B0683
Title: Model-based ordination of multivariate vegetation percent cover data
Authors: Pekka Korhonen - University of Jyvaskyla (Finland) [presenting]
Sara Taskinen - University of Jyvaskyla (Finland)
Jenni Niku - University of Jyvaskyla (Finland)
Bert van der Veen - Norwegian University of Science and Technology (Norway)
Francis Hui - The Australian National University (Australia)
Abstract: In recent years, model-based ordination of ecological community data has gained a lot of popularity among practitioners due to the greater availability and utilization of computational resources. In particular, the family of generalized linear latent variable models (GLLVM), a factor-analytic and rank-reduced form of mixed effect models, has proven to be both accurate and computationally efficient when paired with techniques of variational inference and automatic differentiation. GLLVMs have been implemented and used for many response types common to community ecology: presence-absence, biomass, overdispersed and/or zero-inflated counts. The aim is to extend this list to include vegetation cover data. A big challenge with such data comes from the fact that it is often very sparse. The beta distribution, typically used for responses in (0,1), cannot account for zeros (or ones). Thus, some form of augmentation is needed. Two methods are compared, a hurdle beta model and the more recently proposed ordered beta model. Comparisons include simulation studies where the Procrustes errors of the latent variables are assessed and studied based on real data sets to compare predictive performance. In addition to the augmented beta models, the comparisons also include the beta model on shifted responses, the binary model on presence-absences and the ordinal model on cover classes as benchmarks.
