Testing of ambient noise forecasting algorithms with LADC noise data

Abstract

Long-term, omnidirectional underwater acoustic noise was measured by three buoys in the northern Gulf of Mexico over a 36-day period during the summer of 2001 as part of the Littoral Acoustic Demonstration Center project, phase I. This extensive data set is used to develop and evaluate algorithms for near-term temporal prediction of noise, based on immediate in situ data. Noise time series data in one-third-octave bands from 10 to 5000 Hz are processed to extract characterizing parameters such as mean level, variance, and temporal coherence across all frequency bands. The fluctuation spectrum is analyzed to determine how the noise variability is distributed with temporal period, and eigenvectors based on the fluctuation covariance matrix are used to identify underlying basic properties of noise variation. These tools are then used to predict future values of noise based on measurements at preceding times. [Research supported by ONR.]