The combination of typhoon-induced storm surges and astronomical tides can result in extreme seawater levels and disastrous effects on coastal socioeconomic systems. The construction of an appropriate wind field has consistently been a challenge in storm tide forecasting and disaster warning. In this study, we optimized a nonlinear regression formula based on the C15 model to determine the maximum wind radius. The simulation based on the improvement showed good accuracy for storm tides during super typhoon Mangkhut (WP262018), Saola (WP092023), and severe typhoon Hato (WP152017). The correlation coefficients were in the 0.94–0.98 range, and the peak bias was less than 5cm. The trough errors were significantly reduced compared to other wind fields. Owing to the importance and lack of the maximum wind radius (Rmax), we attempted to predict Rmax using an LSTM (Long Short-Term Memory) neural network for forecasting storm tides during strong typhoons. Constrained LSTM showed good performance in hours 6–48, and effectively enhanced the forecasting capability of storm tides during strong typhoons. The workflows and methods used herein have broad applications in improving the forecasting accuracy of strong typhoon-induced storm tides.

中文翻译：

---

使用人工智能和物理模型预测强台风期间的风暴潮


台风引起的风暴潮和天文潮汐的结合可能导致极端海水水位并对沿海社会经济系统造成灾难性影响。建设合适的风场一直是风暴潮预报和灾害预警的一个挑战。在本研究中，我们基于C15模型优化了非线性回归公式来确定最大风半径。基于改进的模拟显示了超强台风“山竹”（WP262018）、“萨乌拉”（WP092023）和强台风“天鸽”（WP152017）期间风暴潮的良好精度。相关系数在0.94-0.98范围内，峰值偏差小于5cm。与其他风场相比，波谷误差显着减小。由于最大风半径 (Rmax) 的重要性和缺乏，我们尝试使用 LSTM（长短期记忆）神经网络来预测 Rmax，以预测强台风期间的风暴潮。约束LSTM在6-48小时表现良好，有效增强了强台风期间风暴潮的预报能力。本文使用的工作流程和方法在提高强台风引发的风暴潮的预报精度方面具有广泛的应用。