Greenwashing detection in published forest-carbon casework
Independent cross-checking of forest-carbon project claims against the satellite record: verifying stated cover, detecting post-crediting clearings, and testing whether reference regions were drawn to inflate counterfactual deforestation rates.
Sensors
- Landsat 5/7/8/9 TM/ETM+/OLI: 30 m multispectral resolution, 16-day revisit per satellite, archive back to 1984. The only freely available record long enough to reconstruct historical deforestation rates over the 10-to-20-year reference windows that REDD+ methodologies typically require. Band combinations including SWIR allow forest/non-forest discrimination even through light haze.
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. Useful for confirming clearing boundaries and assessing canopy condition at the project perimeter. Archive starts October 2015, which limits its use for historical baselines but makes it well-suited for post-crediting-date change detection.
- Planet SuperDove: 3 m resolution, near-daily revisit globally. No free archive, but commercially licensed imagery can resolve individual clearings, track logging road extension and confirm the presence or absence of canopy at specific coordinates where a project claims standing forest. Useful for ground-truthing suspicious pixels flagged by coarser sensors.
- Hansen et al. Global Forest Watch tree-cover loss product: Annual 30 m tree-cover loss derived from Landsat, published by the University of Maryland and hosted on Global Forest Watch. The canonical public dataset for historical deforestation rates. Its year-of-loss layer is the standard reference for assessing whether a project's claimed baseline period matches observed loss, and whether loss accelerated or decelerated after crediting began.
- Sentinel-1 SAR C-band: 10 m resolution, 6-to-12-day revisit, cloud-penetrating. Tropical projects are frequently cloud-obscured during the wet season, precisely when clearing activity can be concealed from optical sensors. SAR coherence loss and backscatter change can confirm clearing events that optical imagery missed.
What the registry says versus what the archive shows
Forest-carbon registries such as Verra's VCS and Gold Standard publish project design documents, monitoring reports and credit issuance records. These documents make specific, auditable claims: the forest cover at a given date, the deforestation rate in a reference region, the project boundary coordinates, and the volume of carbon credited. Every one of those claims has a spatial and temporal footprint that the satellite archive can interrogate independently.
The methodological starting point is not a blank map. It is the project document itself. Analysts extract the stated crediting start date, the project boundary polygon, and the reference region used to calculate the counterfactual deforestation rate. Those three elements define the questions the satellite record must answer: Was the claimed forest cover actually present at the start date? Has any clearing occurred inside the boundary since that date? And does the reference region's historical deforestation rate reflect a credible counterfactual, or was it selected to make the threat look larger than it was?
Reconstructing the historical record
Landsat's archive back to 1984 is the backbone of any historical baseline audit. The Hansen et al. annual tree-cover loss product, derived from that archive, provides a ready-made year-of-loss layer at 30 m resolution. For a project claiming a baseline deforestation rate of, say, 2 percent per year over a ten-year reference period, the audit simply calculates the observed loss rate from the Hansen product within the stated reference region for those exact years. Discrepancies between the claimed rate and the observed rate are immediately visible.
Reference region selection is where many projects introduce the largest distortions. A developer can, within the latitude of some methodologies, draw a reference region that captures a historically high-deforestation frontier rather than the landscape surrounding the actual project. Spatial analysis of the Hansen product across alternative plausible reference regions, drawn using objective criteria such as distance bands, watershed boundaries or road-access buffers, can test whether the developer's choice was an outlier. This is not a novel technique. Academics including West et al. (2020, published in Science) applied exactly this logic to a large sample of REDD+ projects and found that many claimed counterfactual deforestation rates substantially exceeded rates in matched control areas.
Detecting clearings that should not exist
Once a project is registered and credits begin to be issued, any deforestation inside the project boundary is supposed to trigger a buffer deduction from the credit pool or, in severe cases, invalidation of issued credits. Detecting such clearings is straightforward in principle. The analyst takes the project boundary polygon and runs a change-detection time series from the crediting start date to the present, using Landsat and Sentinel-2 in combination. Clearings above roughly 0.1 hectares are detectable at Sentinel-2's 10 m resolution. Smaller disturbances, including selective logging tracks, may require Planet SuperDove imagery at 3 m.
Cloud cover is the principal practical obstacle in humid tropical forests. A project area in the Congo Basin or the western Amazon may have fewer than 20 cloud-free Landsat observations per year. The correct response is to combine optical and SAR data rather than to accept optical gaps as absences of evidence. Sentinel-1 C-band SAR penetrates cloud and detects the loss of canopy structure through coherence change. A clearing that is invisible in optical imagery for three months of wet season can still be dated to within a few weeks using SAR coherence time series.
One honest limit: SAR cannot reliably distinguish forest degradation from outright clearing at the scale of a few pixels. Where the distinction matters for credit accounting, high-resolution optical confirmation from Planet or commercial tasking is necessary.
Testing the forest-cover claim at the crediting start date
Some projects have been crediting carbon from land that was already degraded or partially cleared at the stated start date. The fix is simple. The analyst pulls the Landsat scene closest to the crediting start date, calculates NDVI and a forest/non-forest classification, and compares it to the canopy cover percentage stated in the project document. Where the document claims, for example, 85 percent canopy cover and the contemporaneous Landsat scene shows a patchwork of degraded scrub, the discrepancy is documented with the scene metadata as evidence.
