XAI4Extremes: An interpretable machine learning framework for understanding extreme-weather precursors under climate change

2025-03-11

AI Summary

This study addresses the challenge of identifying physically interpretable precursors of extreme weather events—specifically, heatwaves over the Indochina Peninsula—and characterizing their evolution under climate change. We propose an explainable machine learning framework that integrates deep neural networks with Layer-wise Relevance Propagation (LRP), trained on ERA5 reanalysis data. Leveraging spatiotemporal sliding windows and multi-horizon sliding time-binning feature engineering, the framework generates meteorological relevance maps to enable both physical interpretability of precursory signals and multi-temporal climate attribution. Our key contribution is the first quantitative identification of a synergistic precursor pattern: anomalous westward extension of the subtropical high coupled with enhanced convective inhibition 7–15 days prior to heatwave onset. This finding both corroborates classical dynamical theories and reveals previously unrecognized mechanisms, thereby bridging data-driven discovery with climate dynamical understanding. The approach establishes a novel paradigm for subseasonal extreme-event prediction and process-level climate attribution.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Extreme weather events are increasing in frequency and intensity due to climate change. This, in turn, is exacting a significant toll in communities worldwide. While prediction skills are increasing with advances in numerical weather prediction and artificial intelligence tools, extreme weather still present challenges. More specifically, identifying the precursors of such extreme weather events and how these precursors may evolve under climate change remain unclear. In this paper, we propose to use post-hoc interpretability methods to construct relevance weather maps that show the key extreme-weather precursors identified by deep learning models. We then compare this machine view with existing domain knowledge to understand whether deep learning models identified patterns in data that may enrich our understanding of extreme-weather precursors. We finally bin these relevant maps into different multi-year time periods to understand the role that climate change is having on these precursors. The experiments are carried out on Indochina heatwaves, but the methodology can be readily extended to other extreme weather events worldwide.

Problem

• Identifying precursors of extreme weather events under climate change.
• Using interpretable machine learning to analyze extreme-weather precursors.
• Assessing climate change impact on extreme-weather precursors over time.

Innovation

• Post-hoc interpretability methods for relevance maps
• Deep learning models identify extreme-weather precursors
• Binning maps to analyze climate change impact