Habitat use models of spatially auto-correlated data: a case study of the common bottlenose dolphin, <i>Tursiops truncatus truncatus</i>, in southeastern Brazil

A common approach to studying habitat use in cetaceans is to conduct line-transect surveys, to investigate their distribution. In developing countries, there are limited resources for data collection. One solution is to employ field surveys to collect a wide range of ecological and behavioural data, for which a haphazard sampling schedule is adopted, to optimize the cost–benefit ratio. As with line-transect surveys, the haphazard sampling may lead to spatial autocorrelation (SAC), an overlooked problem in ecology. Here, we investigated common bottlenose dolphins (<i>Tursiops truncatus truncatus</i>) habitat use on an upwelling area and tested an approach that can improve model-based inference on auto-correlated data. We collected data in Cabo Frio, Rio de Janeiro, photo-identified 429 individuals and compared the predictions and model coefficients of standard generalized linear model (GLM) without correcting for spatial autocorrelation with a spatial eigenvector generalized linear model (SEV-GLM) which compensates for SAC. Our best SEV-GLM predicted dolphins are more likely to occur on cold waters with increased chlorophyll concentration, indicating dolphins are influenced by the upwelling. Moreover, by correcting for SAC, our models had a better fit to data, magnified the relevance of significant variables and showed smaller and less clumped residuals than when not correcting.