Graph Neural Networks for Enhancing Ensemble Forecasts of Extreme Rainfall

2025-04-07

AI Summary

Climate change intensifies extreme precipitation, yet ensemble forecasts exhibit substantial biases in representing its spatial dependence and tail behavior. To address this, we propose the first graph neural network (GNN)-based post-processing framework tailored for extreme precipitation: meteorological graphs are constructed using geographic proximity and watershed topology; extreme-value distributions are explicitly modeled; and an extreme-value-theory-driven loss function is designed to improve tail calibration. This approach overcomes key limitations of conventional statistical post-processing—namely, its inability to adequately capture precipitation’s non-Gaussianity and spatial heterogeneity. Evaluated across multiple flood-prone regions, our method improves the threat score (TS) for extreme rainfall by 18.7% within 5–48-hour lead times, reduces false alarm rate by 23%, and significantly enhances the reliability of probabilistic heavy-precipitation forecasts and associated disaster early-warning capabilities.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Graphical Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Climate change is increasing the occurrence of extreme precipitation events, threatening infrastructure, agriculture, and public safety. Ensemble prediction systems provide probabilistic forecasts but exhibit biases and difficulties in capturing extreme weather. While post-processing techniques aim to enhance forecast accuracy, they rarely focus on precipitation, which exhibits complex spatial dependencies and tail behavior. Our novel framework leverages graph neural networks to post-process ensemble forecasts, specifically modeling the extremes of the underlying distribution. This allows to capture spatial dependencies and improves forecast accuracy for extreme events, thus leading to more reliable forecasts and mitigating risks of extreme precipitation and flooding.

Problem

• Improving extreme rainfall forecasts using graph neural networks
• Addressing biases in ensemble prediction systems for extreme weather
• Modeling spatial dependencies to enhance precipitation forecast accuracy

Innovation

• Graph neural networks post-process ensemble forecasts
• Model extremes of precipitation distribution accurately
• Capture spatial dependencies for reliable forecasts