Sentinel-2-based Forest Disturbance Predictions
This product provides forest disturbance predictions for all of Germany based on Sentinel-2 imagery (spatial resolution: 10 m), using the Transformer model proposed by Schiller et al. (2024). Predictions are generated approximately every five days starting from June 2024 and are continuously updated.
Specifically, the product represents the model’s confidence (ranging from 0 – high confidence in an undisturbed pixel, to 1 – high confidence in a disturbed pixel) that a given pixel has experienced forest disturbance (from any disturbance agent) within the four years prior to the file’s date (inclusive). Sub-pixel disturbances can also be detected, albeit with reduced accuracy (see publication for details). A threshold of 0.5 or higher can be used to convert the output into a binary classification, depending on whether avoiding false positives is prioritized.
While the product offers full national coverage, it is important to note that the model has not been validated in all regions of Germany (e.g., no validation exists for the German Alps).
The Free University Berlin (FUB) collaborates with EO-Lab to make this product available to researchers interested in forest disturbance assessment. Potential applications include nationwide disturbance monitoring, retrospective identification of disturbance hotspots, and near real-time forest change detection.
For detailed information on the model architecture and its evaluation, please refer to the original publication: Schiller et al. (2024). A demonstration of practical usage is available via Jupyter Notebook using R.
Technical specifications:
- Spatial resolution: 10 m
- Spatial coverage: Germany
- Temporal resolution: approx. 5 days
- Temporal coverage: since June 2024 (ongoing)
Usage requirements
- Registration on EO-Lab: yes
- Cloud resources on EO-Lab: yes
- Individual request: yes (to establish access)
Data access
This data is available on request. Please create a support ticket to receive the access information.
More information
Further information on this product and underlying project can be found at:
- https://github.com/ChrSchiller/dl_forest_disturbance
- https://github.com/ChrSchiller/forest_dist_inference
- https://future-forest.eu/
References
Terms of usage
Use of the product is subject to the Creative Commons Attribution 4.0 (CC BY 4.0) license.
