Environmental Engineering Reference
In-Depth Information
Why Species Distribution Modelling?
There are several fundamental challenges to this approach (e.g. first and foremost
that it is correlative; see Vaughan and Ormerod 2005 and Dormann 2007b for a
recent critique), and before jumping into the analysis, it is worth considering
whether SDMs are actually useful and fit for the purpose of our specific problem.
For example, at very small spatial scales, differences in environmental conditions
may be too small to be of predictive value and biotic interactions (competition,
predation) may be of crucial importance. In contrast, at the global scale, data
become so coarse that we “only” model the climate niche and specific habitat
requirements cannot be detected.
On the other hand, SDMs try to extract ecological information from a species
occurrence pattern when and where it matters. Expert knowledge usually cannot
inform us which trait or limitation will be relevant for our problem at hand. We may
know that a palm tree does not survive sub-zero temperatures, but the observed
distribution will tell you that even 10
10 km grid squares with minimum
temperatures well below 0
C harbour this species because of microclimatically
suitable places. Thus, at the spatial resolution under investigation, the physiological
threshold can be misleading even though it may be true. Overall, SDMs are useful
for complementing existing approaches in at least these five areas of research:
1. Small-extent, decision-support for conservation biology (such as Biological
Action Plans: Zabel et al. 2003, and numerous others)
2. Testing specific hypotheses, e.g. on the spatial scale of habitat selection (Graf
et al. 2005; Mackey and Lindenmayer 2001), the species-energy hypothesis
(Lennon et al. 2000) or range-size effects on diversity pattern (Jetz and Rahbek
2002)
3. Generating hypotheses, e.g. on correlation of species traits with environmental
variables (K
uhn et al. 2006), which can then be tested experimentally
4. Identifying hierarchies of environmental drivers (Bjorholm et al. 2005; Borcard
and Legendre 2002; Pearson et al. 2004)
5. Prospective design of surveys, e.g. optimizing sampling schemes for rare species
(Guisan et al. 2006)
€
Now, we shall focus on the technical side and assume that you know what you
are doing, ecologically speaking.
Analysing the geographic distribution of species' occurrence, abundance or
diversity is, essentially, a statistical task. As such, the fundamental ideas and
principles of good statistics apply (and can be found in the excellent but advanced
book of Hastie et al. 2009). There are at least three reasons why methods for
describing or modelling these patterns have reached a higher level of sophistication
than many other fields in ecology. Firstly, biogeographical data sets are nowadays
large (both in terms of number of data points and potentially explanatory variables),
necessitating the use of new statistical strategies. Secondly, species distribution
data typically carry a largish bunch of common intrinsic statistical problems and
