Environmental Engineering Reference
In-Depth Information
precisely, “Were observed changes statistically different from those generated by
random processes alone?” The RwC neutral model was developed, random land-
scapes with identical political and physical boundaries were generated, and a KS
test comparing random with actual landscapes showed that the cumulative dis-
tributions of random versus actual forest patch sizes were indeed, significantly
different. The second question was, “Has there been a significant change in
patterns of association between urban and forested areas?” An examination
of the association matrix, with a
2
of the contingency table, showed that patterns
of urban:forest association were less than expected by chance alone, but the loss of
forest edge was greater than expected from increases in urban areas and decreases
in forest alone.
There are many other hypotheses that could be tested with these data. For
instance, we might wish to know “Did urban growth has result in a rate a loss of
larger forested areas that was greater than expected by chance alone?” Or perhaps
the alternative question would be of interest: “Were smaller patches of forest
absorbed by urban growth at a rate greater than expected by chance alone?” An
additional set of questions could be developed regarding the effect that changes in
agricultural lands might have had on forest loss. The construction of a complex set
of tests that compared the relative impact of agricultural versus urban land use
change is an intriguing possibility left to the curiosity of the reader. Such inquiries
might involve the development of more complex neutral models, or the critical
application of the suite of software tools (not just Qrule) that are now available.
Whatever the approach, reliable results will only be achieved if the analysis adheres
to the statistical principles of hypothesis testing.
w
15.6 Conclusion
Relatively simple models can often “explain” complicated patterns. Indeed, as
Gardner et al. (1987) have demonstrated, real landscapes are sometimes indistin-
guishable from purely random maps. This result does not argue that real landscapes
are produced by simple random processes, but simply defines the conditions in which
more detailed explanations of cause-and-effect cannot be demonstrated. When a
simple random model fails, then a more complicated explanation (or model) may be
tested. For instance, explanations that invoke specific agents of pattern formation due
to the physical template (environmental gradients), biotic processes, or disturbance
regimes (including human actions). We have illustrated one method for employing a
neutral model that considers the effect of landscape constraints on observed patterns.
Although the comparisons made here between this model and actual landscapes were
dramatically different, variations in the shape, extent, and proportions (
p
)ofland
cover types have yet to be fully explored for the RwC model. We hope the combined
process of model development and statistical tests will be helpful in the extension of
these ideas to other landscapes and landscape questions.
