156
Geographic Information Science and Ecological Assessment
ability of various data sets for our modeling purposes, one important criterion was its spatial scale
or resolution.
11.4.2 Scale
Scale, both spatial and temporal, has been an important theme in science generally and ecology
specifically (Allen and Star, 1982; Weins, 1989;
Levin, 1992). Relevant to this chapter is the spatial scale or resolution (grain) and extent of geographical data used for studying Earth system
processes (Meentemeyer, 1989; Lam and Quattrochi, 1992; Ehleringer and Field, 1993; Foody
and Curran, 1994; Stewart et aI., 1996; Quattrochi
and Goodchild, 1997; van Gardingen et aI., 1997).
Whereas Goodchild (1992b) noted that cartographic scale reflects a view of the database and is
not a property of the database itself or of geographical reality, ecological processes may have
characteristic spatiotemporal scales, and spatial
databases may have been collected, modeled, or interpolated at a specific resolution. Therefore, we
must exercise caution and good judgment in using
spatial data to model an ecological distribution,
drive a process model, or combine with data of
greatly different resolution in a cartographic model
(in which digital map layers are combined using
Boolean operators). A global-scale soils database
might be useless for predicting the distribution of
a locally endemic plant in San Diego County or driving the CENTURY model for eastern Colorado (see
Chapter 18). Generalized land ownership data for
the state of California would not be used to design
a habitat conservation area in the city of Escondido.
A 1-meter-resolution DEM with centimeter vertical accuracy, developed from kinematic GPS
(global positioning system) survey, would not be
needed to model energy balance for a large watershed (Dubayah, 1994), but might be necessary to
find suitable breeding habitat for an endangered
bird nesting in a coastal salt marsh (Brewster,
1996).
11.4.3 Error and Accuracy
Spatial data accuracy has been a research focus in
geographical information science for more than a
decade (Goodchild and Gopal, 1989; Veregin,
1989; Lunetta et aI., 1991). Issues of data and metadata quality are important to all types of scientific
data management (Chapter 5), but some aspects of
spatial data are unique in this regard (also see Section 11.6). One is that maps on paper or computer
screens, with their sharp boundaries, interesting
colors, and references to real places, seem to carry
more authority than tables, pie charts, or histograms
when viewed by land managers, decision makers,
or the public (see Monmonier, 1996). This is in
spite of the fact that they represent environmental
phenomena, such as vegetation cover or water quality, with precision and accuracy that vary spatially
and may not be fully specified. Furthermore, a map
may be the output of a model (cartographic overlay, statistical, process, or other) built on several
mapped input variables whose respective errors are
propagated through that model.
Second, maps are models (representations) approximating complex spatial distributions (geographical reality sensu Goodchild, 1992b) that can
vary continuously or with abrupt boundaries. These
approximations are made with some uncertainty.
Although it is highly desirable to quantify the level
of uncertainty, as with confidence limits around an
estimate, it can be very difficult with spatial data
(Goodchild et aI., 1992). But the important conceptual issue is that the "error" in digital maps of
biophysical variables is not so much a cartographic
"mistake" (Goodchild, 1988) as it is uncertainty in
a spatially distributed estimate, akin to uncertainty
or variance in other scientific measurements
(Goodchild, 1994).
A number of methods, both simple and sophisticated, now exist for quantifying error or uncertainty in maps of continuous and categorical variables and in maps resulting from overlay operations
(Goodchild et aI., 1992; Goodchild et aI., 1993;
Heuvelink and Burrough, 1993; Veregin, 1995).
Also, the effects of geographical data uncertainty
on the output of spatially explicit process models
can be explored through simulation and sensitivity
analysis (Lodwick et aI., 1990; Stoms et aI., 1992;
Journel, 1996). Error analysis or sensitivity analysis is an essential part of every EA that uses GIS
data and modeling for spatial decision support, but
one that is frequently overlooked.
11.5 Data Analysis
In the classic definition of GIS presented in the introduction of this chapter, data analysis was deemphasized. This accurately reflects the development of the technology and software, which in large
part placed the challenges of data structures, format, storage, input, and output ahead of analysis
(Goodchild, 1988; 1992a).1t was (and is) often necessary to move data out of a GIS software package
in order to execute statistical or process models.
