Detecting marine mammals with an adaptive sub-sampling recorder in the Bering Sea

Adaptive sub-sampling overcomes the hardware and power limitations of continuous recorders by reducing the amount of data acquired and yet maintains a high probability of detection and classification of targeted signals. This enables data collection for a full year at a high sampling rate from a small, relatively lightweight recorder. The Passive Aquatic Listener (PAL) collects short, high temporal resolution (100 kHz) time series, called "sound bites" at intervals on the order of minutes. Power spectra are stored and provide a description of the background physical environment with a temporal resolution of minutes. Sound bites are only saved if they contain targeted signals of interest, e.g. marine mammals, and can be used to provide an audio confirmation for these sound sources. The estimated probability of detecting a particular marine mammal species depends on, among other physical propagation characteristics, (1) vocalization characteristics for a particular species or activity, (2) an estimate of the percentage of time that active sound is produced, and (3) an estimate of the duration of a typical calling bout. A PAL deployment in the Bering Sea demonstrates the detection of marine mammal signals within a dynamic background sound field. In addition to the automatic detections, post-processing of the sound bites revealed secondary marine mammal detections. These secondary detections demonstrate the wealth of information that can be obtained from short sound clips. Species with high vocalization rates and long calling bouts (fin whales, bowhead whales, and bearded seals) were more likely to be recorded incidentally compared to species with low vocalization rates and short calling bouts (right whales). Acknowledging that physical propagation characteristics associated with detectabiliry must also be addressed, this work illustrates how the combination of spectral data and short sound clips obtained with small, low duty cycle recorders can provide the information needed to address questions of species richness, temporal presence of animals, and response to changes in the environment without the large volume of data obtained by continuous recorders.