Tree cover loss data by country: the satellite record, delivered as rows

The Global Nature Watch (formerly Global Forest Watch) Open Data Portal from the World Resources Institute is the working answer. Its machine-readable catalog held 382 dataset entries at the August 2026 research pass, 360 of them shipping data products, and its flagship umd_tree_cover_loss layer - version 1.13 in the current release - exposes 33 raster and cloud-optimized GeoTIFF assets covering global land area at 30 m resolution.

You do not have to sum pixels yourself. The same platform aggregates that raster into summary tables keyed to ISO country codes, admin-1 divisions and admin-2 divisions, so every land area outside Antarctica and other Arctic islands carries an annual loss series. Country summary tables run on the order of tens of thousands of rows each.

In Datadory's catalog the dataset scores 8/10 against a 7.81 average across 1,744 datasets, inside a 20-record forest-products slice that is otherwise government statistics - production, trade, prices, inventories. This is the layer that observes the trees themselves, pixel by pixel, and it sits beside four alert families - GLAD, integrated, MODIS and VIIRS - whose 64 detection tables run at country, admin-1 and admin-2 grain in daily and weekly windows.

Three time windows matter, and they are not interchangeable:

1. Tree cover loss, 2001-2025 - umd_tree_cover_loss v1.13 is annual and current through 2025.
2. Tree cover gain, 2000-2012 - a single-period companion layer; there is no year-by-year gain series, so annual gain cannot be netted against annual loss.
3. Carbon flux, 2001-2024 - pairs each year of loss with estimated greenhouse-gas emissions.

A fourth layer fills the baseline: tropical tree cover mapped at 10 m for the 2020 baseline year, covering the tropics rather than the whole globe. Alert systems then sit on top - GLAD, integrated, MODIS and VIIRS publish near-real-time detections through daily and weekly alert tables, so an active clearance shows up in alerts months before the next annual raster release.

For years before 2001 there is no satellite-equivalent series in this slice. FAO reporting reaches back to 1990, but those are country-reported benchmarks rather than pixel-derived measurements.

Each summary row binds a location key to a year of loss. At the top level the key is the ISO code; drill down and the same structure repeats for admin-1 (states, provinces) and admin-2 (districts, regencies), which means a query written for Brazil runs unchanged for a single Indonesian regency. The underlying raster resolves individual 30-meter pixels, and the tabular summaries are simply those pixels rolled up to each geography.

The same underlying events read two ways, and the choice is analytical rather than mechanical: tabular summaries keyed on ISO and admin codes join straight into trade, supplier and portfolio tables, while the 30 m raster answers custom geometries - a concession polygon, a mill supply shed, a watershed - where pre-aggregated boundaries stop short.

Two conditioning fields ride on nearly every row, and both change the number you get. The canopy-density threshold - tcd0 through tcd75 - decides what counts as a tree in the first place, and totals move materially between tcd10 and tcd50. The dominant driver class separates commodity-driven deforestation, shifting agriculture, forestry, wildfire and urbanization, which is the difference between a supply-chain incident and a lightning strike in your escalation queue. Alongside them sit planted-forest type, an Intact Forest Landscapes flag and a year-2000 biomass-and-carbon balance sheet carried on the same keys.

Three artifacts from the August 2026 research pass, exactly as cataloged:

# dataset: umd_tree_cover_loss
subtitle    : Tree cover loss - annual, 30 m, global, UMD GLAD
version     : v1.13        latest: yes        updated_on: 2026-08-04
assets      : 33           types: Raster tile set - COG
band        : year         values: 1=2001 ... 25=2025
thresholds  : tcd0 - tcd10 - tcd15 - tcd20 - tcd25 - tcd30 - tcd50 - tcd75

# aggregation table: gadm__tcl__iso_summary
grain       : country (ISO 3166-1 alpha-3)
version     : v20260407
columns     : 36

# lineage recorded on the flagship layer
citation    : Hansen et al., 2013. High-Resolution Global Maps of
              21st-Century Forest Cover Change. Science 342:850-53

Read the band line twice, because it is doing the heavy lifting: one coded year per pixel, code 1 for 2001 through code 25 for 2025, which is also why history moves - each annual re-release can revise earlier epochs, and v1.13 is simply the current statement. Pin the version beside any trend you publish, because a slope computed on one release will not reproduce byte-for-byte on the next.

