Metric Methodology

These are the headline values most consumers read first. Available at every level (plots, sheds, facilities) unless noted. All deforestation fields are emitted flat on the response attributes object — the metric endpoints do not nest EUDR fields under a separate sub-object. (The ingestion-results endpoint GET /ingestions/asset_results does nest them under attributes.eudr via a separate aggregator — see EUDR Compliance. The metric endpoints described on this page are flat.)

Filtering to EUDR-equivalent rows: pass filter[monitoring_start_date]=gte:2020-12-17&filter[monitoring_start_date]=lte:2021-01-14 to scope to rows where is_compliant is guaranteed non-null. A narrower filter[monitoring_start_date]=eq:2020-12-31 also works if you want the cut-off-anchored row specifically. The bare filter[is_compliant]=eq:true|false filter already excludes nulls inherently (an equality filter never matches null), so combining it with a date filter is only necessary when you want to include the cut-off-window rows that happen to be null due to missing data.

Sparse fieldsets: when using fields[deforestation]=..., list the deforestation fields by their flat names (e.g. noncompliance_area, is_compliant, deforestation_alert). There is no nested eudr.* accessor on metric responses.

# The regulatory EUDR view (is_compliant is meaningful here)
curl '/metrics/plots/deforestation?filter[deforestation_period]=eq:eudr' \
  -H "x-api-key: $KEY"

# Deforestation since the user's own start_date
curl '/metrics/plots/deforestation?filter[deforestation_period]=eq:user' \
  -H "x-api-key: $KEY"

{
  "data": {
    "type": "deforestation",
    "id": "…",
    "attributes": {
      "noncompliance_area": 0.32,
      "noncompliance_area_perc": 0.018,
      "is_compliant": true,
      "noncompliance_plot_count": 0,
      "plot_count": 18,
      "forest_area": 14.6,
      "area": 17.8,
      "agreement": { "all": 0.9, "tmf": 0.8, "glad": 0.95, "dynamic_world": 0.85 },
      "overlap": { "natural_forest": 0.7, "wdpa": 0.0 },
      "annual_defor": { "defor_ha_2023": 0.12, "defor_ha_rmse_2023": 0.04 },
      "uncertainty_rrmse": 0.11,
      "thumbnail_url": "https://…",
      "deforestation_alert": "critical",
      "deforestation_confidence": "very high",
      "monitoring_start_date": "2020-12-31T00:00:00Z",
      "monitoring_end_date": "2024-12-31T00:00:00Z",
      "ingestion_date": "2026-04-29T08:14:31Z"
    }
  }
}

Five detection systems run in the Standard methodology and are combined to maximise sensitivity while controlling false positives. A sixth system — Epoch CCDC (Dynamic World) — is an optional enhanced check that is added to the ensemble only when the Enhanced methodology is enabled (opt-in via metric_options.enhanced_deforestation: true); it does not run by default.

Additionally, Epoch's Biomass Change Trends (10 m, global) capture gradual degradation over timescales longer than a year, complementing the abrupt change detection systems above. This ensures all types of forest loss (deforestation and degradation) are captured, aligning with the EU's definition of deforestation under EUDR.

The API returns agreement between detection systems so users can assess confidence based on multi-source consensus.

Under EUDR, only the loss of natural forests constitutes non-compliance — harvesting in legally-managed timber plantations or licensed managed forests is explicitly outside the regulation's scope. Detected deforestation events that fall within legally-classified managed or publicly-owned forest areas are excluded from the noncompliance_area calculation. The following datasets currently provide these exemptions:

To the extent that authoritative national or subnational managed forest and tenure datasets become available for additional countries, they will be progressively integrated into the exemption layer.

1. Change detection: Combine signals from the Standard detection systems (GLAD-L, GLAD-S2, RADD, OPERA DIST, TMF) over the monitoring period (from the EUDR cut-off date to the end date). When the Enhanced methodology is enabled (enhanced_deforestation: true), the optional CCDC (Dynamic World) time-series layer is added to the ensemble. Pixels where at least one active source indicates forest loss are flagged.
2. Natural forest filter: Only pixels classified as natural forest at the EUDR cut-off date (2020-12-31) are counted. Deforestation outside natural forest is excluded so the metric aligns with EUDR's focus on natural forest.
3. Managed forest exemptions: Exclude pixels in legally-classified managed or publicly-owned forest areas covered by the tenure and ownership datasets above, regardless of detected change.
4. Spatio-temporal sieving: Applied at plot level and in batch/aggregate processing (supply-shed, statistical-region). Connected components of deforested pixels are identified across the full monitoring window; only groups whose combined area meets the minimum threshold (0.5 ha, matching the EUDR critical threshold) are retained. This ensures incremental loss across years is counted while scattered noise is excluded.
5. Weighted area: Partial pixels at boundaries are weighted by the proportion inside the geometry to avoid over- or under-counting.

Commodity presence is used in supply shed analysis (to identify commodity plots within a facility's sourcing area) and in confidence scoring (commodity overlap boosts confidence from high to very high).

The API exposes the alert as the top-level deforestation_alert field on every deforestation response.

The API exposes the band as the top-level deforestation_confidence field on every deforestation response.

