deadtrees.earth Maps: Tree Mortality and Disturbance Mapping from Sentinel-2 Timeseries Across The Globe

Forest disturbances and excess tree mortality are increasingly reported worldwide, yet satellite monitoring is still often limited to coarse-resolution, binary forest loss products that miss fine-scale mortality where only a few trees decline within an otherwise intact canopy. This limits our ability to quantify emerging disturbance dynamics, compare regions with consistent metrics, and identify early signals of larger forest change.

We present yearly, wall-to-wall maps (2018–2025) of fractional forest cover and fractional standing deadwood cover at 10 m resolution for Europe and beyond. We use these maps to provide a first Europe-wide quantitative overview of recent mortality patterns, summarizing the spatial distribution of elevated standing deadwood and its year-to-year dynamics from 2018 to 2025. These map products enable new analyses of disturbance dynamics at unprecedented spatial detail: tracking year-over-year mortality progression patterns; distinguishing general tree removal from trees dying standing by jointly analyzing forest and deadwood fractions; and quantifying subtle early-stage disturbance signals before they aggregate into larger forest change.

The maps link centimeter-scale aerial reference data from the crowd-sourced deadtrees.earth drone archive with multi-year Sentinel-2 reflectance time series: tree and standing-deadwood masks are derived on drone orthophotos using semantic segmentation, aggregated to sub-pixel cover fractions, and used to train a per-pixel computer vision model that translates reflectance signatures into annual forest and standing deadwood cover. As the deadtrees.earth drone archive continues to grow, its automated processing pipelines can feed regular model retraining, allowing the maps and models to be iteratively improved in space and time with each new contribution.