Table 1.
General characteristics of our three case study sites.
We calculated deforested area here using the Global Forest Change dataset.
Table 2.
Validation results for our classification maps.
Fig 1.
Classified land cover maps for Amboró and Carrasco National Parks in Bolivia, Jamanxim national forest in Brazil, and Tambopata National reserve and bahuaja-Sonene National Park in Peru.
Black lines indicate park boundaries, with a 20km external buffer. Rio Novo National Park and the Ambiental Do Tapajós Protected Areas park borders Jamanxim to the west, Rio Grande Valles Crucenos borders Amboró National Park to the southeast, and Madidi National Park borders Bahuaja-Sonene to the east. Maps derived from Landsat 5 (TM), Landsat 7 (ETM+), and Landsat 8 (OLI) imagery, courtesy of the U.S. Geological Survey.
Table 3.
Detected deforestation (2008–2018) between our and Global Forest Change’s land cover maps.
Fig 2.
Sankey graphs demonstrating 2008–2018 land cover transitions.
Flow widths represent proportions of land area transitioning and colors follow the final land classification assignment. Numbers below each graph represent the percent of pixels within that land area that transitioned. If a transition frequency (i.e. water to urban) accounted for less than 1% of all transitions, we did not graph it, for visual simplicity. Land classes: F: forest, A: agriculture (and pasture), S: bare soil, We: wetland, Wa: water, D: desert, U: urban.
Fig 3.
Spatial comparison of our and Global Forest Change’s maps in the Bolivia case study.
Panel A shows the entire study area (with the protected area borders in black), panels B and C show two close-up regions (labeled 1 and 2 in panel A), and panel D quantifies types of disagreement by percent area. Grey indicates where both maps agreed there was no land cover change between 2008 and 2018. Our maps were derived from Landsat 5 (TM), Landsat 7 (ETM+), and Landsat 8 (OLI) imagery, courtesy of the U.S. Geological Survey. Global Forest Change source: Hansen/UMD/Google/USGS/NASA.
Fig 4.
Spatial comparison of our and Global Forest Change’s maps in the Brazil case study.
Panel A shows the entire study area (with the protected area border in black), panels B and C show two close-up regions (labeled 1 and 2 in panel A), and panel D quantifies types of disagreement by percent area. Grey indicates where both maps agreed there was no land cover change between 2008 and 2018. Our maps were derived from Landsat 5 (TM), Landsat 7 (ETM+), and Landsat 8 (OLI) imagery, courtesy of the U.S. Geological Survey. Global Forest Change source: Hansen/UMD/Google/USGS/NASA.
Fig 5.
Spatial comparison of our and Global Forest Change’s maps in the Peru case study.
Panel A shows the entire study area (with the protected area borders in black), panel B shows a close-up region to demonstrate details, and panel C quantifies types of disagreement by percent area. Grey indicates where both maps agreed there was no land cover change between 2008 and 2018. Our maps were derived from Landsat 5 (TM), Landsat 7 (ETM+), and Landsat 8 (OLI) imagery, courtesy of the U.S. Geological Survey. Global Forest Change source: Hansen/UMD/Google/USGS/NASA.
Fig 6.
Explanations for dissimilarity among our and Global Forest Change’s maps.
N represents the sampling points (out of 600 total) that fell into each explanation of dissimilarity (excluding desert points). We additionally show the proportion of pixels Global Forest Change marked as deforested where our maps detected no change, pixels that our maps marked as deforested where GFC detected no change, and pixels where GFC had no data for forest cover in 2000 or forest loss.