Forecasting tropical cyclones wave height using bidirectional gated recurrent unit

A bidirectional Gated recurrent units (BiGRU) network is proposed for the prediction of wave height during tropical cyclones (TC). We used a data set of 28 TC events collected from 14 buoys in different environments over the past 9 years. We use buoy data and TC data collected from 27 TC events for training and use six different parameters to predict the wave height in different lead time, the trained model was used to predict the wave heights of 10 buoys during a new typhoon, compared to machine learning models, the results illustrate that BiGRU's predictive performance is stable, especially for prediction 24 hours in advance, and the model can still be effectively used for real-time wave height prediction when the performance of traditional machine learning methods is severely degraded. In terms of long-term prediction, the model's performance exceeds that of existing methods.

Automated summary

A bidirectional Gated Recurrent Units (BiGRU) network is introduced for predicting wave height during tropical cyclones, utilizing buoy and TC data for training, the model's performance is found to be stable and superior compared to traditional machine learning methods, especially for 24-hour advance prediction and long-term forecasting.