ESG deforestation claim verification in agricultural supply chains
Satellite SAR and optical time-series can independently verify whether forest cover in commodity sourcing areas has been removed, and when. This page explains the physics, the resolution limits, and the attribution gap that no sensor alone can close.
Sensors
- Sentinel-1 SAR (C-band, ESA): 10 m ground resolution in Interferometric Wide Swath mode, 6-day revisit at the equator (12-day per satellite, two-satellite constellation). C-band backscatter drops sharply when forest volume-scattering structure is removed, making canopy loss detectable through cloud cover within one to two repeat cycles of the clearing event. Cannot distinguish crop type from bare soil without optical confirmation.
- Sentinel-2 MSI (ESA): 10 m resolution in visible and near-infrared bands, 5-day revisit at mid-latitudes. NDVI and NBR indices confirm land-use conversion to agriculture after clearing, but cloud cover in humid tropical regions can delay usable optical acquisition by weeks to months. Red-edge bands (20 m) help distinguish early regrowth from standing forest.
- Landsat 8/9 OLI (USGS/NASA): 30 m resolution, 8-day combined revisit from the two-satellite pair. The long archive (Landsat 5 back to 1984) is essential for establishing pre-disturbance forest baselines and detecting historical clearing that predates corporate commitments. Coarser resolution means clearings smaller than roughly one to two hectares are unreliable to detect.
- Planet SuperDove: 3 m resolution, near-daily revisit globally. Resolves individual field boundaries and can reduce the temporal lag between a clearing event and optical confirmation to one or two days in cloud-free conditions. Commercial licence required; archive depth and pricing vary by area of interest.
What a missing canopy looks like to a radar
C-band radar, as used by Sentinel-1, interacts with forest structure through volume scattering: the signal bounces repeatedly between branches, trunks and the ground before returning to the sensor. Intact tropical forest produces a characteristic high-backscatter signature in cross-polarisation (VH). Remove the canopy and that volume-scattering mechanism collapses. The backscatter drops by several decibels within the first repeat pass after clearing, typically within six to twelve days at the equator.
This physics-based response is what makes SAR the primary detection tool in persistently cloudy regions such as the Amazon basin, the Congo Basin and Southeast Asian palm-oil landscapes. Cloud is transparent to C-band. A corporate sustainability report claiming no clearing occurred during a wet season can be checked against a SAR time-series that was acquiring data throughout. The signal does not lie, though interpreting it correctly requires careful handling of incidence-angle variation and seasonal moisture effects on soil backscatter.
The minimum clearing size problem
Resolution sets a hard floor on what is detectable. At Sentinel-1's 10 m pixel size, a single pixel covers 100 square metres. In practice, reliable detection of a discrete clearing event requires the disturbed area to span several pixels, placing the practical minimum at roughly 0.5 to 1 hectare for SAR-based methods, depending on the surrounding forest contrast and the algorithm used. Landsat at 30 m pushes that floor closer to 2 to 4 hectares. Planet SuperDove at 3 m can resolve clearings of a few hundred square metres, but at the cost of a commercial licence and without the long historical baseline that Landsat provides.
This matters in supply-chain contexts because smallholder conversion, which accounts for a significant share of tropical deforestation in some commodities, often proceeds in fragments of one to three hectares. A monitoring programme using only Landsat will systematically miss a portion of that activity. Honest reporting to an ESG auditor should state which sensor stack was used and what minimum clearing size the analysis can reliably detect, rather than asserting blanket coverage.
Dating the event and the optical confirmation lag
SAR gives a detection window: clearing occurred between the last clean acquisition showing intact canopy and the first acquisition showing the backscatter drop. With a six-day Sentinel-1 revisit, the uncertainty on the clearing date is typically under two weeks. That is precise enough to establish whether a clearing happened before or after a company's stated commitment date, which is often the critical legal and regulatory question.
Optical confirmation via NDVI or the Normalised Burn Ratio takes longer. In the humid tropics, persistent cloud cover can mean that a Sentinel-2 or Landsat scene showing the cleared area at full quality does not arrive for weeks or months after the SAR detection. During that window the land may already be under crop cover, making it harder to characterise the conversion type. Fusing SAR detection with the next available optical scene, rather than waiting for optical alone, is the standard approach in published alert systems such as the Global Land Analysis and Discovery (GLAD) alerts from the University of Maryland.
The attribution gap: pixels do not name companies
This is the hardest part of supply-chain verification, and any provider who glosses over it is selling you something incomplete. Satellite data can establish that forest cover was removed at a specific location on a specific date. It cannot, by itself, identify which farm, which trader, or which corporate entity is responsible.
