EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

The Critical Role of Surge Autocorrelation in Data‐Driven Nowcasting of Typhoon‐Induced Storm Surge

Yujia Wang, Dingqi Yang, Ka‐Veng Yuen, Huabin Shi

Water Resources Research · 2026-04 · 62 (4)

DOI: 10.1029/2025wr040850 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Water Resources Research, online erschienen am 20.04.2026. Verfasst von Yujia Wang und 3 weiteren Autor:innen. Die Arbeit wurde bisher 3-mal zitiert (Crossref) und nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Southern Marine Science and Engineering Guangdong Laboratory, Research Services and Knowledge Transfer Office, University of Macau.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Typhoon‐induced storm surges have been the most destructive disasters in coastal regions. The nowcasting of storm surge, in which the prediction lead time is not greater than 3 hr, is critical to disaster emergency response and recovery, enabling governments to make decisions and take actions according to the latest situation and prediction. Machine learning (ML) methods have been increasingly applied to storm surge predictions, but few ML studies focused on short‐term nowcasts. Moreover, conflicting conclusions often arise regarding the superiority of ML methods and model input variables for storm surge predictions. This study highlights the significance of surge autocorrelation, that is, the correlation of surge height with its local historical values, in the design of ML models for storm surge nowcasts. It is shown in the comparative analyses of model input variables that, independent of the adopted ML methods, models relying solely on historical surge heights as inputs are sufficient to achieve reasonably accurate nowcasts of surge height. Conversely, models without historical surge heights in their inputs yield comparable prediction errors. Involving surge autocorrelation, simple multilayer perceptron models perform even better than sequential‐compatible recurrent neural network category of models. Moreover, it is illustrated that the surge autocorrelation inhibits the performance of Transformer and XGBoost models for storm surge nowcasts. This study shows the possibility to achieve efficient and accurate nowcasts (a lead time under 3 hr) of localized storm surge only based on limited observations by local tide gauge, supporting decision‐making in disaster emergency response and recovery within small‐scale coastal regions.

Förderung

  • Southern Marine Science and Engineering Guangdong Laboratory
  • Research Services and Knowledge Transfer Office, University of Macau

Zitieren

APA

Wang, Y., Yang, D., Yuen, K., & Shi, H. (2026). The Critical Role of Surge Autocorrelation in Data‐Driven Nowcasting of Typhoon‐Induced Storm Surge. Water Resources Research, 62(4). https://doi.org/10.1029/2025wr040850

Vancouver

Wang Y, Yang D, Yuen K, Shi H. The Critical Role of Surge Autocorrelation in Data‐Driven Nowcasting of Typhoon‐Induced Storm Surge. Water Resources Research. 2026;62(4). doi:10.1029/2025wr040850

BibTeX
@article{Wang2026,
  author = {Wang, Yujia and Yang, Dingqi and Yuen, Ka‐Veng and Shi, Huabin},
  title = {The Critical Role of Surge Autocorrelation in Data‐Driven Nowcasting of Typhoon‐Induced Storm Surge},
  journal = {Water Resources Research},
  year = {2026},
  volume = {62},
  number = {4},
  doi = {10.1029/2025wr040850}
}
Kostenlos starten →