Quantitative evaluation of standard and enhanced feature extraction from individual biological sounds recorded in coral reef environments

Passive acoustic monitoring of three protected coral reef habitats south of St. John Island in the US Virgin Islands was conducted in April, 2019. Four wideband Autonomous Passive Acoustic Monitoring (APAM) packages, each with 12-element hydrophone arrays, were deployed on the 10-m-deep seafloor and recorded continuously over the 9-day experiment. Previously published methods for robust feature extraction of individual biological sounds [Mellinger and Bradbury, 2007] have been applied to these data and compared with enhanced approaches to feature extraction. This talk presents results of quantitative evaluations of the performance of these methods. Feature extraction is categorized as a parameter estimation problem whose performance is quantified by the statistical distribution of estimates, typically summarized by the bias and variance of these distributions. Results for a simulated case first are presented, and then those from the APAM-recorded data. The enhanced approaches show significant improvement over the standard approaches for several of the extracted features.