Estimating Change in Humpback Whale Song Repertoire Through Occupancy Modeling and Phrase Transition Analysis

There are 14 distinct population segments of humpback whales worldwide, each known to sing its own unique, population-specific song. The structure of these songs is hierarchical: songs are made up of themes, themes consist of phrases, and phrases are composed of individual units. Over time, the occurrence and sequence of these components gradually change. These slow, progressive changes drive song evolution within a population through cultural transmission. Additionally, song changes can spread between populations, with the extent of transmission depending on the level of connectivity between them. Therefore, understanding song change is important in conservation, as the degree of song change can reflect the degree of population connectivity, but measuring and quantifying the degree of change is often challenging due to the complexity of the song itself and discrepancies in recording effort. In this thesis I approach the study of humpback whale song change using two methods: (1) I test the application of occupancy modeling to estimate the pace of song change by focusing on the smallest component of the song, the unit, across 17 years, and (2) I use a transition probabilities approach to measure changes in song phrase occurrence and sequence across two breeding seasons. The first approach accounts for differences in sampling effort, which can influence unit detection, and was applied to recordings from the Breeding Stock G (BSG) humpback whale population. The second approach involves an exhaustive sampling effort using passive acoustic monitoring that spans the beginning, middle, and end of the breeding season. This method was applied to recordings from the Central American (CA) population. Both populations have important breeding areas in Panama and Costa Rica, where the data was collected. The results of this thesis indicate the plausibility of combining both methods in the study of humpback whale song change. The occupancy modeling approach effectively tracked the presence and absence of BSG whales’ song units over time and identified recording duration and sea surface temperature as the most influential features on probability detection. The transition probability approach provided insights into both within-season and between-season song variation in Central American (CA) humpback whales. Unlike the gradual changes typically observed in other populations, seasonal changes in the CA whales' songs were rapid. In Season 2, 70% of the phrases were entirely new. In season 1, phrase transitions were more variable at the beginning of the breeding season, while in season 2 the song remained relatively stable. The combination of both approaches can significantly improve our understanding of song changes and allow researchers to predict the direction of such changes in future.