Integrating multiple observation data sources to facilitate movement modelling of killer whale pods

Animal movement modelling is vital to the conservation and protection of animal populations across the world, but models must be flexible and diverse to account for the different ways animals move and the different methods we use to observe them. Most widespread techniques for modelling movement rely on observation data that are at least fairly consistent in their temporal sampling rate and measurement error variance (e.g., from electronic tracking devices). Animal populations lacking such data require modifications to traditional models if their locomotive behavior is to be sufficiently understood. We develop such a model and fit it to a diverse collection of observation data tracking the movement paths of Southern Resident killer whales in the Salish Sea, a core region within their distribution. We expanded upon existing continuous-time models for animal movement, which generate results that are invariant to the sampling rate of the input data, by assuming separate measurement error distributions for different data types, applying a sub-algorithm to automatically detect and downweight potential outliers, and incorporating the movements of a group of animals into one single model. Our model effectively handles irregular time gaps, different observation methods (and associated error distributions), and even the group-oriented fashion with which the whales move. Despite most observations merely indicating the presence of some individuals in the group, or simply reporting the presence of the entire group without location estimates for each individual, our model was still able to parameterize qualities of the whales’ movements, including average speed, directional autocorrelation, and spatial “spacing” between pod members. Our results provide quantitative support for existing estimates of movement speed and directionality for this population and can be used to generate realistic and flexible simulations of killer whale movement paths. Our methodological developments expand the applicability of continuous-time movement modelling to a broader array of animal populations where traditional data collection techniques are not possible.