Confidence ranges from very low to very high (or no detection). It is advised to visually inspect thumbnails for very low, low, and medium cases, as these may be due to georeferencing issues or false positives. The API may return low confidence even when noncompliance_area == 0 if the plot overlaps natural forest or protected areas (WDPA), as a conservative risk flag.

The confidence hierarchy reflects that forest overlap is more reliable than commodity detection. Forest overlap is required for a critical alert; commodity presence boosts confidence from high to very high but is not required. If no commodity information is available (e.g. collection_hash not provided), the system defaults to assuming commodity presence for backward compatibility.

For confidence levels from low to high, it is recommended to manually inspect the thumbnails (see Thumbnail Interpretation). The underlying values used to determine confidence are available in the API response, so operators can build their own rule-based systems for reporting.

The categorization uses absolute area thresholds (no percentage-based checks) for consistent classification across plots of varying sizes:

• Critical threshold: noncompliance_area > 0.5 ha AND natural forest overlap — both conditions required
• Non-critical threshold: noncompliance_area > 0.25 ha (absolute area only) OR has natural forest overlap but not meeting critical threshold

Important distinction:

• Critical alert = area > 0.5 ha AND natural forest overlap (both required)
• Non-critical alert = area > 0.5 ha but no forest overlap, OR area > 0.25 ha, OR forest overlap but not meeting critical threshold
• Large deforestation (> 0.5 ha) without forest overlap is still significant but gets non-critical alert (e.g. deforestation in existing plantations, degraded areas)
• No deforestation = no detection from any source

The EUDR defines "forest" as land spanning more than 0.5 hectares with trees higher than 5 meters and a canopy cover of more than 10%, or trees able to reach those thresholds in situ. The Epoch API distinguishes natural forests from commodity forests using multiple authoritative datasets.

The natural forest check requires any overlap (> 0 ha) with natural forest or protected areas at the reference date (2020). It combines:

1. Nature Trace Forest Typology 2020 (FORTY) — Primary Forest + Naturally Regenerating Forest classes (global, 10 m)
2. FDP Forest Persistence (global, adaptive thresholds)
3. WDPA protected areas

The check does not impose strict percentage thresholds — any overlap is sufficient. This ensures sensitivity without being overly restrictive.

The commodity check verifies that the target commodity is present within the plot using FDP commodity probability models. Its role in the confidence hierarchy:

1. Very high: Critical + forest + commodity (all indicators) → critical alert
2. High: Critical + forest (forest is most reliable, even without commodity) → critical alert
3. Medium: Critical + commodity (commodity present but forest missing) → non-critical alert
4. Low: Critical only (missing both indicators) → non-critical alert

Commodity presence is optional for critical alerts — it acts as a confidence booster rather than a requirement. This accounts for accuracy limitations in commodity detection layers.

UNEP's World Database on Protected Areas (WDPA) is overlaid with detected changes to flag overlap with protected areas and assess legality. A distinction is made between IUCN/CBD-defined protected areas (governed by "legal or effective" means) and those that do not fit the definition.

Below-ground biomass (BGB) is estimated through a root-to-shoot ratio applied to above-ground estimates.

All *_tco2ehayear variants provide the per-hectare normalised equivalents (e.g. nonluc_per_ha, total_per_ha). Per-component uncertainty fields (luc_uncertainty_sd, removal_uncertainty_sd, biogenic_uncertainty_sd) provide standard deviations.

Naming note: at the facility level the annual_lc plot/shed object is replaced by annual_luc, which carries annual LUC emissions in tCO2e/year rather than a land-cover snapshot. Don't conflate the two — the facility emissions resource has no annual_lc field.

• Area calculations are weighted (partial pixel overlap accounted for)
• Biomass can be modelled from 2000 onwards (30 m before 2017, 10 m after)
• AGB model uncertainty stems from GEDI footprint positional accuracy, DEM accuracy, slope, and allometric equation representativeness
• Independent field validation is recommended for carbon accounting use cases

The water stress index is computed as a weighted composite of three sub-indices:

1. Precipitation Deficit Index (weight: 40%)

Computed from CHIRPS precipitation data as the ratio of observed precipitation to long-term potential evapotranspiration (P/PET). A ratio below 1.0 indicates that precipitation is insufficient to meet atmospheric evaporative demand. The index is normalized to a 0–1 scale using historical percentiles for the region and season.

2. Soil Moisture Anomaly Index (weight: 40%)

Computed from SMAP L4 root-zone soil moisture. For each pixel and time step, the observed soil moisture is compared to the climatological distribution (2015–present baseline). Negative anomalies (drier than normal) increase the stress score. The index is normalized to 0–1 using z-scores mapped to percentiles.

3. ET Anomaly Index (weight: 20%)

Computed from MODIS actual ET relative to climatological ET for the same location and time of year. Negative ET anomalies (less evapotranspiration than expected) indicate vegetation water stress — plants are transpiring less due to water limitation. Normalized to 0–1.

Composite Index Calculation:

water_stress_index = 0.40 * precipitation_deficit + 0.40 * soil_moisture_anomaly + 0.20 * et_anomaly

Each sub-index is clipped to the 0–1 range before weighting.