Description
In recent years there has been an uptick in the volume of research addressing the social behaviors of animals, specifically non-herd movement patterns. There have been many notable approaches, mostly centered around the modeling of interconnected networks of animals. Developing such models has been favored as the social relations of animals in a group cannot be easily disregarded when attempting to analyze their movement patterns. Subsequently, creation of hierarchical Bayesian approaches has emerged, which seek to model the social relations and their influence on movement patterns. A primary source of information about social behavior has been movement data which has become significantly more accessible in recent years as aerial drone footage has become more omnipresent. This thesis contends with the social movement patterns of Long-Beaked Common dolphins. An unfortunate drawback of using aerial movement data with dolphins is the matter of labeling. As individuals are quick to move around, they often move outside a drone’s field of view and are subsequently relabeled with a new label upon reentry. This leads to concerns over multiply labeled movement data, which recent works have been addressing, and which is rather computationally taxing. The computational challenges of prior works was the impetus of the current work, which seeks to use subsampling of small groups of individuals to infer results for underlying behavioral parameters for an entire population. An augmented simple random sample was used to obtain six different sample sizes ranging from as few as three concurrently tracked individuals to as many as eight. Bayesian hierarchical models were fit to the data for each of these different sample sizes, and posterior distributions were compared. The primary objective was to see how small of a subsample would yield meaningful posterior distributions. There were on average 40 individuals present throughout the observation period. It was found that a subsample of six individuals was sufficient to obtain adequate posteriors considered representative of the entire network of individuals present. Thus, this sampling approach has further developed the possibility of modeling larger groups of dolphins than previously attempted.
