• Home
• Program
• Search
• Author Index
• Sponsors
• Committees
• Contact Us
• Tech Support

Session:

Posters

Abstract No.:

P-034

Title:

Uncertainty in species distribution modelling ? the use of Mahalanobis distance

Author(s):

J.O. Skøien, European Commission, Joint Research Centre (IT)
G. Dubois, European Commission, Joint Research Centre (IT)
G. B. M. Heuvelink, Wageningen University, Land Dynamics Group (NL)
M. Schulz, European Commission, Joint Research Centre (IT)

Abstract:

The Mahalanobis distance is a simple and commonly used method for species distribution modelling or niche modelling. Assuming that the range of a species depends on a set of environmental indicators, such as elevation, climate and land cover, we compute the mean vector and covariance matrix for these indicators at the training locations, typically a protected area or locations where a species has been observed. The Mahalanobis distance can then be used as a similarity measure for these indicators for a larger region. Locations with high similarity can either be locations where it is likely to find the species, or locations which can be used for relocation of species at threatened locations. Traditional niche modelling using the Mahalanobis distance does not take uncertainty of the input variables into account. In this paper, we discuss how to address uncertainty in the environmental indicators and how this influences the resulting measure of similarity. In case uncertainties in input data are not documented, we will present some suggestions on how the spatial distribution (mean, variance, spatial correlation) of these uncertainties can be derived from the existing data.We will analyze the uncertainty both with analytical methods and a Monte Carlo approach. For the Monte Carlo approach we use sequential Gaussian cosimulations for generation of the realizations of the environmental indicators. Inclusion of uncertainty will typically reduce the areas with high similarity and, at the same time, increase the areas with lower similarities. The approach discussed here was implemented in a Web Processing Service called eHabitat. While such a web based modelling service highlights the challenge of passing uncertain information in a web based model environment (the Model Web), it also shows the advantages of having access to enhanced discovery tools allowing the use of different data sets.