Deep learning based bias correction model for numerical simulations of wave spectra

30 de dezembro de 2025

To accurately characterize the energy distribution under different frequency ranges, this paper developed a deep learning-based wave spectral bias correction model for numerical simulations of wave spectra, which addresses the shape of the numerically predicted spectra. This model preprocesses the measured wave spectra using a Gaussian filter, and the frequency of the numerical simulation results from WAVEWATCH III is matched using the interpolation method. Afterwards, the simulated wave spectra serve as the input to the proposed model, and discrepancies between the numerical simulation results and buoy measurement data are fitted using the neural network model. Moreover, the correction performance of the model is validated using measured data in terms of spectral shape, significant wave height, peak frequency, and peak energy density. With the proposed method the mean absolute percentage error of those parameters of decreases by 5 %–15 % and the spectral correlation for the Hawaiian waters is no less than 0.93. The results demonstrate that the proposed model can more accurately characterize the wave energy distribution and significantly improve the accuracy of wave simulations.

Autores da comunidade :

• 

  C

  Carlos Guedes Soares

  ist11869

  Afiliações e fontes

  IST > CENTEC

  Centro de Engenharia e Tecnologia Naval e Oceânica

AO - Versão original do autor

Elsevier

https://www.sciencedirect.com/science/article/abs/pii/S0029801825028586?via%3Dihub

Ocean Engineering

123175

342

Part 4

0029-8018

10.1016/j.oceaneng.2025.123175

other-engineering-and-technologies - Outras Ciências da Engenharia e Tecnologias

Palavras-chave

• Wave Spectra
• Bias Correction
• Buoy Measurement
• WAVEWATCH III
• Deep Learning

eng - Inglês

Entidade financiadora da bolsa/projeto

Fundação para a Ciência e a Tecnologia

Identificador da Entidade Financiadora: https://doi.org/10.13039/501100001871

Tipo de identificador da Entidade Financiadora: Crossref Funder

Número de bolsa/projeto: UIDB-UIDP-00134-2020