Cattle ranching expansion pressure monitoring around forest carbon projects
Cattle ranching is the dominant proximate driver of tropical deforestation in Latin America and the principal leakage risk for forest carbon projects. Sentinel-2 phenological composites and Sentinel-1 SAR texture metrics can distinguish actively managed pasture from secondary regrowth, quantifying expansion pressure in project buffer zones as Verra's VM0015 requires.
Sensors
- Sentinel-2 MSI: 10 m resolution in visible and near-infrared bands, 5-day revisit at the equator with both satellites. Multi-temporal composites across the dry season (typically June to October in the Brazilian Amazon) produce phenological curves that separate actively grazed pasture, which shows a characteristic flush-and-brown cycle, from closed-canopy forest and from woody secondary regrowth.
- Sentinel-1 SAR (C-band, IW mode): 6-day repeat, 10 m ground range resolution, all-weather imaging. Backscatter texture metrics (variance, GLCM contrast) distinguish the smooth, low-backscatter surface of short grazed grass from the high, spatially heterogeneous return of forest canopy. Particularly useful in the wet season when optical cloud cover is near-total across large parts of Amazonia.
- Landsat Collection 2 ARD: 30 m resolution, 16-day revisit per satellite (8-day with combined Landsat 8 and 9 operations). Analysis-Ready Data products include surface reflectance and pixel-quality flags. The archive extends to 1972, enabling multi-decade baseline construction for additionality assessments and historical leakage belt characterisation.
- MapBiomas Annual Land-Cover Series: Annual 30 m land-cover classifications for Brazil and several other Latin American countries, produced by a multi-institution consortium using Landsat time series and Google Earth Engine. Publicly released classifications distinguish pasture, agriculture, secondary vegetation and forest classes, providing a peer-reviewed reference layer for buffer-zone stratification.
Why ranching pressure is a leakage problem, not just a land-cover problem
Forest carbon projects under Verra's VM0015 and related REDD+ methodologies must account for leakage: the displacement of deforestation from the project area to its surroundings. Cattle ranching is not a background process. It is an economically rational response to land prices, beef demand and road access, and it moves. When a project boundary is drawn, the clearing pressure does not disappear; it shifts to the buffer zone and sometimes to the far side of it.
Monitoring that boundary is therefore a quantitative obligation, not an audit nicety. VM0015 requires periodic assessment of land-use change in a defined leakage belt. Satellite time series are the only cost-effective way to do this at the spatial resolution and revisit frequency the methodology demands. The challenge is not detecting large clearings, which any moderate-resolution sensor can do. The challenge is classifying what replaces forest once cleared, and detecting the slow encroachment of pasture into degraded forest margins before a full clearing event occurs.
What a phenological composite gives away about pasture management
Actively managed cattle pasture has a spectral life cycle that closed-canopy forest and natural grassland do not share. In the Brazilian Amazon dry season, pastureland grasses dry and brown, producing a pronounced drop in the normalised difference vegetation index (NDVI) and a rise in the shortwave-infrared reflectance captured by Sentinel-2 bands 11 and 12. Managed pastures are also often burned or cut at the end of the dry season, producing a brief, sharp NDVI minimum followed by a rapid green flush. This flush-and-brown signature, extracted from a dense time series of cloud-free Sentinel-2 observations across a full annual cycle, is a reliable discriminator.
Secondary regrowth complicates the picture. Young secondary forest (two to five years old) can have a similar canopy height and NDVI profile to low-productivity pasture with scattered woody invaders. Phenological amplitude alone is insufficient to separate them reliably at this stage. Adding Sentinel-1 backscatter texture resolves much of the ambiguity: woody secondary regrowth produces higher and more spatially variable C-band backscatter than short grass, even before canopy closure. The combination of optical phenology and SAR texture is now standard practice in published Brazilian land-cover mapping, including the MapBiomas pasture layer.
The resolution floor and the degraded-forest ambiguity
Sentinel-2 at 10 m resolves individual paddocks and fence lines, which is adequate for buffer-zone pressure mapping. The more persistent problem is thematic, not spatial. Degraded forest with a broken canopy, sparse understorey and compacted soil from cattle trampling can produce spectral signatures that overlap with low-productivity pasture at any optical resolution. This is not a sensor failure; it is a genuine ecological continuum. Cattle regularly graze inside degraded forest fragments in Amazonia, creating a mixed land cover that no classifier handles cleanly.
Honest practice requires reporting a confusion probability for pixels in this transition zone rather than forcing a binary classification. Published accuracy assessments for pasture mapping in Amazonia typically report overall accuracies of 85 to 92 percent at the class level, with the pasture-versus-degraded-forest boundary accounting for the majority of errors. Buyers of MRV analytics should ask for class-level confusion matrices, not headline accuracy figures.
Quantifying expansion rate and directionality in the buffer zone
Detecting pasture is necessary but not sufficient for leakage accounting. The methodology requires quantifying the rate and direction of expansion relative to the project boundary. This means computing annual land-cover transitions from a consistent time series, then calculating net pasture gain within the leakage belt stratified by distance band from the project perimeter.
