What this research found
Annual tree-cover loss across a 10-degree by 10-degree slice of the western Amazon Basin, spanning parts of Peru, Colombia, Ecuador and western Brazil, was quantified from the Hansen Global Forest Change record for 2001 through 2023. Rather than assigning every pixel the nominal 30-metre footprint, the exact ground area of each cell in a 1.6-billion-pixel grid was computed on the WGS84 ellipsoid, removing a systematic bias of roughly 16%. The tile lost 6,071,069 hectares of forest over the 23 years, 5.16% of its year-2000 baseline, with 2016 the worst single year at 520,938 hectares.
- Gross tree-cover loss across the tile totalled 6,071,069 hectares between 2001 and 2023, equal to 5.16% of a year-2000 forest baseline of 117,587,905 hectares, and averaging about 263,959 hectares a year.
- 2016 was the worst single year at 520,938 hectares, or 0.443% of the baseline — roughly double the long-term mean. That timing coincides with the exceptional 2015 to 2016 El Niño drought, during which escaped land-clearing fires and drought-stressed canopies amplify the extent of detectable stand-replacing loss.
- Loss has intensified rather than spiked once. The first decade of the record averaged about 211,073 hectares a year against about 344,950 for 2014 to 2023, a 63% increase, and the five most recent years averaged about 359,394. 2022 at 470,151.97 hectares and 2021 at 415,619.51 hectares rank second and third across the whole series.
- The earlier record was comparatively quiet apart from a 2005 spike of 398,328.55 hectares that lines up with that year's severe Amazon drought. Loss fell below 150,000 hectares in each of 2007, 2008 and 2009.
- Pixel-area handling changes the answer materially. Treating each 0.00025-degree cell as the commonly quoted nominal 0.09 hectares would overstate area by roughly 16% against its true equatorial footprint of 0.076932 hectares. The exact ellipsoidal calculation agreed with an independent equal-area projection to 0.000174% and with a geodesic reference to machine precision.
How it was done
Two co-registered Landsat-derived layers for the tile — year-2000 percent canopy cover and year of loss — were pulled from the University of Maryland repository and checked against the published grid specification, with checksums and byte counts confirmed. Forest was defined as any pixel carrying at least 30% canopy cover in 2000, which covered roughly 96% of the tile. Because the grid is fixed in angular units, the exact ground area of every cell was computed with a closed-form latitudinal-zone integral on the WGS84 ellipsoid, tabulated once per raster row, and used to area-weight each loss pixel. The two grids were never loaded whole: they were streamed in windows of 1,000 rows and accumulated by loss year in a single pass, finishing in about 38 seconds under half a gigabyte of memory. The area model was then validated against the South America Albers Equal-Area Conic projection and against a geodesic ellipsoidal calculation.
Data sources
- Hansen Global Forest Change GFC-2023 v1.11, granule 00N_070W — year-2000 canopy cover and year-of-loss layers, 40,000 by 40,000 pixels at 0.00025-degree resolution
- Hansen et al., Science 342:850 (2013) — the original 30-metre global maps of twenty-first-century forest cover change
- WGS84 reference ellipsoid constants and the South America Albers Equal-Area Conic coordinate system, used for the area model and its validation
- Karney, Journal of Geodesy 87:43 (2013) — geodesic algorithms used as an independent area check
Limitations
These are gross loss figures rather than net change: the underlying product maps gain only as a single multi-year layer, so regrowth is not subtracted anywhere. Tree cover is also a biophysical measure that does not separate natural forest from plantations or tree crops, both baseline and loss shift with the chosen canopy threshold, and loss timing in individual pixels can be uncertain by a year, which reallocates area between adjacent years while leaving the 23-year total intact.
How this research was produced
K-Dense Web planned and ran this environment investigation end to end — gathering the sources, carrying out the analysis, producing the figures, and drafting the report. The full session transcript, including every intermediate step, is available to view.
