Environmental Engineering Reference
In-Depth Information
changes with urbanization, one proposed indicator of urban
resilience, the biodiversity of a site or geographic area, is often
negatively affected as habitats are altered or removed entirely.
Human activities change the selective forces acting upon the biota
existing on a site, causing some species to decline and eventually
go extinct while others expand, colonize, and thrive in these same
areas (Marzluff, 2001; McKinney, 2002).
Many researchers are interested in how the biodiversity of
an area will change with urbanization (Schumaker et al ., 2004;
Hepinstall, Marzluff and Alberti, 2009). To determine how an
area's biota may respond to land cover changes, several steps
are required. First, the habitat requirements of a species must
be known, generally the result of field studies conducted to
document existing species occupancy in relation to vegetation
types and structure (or more generally, land cover). From these
habitat relationships we can develop a habitat model. Modeling
a species' habitat, however, is not a trivial task as each species
has specific habitat requirements that may differ: (1) at different
times of the year (i.e., breeding versus non-breeding season); (2)
in different portions of the species range; and (3) depending on if
the species is declining and not occupying all ''suitable'' habitat
or is increasing and occupying marginal habitat. Second, once
the habitat requirements of a species are determined, we must be
able to map these habitats across large areas. Third, we must be
able to predict and map how these habitats will likely change in
the future so that we may predict how species will likely change
in response to future habitat availability.
While these steps may seem simple, the reality is much more
complicated. For example, species occupancy of a site (MacKen-
zie et al ., 2006) while determined by many factors, is primarily
related to the amount of habitat present on a site. Terrestrial
vertebrate habitats can be mapped through the use of remotely
sensed data and while the focus of such work has generally been
on non-urban areas (Villard, Trzcinski and Merriam, 1999),
increasingly studies have focused on urban ecological relation-
ships (Blair, 1996, 2004; Donnelly and Marzluff, 2004, 2006;
Marzluff, 2005). Many studies have clearly documented that as
land cover composition and configuration change, habitat for
species changes as well. However, some species respond to the
direct loss of habitat while others are also sensitive to change
in the spatial arrangement of remaining habitat (e.g., fragmen-
tation, isolation, distance between patches of habitat, amount
of edge versus interior habitat). While there is a rich literature
of studies investigating the effects of habitat loss and fragmen-
tation on species occupancy (McGarigal and McComb, 1995;
Fahrig, 1999, 2003; Villard, Trzcinski and Merriam, 1999; Betts
et al ., 2006), the specific form (i.e., statistical distribution) of
these relationships is often unknown.
Once empirical data relating species with habitat are available,
models of species occupancy can be developed. Here I discuss two
basic types of spatially explicit models relating species to habitat.
With one type, species-habitat association models, species are
associated with different vegetation types through a set of rules
such as those used in habitat suitability index (HSI) models
and Gap analyses (Scott et al ., 1993) where simple deductive
relationships between habitat elements (e.g., canopy closure,
tree height, tree species) and a species use of these elements is
scaled to an index ranging from 0 - 100. These rule-based mod-
els are quite general and suffer from many problems (Roloff
and Kernohan, 1999). The second general category of models is
those that develop statistical relationships between field obser-
vations and explanatory variables (i.e., inductive models). The
statistical form of the model can be quite varied depending on
the response variables. Many different statistical techniques are
in common use today including logistic regression (Betts, Forbes
and Diamond, 2007, Hepinstall, Marzluff and Alberti, 2009),
classification and regression trees and random forests (Prasad,
Iverson and Liaw, 2006; Cutler et al ., 2007), and climate enve-
lope models (Pearson and Dawson, 2003), to name but a few of
the more common techniques. Several review articles have been
published that categorize modeling approaches (Guisan and
Zimmermann, 2000; Guisan, Edwards and Hastie, 2002,). These
two broad types of modeling approaches each have their place.
Rule-based models may often be preferable to statistical models
because of the ease of interpretation and the ability to illustrate
basic principles. Statistical models, conversely, may allow the
exploration of specific hypothesis such as the relative influence
of habitat amount versus habitat configuration in determining
species presence on a site. Statistical models, however, also suffer
from errors and error propagation (Conroy et al ., 1995) and diffi-
culty in interpreting model results. For example, spatially explicit
population models (SEPMs) attempt to relate species habitat
requirements to demographic rates (survival, reproduction, dis-
persal) to model populations (Dunning Jr. et al ., 1995). The
added complexity of SEPMs necessary to model demographic
processes requires many more parameter estimates and there-
fore includes many more potential sources of error. While these
types of models represent an improvement in understanding the
complexity of ecological systems, often the data requirements
preclude their application for more than a few select species.
Models that are derived from empirical data, but do not
attempt to model specific demographic processes, are more likely
to be available for coupling with land cover change models.
For example, Schumaker et al . (2004), developed species-habitat
models for 279 species present in their western Oregon study area
that were subsequently used to rank different future scenarios
of land cover change (Hulse, Branscob and Payne, 2004). They
used rule-based models of species-habitat associations because
it was impossible to develop more specific habitat models for
each species from existing data. Instead, expert knowledge was
used to develop suitability rankings from 1 to 10 for each of 34
habitats for each species. These rankings were then modified by
up to 50 different adjacency rules to incorporate the importance
of landscape context.
If the goal is to model a smaller subset of species, it may
be possible to develop more refined species models to use as
input into predictions of land cover change. For example, Hepin-
stall, Marzluff and Alberti (2009) used 6437 point count surveys
across 992 locations in 139 study sites to develop linear regression
models predicting community measures (species richness, guild-
specific community richness) and population measures (relative
abundance of individual species) for a subset of the avian com-
munity found on the sites. The regression models used predictor
variables that were output from land use and land cover models
that had been developed concurrently (Hepinstall, Alberti and
Marzluff, 2008).
25.1.3 Predicting future land use and
land cover
The next step in the process is to develop a model that will
produce a spatially explicit map of the future. Many different
Search WWH ::




Custom Search