The Xiamen Symposium on Marine Environmental Sciences (XMAS 2025)

This study utilizes satellite altimetry observations and employs the ConvLSTM (Convolutional Long Short-Term Memory) model to predict sea level anomaly (SLA) in the Kuroshio Extension (KE) region. The ConvLSTM involves both spatial features and spatiotemporal relationships of data, enabling rapid and accurate predictions. The results demonstrate that the ConvLSTM performs well and be effective in predicting SLA fields in the KE region. The regional average Root Mean Square Error (RMSE) increases rapidly in the first 30 days lead time but maintains relatively consistent prediction error levels beyond 30 days. The performance of ConvLSTM shows spatial difference, with higher errors located in regions with strong ocean fronts and energetic eddy activities. Analyses indicate that the ConvLSTM has successfully captured the dynamics of Rossby waves, resulting in favorable prediction outcomes. This study provides valuable insights into predicting oceanic physical quantities using the ConvLSTM model.

sea level anomaly prediction, Kuroshio Extension, deep learning, ConvLSTM