The analysis of spatial data: individual movements and species and community models
Otso Ovaskainen, University of Helsinki
Most observational data sets in ecological research have a spatial component, but analysis of spatial data is challenging. Observations that are made in nearby locations are often similar and consequently the data points are not independent of each other. The presence of spatial autocorrelation can be considered as trouble, as simple statistical tests assuming independence are not valid, or as an opportunity of learning about the biological processes creating a given level of autocorrelation. One point in case is research on animal movement. Consecutive locations in an animal track are necessarily correlated, invalidating e.g. the assumption of independence in models of habitat use. I discuss how this problem can be avoided by analyzing animal movement data either with state-space models or with randomization tests. I also discuss how the level of autocorrelation (or persistence in direction) as well as long-term movement behavior can be summarized for a wide family a models with two parameters only: the characteristic spatial and temporal scales of movement. I then move to species distribution modeling, which I extend to community-level models in two different ways. First, I describe a multivariate regression model that can be used to ask if some species combinations occur more or less often together than by expected by random. This approach is suited for cases where there are much data on few species. Second, for the opposite case of few data on many species, I describe how statistical inference can be improved by gluing the species-specific models together with a higher-level community model.