International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
means that the spatial arrangement of the classes varies up to
MA yy and remains constant for larger MA values. However,
the intensity at which classes vary differs between the three
LCM classes. The class mainly logged forest (mLF) is less
sensitive for MA than the class mainly heavily logged forest
(mHLF), which on his turn is less sensitive than the class
mainly shrub (mSH).
3.4 Variability and arrangement of forest cover pattern
(SIDI & LSI)
Figure 8 shows the LCM classification results as measured by
the two landscape-related metrics SIDI (Simpson’s diversity
index) and LSI (landscape shape index). Simpson's index
literally means the probability that any two pixels selected at
random would be different LCM classes. For the p1990 image,
SIDI remains constant up to MA, meaning that the relative
proportions of the LCM classes do not change. For the more
fragmented pl996 image, however, SIDI shows a distinct
decrease. between. MA,, and MA 4, This means that the
dominance of one or a few LCM classes increases. Both images
show a distinct decrease in SIDI between MA, and MA 50.
Simpson's Diversity Index — p1990
& p1996
0,79 -
iH.
e eJ us 3
0,77 ? 6 2 Án.
0,75 ee
7
a NE E.
5 0,73
0,71
0,69 ee rate FE a
0,67 T T T T T T T T T T T T T
5 e? NS P S S S > S S SH S S e
Minimum Area (Ha)
Landscape Shape Index — pis
S
» Ss $
S
©
$9 4 9
se 95
PP aS
aD S
Minimum Area (Ha)
Figure 8: LCM classification results as measured by the two
landscape pattern metrics SIDI (Simpson's diversity index) and
LSI (landscape shape index) for forest cover pattern.
The metric LSI measures the perimeter-to-area ratio for the
entire landscape. An increase in LSI means that patches
become increasingly disaggregated, thus the spatial
heterogeneity of the landscape increases. Overall, LSI is higher
for the p1996 image than for the p1990 image. Thus, p1996 has
a more heterogeneous landscape than p1990. For both images,
LSI decreases when MA increases. Specifically, LSI shows a
794
significant decrease between MA ; ; and MA ,o for both images:
in this range LSI shows a 50% decrease. In addition, spatial
heterogeneity of the p1990 image and the more fragmented
p1996 image shows a similar trend for the entire MA range.
4. DISCUSSION
The setting for the minimum-area (MA) in LCM classification
has an impact on classification outcome. especially from MA; «
to MA 5. However, for MA values up to 150 ha the total area
of forest cover (mLF + mHLF) remained constant for both the
pl990 image and the more fragmented pl996 image. This
finding is very useful for assessing the area of forest cover in a
certain area. Forest area is an easily understood baseline figure
that provides the first indication of the relative importance of
forests in a country or region (FAO, 2001). In addition, the
results show that according to Van Beurden and Douven
(1999), decision-makers (at different decision-levels) can make
sensible consistent decisions on forest area, because several
levels of aggregation provide the same forest area. However, in
the more fragmented p1996 image, PLAND (percentage of
landscape) of the LCM class mainly shrub (mSH) did not
remain constant for MA values up to 150 ha. Realizing that
HLF and SH share very similar spectral signatures. it depends
very much on the accuracy of the input level (elementary
objects) whether the total area of forest cover is independent of
MA. MA has an effect on NP (number of patches), meaning
that it has an effect on the spatial arrangement of forest cover.
Consequently, from a decision-making point of view it is
important that maps are produced with identical MA values,
when comparing temporally different images. In addition, the
indices NP and KHAT seem to be a useful combination to
indicate which classes are causing significant differences
between classification results.
Simpson's diversity index (SIDI) remained constant for the
pl990 image. However, for the pl1996 image it showed a
distinct decrease between MA », and MA. This difference in
SIDI between the two images can be explained by combining
the results of PLAND and SIDI. PLAND of all seven LCM
classes did not change along the entire MA range (i.e., MA up
to 400 ha) for the p1990 image. However, for the p1996 image.
PLAND did change for the LCM class mainly shrub, especially
at the range between MA, and MA, for which SIDI changed.
A constant PLAND and a constant SIDI mean that the LCM
classes in the image have their own specific composition in the
landscape. Nevertheless, for certain classes like mainly shrub
for which PLAND changed, SIDI quantifies this compositional
change. In this respect, SIDI is a useful index to quantify
changes in LCM classifications in addition to standard PLAND
values. The indices KHAT and LSI. showing a similar trend,
are complementary to each other. From a decision-point of
view, such a combination seems to be useful to indicate which
classification results differ significantly.
5. CONCLUSIONS
From the classification results it can be concluded that the
LCM classification concept provided explanatory information
on vegetation changes in tropical rainforest areas. Based on the
findings of the LCM parameter spatial size, three conclusions
can be made towards decision-making:
I. Decision-makers at different decision-levels can
make sensible consistent decisions on forest area,
because several levels of aggregation provided the
same total forest area.
Intern
un
19°
agi
use
lev
gen
