Classification Approaches for Modeling Animal Species Distributions

• Activity Number: 261
• Type: Topic Contributed
• Date/Time: Tuesday, August 10, 2004 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Thomas C. Edwards*+
• Companies: Utah State University
• Keywords: classification models ; species presence absence ; gap analysis ; conservation planning

Abstract

Classification techniques are powerful tools for modeling linkages of animal species with habitat type, especially when coupled with spatially explicit environmental data structures. Unfortunately, a common problem confronting such models is adequate presence absence data for model building. Too often absence data, irrespective of whether collected from a design or purposive-based approach, are lacking, essentially eliminating the application of classification tools for modeling purposes. Several different approaches exist for generating so called "pseudo-absence" data, each with its own assumptions and resultant impacts on the final classification models. I evaluate the effect of two different approaches for generating "pseudo-absence" data on resultant animal distribution models, comparing: (1) random absences generated from within the known range of collected presence data, and (2) a weighted approach based on density isoclines of presence data. Example species for analysis and evaluation come from the Utah Gap Analysis data structure, as well as other environmental data structures used in ongoing conservation planning.