FORECASTING SEA LEVEL CHANGES USING HYBRID ARIMA-RADIAL BASIS FUNCTION NEURAL NETWORK METHODS

Abstract

Understanding sea level variability is crucial for ensuring the safety of tourists, particularly in marine tourism areas like Marina Ancol Beach in North Jakarta. Climate change has led to rising sea levels, significantly impacting coastal regions. Accurate predictions of sea level are essential for anticipating tidal flooding, which occurs when seawater inundates these areas. Short-term sea level fluctuations are influenced by both linear tidal patterns and nonlinear local effects, making accurate forecasting challenging when using a single modeling approach. This study proposes a hybrid forecasting method that combines the Autoregressive Integrated Moving Average (ARIMA) model to capture linear temporal structures and a Radial Basis Function Neural Network (RBFNN) to model nonlinear patterns present in the residuals. Hourly sea level data consisting of 17,520 observations collected from January 2021 to December 2022 were analyzed. The proposed hybrid ARIMA–RBFNN model achieved a Mean Absolute Percentage Error (MAPE) of 2.74%, slightly outperforming the ARIMA model, which yielded a MAPE of 2.76%. The model provides accurate 24-hour sea level forecasts for Marina Ancol Beach, offering timely information that can support local authorities in anticipating and mitigating tidal flooding events.