GIS and statistical software systems have lacked
Geographic Information Science and Ecological Assessment
ability of various data sets for our modeling purposes, one important criterion was its spatial scale
or resolution.
11.4.2 Scale
Scale, both spatial and temporal, has been an important theme in science generally and ecology
specifically (Allen and Star, 1982; Weins, 1989;
Levin, 1992). Relevant to this chapter is the spatial scale or resolution (grain) and extent of geographical data used for studying Earth system
processes (Meentemeyer, 1989; Lam and Quattrochi, 1992; Ehleringer and Field, 1993; Foody
and Curran, 1994; Stewart et aI., 1996; Quattrochi
and Goodchild, 1997; van Gardingen et aI., 1997).
Whereas Goodchild (1992b) noted that cartographic scale reflects a view of the database and is
not a property of the database itself or of geographical reality, ecological processes may have
characteristic spatiotemporal scales, and spatial
databases may have been collected, modeled, or interpolated at a specific resolution. Therefore, we
must exercise caution and good judgment in using
spatial data to model an ecological distribution,
drive a process model, or combine with data of
greatly different resolution in a cartographic model
(in which digital map layers are combined using
Boolean operators). A global-scale soils database
might be useless for predicting the distribution of
a locally endemic plant in San Diego County or driving the CENTURY model for eastern Colorado (see
Chapter 18). Generalized land ownership data for
the state of California would not be used to design
a habitat conservation area in the city of Escondido.
A 1-meter-resolution DEM with centimeter vertical accuracy, developed from kinematic GPS
(global positioning system) survey, would not be
needed to model energy balance for a large watershed (Dubayah, 1994), but might be necessary to
find suitable breeding habitat for an endangered
bird nesting in a coastal salt marsh (Brewster,
1996).
11.4.3 Error and Accuracy
Spatial data accuracy has been a research focus in
geographical information science for more than a
decade (Goodchild and Gopal, 1989; Veregin,
1989; Lunetta et aI., 1991). Issues of data and metadata quality are important to all types of scientific
data management (Chapter 5), but some aspects of
spatial data are unique in this regard (also see Section 11.6). One is that maps on paper or computer
screens, with their sharp boundaries, interesting
colors, and references to real places, seem to carry
more authority than tables, pie charts, or histograms
when viewed by land managers, decision makers,
or the public (see Monmonier, 1996). This is in
spite of the fact that they represent environmental
phenomena, such as vegetation cover or water quality, with precision and accuracy that vary spatially
and may not be fully specified. Furthermore, a map
may be the output of a model (cartographic overlay, statistical, process, or other) built on several
mapped input variables whose respective errors are
propagated through that model.
Second, maps are models (representations) approximating complex spatial distributions (geographical reality sensu Goodchild, 1992b) that can
vary continuously or with abrupt boundaries. These
approximations are made with some uncertainty.
Although it is highly desirable to quantify the level
of uncertainty, as with confidence limits around an
estimate, it can be very difficult with spatial data
(Goodchild et aI., 1992). But the important conceptual issue is that the "error" in digital maps of
biophysical variables is not so much a cartographic
"mistake" (Goodchild, 1988) as it is uncertainty in
a spatially distributed estimate, akin to uncertainty
or variance in other scientific measurements
(Goodchild, 1994).
A number of methods, both simple and sophisticated, now exist for quantifying error or uncertainty in maps of continuous and categorical variables and in maps resulting from overlay operations
(Goodchild et aI., 1992; Goodchild et aI., 1993;
Heuvelink and Burrough, 1993; Veregin, 1995).
Also, the effects of geographical data uncertainty
on the output of spatially explicit process models
can be explored through simulation and sensitivity
analysis (Lodwick et aI., 1990; Stoms et aI., 1992;
Journel, 1996). Error analysis or sensitivity analysis is an essential part of every EA that uses GIS
data and modeling for spatial decision support, but
one that is frequently overlooked.
11.5 Data Analysis
In the classic definition of GIS presented in the introduction of this chapter, data analysis was deemphasized. This accurately reflects the development of the technology and software, which in large
part placed the challenges of data structures, format, storage, input, and output ahead of analysis
(Goodchild, 1988; 1992a).1t was (and is) often necessary to move data out of a GIS software package
in order to execute statistical or process models.
GIS and statistical software systems have lacked