Paper · 2026 · Open access

CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai

Xuanwen Pan, Yoichi Shimomura, Sichen Tao, Masatoschi Kawai, Keichi Takahashi, Hiroyuki Takizawa

Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, 177–185 · Published 22 January 2026 · DOI 10.1145/3784828.3785258

Open full textCite this paper

In brief

“CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai” is a 2026 paper by Xuanwen Pan and 5 co-authors, published in Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops. OpenAlex counts 2 citations as of 20 November 2026, which puts it in the top 6% of works published in 2026 by citations. It is open access under a CC BY licence: the full text is free to read.

Citations
2
Most cited in
2026 (2)
Field-weighted impact
20.55
References
22
Authors
6

Abstract

Accurate weather forecasting is essential for public safety, economic planning, and day-to-day decision making. Recently, machinelearning (ML) weather models have rapidly improved medium-range forecasting skills while requiring a fraction of the computational cost of traditional numerical weather prediction (NWP). However, most ML systems, such as GraphCast, have focused on global grids at quarter-degree resolution and above. This paper proposes a city-scale forecasting framework, CityScaleCast, that leverages GraphCast fields as dynamic boundary conditions for a spatiotemporal graph neural network (ST-GNN) to produce ∼ 5 km forecasts over Sendai, Japan. A regional graph is constructed for (35°–41°N, 139°–143°E) with multi-scale spatial edges and temporal links, and forecasts are rolled autoregressively while injecting global context along the domain boundary. CityScaleCast explicitly couples a regional multi-scale ST-GNN with GraphCast-derived dynamic boundary conditions and an autoregressive sliding-window rollout. An efficient training stack is described—distributed training with streaming data pipelines—to sustain high throughput on commodity multi-GPU nodes under modest memory budgets. Evaluation compares against state-of-the-art baselines, including GraphCast global forecasts (native grid), regional U–Net and LightGBM learners, using mean absolute error and root mean squared error across core surface variables; training and inference time are also reported to quantify cost–accuracy trade-offs. CityScaleCast combines GraphCast-derived dynamic boundary conditioning with a regional spatiotemporal GNN to deliver city-scale 3–24 h forecasts that surpass a reprojected GraphCast reference on near-surface variables (T2M, MSLP, U10/V10) while achieving markedly lower training and inference cost on commodity GPUs, at the expense of a small degradation on intermittent precipitation.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Pan, X., Shimomura, Y., Tao, S., Kawai, M., Takahashi, K., & Takizawa, H. (2026). CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai. Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, 177–185. https://doi.org/10.1145/3784828.3785258

MLA 9

Pan, Xuanwen, et al. “CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai.” Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, 2026, pp. 177–185. https://doi.org/10.1145/3784828.3785258.

Chicago

Pan, Xuanwen, Yoichi Shimomura, Sichen Tao, Masatoschi Kawai, Keichi Takahashi, and Hiroyuki Takizawa. 2026. “CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai.” Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops: 177–185. https://doi.org/10.1145/3784828.3785258.

Harvard

Pan, X. et al. (2026) ‘CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai’, Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, pp. 177–185. doi:10.1145/3784828.3785258.

IEEE

X. Pan, Y. Shimomura, S. Tao, M. Kawai, K. Takahashi, and H. Takizawa, “CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai,” Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, pp. 177–185, 2026, doi: 10.1145/3784828.3785258.

BibTeX

@article{pan2026cityscalecast,
  title     = {{CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai}},
  author    = {Pan, Xuanwen and Shimomura, Yoichi and Tao, Sichen and Kawai, Masatoschi and Takahashi, Keichi and Takizawa, Hiroyuki},
  journal   = {Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops},
  year      = {2026},
  pages     = {177--185},
  doi       = {10.1145/3784828.3785258},
  url       = {https://doi.org/10.1145/3784828.3785258}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Earth and Planetary Sciences › Atmospheric Science › Meteorological Phenomena and Simulations

  • spatio-temporal graph neural networks
  • GraphCast
  • dynamic boundary conditions
  • multi-step forecasting
  • regional weather model
  • Sendai
  • precipitation forecasting

Work with this paper

Most-cited papers in Earth and Planetary Sciences

All highly cited Earth and Planetary Sciences papers

Questions about this paper

How do I cite this paper in APA style?

Pan, X., Shimomura, Y., Tao, S., Kawai, M., Takahashi, K., & Takizawa, H. (2026). CityScaleCast: Spatiotemporal GNN for City-Scale Weather Prediction with GraphCast-Guided Parallel Modeling and Multi-Step Forecasting in Sendai. Proceedings of the Supercomputing Asia and International Conference on High Performance Computing in Asia Pacific Region Workshops, 177–185. https://doi.org/10.1145/3784828.3785258

How many times has this paper been cited?

2 times, according to OpenAlex on 20 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access: sign in and open the full text from this page.

Who wrote this paper?

Xuanwen Pan, Yoichi Shimomura, Sichen Tao, Masatoschi Kawai, Keichi Takahashi, Hiroyuki Takizawa.

What is the DOI of this paper?

10.1145/3784828.3785258, permanently at https://doi.org/10.1145/3784828.3785258. Sign in and open it from this page.

Bibliographic data from OpenAlex (CC0), last checked 20 November 2026. ResearchGuru does not host this paper; once you are signed in, the links open it at the publisher or the repository that does.