Assessing Health of Ecological Systems: Models for Spatially Correlated, Multivariate Ordinal Data

• Activity Number: 21
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Jennifer Hoeting*+ and Erin Schliep
• Companies: Colorado State University and Colorado State University
• Keywords: Bayesian ; spatial ; ordinal ; MCMC ; categorical

Abstract

We examine the problem of spatially correlated, multivariate ordinal data in the assessment of health of an ecological system. A common approach is to use an index to combine various biological or physiochemical metrics that are typically ordinal-scaled variables. For example, stream health is often measured by the Index of Biotic Integrity (IBI), which is the weighted-sum of a collection of metrics at a site. Indices such as these are criticized because the weights assigned to each metric are determined arbitrarily. In this work, we overcome this shortcoming. We propose a latent model for multivariate ordinal data and consider several extensions. We demonstrate our methods using a US Environmental Protection Agency supported study of wetland health in Colorado. Our aim is to predict the spatial distribution of wetland type and ecological condition. In achieving this goal, we estimate the unobserved, or latent, health of each wetland location as predicted by a suite of potential threats and stressors.