A New Approach for the Reconstruction of Significant Wave Height Time-Series

A Neural Network (NN) model is proposed for the reconstruction of significant wave height time series, without any increase of the error of the NN output with the number of reconstructed data. The input of the NN model are correlated data, obtained from nearby stations: no data of the same series we are modelling are used. A weighted error function during the learning phase is also considered to improve the modelling of the higher significant wave height. Furthermore the equivalent triangular storm model is applied to test the ability of the NN model to reconstruct the sea storms. The comparison between actual data of a NOAA buoy moored off San Francisco (California) and the data reconstructed by NN model shows a good agreement, both during calm time periods and during storms.Copyright © 2002 by ASME