EDDA-ConvLSTM: Encoder-Decoder Dual Attention ConvLSTM for Moroccan Coastal Sea Surface Temperature Prediction

This study presents an advanced encoder-decoder dual attention convolutional long short-term memory (ConvLSTM) model designed to predict sea surface temperatures (SSTs) along the Moroccan coastline, a region characterized by complex oceanographic dynamics. Our model leverages the power of convolutional operations to capture spatial dependencies and the LSTM architecture to model temporal sequences while incorporating a dual attention mechanism to enhance feature selection through contextual attention vectors, which allow the model to focus on the most significant spatial and temporal features in the data input and spatial attention to adaptively weigh the importance of different regions in the study area. In addition, the encoder-decoder architecture efficiently processes multidimensional oceanographic data, leveraging spatial and temporal correlations. Results demonstrate significant improvements in prediction accuracy and computational efficiency compared to traditional single-attention models.