Environmental Engineering Reference
In-Depth Information
Table 15.1 Input and output files used or created by Qrule
Routine
Unit Name
Purpose
main.f90
10
rulerun.log
Output file recording programme input and statistical summary
of results
main.f90
11
patch_cfd.
dat
Output data file of cumulative frequency of patch sizes
(compressed mode)
main.f90
12
variable
User-defined name of output map of individual habitat patches
main.f90
13
variable
User-defined name of output map of rank-ordered sizes of
habitat patches
main.f90
14
variable
User defined output map labeled by habitat type
genmap.
f90
15
variable
User named input file of landscape map for analysis by Qrule
main.f90
16
assmat.dat
Output file of adjacency matrices
main.f90
17
stats.csv
Comma delimited output file of summary statistics (landscape
metrics)
genmap.
f90
19
arcgrid.map Output file of Qrule generated map for input into ArcInfo
main.f90
20
lacun.dat
Output file of summary results from lacunarity analysis
landscape pattern lies beyond the 95% confidence region generated by a sufficiently
large set of Monte Carlo iterations, then one may be confident that the observed
patterns are statistically different from the random patterns at
0.05 (Gardner
and O'Neill 1990; Pearson and Gardner 1997). This statistical comparison is, of
course, subject to Type II error (Zar 1996) when multiple metrics are employed to
describe landscape patterns. One may avoid this important pitfall by first forming a
specific question per the example of Krummel et al. (1987), selecting a single
appropriate metric (or limited subset of metrics) and making the appropriate
statistical test(s). Multivariate approaches reduce the dimensionality of the analysis
(Fauth et al. 2000), providing a more succinct summary and avoiding the problem
of correlated parameters (Riitters et al. 1995; Wang and Malanson 2007), but the
utility of multivariate statistics still depends on the formation of a specific testable
a priori
question.
There is extensive literature in landscape ecology that has focused on the
development and interpretation of landscape metrics (e.g. Gustafson 1998; Hargis
et al. 1998; Li et al. 2005; Neel et al. 2004; O'Neill et al. 1987; Wickham and
Riitters 1995). However, the usefulness of robust statistical testing for comparing
neutral models with actual landscapes remains under-appreciated. Consider the
direct effects of habitat fragmentation on the frequency distribution of patch sizes
for a given cover type. The effects of a small amount of habitat loss will have
dramatically different effects depending on
p
, the amount of habitat that exists on
the landscape: when
p
is high, the effects of habitat loss on the frequency distribu-
tion of cluster sizes is small, however, when
p
is ~0.6 and near the critical threshold
defined by percolation theory, small changes will have dramatic effects on the
frequency distribution of cluster sizes (Gardner et al. 1987). Landscape metrics
are generally poor indicators of these effects because they are usually a single-
numbered, averaged value (see Li et al. 2009 for an exception). The alternative is an
a ΒΌ
