Environmental Engineering Reference
In-Depth Information
Fig. 15.2 The
shaded
area represents the Piedmont region of Maryland
forest and urban areas that occurred over this time period. The high precision in
estimates of Sav (and other landscape metrics) is due, in part, to the large landscape
with an equally large number of patches. If one assumes that the confidence intervals
around the actual land cover data (Table
15.2
) can be estimated with equal confi-
dence, then the observed changes in the characteristic size of forest and urban patches
(Sav) should be regarded as significant. Bootstrap estimates of confidence intervals
could be performed to provide a quantitative verification of this assertion.
Other landscape metrics generated by Qrule (or by other similar software
products) are interesting to examine and can provide insights into the change in
landscape pattern with time. However, hypothesis testing using multiple metrics is
dangerous because of the strong correlations that exist among metrics (Riitters et al.
1995; Wang and Malanson 2007). Figure
15.4
uses the “splom” procedure of R (see
Appendix) to illustrate correlations among metrics for the RwC simulations for
forested land cover in 1992. The ten Monte Carlo simulations produced 4,091,630
forest patches with the largest patch of 25.3 ha and a characteristic patch size (Sav)
of 1.77 ha. Figure
15.4
illustrates three strong relationships that exist among
metrics: the amount of edge on the largest cluster (LC.ed) was perfectly correlated
(
r
1.0) with the size of the largest cluster (LC.sz); the total amount of edge
(T.edg) was highly correlated (
r
¼
0.99) with the total number of forested pixels
(S.frq); and the correlation length of patches (C.len) was highly correlated (
r
¼
¼
0.88) with the characteristic patch size (Sav). These values were all significant at
