Estimation of wave spectral shapes using ANN

Representation of surface spectral density against various frequencies of ocean waves has applications in structural design, laboratory wave simulations and coastal wave modeling. Estimation of such a wave spectrum is currently made with the help of empirical relations, like, Pierson-Muskowitz and JONSWAP. These relationships were proposed using statistical curve fitting to observed data. This paper attempts to provide an alternative to such empirical theoretical spectra in the form of neural networks. Networks were developed in order to estimate shapes of the wave spectra out of the given values of representative wave height, period, spectral width and peakedeness parameter. The data collected by wave rider buoys, deployed at stations off the US as well as the Indian coast have been analyzed. The neural network-based spectral estimations were found to be more close to the 'true' spectra than the traditional empirical ones. It is argued that they may be used instead as substitute.