Typhoon wave height prediction based on BO-LSTM and Pangu-Weather models

Chen, Yongping, Qin, Zhipeng, Xu, Zhenshan, Xu, Xiaowu, Pan, Shunqi ORCID: https://orcid.org/0000-0001-8252-5991 and Wang, Jinghua 2026. Typhoon wave height prediction based on BO-LSTM and Pangu-Weather models. Engineering Applications of Computational Fluid Mechanics 20 (1) , 2658296. 10.1080/19942060.2026.2658296

PDF - Published Version Available under License Creative Commons Attribution. Download (7MB)

Abstract

Under the impacts of climate change, such as rising sea levels and intensifying storms, disasters caused by extreme waves are becoming increasingly severe, necessitating in-depth research on intelligent methods for typhoon wave forecasting. This study proposes two intelligent wave height prediction schemes tailored for two specific temporal scales. The first is a short-term forecasting scheme based on the TCWiSE model to generate numerous synthetic typhoons, and then provide surface forcing for the SWAN model to calculate significant wave height at monitoring stations during typhoon events. The results are used to construct a typhoon wave sample database. Subsequently, the BO-LSTM machine-learning model is trained and tested using this database. It is demonstrated that this proposed short-term scheme is capable of predicting the typhoon waves at a single station highly efficiently (within seconds), delivering accurate results with RMSE of 0.02, 0.09, 0.19, and 0.42 m for 1, 3, 6, and 12 h forecasts, respectively, highlighting its advantages of high accuracy and computational efficiency. The second scheme is to further extend the forecast horizon, for a longer term forecasting approach. This method uses the Pangu-Weather model to predict longer term typhoon wind fields, using empirical correction, and then use the SWAN model to calculate the corresponding typhoon waves. The results indicate that this scheme is capable of predicting 7-day typhoon wave heights with a minute-level computational efficiency at an RMSE of 0.76 m. While maintaining reasonable accuracy and efficiency, this approach significantly enhances the forecast horizon compared to conventional methods.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Additional Information:	License information from Publisher: LICENSE 1: URL: http://creativecommons.org/licenses/by/4.0/, Start Date: 2026-04-22
Publisher:	Taylor and Francis Group
ISSN:	1994-2060
Date of First Compliant Deposit:	7 May 2026
Date of Acceptance:	7 April 2026
Last Modified:	07 May 2026 09:15
URI:	https://orca.cardiff.ac.uk/id/eprint/186847

Actions (repository staff only)

Edit Item