Afforestation, reforestation and revegetation project canopy monitoring
Satellite time-series can verify whether ARR carbon projects are developing canopy cover at the modelled rate, but sensor resolution, cloud cover and the plantation-versus-regeneration ambiguity all constrain what can honestly be concluded.
Sensors
- Planet SuperDove: 3 m resolution, 8 spectral bands including red-edge, daily revisit over most land areas. Sufficient to resolve individual tree crowns once canopy reaches roughly 5 m diameter; archive from 2021 for SuperDove, earlier for earlier Dove generations at 3–5 m.
- Sentinel-2 MSI: 10 m optical bands (B2, B3, B4, B8) and 20 m red-edge and SWIR bands, 5-day revisit at mid-latitudes. Reliable for tracking fractional canopy cover at stand level once cover exceeds roughly 10–15%; individual sapling detection is not possible. Free archive from 2015.
- GEDI (Global Ecosystem Dynamics Investigation): Spaceborne lidar operating from the ISS, sampling at 25 m footprints on a non-contiguous grid. Provides canopy height and vertical structure profiles. Coverage limited to latitudes 51.6° N/S; not a wall-to-wall map but a statistically powerful calibration layer. Mission data publicly available from 2019.
- WorldView-3: 0.3 m panchromatic, 1.24 m multispectral, 8 VNIR and 8 SWIR bands. Allows individual sapling crown detection and early mortality identification, but revisit is tasking-dependent (typically 1–4.5 days for any given point) and cost per km² limits systematic coverage to audit samples.
- Sentinel-1 SAR (C-band): 6-day repeat interferometric SAR at 10 m resolution. Penetrates cloud, which matters in humid tropics where optical data can be unavailable for months. Backscatter change tracks canopy development qualitatively; InSAR coherence loss can signal new vegetation establishment, though interpretation requires care.
What the verification question actually is
An ARR carbon project earns credits by claiming that trees were planted, survived and are sequestering carbon at a modelled rate. The verification question is therefore threefold: did planting happen at the claimed location and density, are the trees still alive, and is the canopy developing on the trajectory the project model assumed? These are distinct questions that require different observables and different sensors.
Satellite imagery can answer the first question with high confidence once canopy closure is sufficient for detection. The second and third questions are harder. A stand can look green from orbit while suffering root disease or water stress that will produce delayed mortality. The honest position is that satellite monitoring is a strong screen for gross fraud and a useful tracker of canopy development trajectories, not a substitute for ground-truth sampling at the parcel level.
Minimum detectable canopy: where each sensor draws the line
At 10 m resolution, Sentinel-2 cannot resolve a sapling. A pixel is green or not green, and a newly planted stand of trees at 2 m spacing with 0.5 m crowns occupies perhaps 5% of a pixel's area. The NDVI signal is dominated by understorey and soil. Sentinel-2 becomes reliably informative for ARR monitoring only when fractional canopy cover at stand level approaches 15–20%, which for fast-growing species in humid climates may occur within two to four years of planting, but for dryland species can take a decade.
Planet SuperDove at 3 m resolution changes the arithmetic considerably. A tree crown of roughly 3–5 m diameter begins to occupy a meaningful fraction of a pixel, making individual crown detection feasible for mature saplings. The red-edge band (roughly 700–730 nm) is particularly useful: it responds to chlorophyll content in a way that separates healthy canopy from stressed or dead crowns more cleanly than visible-band NDVI alone. Even so, crowns smaller than 2 m diameter in a 3 m pixel will be mixed with soil signal and are not reliably detected.
WorldView-3 at 0.3 m panchromatic resolution can detect individual saplings with crowns of 1 m or larger, making it the only spaceborne optical option for early-stage mortality surveys. The practical constraint is cost and revisit. A systematic wall-to-wall survey of a 10,000-hectare project at WorldView-3 resolution is expensive; the practical use is stratified audit sampling, where a fraction of the project area is imaged at high resolution to validate the lower-resolution time series.