The lineage matters just as much. The layer descends from Hansen et al. (2013), the most-cited single map in forest-change work, so a loss figure traced to it arrives in an ESG disclosure with provenance attached rather than asserted.

Satellite loss and reported forest change answer different questions, and ESG reviewers increasingly ask for both. The FAO Global Forest Resources Assessment (FRA) Data Platform publishes country reports for 236 countries and areas across nine themes - forest extent, growing stock, biomass and carbon, ownership, disturbances and more - at reference years 1990, 2000, 2010, 2015, 2020 and 2025. Each of the roughly 18 reporting tables per country resolves to benchmark values with the unit stated - thousand hectares for area, million cubic metres over bark for growing stock - so one country joins cleanly to the next.

Granularity is the difference. FRA values are national-level benchmark figures per indicator table - not gridded, not subnational outside the narrative country reports - while the UMD layer works at 30 m anywhere on land. Use FRA for the denominator ("this province's loss equals X percent of national forest extent") and national correspondents' own definitions, and the satellite series for speed and subnational drill-down.

Trade closes the loop. FAOSTAT Forestry tracks production, imports and exports for 104 forest product items across 285 area codes from 1961 to 2024 in roughly 2.46 million rows per release, letting you test whether a loss signal coincides with rising roundwood or sawnwood shipments. For European detail, Eurostat's FOR_BASIC cube serves 144,153 observations of roundwood, chips, pellets and recovered wood from 1988 to 2024.

A country-level spike means more once you know who operates in the landscape. FSC Connect, the Forest Stewardship Council's public registry (dataset profile), lists forest management and chain-of-custody certificates across five record families with holders in 165 countries, searchable by organization, certificate code, country, product category or species. It refreshes continuously - research pulls recorded a platform lastUpdate timestamp of August 21, 2026 - though certification bodies rather than FSC keep entries current.

National inventories give the official counter-narrative. India's State of Forest Report (dataset profile) has run biennially since 1991 - about fifteen editions through ISFR 2023 - assessing forest cover, growing stock and carbon state by state from an NFI sample base of roughly 12,000 forest plots and 20,000 trees-outside-forests plots per cycle. Natural Resources Canada (dataset profile) maintains roughly 22 indicators across ten categories, including harvest and disturbance series from 2000 onward, for Canada plus every province and territory.

Where a jurisdiction publishes its own harvest registry, polygons settle disputes that annual rasters cannot: British Columbia's forestry catalogue (dataset profile) holds about 580 of the province's roughly 3,300 records matching a 'forest' search, including consolidated cutblock polygons reaching back decades, each attributed to its custodian ministry branch.

Ranking the checks keeps the screen auditable. A practical order:

1. Country screen - rank every sourcing country by umd_tree_cover_loss totals, 2001-2025, and flag any origin above your risk threshold.
2. Admin-1 and admin-2 drill-down - narrow flagged countries to the provinces or districts where your suppliers actually operate.
3. Alert watch - put flagged jurisdictions on GLAD or integrated alert watch so new clearance surfaces within days.
4. Certification lookup - search FSC Connect for chain-of-custody holders among affected suppliers across its 165-country footprint.
5. Official reconciliation - compare against FRA country-reported figures and national inventories before escalating, since plantation harvest and fire register as loss too.

Every step runs on rows in one typed shape - location keys, years, hectares, driver classes - so the screen assembles in a single schema instead of five, and each check carries the layer version and canopy threshold it was computed on.