A groundbreaking method for predicting storm surges has emerged from the Yantai Institute of Coastal Zone Research, led by Prof. Mao Miaohua. This innovative approach is set to transform how we model typhoon wind fields by utilizing a hybrid wind field.

The research team ingeniously developed four machine learning (ML) models that, when integrated with the Finite Volume Community Ocean Model (FVCOM-ML), have significantly enhanced the accuracy of storm surge forecasting. These advancements were highlighted in a recent publication in the Journal of Geophysical Research: Machine Learning and Computation.

Timely and precise storm surge predictions are more critical than ever for protecting coastal zones and implementing effective risk reduction strategies. The Bohai Sea—a historically less impacted area—has recently seen a surge in typhoon activity, making accurate forecasting essential to safeguard lives and property.

In this study, the researchers innovatively combined a reanalysis wind field with the Holland model to create their hybrid wind field. By leveraging numerical simulations conducted with the Advanced Circulation Model, they successfully developed machine learning models capable of compensating for previously missing data on storm surges.

These integrated methods not only boost accuracy but also reduce uncertainty, providing forecasts for storm surges with lead times of 6, 12, and 18 hours. Remarkably, the integrated model's accuracy for both single-site and multi-step storm surges surpassed that of individual models by over 30%. This is especially notable for multi-site multi-step predictions, where the integrated model showed considerable superiority.

With techniques like FVCOM-CNN-LSTM and FVCOM-ConvLSTM, this research stands out from traditional methods. Unlike conventional machine learning that relies solely on objective functions, FVCOM-ML models actual physical residuals, effectively addressing the uncertainties typically observed in standard methodologies.

The models demonstrated exceptionally high prediction capabilities in both Bohai and Laizhou Bays, outperforming Liaodong Bay in terms of storm surge forecasting. Overall, the integrated models improve prediction accuracy in the Bohai Sea by an impressive 18%, marking a significant advancement in storm surge prediction technology.