Closing the attribution gap requires overlaying the detected clearing with cadastral boundaries, concession maps, mill-catchment polygons, and supply-chain declarations. In Brazil, the CAR (Cadastro Ambiental Rural) rural property registry provides georeferenced farm boundaries that can be matched to clearing detections. In Indonesia, concession boundaries are published, though completeness and currency vary. In West Africa, equivalent cadastral data is often sparse or not publicly available. The satellite layer is necessary but not sufficient. A credible verification workflow combines remote-sensing detection with supply-chain documentation review, and the final report should be explicit about which parcels could and could not be attributed.
What regulators and auditors are actually asking for
The EU Deforestation Regulation (EUDR), which applies to cattle, cocoa, coffee, palm oil, soya, wood and derived products placed on the EU market, requires operators to demonstrate due diligence showing that goods are deforestation-free with reference to a 31 December 2020 cut-off date. That is a retrospective verification problem, which is exactly what Landsat's archive back to 1984 and the GLAD forest-loss dataset are built for.
Financial institutions applying the Equator Principles or sector-specific ESG frameworks face a related but forward-looking question: is ongoing sourcing from a declared catchment area remaining deforestation-free? That requires a monitoring programme with defined revisit cadence, alert thresholds, and documented response protocols, not a one-off snapshot. Satellize runs this kind of time-series monitoring for clients who need an independent, non-affiliated view of their supply-chain exposure. The Tonga crop-estimation programme uses a comparable multi-temporal optical workflow, adapted to a different commodity context.
Archive depth and what it reveals about historical baselines
Establishing whether a landscape was forested before a corporate commitment requires a credible historical baseline. The GLAD annual forest-loss dataset, derived from Landsat, provides year-by-year global tree-cover loss back to 2000 at 30 m resolution. This is the most widely cited public reference for deforestation auditing. It distinguishes loss from plantation harvest versus primary forest loss, though the distinction relies on contextual data and carries its own uncertainty.
For areas where finer resolution or more recent baselines are needed, Sentinel-2 provides a consistent archive from 2017 onwards, and Planet's commercial archive extends to 2016 in many regions. The practical implication: a company claiming its sourcing area has been deforestation-free since 2015 can be checked against multiple independent archives. Gaps in coverage due to cloud are documented in the metadata and should be reported as uncertainty rather than treated as absence of evidence.
Typical figures
| SAR spatial resolution (Sentinel-1 IW mode) | 10 m |
| Optical spatial resolution (Sentinel-2) | 10 m (visible/NIR), 20 m (red-edge/SWIR) |
| Optical spatial resolution (Landsat 8/9 OLI) | 30 m |
| Commercial optical resolution (Planet SuperDove) | 3 m |
| SAR revisit (Sentinel-1, two-satellite) | 6 days at equator |
| Optical revisit (Sentinel-2, two-satellite) | 5 days at mid-latitudes; cloud-free acquisition highly variable in tropics |
| Minimum reliably detectable clearing (SAR) | ~0.5–1 ha (Sentinel-1); ~2–4 ha (Landsat) |
| Detection lag (SAR) | One to two repeat cycles after clearing; typically under 12 days |
| Archive depth | Landsat from 1984; Sentinel-1/2 from 2014/2017; GLAD loss dataset from 2000 |
| Delivery formats | GeoTIFF change layers, vector polygon alerts, dated event log, PDF verification report |
Analytics Satellize can run
| Forest-cover baseline map | Multi-temporal Landsat or Sentinel-2 classification using NDVI and spectral mixture analysis, cross-referenced to GLAD annual loss dataset | GeoTIFF raster with pre-commitment forest extent; PDF summary with area statistics by sourcing zone |
| SAR-based clearing alert | Sentinel-1 VH backscatter change detection using threshold or z-score method on rolling time-series stack | Vector polygon alert layer with detection date range, backscatter delta, and confidence flag; delivered within 48 hours of acquisition |
| Optical land-use conversion confirmation | NDVI and NBR differencing on Sentinel-2 or Planet scenes acquired after SAR alert; visual analyst review for crop-type characterisation | Annotated image chip with conversion classification (bare soil, early crop, plantation) and acquisition date |
| Dated deforestation event log | Fusion of SAR detection window with optical confirmation date; event attributed to cadastral parcel where boundary data is available | Tabular event log (CSV/GeoJSON) with coordinates, estimated clearing date range, area in hectares, and parcel ID where attributable |
| Sourcing-area compliance status report | Overlay of clearing detections against declared supply-chain catchment polygons, with EUDR cut-off date (31 Dec 2020) applied as temporal filter | PDF verification report suitable for ESG disclosure or regulator submission, with explicit statement of detection limits and unattributable parcels |
| Ongoing monitoring subscription | Automated SAR and optical change detection on defined area of interest, with quarterly analyst review and annual trend summary | Alert feed (email or API), quarterly GIS layer update, annual compliance summary report |
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.