A new pan-European wildfire-probability dataset developed by Alpha-Klima in partnership with IBM Research. Annual probabilities at 2.5 × 2.5 km resolution across Europe, covering historical 2001–2023 and projections to 2050 under RCP 4.5 and RCP 8.5.
Wildfires are becoming an increasingly material source of physical climate risk for financial institutions and public agencies. To support hotspot identification, resilience planning, and climate-risk analysis, Alpha-Klima (in partnership with IBM Research) released a new probabilistic, pan-European wildfire occurrence map.
The dataset provides annual probabilities that at least one wildfire occurs in each grid cell across Europe, at 2.5 km × 2.5 km resolution, for both historical and forward-looking climate scenarios.
At a glance
- Coverage: 36 European countries
- Resolution: 2.5 × 2.5 km
- Time span: historical 2001 to 2022, projections 2023 to 2050
- Scenarios: RCP 4.5 and RCP 8.5
- Delivery: provided on request under a CC BY-NC license
What the dataset provides
This dataset delivers annual probabilities of wildfire occurrence per grid cell across Europe. It is suitable for screening and prioritisation workflows where decision-makers need a consistent view across geographies, portfolios, or administrative regions.
You can use it to support discussions around exposure, concentration of risk, and forward-looking changes under different climate pathways.
Inputs
The modelling pipeline combines climate indicators, observed fire history, and anthropogenic features.
- Copernicus Fire Weather Index v1.0
- MODIS Burned Area
- HARCI-EU infrastructure layers
Method
The probability model is built using XGBoost with physical monotonic constraints. The pipeline is fully reproducible and aligned with IBM Research’s CLAIMED framework, supporting transparent experimentation and repeatable results.
Preview and access
An interactive preview map is available on the Alpha-Klima platform. The full dataset is provided on request under a CC BY-NC license.
Citation
For reference and citation purposes, the dataset is registered on Zenodo under the following DOI:
https://zenodo.org/records/15123977
