PARENT SESSION
Thursday, August 10, 1:30-5:00 pm
COS 91 - Spatial ecology II
Mississippi, Mezzanine Level, Cook Convention Center
Presiders: V Chavez-Varela and B van Wesenbeeck

Dealing with spatial autocorrelation in ecological data: a review and a user's guide.

Dormann, Carsten^*,1, ¹ UFZ Centre for Environmental Research, Leipzig, Leipzig, Saxony, Germany

ABSTRACT- Spatial data present a challenge in ecology. While some inroads have been made with respect to allowing non-statistician to analyse spatial data, these methods are by no means widely available or common standard. To illustrate the effect of spatial autocorrelation on species distribution analysis, I review those published studies that compared spatial and non-spatial models. The type of organism investigated had surprisingly little influence on the discrepency between the two model types. Next, I present the results of a workshop on Analysing Spatial Ecological Data, where several different methods to incorporate spatial autocorrelation into statistical models were compared, both for normally distributed and non-normally distributed data (autocovariate regression, autoregressive models, generalized least squares, generalised estimation equations, spatial filtering, Bayesian models, as well as generalised additive models). These different methods were evaluated on a common data set with known statistical properties. Their performance as well as their assumptions will be discussed and a decision tree on which method to use for which data will be presented. For all these methods the model syntax (in the free statistical software R: www.r-project.org) is available, making spatial analyses accessible also to non-statisticians.

Key words: species distribution, statistics, spatial autocorrelation