STCFormer: Adaptive Spatio-Temporal Modeling with Dynamic Cluster Transformer for Station-based Weather Forecasting

2026-09-30

AI Summary

This study addresses the challenges of modeling complex spatial dependencies, the neglect of local dynamics by fixed groupings, and the lack of global context in site-specific meteorological forecasting. To this end, we propose an adaptive spatiotemporal Transformer. The method introduces dynamic station grouping within time slices, integrating cluster-guided attention with regional state summaries to achieve synergistic local-global modeling. Furthermore, InfoLoss is incorporated to optimize interaction efficiency, and a theoretical robustness analysis is provided via Lipschitz bounds. Extensive evaluations across 48 comparative settings—spanning eight tasks on three real-world datasets—demonstrate that our approach ranks among the top two in 47 settings and consistently achieves the lowest 24-hour MSE, thereby validating its effectiveness and superiority.

Abstract

Station-based weather forecasting supports daily life and economic activity, yet accurate forecasts require modeling complex spatial dependencies among stations. Recent clustering-based selective modeling offers a promising alternative to dense inter-station interactions. However, a grouping shared across an observation window may obscure local changes in station relationships, while intra-cluster interactions alone may miss important global context. The theoretical advantages of selective interactions over dense connectivity also remain insufficiently understood. We therefore propose STCFormer, an adaptive spatio-temporal Transformer that dynamically groups stations according to their local evolution within each temporal patch. Its Cluster-Guided Attention Block combines fine-grained local attention within clusters and global attention over regional state summaries, allowing each station to access information beyond its own cluster. We further show that a derived Lipschitz upper bound for cluster-conditioned local attention is no larger than its fully connected counterpart, explaining a potential robustness benefit and motivating the design of InfoLoss. Experiments on three real-world weather datasets spanning eight temperature and wind forecasting tasks show that STCFormer achieves the lowest 24-hour mean squared error on all eight tasks and ranks first or second in 47 of 48 comparisons across metrics and forecasting horizons. Ablations and case studies further confirm the benefits of locally adaptive grouping and complementary local-global interactions. Our code can be obtained at https://github.com/hnu-vis/STCFormer.

Problem

• Station-based Weather Forecasting
• Spatio-Temporal Modeling
• Dynamic Clustering
• Spatial Dependencies
• Selective Interaction

Innovation

• Dynamic Cluster Transformer
• Cluster-Guided Attention
• Adaptive Spatio-Temporal Modeling
• Lipschitz Upper Bound
• InfoLoss