Near-real-time deforestation alerts for leakage monitoring
REDD+ and VCS VM0015 require deforestation monitoring inside project boundaries and across displacement-leakage belts. Optical and SAR time-series together can detect clearing to roughly 0.5 ha, but revisit gaps, agricultural burn scars and cloud cover impose real limits that any credible MRV system must account for.
Sensors
- Sentinel-2 MSI: 10 m resolution in red, green, blue and near-infrared bands; 5-day revisit at the equator under cloud-free conditions. NDVI and NBR change detection is the primary optical signal for cleared area mapping. Effective revisit in persistently cloudy humid tropics often stretches to 20-40 days per usable acquisition, which is the binding latency constraint.
- Sentinel-1 SAR (C-band): 10-20 m resolution in IW mode; 6-12 day revisit depending on latitude. C-band backscatter drops sharply when closed-canopy forest is replaced by bare soil or low regrowth, providing cloud-penetrating confirmation of optical alerts. Sensitivity to forest structure degrades in dense secondary regrowth within weeks of clearing.
- Planet SuperDove: 3-5 m resolution across 8 spectral bands including red-edge; daily revisit in principle, though cloud and tasking geometry reduce effective tropical revisit to several days. The higher spatial resolution reduces the minimum mappable clearing size to roughly 0.1-0.2 ha, which matters for detecting selective encroachment at leakage-belt edges. Requires a commercial licence.
- RADARSAT Constellation Mission (RCM): C-band SAR at 3-100 m resolution depending on mode; three-satellite constellation achieves daily revisit at mid-to-high latitudes and 4-day revisit near the equator. The compact polarimetry mode offers improved discrimination between bare soil, low herbaceous cover and intact forest compared with single-polarisation C-band, reducing commission errors from agricultural activity.
- MODIS / VIIRS (Terra, Aqua, Suomi-NPP): 250-375 m resolution with daily global coverage. Too coarse to map individual clearings below roughly 5-10 ha, but the GLAD and FORMA alert systems built on these sensors provide a rapid regional signal useful for prioritising where to direct higher-resolution tasking. Latency from acquisition to public alert is typically 1-8 days.
What the standards actually require you to detect
REDD+ methodologies and the Verra VCS VM0015 framework impose two spatially distinct monitoring obligations. The first is straightforward: detect forest clearing inside the project boundary and attribute it to the relevant carbon pool. The second is harder. Leakage monitoring requires a surrounding belt, typically extending 10-50 km beyond the project edge depending on the methodology, where displacement of deforestation pressure must also be tracked. That belt is often larger in area than the project itself, and it receives far less attention from project developers.
The minimum mappable unit matters enormously here. VM0015 does not specify a single threshold, but auditors generally expect clearings above 0.5 ha to be detected and attributed. At Sentinel-2's 10 m resolution, 0.5 ha corresponds to roughly 50 pixels, which is detectable in principle but vulnerable to mixed-pixel effects at forest edges and to commission errors from spectrally similar disturbances such as agricultural burns. Smaller clearings, the kind that characterise incremental encroachment, require Planet-class resolution or field verification.
Why cloud is a structural problem, not a weather inconvenience
The Amazon basin, the Congo basin and the forests of Southeast Asia, where the majority of REDD+ carbon credits originate, share a common characteristic: persistent cloud cover during the wet season, which is also when agricultural expansion often accelerates ahead of the dry-season burn. A Sentinel-2 scene over Pará state in Brazil may be cloud-free fewer than 30 days per year in some pixels. A 5-day revisit sensor that is obscured 90 percent of the time delivers an effective revisit of 50 days. That is the honest latency floor for optical-only systems in the worst-affected zones.
SAR is the practical answer. Sentinel-1 C-band backscatter from intact closed-canopy tropical forest is relatively stable, typically in the range of -6 to -10 dB in VV polarisation, and drops by 2-4 dB after clearing as the volumetric scattering from canopy is removed. That signal is detectable through cloud. The limitation is that C-band penetrates only the top metre or two of canopy, so it responds to the clearing event itself rather than to the subtler structural changes that precede it. It also struggles to distinguish a freshly cleared field from a flooded rice paddy or a harvested oil palm block without additional context.
Commission errors: when a burn scar looks like deforestation
The single largest source of false alerts in tropical deforestation monitoring is the agricultural burn scar. Burning of crop residues, pasture management fires and shifting cultivation all produce sharp NDVI drops and elevated NBR change scores that are spectrally similar to forest clearing. In leakage belts that border agricultural smallholder landscapes, commission error rates in NDVI-only systems can reach 15-30 percent without additional filtering.
Several published approaches reduce this. Temporal trajectory analysis, fitting a harmonic model to the NDVI time series and flagging deviations that persist beyond 30-60 days, distinguishes the rapid recovery of a burn scar from the sustained low-NDVI signature of a cleared and bare or replanted area. SAR coherence change detection adds a second independent signal: a burn scar on standing stubble maintains some coherence between repeat passes, whereas a cleared and tilled field does not. Neither method eliminates the problem. Both reduce it to a manageable level if the time series is dense enough, which returns the argument to revisit frequency.
Building a leakage-belt monitoring system that auditors will accept
A credible leakage-monitoring system needs a documented, reproducible baseline. That means defining the leakage belt boundary in GIS, archiving the pre-project forest cover map with a stated minimum mapping unit, and running the same change-detection algorithm consistently across the monitoring period. Ad hoc visual inspection of imagery does not satisfy a third-party verification audit.
