Modelling sea level variations in the black sea using deep learning method

One of the most devastating consequences of global climate change is sea level rise. Accelerated by thermal expansion and the melting of glaciers, this process destroys coastal ecosystems, threatens infrastructure, and leads to the displacement of millions of people. To mitigate these risks, it is of great importance to model sea level changes precisely and to produce reliable future predictions. However, due to the limitations of the methods used in the current literature, there is a need for stronger and more generalizable models. In this context, the Black Sea is of particular importance due to its geostrategic position, dense coastal settlements, ecological sensitivity, and vulnerability to climate change. In this doctoral thesis, satellite radar altimetry data were used to model sea level changes in the Black Sea. In addition to the commonly used Stacked LSTM and Damped Persistence methods in the literature, ConvLSTM and Auto-Encoder LSTM architectures were applied for this purpose for the first time. Furthermore, instead of manual hyperparameter optimization, Bayesian automatic hyperparameter optimization was employed in sea level modelling studies for the first time. The outcome of the thesis is a sea level early warning system that retrieves up-to-date data from international data centers and offers 15-day forecasts through a user-friendly interface. As the first comprehensive study on sea level modelling in Turkey, this thesis introduces novel deep learning architectures to the literature and provides original contributions by transforming academic outputs into an operational early warning system.