Environmental Engineering Reference
In-Depth Information
lead the National Research Council to declare this problem an urgent scientific
challenge (NRC 2003).
There are many reasons why the study of ecosystem change at landscape studies
is a daunting task: Ecosystem dynamics result from complex interactions of numer-
ous physical and biological attributes that vary in space and time. Each landscape
has a unique, and often unknown history whose effects may be subtle and long
lasting and experimental manipulations of landscapes are nearly impossible (the
“N
1” problem). Consequently, models of landscape dynamics have become an
integral part of landscape ecology. However, the use of a landscape model presents
its own special issues. Each new problem requires the selection of an appropriate
(and often new) model; the parameters required by the model are often difficult to
know or are estimated with high uncertainty; and the errors associated with
predictions are impossible to determine. Hagen-Zankder and Lajoie (2008) and
Gardner and Urban (2007) have noted an additional problem: the geographic
constraints of landscapes (i.e. the boundary conditions of map extent, the presence
of rivers and lakes, etc.) may restrict the dynamic range of model output, resulting
in alternative model formulations producing similar sets of predictions. These
constraints result in confusion with regards to the causes and effects of altered
landscape dynamics. Under all of these conditions a neutral modeling approach can
be of great value.
The ideal neutral landscape model is a simple construct, which generates
landscape pattern without including specific physical and biological processes of
interest. The magnitude of difference of landscape data, or more complex process-
based models from the neutral model results is a measure of the important role that
these processes may play in the development of landscape pattern (Gardner et al.
1987; Pearson and Gardner 1997). The simplest neutral model is a random map
(Gardner et al. 1987) that produces patterns based on a single parameter,
p
i
, the
fraction of the map occupied by each habitat type,
i
. Although the appearance of
random maps obviously differs from actual landscapes, the lessons learned have
been diverse and significant (see Gardner and Urban 2007 for a review).
The background for the development of neutral landscape models has a number
of key elements. The first occurred when colleagues at Oak Ridge National Labora-
tory began analyzing USGS digital land use and land cover (LUDA) data (Krummel
et al. 1987). The USGS LUDA database (Fegeas et al. 1983) originated from NASA
U2/RB-57 high-altitude aerial photo coverage in 1973, which were subsequently
hand-delineated into 1 of 37 land cover categories. For the first time, these data
provided a spatial description of habitat distribution for the entire U.S. Krummel
et al. (1987) selected the Natchez Quadrangle for analysis, 1:250,000 quadrangle
composed of 24 separate sections. To analyze this quadrangle a special computer
programme was written (remember, this was pre Arc-Info days!) to reformat the arc
and node topology and remove section boundaries. It was evident that the shape of
habitat boundaries changed with scale. George Sugihara suggested that fractal
geometry could be used to characterize patch shape. The large size of the data
set allowed a moving-window regression analysis to detect the scale at which the
fractal index changed.
¼
