Parameterization of the oceanic ambient noise using orthogonal decomposition

Utility of the empirical orthogonal function (EOF) decomposition for the processing and prediction of ambient noise in the open ocean is assessed. Using ambient noise observations in the Southern Ocean, it is found that 89%, 96%, and 98% of the ambient noise spectrum can be explained by the first, two, and three EOFs, respectively. This provides motivation for using singular value decomposition for the analysis, and pre-processing of the ambient noise observations, which usually include outliers and gaps. It is also shown that EOF-based re-filling of observational gaps provides approximately 35% higher accuracy than the conventional linear interpolation. EOFs significantly decrease the number of the regression parameters required for the ambient noise prediction using wind speed or other predictors. This allows for the prediction of the continuous ambient noise spectrum in computationally efficient ways and avoids subdividing ambient noise spectra into a limited number of frequency bands. It is also found that the accuracy of the wind-based ambient noise prediction is essentially controlled by the prediction of the frequency-averaged ambient noise magnitude so that accurate prediction of the ambient noise magnitude would decrease the RMS of the ambient noise spectral prediction from 4 to 1.6 dB.