COS 28-8 - Choosing biologically meaningful variables for modeling species distributions

Tuesday, August 3, 2010: 10:30 AM
412, David L Lawrence Convention Center
Leone M. Brown, Odum School of Ecology, University of Georgia, Athens, GA, Jorge Velasquez-Tibata, Ecology and Evolution, Stony Brook University, Stony Brook, NY, Wolfgang Buermann, Center for Tropical Research, UCLA Institute of Environment, Los Angeles, CA and Catherine Graham, Dept. Ecology and Evolution, Stony Brook University, NY
Background/Question/Methods

Species distribution models (SDMs) are widely used in ecology and conservation, and are useful for addressing questions about species distributions under past, current and future conditions. Initially built using only environmental information, SDMs now often include species-specific information. However, for rare species in poorly-studied regions, we lack such information. Further, as data layer availability increases, modelers must decide which among multiple variables are relevant to the study species. One question is choosing variables meaningful for species about which little is known. We addressed this by considering two major processes that may influence species distributions: historical or biogeographical factors, and habitat choice. We chose bioregion demarcations and current vegetation to represent these processes. We modeled 50 threatened Colombian bird species, using a suite of bioclimatic and remote sensing (RS) variables. We also ran models based on points located only within specific bioregions or vegetation types. We compared the variables most often selected for determining species distribution models to those most often chosen for vegetation type and bioregion models, expecting an influence of the aforementioned processes on species distributions to be evident based on the variables most influential in a particular bioregion or vegetation type, versus those most influential for species distributions.

Results/Conclusions

Assuming our variables accurately capture these processes, species distributions were influenced by a combination of broader variables, and variables specific to the suite of species modeled. Percent tree cover was most often chosen for vegetation models, and elevation for bioregion models. Measures of canopy complexity were most often chosen for species models, followed by elevation and percent tree cover. The variable chosen most often for species models did not change even when all points were found within one bioregion. This suggests species distributions are most influenced by vegetation structure, but variables representing processes hypothesized to drive species distributions also play a role. This supports previous literature documenting the importance of vegetation structure for bird species, and the utility of high-resolution RS variables in modeling species distributions. Lastly, conservation organizations often use habitat types or other environmental surrogates to prioritize regions for conservation, but aiming to conserve specific regions or broad landscapes may not conserve environmental aspects needed for species persistence. Considering variables representing characteristics of species and processes leading to their distributions may improve the utility of species distribution models in conservation planning by capturing information relevant to habitat complexities upon which taxa, such as birds, rely.

Copyright © . All rights reserved.
Banner photo by Flickr user greg westfall.