leaves, maintains relatively high coherence over short intervals. Remove the canopy and the scattering geometry changes abruptly. Coherence drops. That drop is detectable even when cloud prevents any optical observation.

L-band wavelengths (around 23 cm for ALOS-2 PALSAR-2) penetrate cloud completely and interact with the woody structural elements of the canopy rather than just leaves. This makes L-band coherence sensitive to the removal of trees rather than merely to seasonal leaf change, which can confuse shorter-wavelength sensors. The Global Forest Watch alert system, operated by the University of Maryland, uses a combination of Landsat and MODIS optical data for its primary disturbance alerts; SAR coherence methods documented in peer-reviewed literature extend that detection capability into cloud-persistent conditions where optical data gaps can last weeks.

Sentinel-1's C-band adds temporal density. Six-day revisit means that a clearing event can be bracketed to within days rather than the 14-day PALSAR-2 cycle, which matters when a bond covenant requires notification within a defined window.

Where the method reaches its limits

Selective logging is the hard case. When individual trees are felled without removing the surrounding canopy, the gap in the canopy may be smaller than the sensor's resolution cell. At 10–25 m pixel sizes, a single felled tree is invisible. Helicopter logging or high-grading operations that extract only the highest-value stems can remove substantial timber volume while leaving the canopy largely intact from above. This is a genuine detection limit, not a caveat to be buried in footnotes.

Regrowth is the second constraint. Tropical secondary vegetation can re-establish enough canopy structure within 12 to 24 months to partially restore coherence and NDVI signals. A clearing that occurred two years before a bond audit may show only attenuated disturbance signatures. Historical archive analysis using Landsat's 50-year record can recover older events, but the dating confidence degrades with time and regrowth density.

Concession boundary accuracy also matters. If the legal boundary shapefile is imprecise, a clearing just outside the boundary may appear inside it, or vice versa. Ground-truth GPS surveys of boundary markers are the only remedy for that source of error.

The detection workflow in practice

The standard approach is a change-detection stack built on pre-bond-issuance imagery as baseline. For each monitoring period, new SAR acquisitions are processed for coherence change relative to that baseline and to the immediately preceding acquisition. Pixels showing coherence loss above a threshold, typically calibrated against known clearing events in the same forest type, are flagged as candidate disturbances.

Optical data from Sentinel-2 and Landsat then serve as confirmation. Where cloud-free optical imagery is available in the same window, NDVI change and bare-soil indices either confirm or dismiss the SAR flag. Where optical data are absent, the SAR signal stands alone, with a lower confidence rating attached. The output is a polygon layer of candidate disturbances, clipped to the concession boundary, with area estimates, confidence scores and acquisition dates.

For bond reporting, the relevant output is not a pixel map but a structured summary: total area of confirmed disturbance within the concession boundary during the reporting period, area of unconfirmed (SAR-only) candidate disturbance, and a comparison against the baseline forest state at bond issuance. That is what a trustee or independent verifier can act on.

ESG supply-chain audits and the same data, different question

For ESG auditors verifying no-deforestation commitments in timber or agricultural supply chains, the question is slightly different from bond collateral verification. The auditor wants to know whether a supplier's concession, or the landscape around it, shows clearing that could be attributed to the supplier's operations. The sensor stack is identical; the boundary layer changes from a bond concession to a supplier's declared sourcing area.

The Global Forest Watch platform publishes near-real-time deforestation alerts based on Landsat and MODIS data, which provide a publicly auditable baseline. A more rigorous supply-chain audit adds SAR coherence analysis to catch events that occurred during cloud-persistent periods between optical alerts. The combination reduces the window during which an event can remain undetected.

Satellize runs this kind of analysis on open constellations, supplemented by commercial tasking where higher resolution or shorter revisit is needed. The Tonga crop-estimation programme uses a similar multi-sensor fusion approach, though over agricultural rather than forest land. For a prospective bond trustee or ESG programme manager, the practical next step is defining the concession boundary dataset and the reporting cadence before any sensor tasking begins.

Typical figures

Primary SAR spatial resolution	6 m (PALSAR-2 Spotlight), 10 m (Sentinel-1 IW), 25 m (PALSAR-2 ScanSAR)
Optical spatial resolution	10 m (Sentinel-2 RGB/NIR), 20 m (Sentinel-2 SWIR/red-edge), 30 m (Landsat OLI)
SAR revisit cadence	6 days at equator (Sentinel-1 dual satellite); 14 days (PALSAR-2, steerable to shorter)
Optical revisit cadence	5 days (Sentinel-2 dual satellite); 16 days (Landsat 8/9 combined ~8 days)
SAR frequency bands	L-band 1.27 GHz (PALSAR-2); C-band 5.4 GHz (Sentinel-1)
Minimum detectable clearing (full canopy removal)	Approximately 0.1–0.5 ha at 10–25 m resolution; selective logging below detection threshold
Cloud penetration	Complete for SAR (all weather); optical data cloud-limited, particularly in humid tropics
Archive depth	Landsat to 1972; Sentinel-1 from 2014; PALSAR-2 from 2014; PALSAR-1 from 2006
Alert latency (after acquisition)	1–5 days for processed disturbance flag; bond-period summary reports on agreed cadence
Delivery formats	GeoTIFF change layers, GeoJSON disturbance polygons, PDF summary report for trustees

Analytics Satellize can run

Concession-boundary disturbance map	SAR coherence change detection (L-band and C-band stack) clipped to legal concession boundary shapefile	GeoJSON polygon layer of candidate clearings with area, date range and confidence score
Bond-period forest-state comparison	Baseline forest mask at bond issuance versus current-period canopy extent using NDVI and SAR backscatter	Structured PDF report quantifying net canopy loss in hectares and percentage of collateral area
Cloud-gap SAR-only alert	Sentinel-1 coherence anomaly detection during periods of sustained optical cloud cover	Timestamped alert feed with candidate disturbance polygons flagged as SAR-only pending optical confirmation
Historical clearing timeline	Landsat archive NDVI change series and PALSAR-1/2 backscatter trend analysis for pre-bond baseline dating	Time-series chart and GIS layer showing clearing events by year back to available archive depth
Regrowth and recovery assessment	Multi-date NDVI trajectory analysis (Sentinel-2, Landsat) to identify areas of secondary regrowth masking prior disturbance	Annotated map distinguishing primary intact forest, confirmed clearings and secondary regrowth by estimated age class
ESG supply-chain sourcing-area audit	Disturbance detection within supplier-declared sourcing polygons using Global Forest Watch alert cross-reference plus SAR coherence	Audit-ready report comparing declared no-deforestation commitment against observed disturbance events with source imagery citations

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.