Characterizing and predicting species distributions across environments and scales: Argentine ant occurrences in the eye of the beholder
Article first published online: 8 OCT 2008
© 2008 The Authors. Journal compilation © 2008 Blackwell Publishing Ltd
Global Ecology and Biogeography
Volume 18, Issue 1, pages 50–63, January 2009
How to Cite
Menke, S. B., Holway, D. A., Fisher, R. N. and Jetz, W. (2009), Characterizing and predicting species distributions across environments and scales: Argentine ant occurrences in the eye of the beholder. Global Ecology and Biogeography, 18: 50–63. doi: 10.1111/j.1466-8238.2008.00420.x
- Issue published online: 9 DEC 2008
- Article first published online: 8 OCT 2008
- Argentine ant;
- ecological niche models;
- Linepithema humile;
- model averaging;
- spatial grain;
- species distribution models
Aim Species distribution models (SDMs) or, more specifically, ecological niche models (ENMs) are a useful and rapidly proliferating tool in ecology and global change biology. ENMs attempt to capture associations between a species and its environment and are often used to draw biological inferences, to predict potential occurrences in unoccupied regions and to forecast future distributions under environmental change. The accuracy of ENMs, however, hinges critically on the quality of occurrence data. ENMs often use haphazardly collected data rather than data collected across the full spectrum of existing environmental conditions. Moreover, it remains unclear how processes affecting ENM predictions operate at different spatial scales. The scale (i.e. grain size) of analysis may be dictated more by the sampling regime than by biologically meaningful processes. The aim of our study is to jointly quantify how issues relating to region and scale affect ENM predictions using an economically important and ecologically damaging invasive species, the Argentine ant (Linepithema humile).
Location California, USA.
Methods We analysed the relationship between sampling sufficiency, regional differences in environmental parameter space and cell size of analysis and resampling environmental layers using two independently collected sets of presence/absence data. Differences in variable importance were determined using model averaging and logistic regression. Model accuracy was measured with area under the curve (AUC) and Cohen's kappa.
Results We first demonstrate that insufficient sampling of environmental parameter space can cause large errors in predicted distributions and biological interpretation. Models performed best when they were parametrized with data that sufficiently sampled environmental parameter space. Second, we show that altering the spatial grain of analysis changes the relative importance of different environmental variables. These changes apparently result from how environmental constraints and the sampling distributions of environmental variables change with spatial grain.
Conclusions These findings have clear relevance for biological inference. Taken together, our results illustrate potentially general limitations for ENMs, especially when such models are used to predict species occurrences in novel environments. We offer basic methodological and conceptual guidelines for appropriate sampling and scale matching.