Results of Models for Nowcasting of Typhoon-Induced Storm Surge

Accurate storm surge prediction is critical for coastal safety and disaster preparedness. This study evaluated the performance of machine learning (ML) models in the nowcasting (0–3h lead time) of tropical cyclone (TC)-induced storm surges, with a focus on the role of sea level autocorrelation. Choosing study area as Quarry Bay Station in Hong Kong during 257 TCs (1968–2021), seven ML models—including Multi-Layer Perceptron (MLP), Recurrent Neural Networks (RNN, LSTM, GRU), Transformer, and XGBoost—were developed and compared. Input variables encompassed historical surge levels, TC characteristics (e.g., central pressure, wind speed), and local meteorological factors (wind speed, atmospheric pressure).