•  
  •  
 

Abstract

This study focuses on assessing and improving the predictive performance of sea surface temperature (SST) in the Gulf of Thailand under unique environmental challenges. Data from the Hybrid Coordinate Ocean Model (HYCOM) were utilized, covering the period from July 1994 to June 2024 at depths ranging from 0 to 30 meters. After cleaning the data using the Interquartile Range (IQR) method, the processed dataset was tested using four models: Recurrent Neural Network (RNN) architectures, including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), as well as Convolutional Neural Networks (CNNs). Model performance was enhanced by tuning the parameters over 1 to 20 epochs. Evaluation metrics, including Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE), were used to assess the models. The results revealed that the GRU model achieved the highest performance with an optimal epoch setting of 6, yielding the lowest prediction error of 0.015 °C. Additionally, the study analyzed the impact of rising sea temperatures on the environment from the past to the present. During El Niño events, significant increases in sea temperatures triggered global coral bleaching in the years 1998, 2010, and 2014 to 2016.

COinS