RECOVERY OF SURFACE WAVES FROM BOTTOM PRESSURE BY NEURAL NETWORK WITH BISPECTRUM

Abstract

 In order to cope with missing data observed by Doppler-type Directional Wavemeter (DWM) and to maintain stable wave observations, we developed a method to recover surface waves from bottom pressure records observed by pressure gauges equipped on DWM. A neural network was used to estimate the transfer function required for the recovery. For the input data of the neural network, bicoherence was used for the purpose of considering the effects of nonlinearity and multidirectionality of coastal waves. Then, the spectrum of the surface waves were estimated based on the estimated transfer functions, and their accuracy was confirmed by comparing the observed and estimated energies in each frequency band. As a result, it was confirmed that the proposed method can accurately recover the water surface in a wide frequency range including long-period waves.