Optimizing passive acoustic systems for marine mammal detection and localization: Application to real-time monitoring north Atlantic right whales in Gulf of St. Lawrence

WhaleMap: a tool to collate and display whale survey results in near real-time

“…It has demonstrably improved conservation outcomes for endangered whales in this region by optimizing research activities, facilitating dynamic risk-mitigation measures, and engaging with the ocean industry and the public. WhaleMap has also already been cited in several scientific publications (Baumgartner et al, 2020;Gervaise et al, 2021;Johnson et al, 2020;Koubrak et al, 2021;Kowarski et al, 2020). We are not aware of any equivalent software in existence.…”

Performance of a deep neural network at detecting North Atlantic right whale upcalls

“…The spectrogram representation was then computed on a dB scale using a window size of 0.256 s, a step size of 0.032 s (88% overlap), and a Hamming window. These parameters have been shown to be optimal for identifying NARW upcalls (Gervaise et al, 2019a) and produce a spectrogram with (time, frequency) dimensions of 94 × 129. We note that the spectrograms were fed to the network in their raw form.…”

Data_Sheet_3.docx

“…Visual review will likely be required to some degree when dealing with acoustic detection of rare species such as North Atlantic right whales and/or for ensuring data quality. Standard performance metrics require evaluation and reporting, such as precision, recall, and accuracy, as well as false detection, false positive, false omission and missed detection rates (Figure 6) (e.g., Baumgartner et al, 2019;Kirsebom et al, 2020;Madhusudhana et al, 2020;Gervaise et al, 2021). PAM software detectors comprise a wide range of custom-built computer programs, aimed at automating the process of detecting target species' calls in a dataset [see review in Bittle and Duncan (2013), Shiu et al (2020), Gervaise et al (2021].…”

“…Standard performance metrics require evaluation and reporting, such as precision, recall, and accuracy, as well as false detection, false positive, false omission and missed detection rates (Figure 6) (e.g., Baumgartner et al, 2019;Kirsebom et al, 2020;Madhusudhana et al, 2020;Gervaise et al, 2021). PAM software detectors comprise a wide range of custom-built computer programs, aimed at automating the process of detecting target species' calls in a dataset [see review in Bittle and Duncan (2013), Shiu et al (2020), Gervaise et al (2021]. In both PAM archival and real-time data analysis, in addition to any software detector(s) used, some level of visual confirmation by an acoustic expert often still remains essential to improve accuracy and minimize error in call type reporting.…”

“…Localizing calling species during the construction phase of offshore wind projects would be very useful for satisfying mitigation requirements regarding the location and distance of the species in question, relative to the sound source (e.g., pile driving location). Localization can be carried out by the placement of multiple fixed or mobile omnidirectional hydrophones arranged in a configuration that allows for localization of the vocalizing animal using the difference in the time of arrival of a call (or calls) on multiple time-synchronized sound recorders (e.g., Stanistreet et al, 2013;Hastie et al, 2014;Risch et al, 2014;Gillespie et al, 2020;Gervaise et al, 2021). It can also be achieved by using multiple sensors that can calculate bearing [e.g., directional autonomous seafloor acoustic recorders (DASARs); Greene et al, 2004;Blackwell et al, 2007;Mathias et al, 2012;Blackwell et al, 2013].…”