Forecasting instantaneous sea level changes with machine learning methods in Aegean Sea

Forecasting instantaneous sea-level is of great importance in terms of determination of geodetic vertical datum and updating, conservation of coastal areas, monitoring coastal ecosystems, maintenance, and planning of coastal structures, monitoring of climate change effects. Traditional methods used for instantaneous sea level estimation are often based on linear assumptions. However, contributors to sea levels are very various and their effects vary from region to region. Generally, they have complex and nonlinear dependence structures. Therefore, nonlinear sea-level cannot be determined with high precision using linear models. Recently, machine learning prediction methods have been frequently used in the modelling of complex dependency structures between variables. Within the scope of this study, to predict the instantaneous sea level with high accuracy and to compare linear estimation methods with nonlinear estimation methods, the Multiple Linear Regression (MLR) linear model, Support Vector Regression (SVR) non-linear model, and Random Forest Regression (RFR) non-linear model algorithms were used, and their prediction performances were compared. As a result of the study, the highest prediction performance for instantaneous sea level was obtained with RFR, and the lowest prediction performance was obtained with the MLR method. As a result, it has been shown that instantaneous sea level can be predicted with high precision using RFR with the features used in this study, and the linear prediction models are insufficient in modelling the complex dependency structure of instantaneous sea level.