STCFormer: Adaptive Spatio-Temporal Modeling with Dynamic Cluster Transformer for Station-based Weather Forecasting
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.
Related
- RATSIFT: Resolution-Adaptive Transformer for Solar Irradiance Forecasting and Temporal Downscaling
- UniHAR: A unified hierarchical transformer with multi-scale temporal modeling for wearable human activity recognition
- Adaptive Information Coordination for Context-Aware International Event Forecasting with Temporal Knowledge Graphs
- Short-term metro passenger flow forecasting via decoupled spatio-temporal modeling with a diffusion-convergence graph and Swin-Enhanced LSTM
- Hopformer: Homogeneity-Pursuit Transformer for Time Series Forecasting
BibTeX
@misc{li2026stcformer,
title = {{STCFormer: Adaptive Spatio-Temporal Modeling with Dynamic Cluster Transformer for Station-based Weather Forecasting}},
author = {Rongwen Li and Haixin Xie and Mingyang Wang and Hongwu Liu and Kun Fang and Changjian Chen and Zhuo Tang and Kenli Li},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.00377},
eprint = {2610.00377},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.00377},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - STCFormer: Adaptive Spatio-Temporal Modeling with Dynamic Cluster Transformer for Station-based Weather Forecasting
AU - Rongwen Li
AU - Haixin Xie
AU - Mingyang Wang
AU - Hongwu Liu
AU - Kun Fang
AU - Changjian Chen
AU - Zhuo Tang
AU - Kenli Li
PY - 2026
DA - 2026/10/01
PB - arXiv
DO - 10.48550/arxiv.2610.00377
UR - https://arxiv.org/abs/2610.00377
AB - 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.
ER -