The GLAD forest alert system, published by the University of Maryland and freely available, provides a useful independent cross-reference layer for Landsat-based alerts at 30 m resolution with roughly weekly latency in cloud-free conditions. Projects using it as a supplementary layer should document its known limitations: 30 m resolution misses clearings below about 0.1 ha, and the alert-to-confirmation lag in cloudy regions can exceed 60 days. Satellize structures its leakage-monitoring analytics to combine the GLAD alert feed with Sentinel-1 SAR confirmation passes, producing a two-source corroborated alert that carries a documented commission-error estimate rather than a binary flag.
For projects seeking registry-grade evidence packages, every alert should carry: the detection date and sensor, the cloud-cover fraction for the relevant scene, the area estimate with uncertainty bounds, and the SAR confirmation date. That audit trail is what separates a monitoring system from a monitoring claim.
Sensor combinations by forest type and budget
No single sensor is optimal across all tropical forest contexts. Humid lowland forests with persistent cloud demand SAR as the primary detection layer, with optical used opportunistically for confirmation and attribution. Seasonally dry forests, including the Cerrado and Miombo woodlands, have longer cloud-free windows and are better served by dense Sentinel-2 time series, where the higher spatial resolution and spectral richness reduce commission errors from fire. Montane forests present a different problem: terrain shadows in SAR imagery can mimic backscatter drops from clearing, requiring careful geometric correction and slope masking.
Budget shapes the architecture. Open-access Sentinel-1 and Sentinel-2 data, combined with GLAD alerts and VIIRS active-fire products, can cover a leakage belt at no imagery cost. The processing and analysis are where cost sits. Adding Planet SuperDove tasking over suspected encroachment hotspots, triggered by a coarser alert, is a cost-effective way to achieve sub-0.2 ha detection without paying for daily commercial coverage of the entire belt. RCM compact polarimetry, where available under a national licence, adds discrimination capability that single-pol Sentinel-1 cannot match.
Latency, archive depth and what to tell your verifier
Alert latency in operational systems ranges from 1-2 days for VIIRS-based coarse detection to 6-15 days for a SAR-confirmed Sentinel-2 alert in a cloudy tropical zone. That is not fast enough to prevent clearing in progress, but it is fast enough to trigger a field inspection before a cleared area is replanted and the evidence is harder to read. The honest message for project developers is that satellite monitoring is a detection system, not a deterrent by itself.
Archive depth is an asset that is often underused. Sentinel-2 data runs back to 2015, Sentinel-1 to 2014, and Landsat provides a consistent 30 m record to 1972. A well-constructed pre-project baseline drawn from this archive is defensible in a way that a single-date forest cover map is not. Verifiers increasingly expect to see the full time series, not just the change layer, and the ability to replay the detection algorithm on historical data is a meaningful quality signal.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2 NIR/Red); 3-5 m (Planet SuperDove); 30 m (Landsat 8/9) |
| SAR spatial resolution | 10-20 m IW mode (Sentinel-1); 3-100 m depending on mode (RCM) |
| Optical revisit (cloud-free, equatorial) | 5 days (Sentinel-2 twin satellites); daily in principle (Planet); 16 days (Landsat) |
| Effective optical revisit (humid tropics) | 20-60 days typical; worst-case pixels fewer than 10 usable scenes per year |
| SAR revisit | 6-12 days (Sentinel-1, latitude-dependent); 1-4 days (RCM three-satellite constellation) |
| Alert latency from clearing event | 1-2 days (VIIRS/MODIS coarse); 6-15 days (SAR-confirmed Sentinel-2, cloudy tropics) |
| Minimum mappable clearing size | ~0.5 ha (Sentinel-2 NDVI); ~0.1-0.2 ha (Planet SuperDove); ~5-10 ha (VIIRS/MODIS) |
| Key spectral bands / radar frequency | Red (665 nm), NIR (842 nm), SWIR (1610, 2190 nm) for optical; C-band 5.4 GHz for Sentinel-1 and RCM |
| Archive depth | Sentinel-2 from 2015; Sentinel-1 from 2014; Landsat from 1972 |
| Delivery formats | GeoTIFF change layers, GeoJSON alert polygons, CSV alert log with metadata, WMS/WMTS feed for GIS integration |
Analytics Satellize can run
| Leakage-belt deforestation alert feed | Bi-temporal and time-series NDVI change detection on Sentinel-2, filtered by temporal persistence to reduce burn-scar commission errors | GeoJSON polygon feed with detection date, area estimate, cloud-cover fraction and confidence tier; updated on each new cloud-free acquisition |
| SAR confirmation layer | Sentinel-1 VV/VH backscatter change detection using log-ratio thresholding over a 30-day reference window, with slope-mask applied in terrain above 15 degrees | Binary confirmed/unconfirmed flag appended to each optical alert polygon within 6-12 days of the triggering acquisition |
| Commission-error probability estimate | VIIRS active-fire coincidence check plus harmonic time-series residual analysis to distinguish persistent clearing from transient burn-scar signal | Per-alert commission-error probability score (0-1 scale) included in alert metadata, with methodology note suitable for verifier review |
| Pre-project baseline forest cover map | Supervised classification of Sentinel-2 and Landsat archive imagery using spectral indices and texture features, with minimum mapping unit of 0.5 ha documented | GeoTIFF forest/non-forest map for user-defined reference period, with accuracy assessment confusion matrix and 95% confidence area bounds |
| Cumulative leakage attribution report | Spatial overlay of confirmed clearing polygons against leakage-belt zones, aggregated by calendar quarter with area-weighted carbon stock estimates from published biomass maps | PDF and spreadsheet report formatted for VCS VM0015 monitoring report annexes, including time-series chart and per-zone summary table |
| High-resolution encroachment investigation mosaic | Planet SuperDove tasking triggered by coarse-resolution alert, orthorectified and panelled as a before/after comparison at 3-5 m resolution | GeoTIFF mosaic pair with georeferenced field-inspection coordinates, suitable for registry submission as supporting evidence |
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.