PARENT SESSION
Contributed Oral Session 20: Modeling Animal Populations
Monday, August 8, 1:30 PM - 5:00 PM, Meeting Room 513 C, Level 5, Palais des congrès de Montréal

Spatial autocorrelation and connectivity measures in species distribution models.

van Teeffelen, Astrid^*,1, Ovaskainen, O¹, Moilanen, Atte¹, Wintle, Brendan², Elith, Jane², ¹ University of Helsinki, Helsinki, Finland² University of Melbourne, Melbourne, Australia

ABSTRACT- Models predicting species distributions are popular tools in ecology, for example in conservation planning. However, several statistical methods used to model species distributions (e.g. logistic regression) assume independent observations, whereas ecological processes are typically spatially autocorrelated. Ignoring spatial autocorrelation in modelling distributions introduces uncertainty in model fit and predicting accuracy. Spatial autocorrelation in species distributions can be due to intern processes such as spatial population dynamics, and/or extern processes, such as spatially autocorrelated environmental factors. Therefore it is important to identify whether different measures can account for different spatial autocorrelation patterns. Autologistic regression models account for spatially correlated observations by including a connectivity measure (autocovariate). The autocovariates found in literature closely resemble connectivity measures used in metapopulation models, which are used to predict colonisation and extinction events. When occupancy data is not available for all sites in a study region, only an approximation of the autocovariate can be computed. I present a study where different measures and their approximations in autologistic models were compared, with respect to their ability to explain aggregated patterns in species occurrence, caused by different processes. For comparison, several simulated datasets and datasets on occurrence of a number of birds in the Netherlands were used in this study. The performance of approximations for the autocovariate strongly depends on the cause of spatial aggregation in the data.

Key words: species distribution models, spatial autocorrelation, autologistic regression model, connectivity measures