Comparative Analysis of SARIMA, FFNN, and Hybrid Models for Sea Surface Temperature Prediction at Enggano Island (2018–2024)

Abstract

Sea Surface Temperature (SST) is a key oceanographic variable that influences fish distribution and the livelihoods of coastal communities. On Enggano Island, where most residents rely on fishing, SST is critical for identifying optimal fishing grounds due to limited accessibility and high operational costs. Accurate modeling and forecasting of SST are therefore essential for effective fisheries management and sustainable resource use. This study analyzes and predicts monthly SST patterns in Enggano Island using Seasonal Autoregressive Integrated Moving Average (SARIMA), Feed Forward Neural Network (FFNN), and Hybrid SARIMA-FFNN models. SARIMA effectively captures linear trends and seasonal variations but struggles with nonlinear dynamics and requires statistical assumptions. Conversely, FFNN models nonlinear relationships without such assumptions but is less efficient in representing linear and seasonal structures. The hybrid SARIMA-FFNN combines the strengths of both approaches, integrating linear-seasonal accuracy with nonlinear adaptability. Monthly SST data from January 2018 to December 2024, covering northern, eastern, southern, and western regions of Enggano Island, were analyzed. Results show that all models achieved high predictive accuracy, with MAPE values below 10%. Based on RMSE, FFNN outperformed the other models across all regions (north: 1.173, east: 0.999, south: 1.245, west: 1.049), confirming FFNN as the most accurate model for SST prediction. Predicted SST values across the four regions exhibited only minor differences, offering fishermen flexibility in selecting fishing grounds. Sustainable fishing strategies should also consider species-specific temperature preferences and other ecological factors influencing fish distribution.

Keywords

• Enggano Island
• FFNN
• Hybrid SARIMA-FFNN
• SARIMA
• Sea Surface Temperature