Significant Wave Height Prediction in the South China Sea Based on the ConvLSTM Algorithm

Deep learning methods have excellent prospects for application in wave forecasting research. This study employed the convolutional LSTM (ConvLSTM) algorithm to predict South China Sea (SCS) significant height (SWH). Three prediction models were established investigate influences of setting different parameters and using multiple training data on effects. Compared with SWH from China–France Ocean Satellite (CFOSAT), WAVEWATCH III (WWIII) pacific islands ocean observing system are accurate enough be used as ConvLSTM-based model. Model A was preliminarily by only WWIII data, 20 sensitivity experiments carried out parameter settings effect A. The experimental results showed that has best when three years hourly input data. With same performance A, B C also more wind shear velocity When making a 24-h forecast, compared root mean square error (RMSE) is decreased 17.6%, correlation coefficient (CC) increased 2.90%, absolute percentage (MAPE) reduced 12.2%. SWH, velocity, direction RMSE 19.0%, CC 2.65%, MAPE 14.8%. As model mainly rely All show greater nearshore area than deep SCS during period typhoon transit without typhoon. Considering wind, will improve prediction.