Abstract Europe’s forests face increasing threats from natural disturbances such as insect outbreaks, pathogens, and windthrow, often aggravated by extreme weather events and followed by subsequent salvage logging. Monitoring these events at high spatial detail is essential for forest management and climate adaptation, yet many remain undetected when using medium-resolution satellite imagery, and manual reporting by authorities is time-consuming and inconsistent. Here we present a high-resolution, deep learning-ready dataset designed for the classification of forest disturbance types. It consists of ~17,500 image patches (500 × 500 pixels at 0.2 m resolution) derived from digital orthophotos of Rhineland-Palatinate, Germany. Each patch includes five channels (red, green, blue, near-infrared, and object height) and a segmentation mask with labeled disturbance classes such as bark beetle damage, clear-cuts, and windthrow. To demonstrate its utility, we apply a deep learning model and assess the contribution of individual channels through ablation analysis. The model achieved an overall accuracy of 88.2%, with near-infrared and object height identified as the most informative channels. The dataset offers a high-resolution resource for advancing deep learning-based forest disturbance monitoring.
Rodríguez-Paulino et al. (Thu,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: