Annual forest cover conditions across the Southwestern Amazon, 2003-2021

A Landsat-based machine learning algorithm (Reygadas et al. 2021, Environmental Research Communications) adapted from Wang et al. (2019, Remote Sensing of Environment) to the Southwestern Amazon was used to map intact forest, degradation, and deforestation in this region on a yearly basis during the 2003-2021 period. Degradation is defined as a long-term process in which forest is negatively affected but it is not converted into another land-cover. In contrast, deforestation is defined as the permanent, or long-term, conversion of forest into non-forest. The algorithm classifies forest covers by training a random forest model with sixty-six metrics derived from six time series variables (i.e., the Normalized Difference Vegetation Index, two shortwave infrared bands, two Normalized Difference Water Indices, and the Soil-Adjusted Vegetation Index) from which eleven descriptive statistics are calculated. As the algorithm uses statistical characteristics of time series to determine the forest conditions in the end of the study period, time series composed of the last 20 years prior to the target year were used in each annual run. A forest mask, composed of all areas covered by forest at least three consecutive years and never covered by water during the 2000-2018 period, was applied to all maps. A data key is included in the description of each file. Note: Although the same algorithm is used in Reygadas et al. (2021), these data differ from those of the manuscript as they are annual and cover a larger area.