Time series analysis of Turkish National Sea Level Monitoring System (TUDES) level data for Amasra Station

Abstract

The observation and prediction of sea level are crucial for various reasons including the vertical datum determination, crustal movement forecasting, oceanographic modeling, and coastal infrastructure planning. In Turkey, a sea level monitoring system has been established by the General Directorate of Mapping and aims to measure sea level. Through the Turkish National Sea Level Monitoring System (TUDES), sea level is monitored using data collected at 20 tide gauge stations at 15-minute intervals. Time series analysis is considered a highly suitable modeling and forecasting method for data that is periodically measured. In this study, time series analysis models including ARIMA, SARIMA, and Holt-Winter's methods were applied using data from the Amasra tide gauge station within the TUDES for the year 2019. Additionally, a prediction for January 2020 at the same station was performed. The results were compared with the measured tide gauge data to assess the performance of the models. Evaluation criteria included the Mean Absolute Percentage Error (MAPE) for the Holt-Winter's method and the corrected Akaike Information Criteria (AICc) for the ARIMA and SARIMA models. The SARIMA (3,0,0) (0,2,2) model with an AICc value of -1307.83, indicating a seasonality of 12, was observed to be the best-performing model.