Geoscience Reference
In-Depth Information
derived from different sources and methods. For example, urban areas typically contain significant
forest cover that is not represented in LU class definitions. Also, landscape metric values describing
fragmentation of forest calculated from a LU/LC map were found to be different from those
calculated from a classification specifically designed to map forest and nonforest. Applications that
target the habitat quality for specific animal species, especially small animals, may not be well
served by aggregated LU classes.
Landscape pattern metrics transform spatial data in complex ways and users need to exercise
caution when interpreting the calculated values. Spatial simulation was a valuable tool for evaluating
the behavior of landscape metrics and their sensitivity to various inputs. Ecotone abruptness can
be detected using existing landscape metrics, but simulation illustrates that new metrics that measure
the variation in boundary element locations are more sensitive to abruptness than existing metrics.
Most measures of spatial pattern are also sensitive to both the composition and configuration of
the landscape. More work is needed to evaluate the various influences of landscape configuration
and composition on metric values.
The results presented here raise several issues for both the users and producers of remote
sensing-based LU and LC products for landscape ecology investigation. First, in addition to the
issue of data accuracy, the user is well advised to consider the appropriateness of class definitions
for a specific application. For example, a general LU/LC classification may not be appropriate for
calculation of landscape metrics in a study of habitat quality for a specific species. For this type
of application, landscape maps may need to be developed specifically for the intended application.
Second, the nature of the spatial transformations taken to compute pattern metrics can have dramatic
implications for the precision of the estimated values. Metrics that require image classification and
patch delineation are subject to greater imprecision than those based on local characterizations of
pattern. Third, the meaning of metric values can be confounded and difficult to interpret. Applica-
tions of landscape metrics that seek empirical relationships between metric values and ecosystem
characteristics may be able to bypass concerns about meaning and instead focus on correlations
with ecosystem outcomes of interest (e.g., based on independent measurements of ecosystem
characteristics). However, when directed toward spatial land management goals (e.g., a less frag-
mented forest), understanding the meaning of metrics is important to improve the probability of
achieving the desired objectives.
16.7 SUMMARY
Landscape pattern metrics have been increasingly applied in support of environmental and
ecological assessment for characterizing the spatial composition and configuration of landscapes
to relate and evaluate ecological function. This chapter summarizes a combination of previously
published and new work that investigates the precision and meaning of spatial landscape pattern
metrics. The work was conducted on landscapes of the upper midwestern U.S. using satellite images,
aerial photographs, and simulated landscapes. By applying a redundant mapping approach, we
assessed and compared the degree of precision in the values of landscape metrics calculated over
landscape subsets. While increasing landscape size had the effect of increasing precision in the
landscape metric estimates, by giving up spatial resolution, postprocessing methods such as filtering
and sieving did not have a consistent effect. Comparing multiple classifications of the same area
that use different class definitions, we demonstrate that conclusions about landscape composition
and configuration are affected by how the landscape classes are defined. Finally, using landscape
simulation experiments, we demonstrate that metric sensitivity to a pattern characteristic of interest
(e.g., ecotone abruptness of forest fragmentation) can be confounded by sensitivitiy to other
landscape characteristics (e.g., landscape patchiness or amount of forest), making direct measure-
ment of the desired characteristic difficult.
