Discriminating and classifying odontocete echolocation clicks in the Hawaiian Islands using machine learning methods

Abstract

Passive acoustic monitoring (PAM) has proven a powerful tool for the study of marine mammals, allowing for documentation of biologically relevant factors such as movement patterns or animal behaviors while remaining largely non-invasive and cost effective. From 2008–2019, a set of PAM recordings covering the frequency band of most toothed whale (odontocete) echolocation clicks were collected at sites off the islands of Hawaiʻi, Kauaʻi, and Pearl and Hermes Reef. However, due to the size of this dataset and the complexity of species-level acoustic classification, multi-year, multi-species analyses had not yet been completed. This study shows how a machine learning toolkit can effectively mitigate this problem by detecting and classifying echolocation clicks using a combination of unsupervised clustering methods and human-mediated analyses. Using these methods, it was possible to distill ten unique echolocation click ‘types’ attributable to regional odontocetes at the genus or species level. In one case, auxiliary sightings and recordings were used to attribute a new click type to the rough-toothed dolphin, Steno bredanensis. Types defined by clustering were then used as input classes in a neural-network based classifier, which was trained, tested, and evaluated on 5-minute binned data segments. Network precision was variable, with lower precision occurring most notably for false killer whales, Pseudorca crassidens, across all sites (35–76%). However, accuracy and recall were high (>96% and >75%, respectively) in all cases except for one type of short-finned pilot whale, Globicephala macrorhynchus, call class at Kauaʻi and Pearl and Hermes Reef (recall >66%). These results emphasize the utility of machine learning in analysis of large PAM datasets. The classifier and timeseries developed here will facilitate further analyses of spatiotemporal patterns of included toothed whales. Broader application of these methods may improve the efficiency of global multi-species PAM data processing for echolocation clicks, which is needed as these datasets continue to grow.

Odontocete spatial patterns and temporal drivers of detection at sites in the Hawaiian islands

“…The false killer whale class presents an opportunity to explore this in some detail. Due to low precision values (many misclassifications with noise), manual review was completed for all false killer whale detections, and this manually modified data set was used moving forward (Ziegenhorn et al, 2022). When comparing this timeseries to the precision‐modified timeseries, we might have used otherwise, it is clear that noise detections would have created artificial patterns.…”

“…These types were identified to species level where possible based on known records and auxiliary data from the region. Then, types were used as classes to train a neural network‐based classifier, which was run on all data to label the entirety of the data set as either one of the echolocation click types or noise (e.g., Frasier, 2021; Ziegenhorn et al, 2022).…”

“…Bins were retained for timeseries only if they had more than a certain number of "true" clicks (>50 for delphinids, >20 for beaked whales and Kogia spp., based on clicking rates for these species). Further detail on network precision and accuracy can be found in Ziegenhorn et al (2022). For false killer whales, neural-network classifications included many false detections from noise sources, which were removed manually in lieu of this process.…”

“…These recordings include data from two subsites at a remote island in the Northwestern Hawaiian Islands (Manawai, otherwise known as Pearl and Hermes Reef), where records of odontocete spatiotemporal trends are limited compared to the Main Hawaiian Islands. A recent study that applied a machine learning toolkit to this data set resulted in labeled data for eight groupings of odontocetes (Ziegenhorn et al, 2022). These groupings included five species‐specific labels: false killer whale ( Pseudorca crassidens ), rough‐toothed dolphin ( Steno bredanensis ), short‐finned pilot whale ( Globicephala macrorhyncus ), Blainville's beaked whale ( Mesoplodon densirostris ), and Cuvier's beaked whale ( Ziphius cavirostris ).…”

“…), temporal analyses and documentation of patterns are relatively novel, as few previous studies have had sufficient detections of these species to describe patterns. In the case of rough‐toothed dolphin, description of spatiotemporal trends based on a novel click type described by Ziegenhorn et al (2022) will represent one of very few descriptions of temporal patterns in this species' behavior. Even in cases where strong patterns in presence have been documented, additional data, particularly data arising from differing methodologies, can prove useful by filling in knowledge gaps.…”