For example, in the map with low contagion
in the second example scenario (case 2, above), the
calculated contagion value is larger than expected
for this reason.
The indices are also affected by variation in
attribute class frequencies (Li and Reynolds 1993).
Recall that the derivations of all maximum values
assumed an equality of attribute frequencies. When
some attributes are relatively more common than
others, then some types of attribute adjacencies are
necessarily more frequent than other types. The
maximum entropy (or minimum contagion) under
the null model can never be realized; there must be
some amount of contagion, and this will vary with
attribute frequencies. This effect was demonstrated
in Gustafson and Parker’s (1992) simulation study.
Cases can be made for considering this to be “real”
contagion, an artifact of unequal-probability sampling,
or an inappropriate application of the contagion
index.
In summary, when the contagion index is small,
it may be inferred that the attribute class frequencies
are more or less equal, and that the frequencies
of same-class adjacencies are about the same as the
frequencies of different-class adjacencies. When
the index is larger, it may due to a real tendency for
clumping, perhaps as a result of variation in attribute
class frequencies. Or it may be caused by a
high frequency of adjacencies between two different
classes.
If entropy (like angular second moment) measures
overall image “texture” (e.g., Haralick et d.
1973; Musick and Grover 1991; Gonzalez and
Woods 1992), then so must its contagion derivatives.
An image with “coarse” texture typically displays
a certain amount of clumpiness (possibly a
result of unequal attribute class frequencies),
whereas an image with “fine” texture does not
(compare the example maps in Fig. 1). Thes
itive connections among pattern indices
ported by large empirical correlations bet
indices of attribute diversity, dominance, contagion,
and texture (Riitters et al. 1995).
Summary
Connectivity and texture are recurrent themes of
spatial analysis in many fields including ecology,
image processing, and statistics. As a result, there
are many map-based measures of the tendency for
attributes to clump or coalesce. The choice among
them will depend n circumstances, including
the particular hypo
maps (e.g., vector
ters (e.g., grain size and extent), and the scale of
analysis (e.g., pixel-level versus patch-level). The
scope and depth of analysis are also important.
Whereas summary indices are needed to study
many aspects of pattern simultaneously (e.g., fragmentation,
patch compactness, and fractal dimension),
more complicated models are needed to partition
the detailed information from any one summary
index. 不需要完全准确的翻译、大概意思差不多就可以了 总之别像那些翻译软件翻译出来的无厘头汉语就成~,谢谢了
别给我用软件翻译的~~