Saltar al contenido
Alpha-Klima
  • Home
  • About Us
  • Physical Risk
  • Transition Risk
  • Regulatory Compliance
  • Platform
  • Contact
  • Events
Alpha-Klima
  • Home
  • About Us
  • Physical Risk
  • Transition Risk
  • Regulatory Compliance
  • Platform
  • Contact
  • Events
Anterior

Alpha-Klima releases its first dataset of EU Fire Probability

Alpha-Klima Wildfire Risk v1.0

– Date: 28 March 2025

– Version: 1.0

– Status: Public release to OS-Climate

– Changes: Initial publication of high-resolution wildfire risk projection dataset for Europe (2001–2050; Scenarios: Historical, RCP4.5, and RCP8.5)


1. Introduction & Alignment with Previous Research

Alpha-Klima Wildfire Risk v1.0 provides high-resolution annual wildfire occurrence probabilities at 2.5 km × 2.5 km (EPSG:3035) for Europe from 2001 through 2050. It leverages a constrained XGBoost classification framework informed by Burger et al. (2024), incorporating:

  • Meteorological Indicators: Means and Maximum of the Fire Weather Index (FWI) from Copernicus.
  • Land Surface Predictors: Annual MODIS land cover data.
  • Anthropogenic Features: Road, railway, and urban-area proximity from harmonized datasets and OpenStreetMap sources.

Key Similarities to Burger et al. (2024)

This dataset has been developed following the guidelines of:

  1. Single-step wildfire occurrence modeling rather than multi-stage burnt-area/duration models.
  2. Constrained gradient boosting for more explainable results (i.e., monotonic relationship with critical predictors like FWI).
  3. 2.5 km resolution covering Europe.
  4. Historical calibration (2001–2022) plus forward-looking scenario analysis (2023–2050).
  5. Explainability emphasis: Partial dependence plots demonstrate how changes in FWI, cropland share, and distance to roads affect fire probability.

New in Alpha-Klima v1.0

  • Additional Urban Data: Incorporating urban extent from OpenStreetMap to refine ignition-risk factors near population centers.
  • Exclusion of some countries: The HARCI-EU dataset does not include Turkey, so Turkey is excluded from the analysis.

2. Methods

2.1 Model Inputs & Preprocessing

  • Meteorological Drivers (FWI):
    – Primary input from the Copernicus Climate Data Store.
    – Historical (2001–2006) plus RCP4.5 and RCP8.5 climate-projection FWI up to 2050.
    – Mean and maximum FWI.
  • Land Cover:
    – Annual MODIS Land Cover (MCD12Q1), 500 m upscaled to 2.5 km.
    – Forest, cropland, grassland, and urban categories. Fractional cover computed using histogram aggregation.
  • Anthropogenic & Socioeconomic:
    – Distance to nearest road, railway, and urban area.
    – Derived from HARCI-EU plus open-access geospatial sources.
    – Monotonic constraints reflect that closer proximity generally increases ignition risk.
  • Historical Fire Observations:
    – NASA MCD64A1 burned-area data used to generate yearly binary “fire occurrence” labels (2001–2023).
  • Grid Harmonization:
    – All inputs aligned to a common 2.5 km × 2.5 km resolution (crs EPSG:3035).
    – Missing values imputed by nearest-neighbor filling.

2.2 Modeling Approach

  1. Single-Step Wildfire Occurrence: The target is a 0/1 label indicating at least one wildfire in a grid cell per year.
  2. Gradient Boosting with Monotonic Constraints: – Implemented via XGBoost with constraints to ensure physically/intuitively consistent relationships (e.g., higher FWI cannot lower predicted fire probability).

2.3 Bias Correction & Scaling

  • Static Inputs beyond 2023: Land cover and infrastructure remain unchanged for future years.
  • Interpolation & Resampling:
    – Bilinear interpolation for continuous variables (e.g., FWI).
    – Majority or nearest-neighbor filters for categorical land cover.

3. Scenario Outputs & Caveats

Historical (2001–2023):
Baseline rates derived from NASA MCD64A1.
Model evaluated extensively to match observed spatiotemporal patterns.

Forward-Looking (2024–2050):
RCP4.5 & RCP8.5 FWI projections from Copernicus.
Results indicate overall increasing fire probability in southern and parts of central Europe, with localized decreases in more humid regions.

Known Limitations:

  1. Exclusion of Turkey: HARCI-EU dataset does not provide coverage for Turkey.
  2. Static Land Cover: Real-world changes (e.g., afforestation) are not captured for scenario analysis.
  3. Infrastructure Assumption: Road/urban footprints remain static for 50 years.
  4. Scenario Scope: Only RCP4.5 and RCP8.5 are included.

4. Datasets Used

  1. Fire Weather Index (FWI)
    – Source: Copernicus Climate Data Store (C3S).
    – Usage: Maximum and mean yearly data used.
    – CDS
  2. Burned Area History
    – Source: NASA FIRMS (MODIS MCD64A1).
    – Usage: Binary labeling of fire occurrence per grid cell per year.
    – NASA FIRMS
  3. Land Cover Classification
    – Source: MODIS MCD12Q1.
    – Usage: Annual land cover categories for vegetation-fuel modeling.
    – MODIS Land Cover
  4. Critical Infrastructure
    – Source: HARCI-EU plus public open datasets (OpenStreetMap).
    – Usage: Road, railway, and urban proximity as ignition-risk proxies.
    – HARCI-EU

5. Licensing & Attribution

License: Published under CC-BY 4.0
Citation:
Alpha-Klima (2025). Wildfire Risk for Europe under RCP4.5 and RCP8.5: 2001–2050 (v1.0). DOI:10.5281/zenodo.15123977


6. Contact

Alpha-Klima Team

Calle José Bardasano Baos 9, 7ºAB, 28016 Madrid, Spain

mailling@alpha-klima.com

csanmillan@arfimaconsulting.com

Por Carlos San Millán|2025-04-02T11:46:53+02:001 de abril de 2025|Hazards, Wildfire|Sin comentarios

Share This Story, Choose Your Platform!

FacebookXRedditLinkedInWhatsAppTelegramTumblrPinterestVkXingCorreo electrónico

Sobre el Autor: Carlos San Millán

Deja tu comentario Cancelar la respuesta

C/ José Bardasano Baos, 9
208016 Madrid
+34 91 1292874
info@afs-services.com

  • Home
  • About Us
  • Physical Risk
  • Transition Risk
  • Regulatory Compliance
  • Platform
  • Contact
  • Events

Subscribirme

  • Privacy Policy
  • Cookie Policy

Proyecto CPP2021-008644, financiado por MCIN/AEI/10.13039/501100011033 y por la Unión Europea-NextGenerationEU/PRTR.

© All rights reserved • Alpha-Klima

Page load link
Ir a Arriba