Ecological modeling in urban environments: predicting changes in biodiversity in response to future urban development - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 344

models have been developed to predict future land use (Schneider
and Pontius, 2001; Parker et al ., 2003; Verburg et al ., 2004; Tang
et al ., 2005; Verburg, Eickhout and Vanmeijl 2008), land cover
(Hepinstall, Alberti and Marzluff, 2008), or combined land use
and land cover. Most ecological processes are influenced by
land cover, so models predicting future land cover are needed.
However, most urban development and transportation models
predict future land development (e.g., land use, development type
and intensity) and not land cover (Parker et al ., 2003; Verburg
et al ., 2004). For example, UrbanSim, an urban development
model based on simulating the interactions between land use,
transportation, and the economy, predicts specific development
events and intensity (i.e., new square feet of commercial land
use, new residential units). While these future events are highly
relevant to the change in urbanization of a geographic region,
they do not directly relate to how the land cover will change
in the amount of impervious area and vegetation (e.g., species,
type, or areal extent) present on a site. Another commonly
used land use model, CLUE-S (Verburg et al ., 2002; Verburg
and Veldkamp, 2004), uses spatial allocation rules to simulate
competition between different land uses for a specific location on
the ground. A third model, GEOMOD2 (Pontius, Huffaker and
Denman, 2004) models future land use based on biophysical and
climate factors (lifezone or ecoregion, elevation, soil moisture,
precipitation) and potential land use. With each of these models,
the output is land use or a combination of land use and land
cover. These outputs must be translated to changes in land cover,
vegetation, or species habitat to be used as input into species
habitat models.
The Land Cover Change Model (LCCM) represents one model
that has been developed to integrate predictions of future land
use (derived from another model: UrbanSim) into predictions
of land cover change (Hepinstall, Alberti and Marzluff, 2008).
Specifically, LCCM uses the output of the UrbanSim economic
development model discussed above as one of many inputs
into a model that predicts changes in land cover. The outputs
from LCCM and UrbanSim are then available as inputs into
biodiversity models (see details below).

25.2 Predicting changes in
land cover and avian
biodiversity for an area
north of Seattle,
Washington

This section details an example an integrated modeling approach
that uses classified remotely sensed data to predict likely changes
in ecological systems in an urban and urbanizing environment.
This process entailed the multiple steps discussed above and
depicted in Fig. 25.1. Observed land cover change was modeled
into the future with LCCM. UrbanSim (www.urbansim.org) was
used to simulate land use change (e.g., change in land use type,
number of residential units, and area of industrial or commercial
buildings). Models of avian biodiversity were developed from
field data using predictor variables that were output by Urban-
Sim and LCCM. Future biodiversity was predicted by applying
the biodiversity model to the predicted future land cover and
land use. Below I go into more detail regarding each step in
the process.

25.2.1 Land cover maps

Land cover maps depicting 14 different land cover classes
were available for 1995, 1999, and 2002 for six counties in
western Washington, USA (Hepinstall-Cymerman, Coe and
Alberti, 2009). These data were developed using Landsat TM
imagery with 30 m cell resolution. Images were segmented
into vegetated, unvegetated, and shadow using spectral mixture
analysis (Rashed et al ., 2003; Powell et al ., 2007). Supervised
classification was used on the different image segments. Multi-
season (i.e., leaf-on and leaf-off) data were used within each year
to separate deciduous and mixed forest from coniferous forest.
Spectral mixture analysis was used to extract three land cover
classes of urban (heavy [ > 80% impervious surface within a
30 m pixel], medium [50 - 80% impervious surface], and light
[20 - 50% impervious surface]). Temporal trajectories were used
to separate bare soil into separate classes of agriculture, clearcut
forest, and cleared for development. Overall classification accu-
racy ranged from 80% to 89% (Hepinstall-Cymerman, Coe and
Alberti, 2009).

25.1.4 Integratingmodels to predict
future biodiversity

Studies that have documented land cover change over time
often link these changes to changes in available habitat for
wildlife species (Pearson, Turner and Drake, 1999; Tworek, 2002;
Turner et al ., 2003; Gude, Hansen and Jones, 2007). Linking
land cover and land use change models to biodiversity can be
accomplished using spatially explicit habitat data and either the
coarse habitat-association models or more sophisticated statis-
tical models discussed above. Often constraints (on gathering
empirical data, computing power, storage) will dictate either few
statistically based high spatial and thematic resolution models
for few species (Dunning Jr. et al ., 1995; Schumaker et al ., 2004;
Hepinstall, Marzluff and Alberti, 2009) or simpler, lower spa-
tial resolution models for many species (White et al ., 1997;
Schumaker et al ., 2004).

25.2.2 Land use changemodel

Both the LCCM (below) and the avian richness models require
variables measuring urban development to run. I derived vari-
ables associated with land use change and intensity of new
development from UrbanSim (Waddell, 2002) using simulations
for 2003 - 2027 (i.e., 25 years into the future) developed for the
Puget Sound Regional Council (P. Waddell, personal communi-
cation) as input to both the LCCM and the avian species richness