Temporal variations in blue whale call types in the northeast Pacific at diel, seasonal, and interannual time scales with tag-derived behavioral context

The blue whale (Balaenoptera musculus) is a highly-migratory, endangered cetacean found throughout the California Current Large Marine Ecosystem. This species is difficult to observe directly for extended periods, making passive acoustic monitoring (PAM) one of the most viable methods for studying their vocal behavior over large temporal and spatial scales. However, the enormity of PAM datasets maintained for many seasons and years to address questions over a variety of temporal scales requires the development of robust automated techniques for identifying specific call types. Furthermore, additional behavioral data streams are required to place identified patterns of blue whale vocal behavior in a broader behavioral context. Here, we present the novel combination of traditional energy detection methods and machine learning classification to minimize both false positives and false negatives in automated call detection. We employ this approach to quantify received blue whale A, B, and D calls in ~4 years (fall 2015 – fall 2019) of nearly-continuous PAM data from a cabled, real-time hydrophone centered in Monterey Bay National Marine Sanctuary (MBNMS). We quantify distinct patterns in call usage at interannual, seasonal, and diel scales. Interannual analyses reveal a strong peak in blue whale vocal activity in 2017. Seasonal analyses demonstrate seasonal covariation of A and B calls, lagging the seasonal cycle in D calls by two months. Seasonal modulation of diel variation in production of B calls is also evident, with a preference for nighttime call production early in summer but little diel variability in late fall. We further contextualize these patterns in vocal behavior via tag-derived behavioral information. In combination with tag-based analysis, we find that anomalous PAM detection rates of D calls are correlated with the formation of anomalously large feeding aggregations of blue whales. Tag-based behavioral analysis also suggests that the PAM-detected seasonal peak and modulation of diel B call behavior is due to late fall behavioral state changes from a feeding to reproductive focus in calling individuals found in MBNMS. Finally, we consider the oceanographic context of these behavioral patterns and their implications for conservation questions regarding this endangered species.