Modelling species distributions using regression quantiles
Authors: Vaz S, Martin CS, Eastwood PD, Ernande B, Carpentier A, Meaden GJ, Coppin F
Publication Year: 2008
Reference: Journal of Applied Ecology, 45(1):204-217 (February 2008) (Published online 18 October 2007)
. Also available as IIASA Interim Report IR-07-056 www.iiasa.ac.at/Admin/PUB/Documents/IR-07-056.pdf
Abstract
Species distribution modelling is an important and well-established tool for conservation planning and resource management. Modelling techniques based on central estimates of species responses to environmental factors do not always provide ecologically meaningful estimates of species -environment relationships and are being increasingly questioned.
Regression quantiles (RQ) can be used to model the upper bounds of species -environment relationships and thus estimate how the environment is limiting the distribution of a species. The resulting models tend to describe potential rather than actual patterns of species distributions.
Model selection based on null hypothesis testing and backward elimination, followed by validation procedures, are proposed here as a general approach for constructing RQ limiting effect models for multiple species.
This approach was applied successfully to 16 of the most abundant marine fish and cephalopods in the eastern English Channel. Most models were validated successfully and null hypothesis testing for model selection proved effective for RQ modelling.
"Synthesis and applications." Modelling the upper bounds of species-habitat relationships enables the detection of the effects of limiting factors on species' responses. Maps showing potential species distributions are also less likely to underestimate species responses' to the environment, and therefore have subsequent benefits for precautionary management.
KEYWORDS: distribution models; geographical information systems; habitat; limiting factors; marine fish
Regression quantiles (RQ) can be used to model the upper bounds of species -environment relationships and thus estimate how the environment is limiting the distribution of a species. The resulting models tend to describe potential rather than actual patterns of species distributions.
Model selection based on null hypothesis testing and backward elimination, followed by validation procedures, are proposed here as a general approach for constructing RQ limiting effect models for multiple species.
This approach was applied successfully to 16 of the most abundant marine fish and cephalopods in the eastern English Channel. Most models were validated successfully and null hypothesis testing for model selection proved effective for RQ modelling.
"Synthesis and applications." Modelling the upper bounds of species-habitat relationships enables the detection of the effects of limiting factors on species' responses. Maps showing potential species distributions are also less likely to underestimate species responses' to the environment, and therefore have subsequent benefits for precautionary management.
KEYWORDS: distribution models; geographical information systems; habitat; limiting factors; marine fish