Continuous monitoring and classification of odontocete echolocation clicks from an ocean observatory

Abstract

Passive acoustic monitoring is a valuable technique for detecting the presence and inferring the activities of odontocetes, but is often limited by power and data-storage constraints of self-contained recorders. Since July 2015, a broadband (10 Hz–128 kHz) hydrophone has been recording continuously at the Monterey Accelerated Research System (MARS), a cabled observatory in Monterey Bay, California, USA. This acoustic record is notable for its combination of duration, bandwidth, and completeness. An automated detector identified more than 200 million unique clicks in this dataset, and extracted a variety of time-domain, spectral, and cepstral features from each click. Depending on the criteria used, clustering algorithms identified 4-8 click classes consistent with those of local odontocete species, including dolphins and beaked whales. Clicking rates were highly variable, with a median of 0.2 clicks minute-1 and a mean of 165 (± 130 standard error) clicks minute-1. Echolocation activity was 1-2 orders of magnitude higher at night than during daytime, and was seasonally higher in fall and winter. Prior work has shown that mesopelagic sound-scattering layers in Monterey Bay are densest at these times of year, suggesting higher availability of prey. Continuous passive monitoring has great potential to improve our understanding of these species’ foraging ecology, especially when integrated with environmental measurements and active acoustic measurements of their prey fields.Passive acoustic monitoring is a valuable technique for detecting the presence and inferring the activities of odontocetes, but is often limited by power and data-storage constraints of self-contained recorders. Since July 2015, a broadband (10 Hz–128 kHz) hydrophone has been recording continuously at the Monterey Accelerated Research System (MARS), a cabled observatory in Monterey Bay, California, USA. This acoustic record is notable for its combination of duration, bandwidth, and completeness. An automated detector identified more than 200 million unique clicks in this dataset, and extracted a variety of time-domain, spectral, and cepstral features from each click. Depending on the criteria used, clustering algorithms identified 4-8 click classes consistent with those of local odontocete species, including dolphins and beaked whales. Clicking rates were highly variable, with a median of 0.2 clicks minute-1 and a mean of 165 (± 130 standard error) clicks minute-1. Echolocation activity was 1-2 orders of ...