ABSTRACT
BACKGROUND: The heterogeneous oceanographic conditions of continental shelf ecosystems result in a three-dimensionally patchy distribution of prey available to upper-trophic level predators. The association of bio-physical conditions with movement patterns of large marine predators has been demonstrated in diverse taxa. However, obtaining subsurface data that are spatio-temporally relevant to the decisions made by benthically-foraging species can be challenging. METHODS: Between 2009 and 2015, grey seals were captured on Sable Island, Nova Scotia, Canada during summer and fall and instrumented with high-resolution archival GPS tags. These tags recorded location data as well as depth (m), temperature (°C), and light level measurements during dives, until animals returned to the haulout site to breed. Hidden Markov models were used to predict apparent foraging along movement tracks for 79 individuals (59 females, 20 males) every 3 h. In situ measurements were used to estimate chlorophyll-a concentration (mg m- 3) and temperature within the upper-water column (50 m) and temperature and depth at the bottom of dives. As chlorophyll-a could only be estimated from 10:00 to 14:00 AST for dive depths ≥50 m, we formulated two generalized linear mixed-effects models to test the association of predicted grey seal behavioural states with oceanographic conditions and phytoplankton biomass: the first representing conditions of the upper-water column likely to influence primary productivity, and a second model including environmental conditions encountered by grey seals at the bottom of dives, when seals were more likely to be foraging. RESULTS: Predicted grey seal behavioural states were associated with fine-scale chlorophyll-a concentrations and other environmental conditions they encountered across the continental shelf. In the Water Column Model, season had no influence on the probability of observing apparent foraging, but chlorophyll-a, upper-water column temperature, and sex did, with females having a greater probability of foraging than males. In the Bottom Conditions Model, again season had no influence on the probability of apparent foraging, but females were over twice as likely as males to be foraging. CONCLUSIONS: The results of this study highlight the value of in situ measurements of oceanographic properties that can be collected at high temporal resolution by animal-borne data loggers. These data provide insight into how inferred behavioural decisions made by large marine predators, such as the grey seal, may be influenced by fine-scale oceanographic conditions.
ABSTRACT
Longitudinal models are commonly used for studying data collected on individuals repeatedly through time. While there are now a variety of such models available (marginal models, mixed effects models, etc.), far fewer options exist for the closely related issue of variable selection. In addition, longitudinal data typically derive from medical or other large-scale studies where often large numbers of potential explanatory variables and hence even larger numbers of candidate models must be considered. Cross-validation is a popular method for variable selection based on the predictive ability of the model. Here, we propose a cross-validation Markov chain Monte Carlo procedure as a general variable selection tool which avoids the need to visit all candidate models. Inclusion of a 'one-standard error' rule provides users with a collection of good models as is often desired. We demonstrate the effectiveness of our procedure both in a simulation setting and in a real application.
