Abstract

Spatial dependency is a common issue in the modeling of ecological data, especially in the surveying of animal distributions. In this paper, we argue that when we make such an inference we should account for both environmental factors and spatial correlation. We show how this can be done by using hierarchical generalized linear models (HGLMs), which allow us to model wide classes of spatial dependencies. In a real data set of reindeer fecal pellet-group count at sample locations in a northern Swedish forest, we found that over 70% of counts were zeros. Analyzing this data set we show that the proposed HGLM-based models can perform better than the other commonly used models, e.g. ordinary Poisson model and spatial hurdle model, in modeling spatially correlated count data with excessive zeros.

Keywords

Excessive spatial correlation, HGLM, pellet-group count, reindeer habitat preference

Published in

Working papers in transport, tourism, information technology and microdata analysis
2013, number: 2013:06
Publisher: Högskolan Dalarna

Authors' information