Monoculture plantation or natural regeneration? The classification problem
Carbon registries treat planted monocultures and natural regeneration differently for baseline and additionality purposes, and some methodologies require evidence of species diversity. Satellite imagery can distinguish the two in favourable conditions. Plantation rows produce a regular spatial texture that is detectable in high-resolution imagery through Fourier or wavelet analysis of crown spacing. Natural regeneration shows irregular crown sizes, variable spacing and spectral heterogeneity across the stand.
The difficulty is that young plantations before canopy closure look almost identical to early natural regeneration from orbit: scattered crowns on bare or grassy ground. The distinction becomes tractable at 3 m resolution once crowns are large enough to show their arrangement. At 10 m resolution it is largely intractable until the stand is mature enough to show texture. Hyperspectral sensors can in principle separate species by leaf biochemistry, but no free-access spaceborne hyperspectral system currently offers the resolution and revisit needed for routine ARR monitoring. PRISMA (ASI) and DESIS offer some capability but with significant coverage and tasking constraints.
The practical answer is to combine spatial texture analysis from Planet or WorldView with field-collected reference polygons of known plantation and regeneration stands, train a classifier on those references, and report the classification with an explicit uncertainty estimate. Presenting a binary plantation-or-regeneration label without that uncertainty is not defensible.
Mortality detection in the first three years
Early mortality is the central verification risk in ARR projects. Planting events are relatively easy to stage or document falsely; survival over three years is harder to fake at scale but also genuinely hard to monitor from orbit. A stand that loses 30% of its trees in year two may still show a rising mean NDVI if surviving trees are growing rapidly.
GEDI lidar addresses this partially. Its 25 m footprint height profiles can detect whether canopy height is increasing over time, and a stand with high mortality will show lower mean canopy height and lower plant area index than a healthy stand of the same age and species. The limitation is spatial sampling: GEDI does not image every hectare. Its orbital track spacing means that a small project area may have only a handful of footprints, which is insufficient for stand-level mortality mapping but adequate for calibrating a Sentinel-2 or Planet-derived canopy cover model.
The most practical mortality-detection workflow combines annual Planet SuperDove composites (to track crown-level change at 3 m) with GEDI height samples (to anchor the biomass trajectory) and stratified WorldView-3 audit imagery (to validate mortality estimates in a sample of parcels). Sentinel-2 provides the long-term baseline and gap-fills cloud-affected periods. None of these alone is sufficient. Satellize structures its ARR analytics around exactly this multi-sensor stack, drawing on the same open-constellation approach it uses for the Kingdom of Tonga crop-estimation programme.
Cloud, seasonality and the tropical monitoring gap
In humid tropical regions, where most large ARR projects are located, persistent cloud cover can render optical sensors effectively blind for months. Sentinel-2 median compositing over a 12-month window can recover a cloud-free view in most tropical areas, but annual compositing means that a mortality event in month three may be invisible until the following year's composite. Planet's higher revisit increases the probability of cloud-free acquisitions but does not eliminate the problem in persistently overcast climates.
Sentinel-1 C-band SAR penetrates cloud and provides 6-day repeat coverage. Backscatter from a developing canopy increases as leaf area grows, and coherence between repeat passes drops as vegetation structure changes. These signals are real but noisy: SAR backscatter is sensitive to soil moisture, wind and precipitation as well as canopy structure, and separating these confounders requires careful temporal filtering. SAR is most useful as a consistency check on optical-derived canopy cover estimates rather than as a primary monitoring signal.
Building a time series that a verifier will accept
Carbon registries and third-party verifiers increasingly expect satellite evidence packages that document the full processing chain: raw imagery provenance, atmospheric correction method, cloud masking approach, the canopy cover algorithm and its validation statistics, and a change-detection record tied to specific dates. A time series that begins before the project start date is essential for establishing the pre-project baseline and demonstrating that the claimed canopy did not exist before planting.
