Characterizing biodiversity in urban areas using remote sensing - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 276

and other data sources. Furthermore, once relationships between
response variables and predictive variables have been established,
the derivedmodels can be used to predict outcomes of alternative
future scenarios.

Furthermore, urbanization has complicated wide-ranging
ecological implications (Grimm et al ., 2008) and in order to
understand and model the ecological processes involved and
their relation to biodiversity, it is necessary to quantify ecolog-
ically significant components of the urban system. Studies that
have used remote sensing to quantify urban land cover in more
detail for this purpose are listed in Section 20.3.2.

20.2.2.1 Predictive environmental
variables in indirect approaches

20.2.2.2 Response variables in indirect
approaches

The predictive environmental variables that can be derived from
remote sensor (and other) data for use in indirect approaches can
include proxies such as coarse land cover classes, or more or less
direct measures of the ecological niche, such as climate, topog-
raphy, hydrology, detailed vegetation types, vegetation structure,
measures of productivity, and disturbance metrics (e.g., Elith
et al ., 2006; Duro et al ., 2007; Saatchi et al ., 2008). Precipitation
data has also been used, for example at 0 . 1 ◦ from NOAA satel-
lites (Pearson et al ., 2007) and 0 . 25 ◦ from the Tropical Rainfall
Mapping Mission (Saatchi et al ., 2008). Building on such vari-
ables, landscape patterns can be quantified via landscape metrics,
using, for example, measures of fragmentation and heterogene-
ity, which can be related to biodiversity response variables (e.g.,
Luoto et al ., 2004).
A common method for obtaining environmental variables
related to biodiversity is land cover classification. Information
on land cover does not correspond directly to biodiversity com-
ponents such as ecosystems, but it can provide very useful basic
information such as the distribution of broad types of forest and
grasslands. However, additional information is often needed in
order to gain ecologicallymeaningful information for biodiversity
studies, such as vegetation types or habitat quality for individual
species, populations or guilds (Groom et al ., 2006; Mucher et al .,
2009). A wide range of land cover classification studies based
on remote sensor data have been undertaken, both for research
purposes, during the collection of environmental data relating to
biodiversity studies, and management (e.g. Gillespie et al ., 2008;
Mucher et al ., 2009).
Low resolution sensors, such as MODIS and AVHRR, with
spatial resolutions of 250 m and 1.1 km respectively, can con-
tribute to biodiversity studies mainly by helping us to understand
ecological processes at regional to global scales, for example
climate-associated vegetation growth patterns (e.g., Lotsch et al .,
2003). These sensors also provide data on temperature, precipi-
tation and fire, which can add to the information derived from
land cover classifications. For urban studies, they have potential
to provide data on global urbanization patterns in relation to this
type of information (Fig. 20.1).
Sensors with intermediate resolutions in the range 10-100 m
include Landsat TM/ETM, SPOT HRV and IRS LISS. Of these,
the NASA Landsat series is the most widely used for biodiversity
related studies due to the ease of obtaining the data, the long time
series and low cost (Gillespie et al ., 2008). Remote sensing studies
using medium resolution sensors with relevance to biodiversity
include classification of landscape types, classification and mon-
itoring of vegetation types, monitoring vegetation degradation
and optimization of land cover information for ecological pur-
poses, from local to regional scales (Groom et al ., 2006; Gillespie
et al ., 2008; Tomppo et al ., 2008; Mucher et al ., 2009). Data
from these sensors have often been used in large-scale land cover
mapping with a general aim to attain 85% accuracy across all
mapping classes (Franklin and Wulder, 2002).

Biodiversity components that can be modeled through indirect
approaches, using remote sensor data, range from detailed pro-
cess models, for example, relating to individual life histories, to
the modeling of broad patterns of biodiversity hotspots (Gontier,
Balfors andMortberg, 2006). Individual-basedmodels (e.g., Top-
ping et al ., 2003) and population viability models (e.g., Akcakaya,
2001) will gain in precision as a result of improvements in remote
sensor data capture and interpretation, but such detailed models
also require a range of parameters to be calibrated before they
can be applied. The modeling of individual movements and
population viability in urban areas and areas becoming urban-
ized is of great importance for improving our knowledge of
biodiversity in this context. However, the application of such
complicated models may be hampered by a lack of information
on parameter values and thresholds, especially in relation to
urban environments (e.g., Gontier, Balfors and Mortberg, 2006).
The modeling of the occurrence of single species is often
called ecological niche or habitat suitability modeling and has
developed rapidly in the field of ecological research (e.g., Scott
et al ., 2002; Guisan and Thuiller, 2005). These models build on
empirical data pertaining to species in the form of presence,
presence-absence or abundance data, or expert knowledge on,
for example, species' habitat preferences, in relation to envi-
ronmental predictors (Gontier, Balfors, and Mortberg, 2006).
Such models have been used for predicting habitat suitability
for single species at the regional level (e.g., Peterson et al ., 2006;
Sober on and Peterson, 2009) and in areas subject to urbanization
(e.g. Mortberg, Balfors and Knol, 2007; Hepinstall, Alberti and
Marzluff, 2008).
In addition, connectivity, indicating the movement potential
for different species on a landscape scale, can be spatiallymodeled
using cost-distance modeling (e.g., Adriansen et al ., 2003) and
graph theory (Townsend et al ., 2009; Zetterberg, Mortberg and
Balfors, 2010). Advances have also been made in the modeling
of species richness and diversity on different scales, in relation to
remote sensor data such as land cover classifications, measures of
productivity, measures of heterogeneity and vegetation indices
(e.g., Luoto et al ., 2004; Leyequien et al ., 2007; Levin et al ., 2007).
The development of spatial ecological models involves
increasingly sophisticated statistical and spatial analyses to study
the distribution of biodiversity components, from univariate or
multiple regression models to general linear models, general
additive models, machine learning methods and geographically
weighted regression analyses, each associated with specific
improvements in accuracy and interpretability (e.g., Scott et al .,
2002; Foody, 2005; Mortberg and Karlstrom, 2005; Guisan and
Thuiller, 2005). All spatial ecological models, once established,
can provide probability maps that can be regarded as spatial
predictions of species distributions and patterns of diversity
within landscapes and regions. These can be applied to the