Bi-LSTM Approach for Seagrass Transplantation Site Detection

Seagrasses form crucial ecosystems globally, supporting diverse communities and providing essential ecological and socioeconomic benefits. However, these ecosystems are under threat, necessitating effective restoration strategies. This study proposed Bidirectional Long Short-Term Memory (Bi-LSTM) approach to identify and evaluate optimal sites for seagrass transplantation. The dataset comprising files containing key environmental parameters such as temperature, salinity, dissolved oxygen, pH, total suspended solids, and bathymetry, was used to train and validate the Bi-LSTM model. The model achieved initial training and validation accuracy are 98.75% and 99.98% respectively, also training and validation loss are 0.0326 and 0.0112 respectively within 10 epochs. A comparative analysis with Echo State Networks (ESN) and One-Dimensional Convolutional Neural Networks (CNN-1D) demonstrated that the Bi-LSTM model significantly outperformed these alternatives, achieving superior prediction accuracy. The model was further validated using environmental data from Dompak Island in Indonesia, with the predictions visualized through QGIS. The Bi-LSTM model exhibited high performance, with an average accuracy of 88.38% and the mean squared error (MSE) of 11.615%. Furthermore, the model’s potential in supporting the strategic restoration and preservation of coastal ecosystems was highlighted. This study provides environmental professionals with a powerful analytical tool for enhancing seagrass cover, water quality, marine habitats, and carbon sequestration processes. Future research could focus on developing a real-time predictive tool integrated with various sensors to further improve the efficiency and accuracy of seagrass transplantation site detection. This would ensure continuous up-dates and immediate feedback based on real-time environmental data.