EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Assessing the predictability of meteorological variables via spatial correlations using echo state networks

Shihori Koyama iD, Daisuke Inoue iD, Hiroaki Yoshida iD, Kazuyuki Aihara iD, Gouhei Tanaka iD

Chaos: An Interdisciplinary Journal of Nonlinear Science · 01.08.2026 · 36 (8)

DOI: 10.1063/5.0324305 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Chaos: An Interdisciplinary Journal of Nonlinear Science, online erschienen am 07.08.2026. Verfasst von Shihori Koyama und 4 weiteren Autor:innen. Beteiligte Einrichtung: Toyota Central R&D Labs., Inc., Aichi 480-1192, Japan. Die Arbeit nennt 51 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Japan Society for the Promotion of Science, Core Research for Evolutional Science and Technology, Moonshot Research and Development Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Collecting spatially dense meteorological observations is essential for accurate climate modeling and prediction. However, such observations are often costly and difficult to obtain. This limitation motivates the prediction of meteorological variables at locations where direct observations are sparse or unavailable, using limited data from distant observation points. In this study, we address a fundamental question: How does the predictability of meteorological variables depend on the spatial distance between the prediction target and the observation point? To cope with limited data, we employ the echo state network, which is a lightweight machine-learning model within the reservoir computing framework. Predictive models are constructed separately for near-surface temperature and atmospheric pressure using time-series data of meteorological variables obtained from a high-quality climate reanalysis dataset. Our analyses not only confirm the qualitative property that prediction accuracy degrades as the distance between the target point and observation point increases but also provide quantitative estimates of the geographical ranges over which predictions can be made within an acceptable margin of error. These results suggest that prior analysis of spatial correlations in limited observational data can be used to estimate meteorological predictability in advance, thereby contributing to the development of climate modeling with improved efficiency in data collection and computation.

Autor:innen und Institutionen

  • Toyota Central R&D Labs., Inc., Aichi 480-1192, Japan.
  • International Research Center for Neurointelligence, The University of Tokyo, Tokyo 113-0033, Japan.
  • Graduate School of Engineering, Nagoya Institute of Technology, Aichi 466-8555, Japan.

ORCID-Profile: Shihori Koyama, Daisuke Inoue, Hiroaki Yoshida, Kazuyuki Aihara, Gouhei Tanaka

Förderung

  • Japan Society for the Promotion of Science
  • Core Research for Evolutional Science and Technology
  • Moonshot Research and Development Program
  • Japan Agency for Medical Research and Development
  • the International Research Center for Neurointelligence
  • Intelligent Mobility Society Design, Social Cooperation Program

Zitieren

APA

Koyama, S., Inoue, D., Yoshida, H., Aihara, K., & Tanaka, G. (2026). Assessing the predictability of meteorological variables via spatial correlations using echo state networks. Chaos: An Interdisciplinary Journal of Nonlinear Science, 36(8). https://doi.org/10.1063/5.0324305

Vancouver

Koyama S, Inoue D, Yoshida H, Aihara K, Tanaka G. Assessing the predictability of meteorological variables via spatial correlations using echo state networks. Chaos: An Interdisciplinary Journal of Nonlinear Science. 2026;36(8). doi:10.1063/5.0324305

BibTeX
@article{Koyama2026,
  author = {Koyama, Shihori and Inoue, Daisuke and Yoshida, Hiroaki and Aihara, Kazuyuki and Tanaka, Gouhei},
  title = {Assessing the predictability of meteorological variables via spatial correlations using echo state networks},
  journal = {Chaos: An Interdisciplinary Journal of Nonlinear Science},
  year = {2026},
  volume = {36},
  number = {8},
  doi = {10.1063/5.0324305}
}
Kostenlos starten →