Sentinel-2 archive depth to 2015 and Landsat archive depth to the 1970s (at 30 m resolution) provide the historical baseline. Planet archive depth varies by location but typically covers 2016 onwards for most land areas. The combination gives verifiers a credible pre-project and post-planting record. Gaps in the time series caused by cloud or sensor outage should be documented explicitly rather than interpolated silently. A verifier who finds an undisclosed gap will trust the whole package less.
Typical figures
| Optical spatial resolution range | 0.3 m (WorldView-3 pan) to 10 m (Sentinel-2 VNIR); 3 m (Planet SuperDove) |
| Revisit frequency | Daily (Planet); 5 days (Sentinel-2 at mid-latitudes); tasking-dependent (WorldView-3, typically 1–4.5 days) |
| SAR revisit (cloud-penetrating) | 6 days (Sentinel-1 C-band, 10 m IW mode) |
| Minimum detectable canopy cover (Sentinel-2) | ~15–20% fractional cover at stand level; individual saplings not detectable |
| Minimum detectable crown (Planet SuperDove) | ~3–5 m crown diameter for reliable detection; crowns below ~2 m are mixed-pixel |
| GEDI canopy height accuracy | ~1 m RMSE for mean canopy height in closed-canopy forest; degrades in sparse canopy; 25 m footprint, non-contiguous sampling |
| Spectral bands relevant to canopy health | Red-edge (~700–730 nm, Planet SuperDove Band 6; Sentinel-2 B5/B6/B7); SWIR (WorldView-3, Sentinel-2 B11/B12) for moisture stress |
| Archive depth | Sentinel-2: 2015 to present; Planet: ~2016 to present (varies by location); Landsat: 1972 to present at 30 m |
| Latency (open constellations) | Sentinel-2 Level-2A: 1–3 days after acquisition; Planet: same-day to 24 hours on standard feed |
| Typical delivery formats | GeoTIFF canopy cover fraction rasters, GeoPackage change polygons, CSV time-series tables per project parcel, PDF evidence report for verifier submission |
Analytics Satellize can run
| Pre-project baseline canopy cover map | Sentinel-2 and Landsat historical archive compositing with fractional cover retrieval (linear spectral unmixing or random forest regression against field reference) | GeoTIFF raster and PDF report documenting canopy state at project start date and up to 10 years prior |
| Annual canopy cover fraction time series per project parcel | Cloud-masked Sentinel-2 annual composites; NDVI and red-edge chlorophyll index time series; change-point detection against modelled growth trajectory | CSV time-series table and GeoPackage polygon layer with per-parcel cover fraction and deviation from model, updated annually |
| High-resolution crown-level mortality survey | Planet SuperDove 3 m annual composites with individual crown segmentation; WorldView-3 audit sample for validation; mortality fraction estimated per stand | GeoTIFF crown-detection raster, stand-level mortality rate table, stratified audit report with confidence intervals |
| Plantation versus natural regeneration classification | Spatial texture analysis (crown spacing regularity) on Planet or WorldView imagery combined with spectral heterogeneity index; supervised classifier trained on field-verified reference polygons | GeoPackage classification layer with per-polygon label and posterior probability score; uncertainty map |
| Canopy height trajectory from GEDI | GEDI Level-2A and Level-2B footprint data filtered for quality flags; mean canopy height and plant area index extracted per project area; compared against species-specific growth models | CSV of GEDI footprint statistics per project area, annotated against expected height-at-age curve, with flagged anomalies |
| SAR-optical consistency check for cloud-affected periods | Sentinel-1 IW backscatter time series (VV/VH polarisation); coherence change detection; cross-validated against optical canopy cover estimates for cloud-free dates | Monthly SAR-derived canopy development index raster; alert flag where SAR and optical signals diverge beyond threshold |
| Verifier-ready evidence package | Full processing-chain documentation per registry requirements (Verra VCS, Gold Standard or equivalent); imagery provenance, atmospheric correction metadata, algorithm validation statistics, gap log | Structured PDF evidence report with embedded GIS layers and processing metadata, formatted for third-party verifier submission |
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.