Time series prediction of sea surface temperature based on BiLSTM model with attention mechanism
2024
Zrira, Nabila | Kamal-Idrissi, Assia | Farssi, Rahma | Khan, Haris Ahmad
With the advancement of technology, ocean observation techniques have become increasingly prevalent in estimating marine variables such as Sea Surface Temperature (SST). This progress has led to a substantial surge in the volume of marine data. Presently, the abundance of available data presents a remarkable opportunity for training predictive models. The prediction of SST poses a challenge due to its temporal-dependent structure and multi-level seasonality. In this study, we propose a deep learning approach that combines the Bidirectional Long Short-Term Memory (BiLSTM) model with the attention mechanism to forecast SST. By leveraging the BiLSTM's ability to effectively capture long-term dependencies through both forward and backward LSTM processing, the attention mechanism accentuates salient features, thereby enhancing the model's evaluation accuracy. To evaluate the effectiveness of the Attention-BiLSTM model in predicting SST, we conducted a case study in the Moroccan Sea, focusing on four distinct regions. We compared the performance of the Attention-BiLSTM model against alternative models such as LSTM, Attention-BiGRU, XGBoost, Random Forest (RF), Support Vector Regression (SVR), and Transformers in forecasting the SST time series. The experimental results unequivocally demonstrate that the Attention-BiLSTM model achieves significantly superior prediction outcomes and is a good candidate for deployment in the field.
Показать больше [+] Меньше [-]Ключевые слова АГРОВОК
Библиографическая информация
Эту запись предоставил Wageningen University & Research