Permanent Forest Conversion

Source: Global Forest Watch (Hansen/UMD) x WRI & Google DeepMind Year: 2021-2025 Category: Transition risks · Ecosystem use Coverage: Global Format: Raster grid (~60 m) Used in risk analysis: Yes — gates Ecosystem use

What it shows

This layer maps where forest was cleared between 2021 and 2025 and the land then changed use for good. It keeps only the loss whose dominant cause is permanent agriculture, hard commodities, or settlements and infrastructure, so it approximates deforestation rather than canopy disturbance. It matters because a site surrounded by land under conversion sits in a landscape under active pressure, whoever does the clearing.

How it is built

The starting point is the University of Maryland tree-cover-loss record, accessed through Global Forest Watch: loss of trees taller than five metres, detected between 2021 and 2025, in zones that carried more than 50% canopy cover in 2000.

A second dataset then filters it, the WRI and Google DeepMind classification of the dominant driver of forest loss on a 1 km grid over 2001-2025. A cell survives the filter only where that dominant driver is permanent agriculture, hard commodities, or settlements and infrastructure. Rotation forestry, wildfire, shifting cultivation and natural disturbance are dropped. On reference sites the filter keeps 0-2% of Scandinavian, Alpine and US rotation forestry against 79-84% of Amazonian and Indonesian conversion.

The window opens in 2021 because the EUDR cut-off is 31 December 2020: loss detected during 2020 predates it.

Each pixel of the roughly 60 m grid holds the fraction of its cell, from 0 to 1, that was converted.

How to read it

A darker cell lost a greater share of its area to permanent conversion. Two limits are worth stating. The driver classification is the dominant class of a 1 km cell over 2001-2025, so it is reliable at landscape scale and only indicative at plot scale. And permanence cannot be observed inside a 2021-2025 window, so the layer infers land use rather than measuring it.

Methodological scope

What the flag measures. Permanent forest conversion inside the analysis buffer around a site. It is a landscape reading, computed on the ground around the site, not on a plot the site owns or sources from.

What the flag does not measure. It is not an EUDR indicator, and nothing read on it establishes EUDR compliance. Three separate objects answer three separate questions:

Read the EUDR flag on the Sourcing screen for what the regulation covers. Read neither flag as conformity: scope is not compliance, and a site carrying no conversion flag is not a site carrying no EUDR obligation.

Why the scopes differ. The layer keeps conversion classes the regulation leaves out: clearing for mines, roads, settlements or industrial sites is permanent conversion, where the regulation is a market rule on seven commodities rather than a general forest-protection instrument. Its 50% canopy threshold is stricter than the regulation's 10% forest definition, which cuts the other way. And the regulation is commodity-aware and plot-level, where this layer reads a landscape buffer around a site.

Class thresholds

The underlying layer models a physical quantity: the fraction of a cell's area, from 0 to 1, that was permanently converted between 2021 and 2025. It is a continuous fraction rather than a set of discrete classes, and the layer is rendered as a linear colour ramp over the full range:

A site is flagged on the Ecosystem use dimension when converted cells cover at least 0.05% of its analysis buffer. A cell counts towards that surface once its converted fraction reaches 0.25, the physical quantum of the source: a 60 m cell aggregates four 30 m Hansen pixels, so 0.25 means "at least one Hansen pixel converted".

How these thresholds were set. This is a Darwin in-house classification. The quantity itself comes from the Hansen / UMD / Global Forest Watch record and the WRI driver classification, but the flag rule is a Darwin choice. It is expressed as a share of the buffer rather than as an absolute area on purpose: buffers run from 5 km for an office to 50 km for a mine, and buffer area grows with the square of the radius, so a fixed area would assert 0.25% of the ground around one and 0.0025% around the other. The 0.05% value is a judgement call and not yet a calibration; it is deliberately close to the fixed 20 ha gate it replaced at the 10 km radius most industrial site types use.

Source

Hansen, M. C., et al. (2013), "High-Resolution Global Maps of 21st-Century Forest Cover Change," Science 342: 850-53. Data version GFC-2025-v1.13, accessed through Global Forest Watch. Licensed CC BY 4.0.

Dominant driver: WRI & Google DeepMind, Global Drivers of Forest Loss at 1 km, v1.3 (2001-2025). Licensed CC BY 4.0.

Comparison with the WWF Risk Filter Suite

This layer maps to WWF Biodiversity Risk Filter indicator S13_2 Forest Canopy Loss, and also contributes to S13_1 Land-Use Change. WWF draws on the same Hansen / Global Forest Watch family. The difference is the driver filter: WWF's indicator counts all canopy loss, where this layer keeps only the loss that a permanent change of land use caused.

Risk analysis

A site is flagged on a dimension by combining a proximity trigger (this layer) with an activity trigger (the entity's ENCORE pressure/service). Proximity only → Potentially material; proximity and the matching ENCORE pressure/service is material → Very material; neither → Not material.

Legend

Symbolised field: Converted fraction (2021-2025)

Generated from darwin/layers/layer-permanent-forest-conversion.toml and risk_indicator_pairs.toml (develop).