Biology Reference
In-Depth Information
where species, or suites of species, are most likely to appear on the landscape.
Because the development of these predictive models can be costly, prioritization is
often employed first to narrow a large list of candidate species to a manageable
number (Fig. 2.2). This is often done at relatively local scales such as parks and
reserves.
The development of predictive models can increase search efficiency by focus-
ing searches on areas that are most likely to be invaded. Predictive models can also
be used to estimate the threat posed by specific species and thus can be integrated
into the prioritization process. Regardless of scale, the goal of predictive models is
to identify sites where invasive species are most likely to occur. Models can be
developed for individual species as well as groups of species (Guisan et al. 1999;
Underwood et al. 2004; Ferrier and Guisan 2006). Good predictive models substan-
tially reduce the enormous amounts of resources required to detect populations
before they become established or before nascent populations begin to expand
(Rejmanek and Pitcairn 2002).
The uses of predictive models in wildlife management and other areas of con-
servation are extensive (Ejrnaes et al. 2002; Scott et al. 2002; Guisan and Thuiller
2005). In contrast, despite a plethora of research predicting what species are likely
to be invasive and what communities are likely to be invaded (Rejmanek 1989;
Reichard and Hamilton 1997; Daehler and Carino 2000; Kolar and Lodge 2001;
Rejmanek et al. 2005; Krivanek and Pysek 2006), the modeling of invasive species
distributions has been relatively limited until only recently (Peterson 2003; Rouget
et al. 2004; Underwood et al. 2004; Thuiller et al. 2008).
2.5.1 Types of Predictive Modeling Approaches
There are two general approaches for predicting which species will likely become
invasive in an area. One is based on decision trees, usually with binary answers
(yes/no) to a series of questions on species biogeography, biology/ecology, and
traits generally considered to be legitimate indicators of invasiveness (Daehler et al.
2004; Pheloung et al. 1999; Reichard and Hamilton 1997). The number of ques-
tions can range from a few (e.g., 7; Reichard and Hamilton 1997) to many (e.g., 50;
Pheloung et al. 1999). In many ways, this approach resembles prioritization with
the use of decision trees and ordinal scores. It is simple in concept and has proven
effective in predicting species likely to colonize a large geographic area (e.g., a
country or state) and become invasive (Krivanek and Pysek 2006).
The other approach is based on statistical models using geo-referenced environ-
mental data at sites where a species is known to occur and, ideally, also where it
does not occur. Standard environmental data are correlated with species distribution
and abundance patterns including climate, topographic, soil, and land cover varia-
bles (Table 2.3). Some of these variables directly influence species distribution pat-
terns (e.g., soil pH, light), while others indirectly influence patterns (e.g., elevation,
aspect). In addition, invasive species biologists have identified other variables that
