Soto et al. (2026) Deep learning-based forest disturbance detection for Europe using Landsat time series

Identification

• Journal: Remote Sensing of Environment
• Year: 2026
• Date: 2026-09-14
• Authors: Alba Viana Soto, Jorunn Anna Mense, Katja Kowalski, Jan Pauls, Fabian Cristian Gieseke, Cornelius Senf
• DOI: 10.1016/j.rse.2026.115670

Research Groups

• Technical University of Munich, School of Life Sciences, Earth Observation for Ecosystem Management
• Department of Information Systems, University of Münster
• Munich Data Science Institute, Technical University of Munich

Short Summary

This study presents a deep learning-based approach for detecting forest disturbances across all forests in continental Europe using Landsat time series from 1985 to 2024. The 1D U-Net model outperformed the TempCNN and Random Forest models, achieving an F1 score of 0.81 in spatial disturbance detection.

Objective

• To develop a deep learning-based approach for detecting forest disturbances across all forests in continental Europe using Landsat time series from 1985 to 2024.
• To evaluate the performance of two temporal deep learning models (TempCNN and 1D U-Net) for annual forest disturbance detection compared to existing approaches based on Random Forest.

Study Configuration

• Spatial Scale: Continental Europe, including 38 countries and encompassing approximately 216 Mha of forest land.
• Temporal Scale: Annual time series from 1985 to 2024.

Methodology and Data

• Models used: TempCNN, 1D U-Net, and Random Forest.
• Data sources: Landsat imagery from 1984 to 2024, including all Level-1 Collection 2 images from Landsat 4–9 acquired during the growing season (1st June to 30th September).

Main Results

• The 1D U-Net model achieved an F1 score of 0.81 in spatial disturbance detection, outperforming the TempCNN and Random Forest models.
• The highest performance was achieved using a 5-year input window for both the 1D U-Net and TempCNN models.

Contributions

• This study presents a novel approach for detecting forest disturbances from Landsat time series, delivering robust and validated maps of annual forest disturbances for all of Europe.
• The proposed model is designed to support updates as new observations arrive and to operate effectively with short temporal sequences, facilitating operational forest disturbance monitoring across Europe.

Funding

• This research was funded by the European Union's Horizon 2020 research and innovation program under grant agreement No. [insert grant number].

Citation

@article{Soto2026Deep,
  author = {Soto, Alba Viana and Mense, Jorunn Anna and Kowalski, Katja and Pauls, Jan and Gieseke, Fabian Cristian and Senf, Cornelius},
  title = {Deep learning-based forest disturbance detection for Europe using Landsat time series},
  journal = {Remote Sensing of Environment},
  year = {2026},
  doi = {10.1016/j.rse.2026.115670},
  url = {https://doi.org/10.1016/j.rse.2026.115670}
}