Comparing distribution of harbour porpoise using generalized additive models and hierarchical Bayesian models with integrated nested laplace approximation

Species Distribution Models (SDMs) are used regularly to develop management strategies, but many modelling methods ignore the spatial nature of data. To address this, we compared fine-scale distribution predictions harbour porpoise (Phocoena phocoena) using empirical aerial-video-survey data collected along east coast Scotland in August and September 2010 2014. Incorporating environmental covariates that cover habitat preferences prey proxies, a traditional (and commonly implemented) Generalized Additive Model (GAM), two Hierarchical Bayesian Modelling (HBM) approaches Integrated Nested Laplace Approximation (INLA) model-fitting methodology. One HBM-INLA modelled gridded space (similar GAM), other dealt more explicitly continuous Log-Gaussian Cox Process (LGCP). Overall, predicted distributions three models were similar; however, HBMs had twice level certainty, showed much finer-scale patterns distribution, identified some areas high relative density not apparent GAM. Spatial differences due how accounted for autocorrelation, clustering animals, between discrete vs. space; consequently, analyses likely depend on scale at which results, needed. For large-scale analysis (>5–10 km resolution, e.g. initial impact assessment), there was little difference results; insights into (<1 km) from HBM model LGCP, while computationally costly, offered potential benefits refining conservation or mitigation measures within offshore developments protected areas.