Physics-informed GNNs improve extreme rainfall forecasts in Thailand

Researchers have developed a novel physics-informed Graph Neural Network (GNN) model combined with extreme-value analysis to enhance long-range extreme rainfall forecasting in Thailand. The model utilizes a graph-structured representation of weather stations and incorporates teleconnections, which are climate indices that influence regional rainfall. This approach, employing an Attention-LSTM architecture and a Spatial Season-aware Generalized Pareto Distribution method for extreme events, demonstrated superior performance compared to existing baselines and the operational SEAS5 system, offering practical improvements for water management decisions. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel GNN approach for extreme weather prediction, potentially improving climate modeling and water resource management.

RANK_REASON This is a research paper published on arXiv detailing a new methodology for rainfall forecasting.

Read on arXiv cs.LG →

• Thailand
• Graph Neural Networks
• Attention-LSTM
• SEAS5
• arXiv

• paper
• other