North Atlantic right whale shift to the Gulf of St. Lawrence in 2015, revealed by long-term passive acoustics

Abstract

This paper contributes to documenting a change in the distribution of North Atlantic right whales Eubalaena glacialis (NARWs) that occurred in the 2010s, when the whales largely abandoned their traditional summering grounds in the Gulf of Maine/Bay of Fundy/Scotian shelf. Data from a year-round passive acoustic monitoring (PAM) network in the Gulf of St. Lawrence were exploited to build the time series of NARW incursions into this inland sea of the Northwest Atlantic, from June 2010 to November 2018. NARWs visited the southern Gulf of St. Lawrence every year from June to January, until ice freeze-up. The earliest detections were made at the end of April and the latest in mid-January. Call occurrence peaked between August and the end of October. NARW contact calls were not detected at the most upstream station at Les Escoumins, in the Lower St. Lawrence estuary, or at the northeastern connection of Belle Isle Strait with the Atlantic, which was monitored from November 2010 to November 2011. The mean daily occurrence of NARWs in the feeding grounds off Gaspé quadrupled after 2015 compared to 2011−2014. Long-term continuous PAM data provided invaluable information to document this marine mammal distribution shift.

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

“…However, by restricting our attention to upcalls with SNR 4.0, the recall of the average model was increased to 85% for the same level of precision while retaining over 95% of the upcalls. Existing algorithms are capable of achieving similar levels of recall, but at the cost of a significantly higher false-positive rate (DCLDE 2013;Simard et al, 2019). Finally, we note that a related work entitled "Deep neural networks for automated detection of marine mammal species" (Shiu et al, 2020) has been published during the review of our paper.…”

“…The DC performance of these classical signal processing methods under actual in situ recording conditions tends to plateau around a detection probability of about 50% (i.e. recall index) when the false detection probability is kept below about 10% (DCLDE 2013;Simard et al, 2019). The objective of the present study is to test if modern machine-learning approaches can break this apparent DC performance ceiling.…”

“…This distribution, however, has changed in the last decade (Davis et al, 2017). The occasional yearly occurrence of a few individuals in the Gulf of St. Lawrence in summer and fall markedly increased in 2015 and a high seasonal occurrence has continued since (Simard et al, 2019). This area is a site of intensive fixedgear fishing.…”

“…Several algorithms, exploiting the time-frequency structure of the call, have been used so far (Baumgartner et al, 2011;Gillespie, 2004;Mellinger, 2004;Simard et al, 2019;Clark, 2006 , 2007;Urazghildiiev et al, 2008). Their performance is dependent of the signal to noise ratio (SNR), which varies with the range of the calling whale, the noise levels and the other biological and instrumentation factors (Gervaise et al, 2019b;Simard et al, 2019). The DC performance of these classical signal processing methods under actual in situ recording conditions tends to plateau around a detection probability of about 50% (i.e.…”

Assessing the lethality of ship strikes on whales using simple biophysical models

“…These smaller vessels generally do not use automatic identification systems (AIS), thus producing less track‐data compared to larger vessels and they are not subject to the regulations to mitigate ship strike lethality (Transport Canada, 2018). This suggests it is a mistake to focus solely on large vessels, particularly given recent observations of right whales in coastal eastern Canadian waters, where smaller vessels are used in local fisheries that are active at times while the whales are present (Davies & Brillant, 2019; Simard, Roy, Giard, & Aulanier, 2019).…”