The 30 m Landsat pixel is not fine enough to resolve individual tree crowns, and forest/non-forest thresholds vary by methodology. The standard Global Forest Watch product uses a 30 percent canopy-cover threshold to define forest. A project may legitimately use a different threshold. The audit must apply the same threshold the project document specifies, not an arbitrary external one. Candour about this matters: the satellite record is strong evidence but it is not a court ruling. It establishes whether a claim is plausible or implausible, and flags cases that warrant a physical site inspection.
Packaging the evidence for auditors and buyers
An independent satellite audit is only as useful as its documentation. The deliverable for a carbon buyer or a registry auditor needs to include the specific scenes used, their acquisition dates and cloud-cover percentages, the classification method applied, the version of any derived product such as the Hansen dataset, and a clear statement of what the satellite record can and cannot resolve. A finding of 'no detectable clearing' is not the same as a finding of 'no clearing occurred'. The distinction must be explicit.
Satellize packages this kind of evidence for clients working in carbon markets, drawing on open constellations and commercial tasking where the resolution demands it. The workflow is the same one applied in the Tonga crop-estimation programme: start from a published claim, define the spatial and temporal scope precisely, run the archive, and report honestly on what the data supports and where uncertainty remains. The output is a structured evidence package with GIS layers, scene provenance, and a written assessment, formatted to sit alongside a verification body's own fieldwork rather than to replace it.
Where satellite evidence reaches its limits
Satellite analysis cannot verify carbon stock volumes directly. It can confirm or contradict the presence of forest cover, detect clearing events, and assess canopy condition as a proxy for degradation. Translating canopy cover into tonnes of carbon requires allometric models and ground-truth data that the satellite record alone cannot supply. This page does not cover stock estimation methods; that is addressed in a sibling page on canopy-structure analysis.
Permanence fraud, where a project reports a reversal event as natural disturbance rather than deliberate clearing, is harder to disprove from space alone. Fire scars, for example, can result from arson or from natural lightning ignition, and the satellite record typically cannot distinguish them. Temporal patterns can be suggestive: a fire that begins precisely at a project boundary and burns inward is statistically unusual. But 'suggestive' is not 'conclusive', and any evidence package should say so plainly.
Typical figures
| Spatial resolution (historical baseline) | 30 m (Landsat TM/ETM+/OLI; Hansen et al. product) |
| Spatial resolution (current-state verification) | 10 m (Sentinel-2 MSI); 3 m (Planet SuperDove, commercial licence) |
| SAR resolution (cloud-penetrating change detection) | 10 m (Sentinel-1 C-band IW mode) |
| Archive depth | 1984 to present (Landsat); October 2015 to present (Sentinel-2); Hansen annual product updated yearly |
| Revisit frequency | 16 days per Landsat satellite; 5 days (Sentinel-2 constellation); near-daily (Planet SuperDove); 6–12 days (Sentinel-1) |
| Minimum detectable clearing (optical) | Approximately 0.09 ha at Sentinel-2 10 m; approximately 0.09 ha at Landsat 30 m for large openings; sub-hectare detection reliability improves significantly below 5 m resolution |
| Cloud-cover limitation | Tropical sites may have fewer than 20 cloud-free Landsat observations per year; SAR used to fill optical gaps |
| Spectral bands used | Red, NIR, SWIR (forest/non-forest NDVI and NBR); C-band VV/VH polarisation (SAR coherence change) |
| Baseline reference product | Hansen et al. Global Forest Watch annual tree-cover loss, 30 m, version updated annually by University of Maryland |
| Delivery format | GeoTIFF change layers, GeoJSON boundary overlays, PDF evidence report with scene provenance table |
Analytics Satellize can run
| Historical deforestation rate audit | Year-of-loss extraction from Hansen et al. product within stated reference region and alternative objectively drawn reference regions; rate comparison against project document claims | Tabular rate comparison with map of reference region alternatives and statistical spread; GIS layer |
| Post-crediting clearing detection report | Optical change detection (Sentinel-2 NDVI differencing and random-forest classifier) combined with Sentinel-1 SAR coherence loss time series from crediting start date to present | Dated clearing polygons with scene provenance, area statistics and cloud-gap log; GeoJSON and PDF |
| Crediting start date forest-cover verification | Landsat scene classification at or near stated start date using project-specified canopy-cover threshold; comparison against claimed cover percentage | Classified raster, cover-percentage estimate with confidence interval, and written assessment of consistency with project document |
| Reference region selection bias assessment | Spatial analysis comparing observed historical loss rates across developer-drawn reference region versus distance-band and road-buffer alternatives; follows methodology described in peer-reviewed literature on REDD+ additionality testing | Multi-scenario rate table, map of alternative reference regions, written interpretation of statistical outlier status |
| SAR-confirmed wet-season clearing log | Sentinel-1 IW coherence time series over project boundary during cloud-obscured months; backscatter change thresholding to flag canopy-loss events | Dated SAR-flagged event polygons with optical confirmation status noted; GeoTIFF and summary table |
| Boundary integrity check | Overlay of project boundary coordinates from registry document against current high-resolution imagery and historical Landsat scenes to confirm boundary was not redrawn post-crediting to exclude cleared areas | Version-comparison map with annotated discrepancies; PDF report section |
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.