Introduction & Objectives

Acoustically, the ocean is interconnected and sound travels great distances. Therefore, operational noise in shallow waters may displace coastal animals and drive these species offshore (e.g. North Atlantic right whales during migration), while Navy and/or oil and gas operations may displace resident shelf-break and deep- water animals from important habitats. However, shifts in distribution of animals may also occur due to ecological changes, independent of anthropogenic activities. Teasing apart the drivers of these changes can only be done through a clear, well designed project with a multi-year dataset that enables inference to be drawn before, during and eventually after operations, as well as between sites with low (controls) and high operational activity.

Acoustically sensitive species such as beaked whales inhabit the North Atlantic shelf break region, and all ESA-listed baleen whales, such as the North Atlantic right whale (Eubalaena glacialis), fin (Balaenoptera physalus), blue (Balaenoptera musculus), and sei whales (Balaenoptera borealis) are known to use this area to different extents. NOAA’s Northeast Fisheries Science Center and Scripps Institution of Oceanography (SIO) collaboratively deployed long-term high-frequency acoustic recording package (HARP) passive acoustic monitoring stations at eight sites along the western North Atlantic shelf break from 2015 to 2019 in coordination with the Bureau of Ocean Energy Management (BOEM). Likewise, the US Navy monitored the shelf break region at 3 to 4 sites from 2007 to 2022. Together these combined efforts bring the total to 11 recording sites spanning the U.S. eastern seaboard, from New England to Georgia.

This work is aimed at moving the analytical component forward on a number of key scientific areas including:

1.  Assessing the seasonal and spatial occurrence of baleen whales
2.  Improving automated classification for beaked whales
3.  Assessing effects of anthropogenic noise on beaked whale vocal activity
4.  Assessing the prevalence of seismic survey noise along the eastern seaboard
5.  Novel broad-scale approach to assessing acoustic niche and anthropogenic contributors, and assessing the utility of new acoustic metrics

Technical Approach

HARPS were deployed at 11 sites by the NEFSC and U.S. Navy along the Atlantic continental shelf break of the U.S. beginning as early as 2015. The sites deployed starting in 2015 include Heezen Canyon, Oceanographer Canyon, Nantucket Canyon (three northernmost sites), Norfolk Canyon, Cape Hatteras, and JAX (U.S. Navy deployments). These were expanded in 2016 to include Wilmington and Babylon Canyons north of Cape Hatteras as well as Gulf Stream, Blake Plateau, and Blake Spur south of Cape Hatteras. Each HARP was programmed to record continuously at a sampling rate of 200 kilohertz (kHz) with 16-bit quantization, providing an effective recording bandwidth from 0.01 to 100 kHz. Further details of HARP design are described in Wiggins and Hildebrand (2007).

Analysis of passive acoustic data utilizes manual analysis or specialized detectors, unsupervised learning and neural networks, and other methods. Programs used include MATLAB, Triton, and others.

Progress & Results

Analysis under this effort varies by year as analytical components for key objectives are advanced. This effort required a system to streamline and automate the process of detecting and classifying beaked whale echolocation clicks using deep-learning neural networks. The classification pipeline consisted of multiple steps targeted to efficiently detect beaked whales, often challenging to detect when other species dominate the soundscape. The steps included: 1) a generic detector to detect clicks above a received-level threshold, 2) a discrimination phase to remove dominant non-beaked whale detections, 3) an unsupervised learning to derive clusters of distinct click types based on similarities in the spectral shape, and 4) a trained deep neural network to classify clusters of echolocation clicks based on spectral shape, interclick interval, and click duration. This improved approach was run through all Atlantic shelf break data. Additionally, a refined statistical approach to investigate the potential impact of mid-frequency active sonar (MFAS) on beaked whale acoustic activity in the western North Atlantic was developed for three beaked whale species, which required data to be classified at a very fine scale resolution. 

Another aspect of this project has compared and contrasted the acoustic dataset with visual surveys from the Atlantic Marine Assessment Program for Protected Species (AMAPPS). This analysis showed that combining these methods are better at providing an accurate representation of species presence, while individually they have the potential to miss the occurrence of certain species.

Finally, ecological species modeling and acoustic niche approaches are being applied to this dataset. Fast and reliable methods are required for biodiversity assessments to determine and compare species richness patterns that can be applied in both accessible and remote habitats. The goal for this component of the project is to apply ecological species modeling (as described in Van Opzeeland and Hillebrand 2020) and acoustic niche approaches (Van Opzeeland and Boebel 2018; Weiss et al. 2021) to our large acoustic data set to apply new techniques for understanding species ecology, community structure and acoustic niche interactions between multispecies groups throughout the shelf break data. Code has been developed to explore species richness and community dissimilarity, in addition to conditional inference trees and acoustic niche plots to explore baleen and odontocete species relationships over space (10 HARPs) and time (3 years: 2016–2019). Multi-year patterns and trends were summarized for each species and site using a modified acoustic niche analysis to visualize when and where specific species are acoustically detected. The “acoustic niche” of each species is also visualized as approximate vocalization range for each species and differentiates species that are low-frequency specialists (e.g., blue whales) from species that are high-frequency specialists (e.g., Gervais’ beaked whale). In separating the acoustic presence of individual species, the acoustic niche analysis illustrates the species that comprise marine mammal communities at each site.

A methods paper describing the beaked whale neural net classifier built for North Atlantic HARP data was submitted to Methods in Ecology and Evolution, and is currently in review. Acoustic data from the 2016 HARP deployments were combined with passive acoustic data collected from a towed hydrophone array as part of the AMAPPS to examine niche partitioning of beaked whales in the western North Atlantic. This project was submitted to Marine Ecology Progress Series and is currently undergoing revisions. 

Presentations & Publications

Combining Spatial and Temporal Acoustic Datasets to Examine the Summer Presence of Beaked Whales Off the East Coast of the U.S. (SMM 2022)

Monitoring the acoustic ecology of the shelf break of Georges Bank, Northwestern Atlantic Ocean: New approaches to visualizing complex acoustic data (Marine Policy, 2021)

Exploring movement patterns and changing distributions of baleen whales in the western North Atlantic using a decade of passive acoustic data (Global Change Biology, 2020)

Spatial and seasonal patterns in acoustic detections of sperm whales (Physeter macrocephalus) along the continental slope in the western North Atlantic Ocean (Endangered Species research, 2018)

Using passive acoustic monitoring to document the distribution of beaked whale species in the western North Atlantic Ocean (Canadian Journal of Fisheries and Aquatic Sciences, 2017)

Long-term passive acoustic recordings track the changing distribution of North Atlantic right whales (Eubalaena glacialis) from 2004 to 2014 (Scientific Reports, 2017)