Objectives and Expected Breakthroughs

in addressing the challenges in AI generalization for hydro-climatic extremes
  • Objective 1: Advance AI methodologies for predicting extremes under non-stationarity regimes

    Breakthrough 1: Predict

    Graphic: KI generated

    Breakthrough PREDICT:

    Develop AI models capable of predicting extremes in space and time, accounting for non-stationary regimes where traditional models fail.

  • Objective 2: Establish uncertainty quantification frameworks for a responsible use of AI in out-of-domain settings

    Breakthrough 2: Quantify Uncertainty

    Graphic: KI generated

    Breakthrough QUANTIFY UNCERTAINTY:

    Develop rigorous uncertainty quantification strategies for DL models to assess their transferability and robustness when applied to unseen regions, climatic conditions, or time periods.

  • Objective 3: Enhance Earth system understanding and predictability of extreme events

    Breakthrough 3: Extract Insights

    Graphic: KI generated

    Breakthrough EXTRACT INSIGHTS:

    Leverage causal modeling, hybrid modeling, and equation discovery to extract scientifically meaningful insights on the drivers and effects of hydro-climatic extremes, improving physical interpretability and predictive power.

  • Objective 4: Translate AI advances into actionable environmental assessment and management

    Breakthrough 4: Demonstrators

    Graphic: KI generated

    Breakthrough DEMONSTRATORS:

    Bridge fundamental AI research with real-world applications by developing generalizable AI demonstrators in the chosen hydro-climatic application domains related to hazards (floods, landslides) and ecosystem responses (phenological shifts, ecosystem fluxes).