Deep Learning-Based Seasonal Forecasting of Oceanographic Parameters Using LSTM Networks

Monitoring oceanographic parameters is essential in understanding climate variability and marine ecosystem dynamics. The study uses an LSTM neural network to model and predict oceanographic parameters, which include wave height, dominant period, average period, mean wave direction, water temperature, and salinity. The dataset obtained from two NOAA buoys (stations 41110 and 41037) between 2010 and 2015 showed seasonal variations. The LSTM model was trained using normalized time-series data and was evaluated using Root Mean Squared Error (RMSE). In comparison with the ARIMA baseline model, which had poor predictive accuracy with negative R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values and large error margins, the LSTM model performed better, especially in wave height and water temperature prediction, with R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> > 0.97 and much lower RMSE values. These findings reflect the strength of deep learning in modelling intricate oceanographic patterns and open the door to more sophisticated predictive models in environmental monitoring and climate change analysis.