[ivory-search id="3463" title="Default Search Form"]
Search
Register here
Search
Type below to search:
OnlineDiscovery - AI for Earth and Sustainability Science Badge available

Fight fire with fire

  • Register

     

    * Register (or log in) to the Neural Network to add this session to your agenda or watch the replay

    • Date
    2 April 2026
    Timeframe
    17:00 - 18:00 CEST
    Duration
    60 minutes
    • Days
      Hours
      Min
      Sec

    The Fire Weather Index is a semi-empirical, process-based model that integrates temperature, humidity, wind speed, and precipitation to estimate potential fire intensity and spread. While widely used in operational forecasting, it is purely weather-driven and limited by simplified processes and limited transferability. But anticipating wildfires is a very difficult problem, so we need more sophisticated methods.

    Wildfires arise from highly non-linear interactions across the atmosphere, biosphere, hydrosphere, and human systems. Deep learning offers a promising alternative by capturing such complex dependencies in a data-driven manner, and this talk will present recent approaches for wildfire forecasting using modern AI methods.

    Key challenges remain: (i) scarcity of positive wildfire labels for training large models, (ii) absence of observed fires does not imply low fire danger, and (iii) the inherently stochastic nature of fire occurrence, which limits deterministic modeling. Addressing these requires techniques such as self-supervised learning, positive-unlabeled learning, and Bayesian deep learning for uncertainty quantification.

    Finally, wildfire dynamics span multiple spatial and temporal scales, from short-term weather variability to longer-term fuel, hydrological, and climate controls, necessitating models that can integrate multi-scale processes.

     

    Session Objectives:

    By the end of this session, participants will be able to:

    • Understand the problem of wildfire forecasting, including how it is defined across different spatial and temporal scales.
    • Recognize the main challenges of wildfire prediction, such as data limitations, stochasticity, and observational biases.
    • Appreciate what data-driven approaches can and cannot achieve in this domain, including their key limitations.
    • Identify where there is room for improvement in current wildfire forecasting methods, particularly through AI.
    • Design workflows integrating AI emulation with downscaling and coupled ocean models for research applications.

     

    Recommended Mastery Level/Prerequisites:

    Intermediate level.

    • Basic understanding of machine learning / deep learning concepts.
    • Familiarity with Earth Observation or geospatial data.
    • General awareness of climate or environmental systems.
    Share this session with your network
    • Session starts in
      Days
      Hours
      Min
      Sec
    Add to Calendar
    Discover more from programme stream
    Discovery - AI for Earth and Sustainability Science
    Fight fire with fire

    Fight fire with fire

    2 April 2026 - 17:00 to 18:00

    Related stories

    Are you sure you want to remove this speaker?