Parameterization of the oceanic ambient noise using orthogonal decomposition.

The article evaluates the application of empirical orthogonal function (EOF) decomposition for processing and predicting ambient noise spectra in the open ocean, using data collected in the Southern Ocean. It demonstrates that one to three EOFs can explain 89% to 98% of the ambient noise variability, enabling efficient gap-filling of missing or corrupted data with approximately 35% higher accuracy than linear interpolation. EOF-based regression models significantly reduce the number of parameters needed to predict ambient noise from wind speed, improving computational efficiency while maintaining prediction accuracy comparable to more complex neural network methods. The study highlights that prediction errors are primarily driven by inaccuracies in forecasting the overall ambient noise magnitude rather than its spectral shape, suggesting that EOF decomposition effectively captures spectral distribution but that improved magnitude prediction is essential for enhanced ambient noise forecasting.