ML-VHRNature-Based SolutionGreen CarbonTier 3 — high assurancemedium confidenceProxy estimate

Machine-Learning Traceability with Very-High-Resolution Imagery

Machine-learning model over high-resolution imagery plus biomass samples; the profile closest to direct measurement — still an estimate without field training data.

traceability programmehigh-integrity MRVsupply-chain verification
Measurement tier
Tier 3 — high assurance
Model confidence
medium
Uncertainty deduction
−10% on net credits

About this methodology

The Machine-Learning Traceability with Very-High-Resolution Imagery profile (ML-VHR) is a Nature-Based Solution and Green Carbon methodology profile used by the Athlas Verity engine to estimate and independently verify carbon outcomes for traceability programme, high-integrity MRV, supply-chain verification. Machine-learning model over high-resolution imagery plus biomass samples; the profile closest to direct measurement — still an estimate without field training data.

Athlas classifies this profile at Tier 3 — high assurance with medium confidence. Because the underlying quantity cannot be observed directly from orbit, the output is treated as a proxy estimate and a larger conservative uncertainty deduction of 10% is applied to net creditable volume. The profile sets parameterisation, dataset bundles and evidence requirements; the core accounting chain is shared across the platform.

Verification is evidence-led. A project mapped to this profile must supply 2 mandatory evidence items plus 4 optional supporting records. Athlas cross-checks the submitted project boundary and documents against 4 Earth-observation and ancillary data layers, then issues a tamper-evident verification report that can be reviewed independently.

How Athlas verifies this

  1. 1

    1. Boundary & evidence intake

    The project boundary (GeoJSON) is ingested and validated against the 6 evidence requirements defined for profile ML-VHR. Ownership, legality and truthfulness declarations are recorded for the audit trail.

  2. 2

    2. Baseline reconstruction

    Reference conditions are reconstructed from a multi-year satellite time series to establish a defensible counterfactual for traceability programme.

  3. 3

    3. Satellite measurement

    The primary signal is measured from COPERNICUS/S2_SR_HARMONIZED (bands B8, B4, B11) with a median reducer at 10 m, combined with the remaining 3 supporting layers for terrain, moisture, climate and land-cover context.

  4. 4

    4. Conservative accounting

    Results are passed through the shared accounting chain using the profile's own parameters, then reduced by the profile uncertainty deduction of 10% (proxy treatment applied) to give a conservative, uncertainty-aware creditable volume.

  5. 5

    5. Report & on-chain anchoring

    A verification report is generated with a tamper-evident hash and anchored immutably, so the Nature-Based Solution result for ML-VHR can be independently reviewed and its retirement tracked.

Satellite & Earth-observation layers

RoleDatasetBandsReducerScale
high res indexCOPERNICUS/S2_SR_HARMONIZEDB8, B4, B11median10 m
biomass samples(optional)LARSE/GEDI/GEDI04_A_002_MONTHLYagbmean25 m
sar structure(optional)COPERNICUS/S1_GRDVHmedian10 m
land coverESA/WorldCover/v200Mapmedian10 m

Profile parameters & provenance

4 recorded parameters1 externally sourced0 calibrated defaults3 Athlas policy

A calibrated default is not a standard value. Before a claim under this profile is marked verified, every calibrated default it relies on must be replaced or justified with project-specific measurement.

carbonFraction
Externally sourced
0.47

Carbon fraction of dry biomass. 0.47 is the temperate/boreal default and 0.5 the tropical default; both live in this range rather than being a single universal constant.

IPCC Guidelines for National Greenhouse Gas Inventories, Volume 4: Agriculture, Forestry and Other Land Use (2006), Ch. 4, Table 4.5 (carbon fraction of aboveground biomass)

modelType
Athlas policy
gradient-boosted-regression

Model family used for the estimate, recorded so the method is reproducible.

trainingSamplesRequired
Athlas policy
30

Minimum number of reference samples before a supervised model may be applied.

capAgbPerHa
Athlas policy
450

Physical plausibility ceiling on aboveground biomass per hectare. Set by us to reject implausible plots; it is a sanity gate, not an ecological measurement.

Externally sourced

The value comes from the cited standard. A reviewer can open the document and check the number.

Calibrated default

A defensible default drawn from the literature, not a standard value. It must be replaced or justified with project-specific measurement before the claim is verified.

Athlas policy

A deliberate Athlas choice — a buffer, deduction or weighting. Not a scientific constant and never presented as one.

Required evidence

  • Project boundary (GeoJSON)

    Required · accepts geojson, json, zip

  • Soil / biomass lab certificate (optional calibration)

    Optional · accepts pdf

  • Baseline / reference study

    Optional · accepts pdf

  • Data truthfulness declaration

    Required · accepts pdf, jpg, png

  • Project field photos (multi)

    Optional · accepts jpg, png, webp

  • Stakeholder / FPIC consultation record

    Optional · accepts pdf

Verification workflow

Frequently asked questions

What is the Machine-Learning Traceability with Very-High-Resolution Imagery (ML-VHR) methodology profile?

It is a Nature-Based Solution and Green Carbon methodology profile in the Athlas Verity library. Machine-learning model over high-resolution imagery plus biomass samples; the profile closest to direct measurement — still an estimate without field training data. It applies to traceability programme, high-integrity MRV, supply-chain verification and sets the parameters, satellite dataset bundle and evidence requirements used to verify a project mapped to it.

What evidence is required for profile ML-VHR?

Mandatory evidence includes: Project boundary (GeoJSON); Data truthfulness declaration. Optional supporting records such as Soil / biomass lab certificate (optional calibration); Baseline / reference study; Project field photos (multi); Stakeholder / FPIC consultation record strengthen the assessment.

How does Athlas verify a ML-VHR project with satellites?

Athlas measures the project using 4 Earth-observation layers, including COPERNICUS/S2_SR_HARMONIZED (B8, B4, B11). Estimates are conservative: a 10% deduction is applied to net credits, because the signal is a proxy rather than a direct measurement.

Which verification domain does ML-VHR belong to?

Profile ML-VHR belongs to the Nature-Based Solution domain. Projects are routed through the corresponding verification workflow and dashboard at the Nature Based Solution verification area.

Related methodology profiles