UC San Diego

Studies of marine mammals using passive acoustic monitoring (PAM) tools are becoming more and more common. This methodology allows for documentation of biologically relevant factors such as movement patterns or animal behaviors while remaining largely non-invasive and cost effective. In the Hawaiian Islands, a set of PAM recordings covering the frequency band of most toothed whale (odontocete) echolocation clicks were collected from 2008-2019 at sites off the islands of Hawaiʻi, Kauaʻi, and Pearl and Hermes Reef (otherwise known as ‘Manawai’). 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. In this dissertation, a machine learning toolkit was used to effectively mitigate this problem by detecting and classifying echolocation clicks using a combination of unsupervised clustering methods and human-mediated analyses. Classified clicks were distilled into timeseries of species’ presence in order to document, and propose reasons for, observed patterns. Habitat modelling employing Generalized Additive Models (GAMs) with and without Generalized Estimating Equations (GEEs) was used to elucidate these trends in combination with oceanographic variables. The machine learning pipeline used distilled eight unique echolocation click types, attributable to eight or more species of odontocetes. Species composition differed amongst considered sites, and this difference was robust to seasonal movement patterns. Temporally, hour of day was the most significant predictor of detection across species and sites, followed by season. When considered in conjunction with sea surface variables, temperature had the strongest relationship to detections. Of the climate indices considered, El Niño Southern Oscillation (ENSO) may have the most effect on species detections at monitored sites. This study demonstrates that PAM is an invaluable tool in studies of oceanic top predators, and that machine learning tools can mitigate issues related to the size and complexity of PAM datasets. Using these tools and habitat modelling analyses, we can gain valuable insights into top predator behavior in relation to temporal variables, surface conditions, and long-term climate indicators.