Prior to CEDM performance, it is necessary to compile a list of species records (i.e. geographic coordinates) and of values of climatic variables which are suitable to describe the actual range of the target species. Fortunately, GIS-maps of environmental variables help to find suitable data sets in many regions (e.g. http://www. worldclim.org). A subsequent step involves the development of a multidimensional view of the climate envelope of a species, which is a challenge given the complex nature of a species' ecological niche (Peterson and Vargas 1993). In simple words, in CEDMs as mentioned here, climatic information for the presence of species localities are summarized to an 'ideal' climatic niche for the target species which is afterwards compared to climatic conditions at localities of query, i.e. from which the presence or absence of the target species is unknown. The results are geographic maps showing different degrees of similarity (interpreted as suitability) of a region with the 'ideal' climatic niche. The selection of a moderate algorithm for the computation of a CEDM depends on the amount of distribution records available and their quality.
One of the earlier applied algorithms for the presence of species data is BIOCLIM (Nix 1986; Busby 1991), as implemented in DIVA-GIS 5.4 (Hijmans et al. 2002; http://www.diva-gis.org). It develops CEDMs by intersecting the ranges inhabited by the species along each environmental axis. More sophisticated algorithms are GARP (Stockwell and Noble 1992; Stockwell and Peters 1999), DOMAIN (Carpenter et al. 1993), Maxent (Phillips et al. 2006) and LIVES (Li and Hilbert 2008). The machine-learning Maxent (see below) is often superior to most other methods and hence becomes more and more distributed (e.g. Elith et al. 2006). If records of apparent species absence are available for modeling other algorithms, such as 'artificial neuronal networks', 'classification and regression trees', 'generalized additive models' or 'generalized dissimilarity models' can be applied. These algorithms for example are implemented in the BIOMOD tool (Thuiller 2003).
Unfortunately, the high degree of endemism in the Pantepui region causes difficulties when using CEDMs at the species level, since a minimum amount of 10-20 distribution records pending on the algorithm applied is at least necessary (e.g. Elith et al. 2006). Looking at the herpetofaunal assemblages, 109 out of 159 (i.e. 97 amphibian and 62 reptile) species are only known from a single tepui (McDiarmid and Donnelly 2005). As a result, even in this relatively well documented group in the region, for most species CEDMs cannot be seriously generated. Since species distributions are commonly related to specific habitats, one possibility is to perform CEDMs for habitats rather than for a single species using a random set of distribution points within a target habitat. Such an approach was well performed by Carnaval and Moritz (2008) for the identification of climatically stable refugia during the Last Glacial Maximum (LGM), 21,000 years BP, in the Brazilian Atlantic forest.
Was this article helpful?