Isabel Joy D. Adriatico, Shaun Tristan Elizer Cuesta, Gerard D. Ompad
Marine heatwaves (MHWs), which are characterized by prolonged periods of anomalously high sea surface temperatures (SST), pose significant threats to marine ecosystems, coastal economies, and global biodiversity. The Philippines—situated in the Coral Triangle—relies heavily on its marine resources, making it vulnerable to these impacts. This study explores the use of deep learning models to predict SST and identify the occurrence of MHWs in the Philippine Sea using data sourced from NOAA Daily OISST. The study evaluated four deep learning architectures: (1) Long Short-Term Memory (LSTM), (2) Convolutional Neural Networks (CNN), (3) a hybrid model combining the strengths of both LSTM and CNN, and (4) Neural Basis Expansion Analysis for Interpretable Time Series (N-BEATS). To optimize performance, all models were fine-tuned using Bayesian optimization (BO). The hybrid model demonstrated a superior performance in SST forecasting, achieving RMSEs of 0.1034 °C at 1-day and 0.0652 °C at 14-day lead times. For MHW detection, N-BEATS achieved an RMSE of 0.0523 °C, with a recall of 98.25%, ensuring reliable identification of rare events. N-BEATS also obtained an F1-score of 97.67%, reflecting a strong balance between precision and recall, minimizing both false positives and false negatives. These results highlight the potential of neural networks in providing accurate SST forecasting and MHW detection to mitigate climate change impacts on the Philippines’ marine ecosystems and coastal communities. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.
University of San Carlos, Cebu, Philippines; University of Copenhagen, Copenhagen, Denmark
