Toward Scalable and Structured Global Station Weather Forecasting

2025-09-10

AI Summary

Global Site Weather Forecasting (GSWF) faces spatiotemporal modeling mismatch: existing approaches often neglect spatial correlations or model them unidirectionally, violating the intrinsic coupling nature of atmospheric systems. To address this, we propose a scalable multi-scale spatiotemporal forecasting model centered on a hierarchical subgraph-structured attention mechanism. This mechanism jointly captures local proximity and global long-range dependencies via dual-path attention—namely, intra-subgraph and inter-subgraph attention—while supporting flexible multi-granularity spatial partitioning and efficient deployment. The model is end-to-end trainable and achieves an average 16.8% improvement in prediction accuracy over state-of-the-art time-series models across multiple benchmarks. Moreover, it significantly reduces inference overhead. Key contributions include: (i) a novel hierarchical subgraph attention architecture that respects physical coupling constraints; (ii) scalable multi-scale spatial representation learning; and (iii) empirical validation demonstrating both superior accuracy and computational efficiency.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Graph-based Machine Learning

Application Category

• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated 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: Efficiency and scalability of Web search engines

Abstract

Global Station Weather Forecasting (GSWF) is a key meteorological research area, critical to energy, aviation, and agriculture. Existing time series forecasting methods often ignore or unidirectionally model spatial correlation when conducting large-scale global station forecasting. This contradicts the intrinsic nature underlying observations of the global weather system, limiting forecast performance. To address this, we propose a novel Spatial Structured Attention Block in this paper. It partitions the spatial graph into a set of subgraphs and instantiates Intra-subgraph Attention to learn local spatial correlation within each subgraph, and aggregates nodes into subgraph representations for message passing among the subgraphs via Inter-subgraph Attention -- considering both spatial proximity and global correlation. Building on this block, we develop a multiscale spatiotemporal forecasting model by progressively expanding subgraph scales. The resulting model is both scalable and able to produce structured spatial correlation, and meanwhile, it is easy to implement. The experimental results show that it can achieve performance improvements up to 16.8% over time series forecasting baselines at low running costs.

Problem

• Modeling spatial correlation in global weather forecasting systems
• Addressing limitations of existing unidirectional spatial modeling approaches
• Developing scalable structured transformers for multi-station weather prediction

Innovation

• Spatial Structured Attention Block partitioning spatial graphs
• Intra-subgraph and Inter-subgraph Attention learning correlations
• Multiscale model progressively expanding subgraph scales