Remote Sensing and Geospatial Analysis for Landscape Pattern Characterization - Landscape Ecology for Sustainable Environment and Culture - page 204

quantitative description of landscape pattern remains difficult due to lack of
generalizations concerning pattern-process relationships for many metrics

(McGarigal 2002 ; Li and Wu 2004 ; Turner 2005 ; Fu et al. 2011 ).
Secondly, given the scale dependence of spatial heterogeneity, the statistic
characteristics of landscape metrics can be affected by spatial extent and scale (e.g.
Turner et al. 1989 ; Wu et al. 2002 ; Corry and Nassauer 2005 ). Turner et al. ( 1989 )
investigated the effects of changing grain size and extent of land cover data on
spatial patterns. Landscape patterns were compared using metrics measuring
diversity, dominance and contagion. They found that the diversity index decreased
linearly as grain size increased, but dominance and contagion did not show such a
linear relationship. While dominance and contagion increased with increasing
spatial extent, diversity showed an erratic response. Corry and Nassauer ( 2005 )
found that the amount of linear habitat patches increased with increasing spatial
resolution of land cover data. When many linear patches are present, metrics
measuring landscape configuration may not be reliable; after resampling the land
cover data into a coarse grain size, landscape configuration metrics became eco-
logically relevant. They found that composition metrics can be more useful in
highly fragmented landscapes. A more comprehensive study was conducted by Wu
et al. ( 2002 ) who examined how some common metrics responded to changing
grain size and extent. Their found that the responses fell into three general cate-
gories: Predictable responses to changing scale with definable, simple scaling
relationship; less predictable, staircase-like responses; and erratic responses

without consistent scaling relationships.
Since landscape metrics are generally derived from categorical maps, the the-
matic resolution of land cover data and the classification scheme can also affect the
statistical properties and behavior of landscape metrics. For example, Corry and
Nassauer ( 2005 ) found that aggregation of land cover classes can reduce the
number of patch types and thus increase the likelihood of contiguity. Huang et al.
( 2006 ) examined the sensitivity of two dozens of metrics to a number of land cover
classes with different spatial patterns. They found that many metrics behaved
predictably with increasing classification detail. At lower class numbers, metrics
were quite sensitive to increasing classification detail. Their studies suggest the
importance of land cover classification scheme in landscape pattern analysis.
Moreover, the quality of input data (i.e. land cover map) can affect the statis-
tical characteristics of landscape metrics. For example, Wickham et al. ( 1997 )
tested the sensitivity of three common metrics (i.e. patch compaction, contagion,
and fractal dimension) to land cover misclassification and differences in land cover
composition. They found that differences in land cover composition need to be
larger than the misclassification error in order to be confident that differences in
landscape metrics are not due to misclassification. Corry and Nassauer ( 2005 )
noted that several data conversion procedures (e.g. vector to raster conversion,
digitization of analogue data, and resampling) often introduce errors in land cover
maps that can further affect the computed metric values. Langford et al. ( 2006 )
found that land cover classification error is not always a good predictor of errors in
landscape metrics but maps with low misclassification rates can yield errors in