Environmental Engineering Reference
In-Depth Information
accordingly several solutions have been tailored to these problems (presence-only
data, low information content of binary data, spatial autocorrelation, multi-collin-
earity, model unidentifiability). Thirdly, species distribution modelling (SDM) is
“sexy”. As habitat of many species is continually lost, as climate changes and as
environmental management becomes a matter of human survival, scientists, deci-
sion makers and the general public look for information and predictions of possible
future scenarios. Consequently, substantial funding (at least for ecological topics)
over the last decade has enabled talented scientists to make a career from SDMs.
Aims of This Chapter
Recent developments have made the field of SDM somewhat complex, diverse and
confusing for the newcomer. The aim of this chapter is thus to (1) provide a recipe
for SDM; (2) briefly discuss a few selected “hot” topics; and (3) give an overview of
challenges of a more ecological modelling type (dispersal, occupancy, biotic inter-
actions, functional variables, evolution, changing limiting resources). I shall restrict
citations to fundamental or specific methodological papers and will therefore have
to ignore the vast amount of good ecological papers that “only” did it right. On the
other hand, I am not aware of any paper on species distribution modelling that could
tick all elements of the recipe below.
13.2 A Species Distribution Modelling Recipe
A good cook needs no recipe. Alas, we are trained more in ecology than statistics.
Moreover, without the right ingredients (a.k.a. data) and tools (software), no dish
will be tasty. Also, I should mention other recipes along this line: see Harrell (2001)
for a generic statistical recipe, and Pearson (2007) and Elith and Leathwick (2009a, b)
for a specific one on SDMs. As for “cooking tools”, I highly recommend using
code-based software so that each step of the analysis is documented and easily repro-
ducible. The functions mentioned in this chapter are all from the free R environment for
statistical programming (R Development Core Team 2008).
The recipe falls into three sections: pre-processing, modelling and model inter-
pretation (Fig.
13.1
). These sections are somewhat arbitrary, but are useful to
structure the whole endeavour. We shall assume that you have your ingredients
well prepared: The observed data are as good as we need them, the explanatory
variables are ecologically relevant and at the same resolution and your statistical
tools are laid out in front of you. A worked example is available at
http://www.
mced-ecology.org
(“Where's the sperm whale?”), which follows the recipe and
provides example data and R-code.
