Geoscience Reference
In-Depth Information
16.3, bottom panel). The largest patch index and the number of patches exhibit the highest coefficient
of variation across the runs, indicating a higher degree of relative error and lower precision. This
finding was consistent with the redundant mapping work described above, and it highlights the
relative instability of metrics that require patch delineation. Both the empirical and simulation work
show that slight changes in the maps of a landscape, as the result of remote sensing image quality
issues or just random perturbations, can result in relatively large variations in the number of patches
identified and, as a result, in the mean patch size.
16.5 DISCUSSION
The results indicated the difficulty involved in distinguishing the effects of changes in the
amount of forest from changes in the pattern of forest. The question is relevant in attempts to
understand the effects of landscape structure on ecological processes. Some have argued that the
concept of fragmentation is meant to include both the amount of forest and its spatial configuration
(Forman, 1997). Others define fragmentation to mean a spatial pattern characteristic of the forest,
independent of the effects of how much forest there is (Trzcinski et al., 1999). If the latter definition
is used, then a measure of forest fragmentation that is not sensitive to the amount of forest is
required. For example, do changes in the pattern of forest have impacts on ecological processes
beyond the effects resulting from changes in the amount of forest? Trzcinski et al. (1999) dealt
with this question by first evaluating the correlation between bird populations and forest amount,
then correlating bird populations with the residuals that resulted from the regression of forest
amount vs. forest pattern. The results indicated that there was little effect of forest pattern on bird
populations independent of forest amount. However, more work is needed to understand the
interactions of land-cover amount and pattern from both the perspective of how to measure pattern
independently and how to understand its independent effects.
16.6 CONCLUSIONS
This chapter summarizes work on the precision and meaning of landscape pattern metrics derived
from remote sensing. The transformations involved in calculating landscape metrics are complex,
and analytical approaches to estimating their uncertainty are likely not to be practical. For that
reason, this study has focused on two approaches to evaluating this propagation. First, we used
redundant mapping of areas and evaluation of the variation in metric values derived from different
imagery acquired near to each other in time. Second, simulation was used to explore the sensitivity
of various metrics to differences in landscapes by controlling certain landscape characteristics.
We determined that uncertainty in input data propagates throughout the calculations and ulti-
mately affects landscape metric precision. The precision of landscape metric values calculated to
measure forest fragmentation is affected by the similarity in seasonal date of the imagery, atmospheric
disturbances in the imagery (clouds and haze), and the amount of forest in the landscape. Metrics
calculated for larger landscapes tend to exhibit less variation, but postprocessing of imagery (e.g.,
through seiving to remove small patches) did not result in increased precision. Landscape metrics
whose calculation required more steps (e.g., image classification and patch delineation) were more
likely to be susceptible to slight variations in the input data. Therefore, patch-based metrics (e.g.,
number of patches and mean patch size) tend to be less precise than boundary-based metrics.
Landscape class definitions, whether intentionally different or different because of the mapping
method used, are important determinants of landscape pattern. It is possible to achieve significantly
different landscape pattern metric values based on different class definitions. This suggests extreme
caution should be used when attempting to compare pattern metric values for landscape maps
