Ecological niche modelling is important for a variety of applications in ecology and conservation (Graham et al. 2004). For example, they attempt to provide detailed distribution cartography related to the spread of invasive species (Thuiller et al. 2005), impacts on climate change (Thomas et al. 2004, Matsui et al. 2009), and spatial patterns of species diversity (Graham et al. 2006, Feldemeyer-Christe et al. 2007). 

Figure 2. Potential distributions of Buxus balearica obtained by the application of MAXENT, MARS, GARP and ENVIRONMENTAL DISTANCE (ENV_DIST)

  As such, ecological niche modelling has been the target for an impressive growth in attention in recent years (Guisan & Zimmermann 2000, Soberón & Peterson 2004, Graham et al. 2004, Thomas et al. 2004, Araújo et al. 2005, Thuiller et al. 2005, Elith et al. 2006, Matsui et al. 2009) placing this technique among emerging new approaches relevant to ecology, biogeography, and conservation biology.

  The basic approach of these models is to combine a set of known occurrences together with prediction variables such as topographic, climatic, edaphic, biogeography and remotely sensed ones. Accurate occurrence data (presence and absence) are rarely available, especially for rare species or inaccessible site locations. Correlative models using species presence and absence locations records for habitat predictions have been referred to as discrimination techniques, while those using only species presence records have been referred to as profile techniques (Caithness 1995).

  The aim of this contribution is to provide a comparison between modelling algorithms to study the habitat distribution of such a fragmented species as Buxus balearica Lam. by evaluating the prediction capability of different spatial distribution models at a local scale.

  The prediction models were tested based on 17 environmental variables. Six methods were compared: Multivariate adaptative regression spline, MARS, Maximum entropy approach to modelling species’ distributions (Maxent), two generic algorithms based on environmental metrics dissimilarity (Bioclim) and Genetic Algorithm for Rule-set Prediction (GARP), and supervised learning methods based on generalized linear classifiers (Support vector machines-SVMs). As a test of the predictive power of the models we used Kappa index.

  Results:

• The predictive power was better for Maxent, following by GARP models. 
• All the other models tested obtained lower accuracy values. 
• By comparing the predictive power of Models, climate variables showed the highest contributions of the environmental variables. 
• The variables with lowest contributions were the terrain isolation models. 
• A test of sensitivity to the reduction in the number of variables obtained an accuracy of over 0.90 by applying just 3 climatic variables (spring rainfall, mean temperature of warmest month, and mean temperature of coldest month). 
• All algorithms produced maps that coincided well with the known distribution of the species.

• References

Navarro-Cerrillo, R. M., Hernandez-Bermejo, J. E. & Hernandez-Clemente, R. (2011). Evaluating models to assess the distribution of Buxus balearica in southern Spain. APPLIED VEGETATION SCIENCE, 14(2), 256-267

Oyonarte, R. M. (2003). Aboveground phytomass models for mayor species in shrub ecosystems of western Andalusia.. Investigación Agraria. Sistemas y Recursos, 12 (3), 47-