Directionality matters. If pasture expansion is concentrated on the side of the project boundary nearest to an existing road or settlement, that is evidence of displacement pressure rather than coincidental regional trends. Road-proximity analysis using OpenStreetMap or government road datasets, combined with the annual pasture-change raster, produces a spatial pressure index that can be reported per monitoring period. Sentinel-1 coherence change detection, which measures how much the radar phase relationship between two acquisitions changes, can flag new clearing events within days of occurrence, providing an early-warning signal before the next optical composite is available.
Fitting into a VM0015 monitoring report
Verra's VM0015 (Methodology for Avoided Unplanned Deforestation) specifies that leakage belt monitoring must use consistent methods across monitoring periods and must document the data sources, classifiers and accuracy assessments used. Satellite-derived pasture-expansion metrics satisfy this requirement when the processing chain is documented and the input data are publicly archived.
Practically, a monitoring package for a single project typically covers the project area plus a leakage belt extending 10 to 50 km outward, depending on the regional deforestation context. Annual land-cover maps derived from Sentinel-2 composites and Sentinel-1 texture layers, cross-referenced against the MapBiomas annual series as an independent validation layer, produce the transition matrices and area statistics the methodology requires. Satellize structures these outputs as GIS layers with attached uncertainty tables, formatted for direct submission to a validation and verification body. The approach is the same one used in the Tonga crop-estimation programme: open-constellation data, documented methods, auditable outputs.
One practical note on timing: Sentinel-2 composites for a full dry season in the Amazon are typically available for processing by November of the same year, meaning a calendar-year monitoring report can be completed within the first quarter of the following year without relying on commercial tasking.
Typical figures
| Optical spatial resolution | 10 m (Sentinel-2 visible and NIR); 30 m (Landsat Collection 2 ARD) |
| SAR spatial resolution | 10 m ground range (Sentinel-1 IW mode) |
| Revisit frequency | 5 days optical (Sentinel-2A+B combined); 6 days SAR (Sentinel-1); 8 days optical (Landsat 8+9 combined) |
| Spectral bands used | Sentinel-2 B2, B3, B4, B8 (NDVI), B11, B12 (SWIR for dry-season browning); Sentinel-1 VV and VH polarisations |
| Minimum detectable clearing | Approximately 0.1 ha at 10 m resolution for complete clearing; degraded-margin encroachment below 0.5 ha carries elevated classification uncertainty |
| Cloud cover limitation | Optical compositing requires 4 to 8 cloud-free observations per pixel per season; SAR provides all-weather gap-fill but does not replace phenological discrimination |
| Archive depth | Sentinel-2: from 2015; Landsat: from 1972; MapBiomas annual series: from 1985 (Brazil) |
| Latency for monitoring outputs | Annual composites available within 6 to 10 weeks of season end; Sentinel-1 change alerts within 1 to 3 days of acquisition |
| Delivery formats | GeoTIFF land-cover rasters, vector transition polygons, confusion matrices in CSV, PDF summary report for VVB submission |
| Coverage | Global; standard leakage belt 10 to 50 km radius around project boundary, configurable |
Analytics Satellize can run
| Annual pasture-extent map with uncertainty class | Random forest classifier trained on Sentinel-2 dry-season phenological composites (NDVI amplitude, SWIR reflectance) and Sentinel-1 GLCM texture metrics; validated against MapBiomas reference layer | GeoTIFF raster with per-pixel class probability; class-level confusion matrix in CSV |
| Net pasture-expansion area by distance band | Annual land-cover transition matrix computed from sequential classified rasters; zonal statistics within concentric leakage-belt rings at 5 km intervals | Tabular area statistics (hectares per transition type per zone) formatted to VM0015 leakage accounting tables |
| Pasture-versus-degraded-forest transition zone flag | Spectral mixture analysis on Sentinel-2 composites combined with SAR backscatter variance; pixels with posterior class probability below 0.75 flagged as ambiguous | Vector polygon layer of ambiguous pixels with recommended field-verification priority score |
| Road-proximity pressure index | Kernel density of new pasture pixels weighted by inverse distance to nearest road or settlement; computed annually within leakage belt | Raster pressure index layer and summary chart for monitoring report narrative |
| Near-real-time clearing alert (SAR-based) | Sentinel-1 coherence change detection between consecutive 6-day repeat passes; thresholded against baseline coherence distribution for forest and pasture classes | Alert GeoJSON feed updated within 3 days of new Sentinel-1 acquisition; email or API push to project manager |
| Multi-decade baseline land-cover trajectory | Landsat Collection 2 ARD time series processed through MapBiomas-compatible classification pipeline; annual maps from 1985 to present for additionality and historical leakage characterisation | Time-series GeoTIFF stack with accompanying area statistics table; suitable for VM0015 historical reference region analysis |
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.