The estimation of monthly mean significant wave heights by using artificial neural network and regression methods

This paper focuses on the prediction of monthly mean significant wave heights from meteorological data by using both artificial neural network (ANN) and regression methods. Waves and meteorological data used in the study were collected by three different buoys located offshore in the Atlantic. Seven different ANN models comprising of various input combinations of monthly mean wind speeds, sea level pressures and air temperature ratios based on hourly observations were performed to evaluate wave height prediction performance of each meteorological parameter for each buoy. The results indicated that the ANN model, having all parameters in the input layer, gave the best prediction performance. In order to obtain wave heights, some empirical formulas were also suggested through the regression analysis in which wave and meteorological parameters were expressed as two dimensionless groups by using π theorem. The results of these suggested formulas were compared with those of the ANN models. From these comparisons, good performances of proposed formulas were verified by the ANN model results under waves and meteorological conditions of the studied buoys.