COMPARISON OF GARCH AND LSTM MODELS FOR PREDICTING SEA SURFACE TEMPERATURE VOLATILITY IN THE SOUTHERN WATERS OF JAVA

Abstract

Indonesia, as an archipelagic country with a maritime economic potential of US$ 1.33 trillion per year, faces serious challenges due to climate change, particularly rising sea surface temperatures (SST) and marine heatwaves (MHW), which impact fisheries productivity and the stability of the Blue Economy. This study aims to predict SST volatility in the region through the development of a forecasting model that compares ARCH-GARCH and Long Short-Term Memory (LSTM) to model non-linear patterns and long-term dependencies. Oceanographic time series datasets were used to train and test the model's performance. The results show that the LSTM model has much higher accuracy than the ARCH–GARCH model in predicting SPL volatility, with lower error values and better ability to capture non-linear patterns and complex sea temperature dynamics. This study confirms that the LSTM approach is more suitable for use as an oceanographic forecasting model, thereby supporting the strengthening of the Blue Economy and the resilience of Indonesia's maritime sector.

Key words: blue economy, sea surface temperature, GARCH, LSTM