Estimating spatially-varying density and time-varying demographics with open population spatial capture-recapture: a photo-ID case study on bottlenose dolphins in Barataria Bay, Louisana, USA
From long-term spatial capture-recapture (SCR) surveys, we can infer a population's dynamics over time and distribution over space. It is becoming more computationally feasible to fit these open population SCR (openSCR) models to large datasets and build complex models with spatially-varying density surfaces and time-varying population dynamics. However, there is limited knowledge on how these methods perform when applied to real data, especially in a maximum likelihood framework. We analyze a multi-year photo-identification survey of common bottlenose dolphins (Tursiops truncatus) in Barataria Bay, Louisiana, USA following the Deepwater Horizon oil spill in 2010. Over 2000 capture histories were collected between 2010 and 2019. Using openSCR methods with non-Euclidean distances, we estimate time-varying population dynamics and a spatially-varying density surface for this population. We show that inference on survival, recruitment, and density over time since the oil spill provides insight into increased mortality after the spill, possible redistribution of the population thereafter, and persistent low survival. Issues in the application are highlighted throughout: possible model misspecification, sensitivity of parameters to model selection, and difficulty in interpreting results due to model assumptions and irregular surveying in time and space. For each issue, we present practical solutions including assessing goodness-of-fit, model-averaging, and clarifying the difference between quantitative results and their qualitative interpretations. Overall, this case study serves as a practical template other analysts can follow and extend; it also highlights the need for further research on the applicability of these methods as we demand richer inference from them.