Deep learning-based forest disturbance detection for Europe using Landsat time series

Forests in Europe are under growing pressure from climate extremes and land use intensification. Long-term monitoring is therefore required to understand disturbance dynamics, yet producing temporally and spatially consistent disturbance maps at continental scales remains challenging. Emerging deep learning models might improve upon existing models by learning spectral-temporal patterns relevant for disturbance detection directly from the underlying data, while simultaneously ignoring noise. Despite their promise, deep learning models have not yet been used for continental-scale disturbance mapping. Here, we implemented two temporal deep learning models (TempCNN and 1D U-Net) for annual forest disturbance detection across all forests of continental Europe using Landsat time series from 1985 to 2024. We evaluated the performance of both models for detecting forest disturbances using short temporal subsequences from the whole time series and compared them to existing approaches based on Random Forest. The 1D U-Net outperformed the TempCNN and Random Forest models across variable classification scenarios. The highest performance was achieved using the 5-year input window. Using the 1D U-Net model, we generated annual forest disturbance maps at 30 m resolution for Europe since 1985, with a total disturbed forest area of 48.5 Mha. Map validation using an independent validation dataset yielded an F1 score of 0.81 in spatial disturbance detection, with balanced commission and omission errors (18.4% and 19.8%, respectively), and less variability in commission and omission errors over time than previous maps and alternative models. Using a temporal convolutional architecture thus led to more consistent disturbance mapping across space and time compared to previous approaches. The proposed model is further designed to support updates as new observations arrive and to operate effectively with short temporal sequences, facilitating operational forest disturbance monitoring across Europe.