Environmental Engineering Reference
In-Depth Information
0.300
1.75
1.70
0.295
1.65
0.290
1.60
0.285
1.55
0.280
1.50
Land Cover: Urban
Land Cover: Forest
Fig. 15.3 A box and whisker plot of Sav contrasting the shifts in the size distribution of urban and
forested patches generated by the RwC model using the 1992 and 2001 data for the Maryland
Piedmont (see text for details)
0.58) among metrics for
edge (LC.eg, T.eg), the total number of clusters (T.cltr), and the fractal index for the
largest cluster (LC.frc). Because landscape metrics are often non-monotonic with
p
(Gardner and Urban 2003), the nature of the correlation structure can be expected
to shift as the level of abundance of land cover type changes. Consequently,
correlations among metrics should always be explored by methods like those
illustrated in Fig.
15.4
before hypothesis testing begins.
Qrule produces an association matrix in the data “assmat.dat” that records the
frequency of adjacency among land cover types for each simulation. The associa-
tion at each time period, and the change in association over the interval 1992-2001
can be statistically tested with the usual
0.01 and d.f.
8. Other strong relationships exist (
r
a ¼
¼
>
2
methods. The expected frequencies are
usually estimated as a function of
p
, the observed frequency of each land cover type.
In fact, this type of test is a neutral model before neutral landscape models were
suggested! Because the simulations reported here were performed using the con-
straints of the landscape (the RwC method), the “expected” values for
w
2
tests the
observed frequencies produced by the RwC simulations. For the large landscapes
considered here, the use of the RwC association values had a minor impact on
results (as noted above). We also know from the above results that the pattern of
association for the actual landscapes will certainly differ from that of the RwC
w
