Hourly-scale coastal sea level modeling in a changing climate using long short-term memory neural network

Area of Science:

• * Environmental Science
• * Data Science
• * Oceanography

Background:

• * Coastal sea level is influenced by complex factors including gravitational forces, weather patterns, and climate change.
• * Accurate sea level estimation is crucial for coastal management and disaster preparedness.
• * Existing models may not fully capture the interplay of short-term and long-term drivers of sea level change.

Purpose of the Study:

• * To develop and evaluate an hourly coastal sea level estimation model using Long Short-Term Memory (LSTM) networks.
• * To integrate various influential factors: solar and lunar gravitational attractions, seasonality, storm surges, and global warming.
• * To assess the impact of input data duration on model performance.

Main Methods:

• * Development of an LSTM-based recurrent neural network model for hourly sea level prediction.
• * Inclusion of astronomical data (relative positions of sun and moon), local meteorological variables (wind speed/direction, MSLP, air temperature), and annual global mean air temperature.
• * Model validation using statistical metrics: mean, Nash-Sutcliffe efficiency (NSE), and root mean square error (RMSE) at the Osaka gauging station.

Main Results:

• * The LSTM model successfully reconstructed the impact of solar and lunar gravitational forces on coastal sea levels.
• * The model accounted for variations in wind speed and mean sea level pressure (MSLP), though underestimation occurred during extreme conditions.
• * Utilizing a longer time series of annual global mean air temperature significantly enhanced model accuracy, achieving an NSE of 0.720.

Conclusions:

• * LSTM networks provide a robust framework for modeling complex coastal sea level dynamics.
• * Integrating diverse data, including long-term climate indicators, is vital for accurate sea level prediction.
• * The study demonstrates the potential of advanced machine learning techniques for improving coastal climate change impact assessments.