International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004 
  
  
  
p 1990 image 
p 1996 image 
  
LF 25.13 mLF | 25.51 1 
  
  
  
  
  
Elementary Composite Elementary Composite 
objects objects objects objects 
mean mean 
LC PLAND |LCM |PLAND|SD |LC |PLAND|LCM |PLAND|SD 
class |in % class |in 96 in 96 | class| in 96 class |in 96 in 96 
.89 
HLF 127.85 mHLF | 26.99 |2.08 
SH 24.60 mSH [24.95 |0.52 
AG 14.77 mAG | 15.44 0.96 
GR. [6.13 mGR | 5.62 0.79 
WA 10.37 mWA | 0.18 0.14 
RI 1.35 mRI 1.32 0.02 
CL 0.00 - - 
  
LF 12125 ImLFE |2126 243 
HLF'|16:30 [mMHLF| 1225 1238 
SH [33.99 | msH | 3820 |360 
AG 114.53 |mAG |1440 |0.74 
GR [11.40 | mGR | 11.60 |2.57 
WA | 1.17 mWA | 1.00 0.13 
R 1.29 mRI 1.29 0.03 
CL. 0.00 - - 
  
  
  
  
  
  
Table 1: Elementary objects versus composite objects expressed in cover percentages (PLAND) for respectivelv the land cover 
classes (LC) and land cover mosaic classes (LCM); the latter are based on the mean PLAND values resulting for the 14 different 
MA values (p1990 image and p1996 image). 
the smallest elementary object up to the largest composite 
object. In addition, the values should also include the 
minimum -area of tree covered land that should be considered 
as ‘forest’, which ranged from 0.01 ha in the Czech Republic to 
100 ha in Papua New Guinea (Lund, 1999). During the 
sensitivity analysis, the LCM parameter mix was kept constant 
(BN=0.55). A total of 28 LCM classifications were carried out 
on two Landsat TM images of the Pelangkaraya study area of 
1990 and 1996 further referred to as p1990 and p1996. 
Five methods were selected to evaluate the LCM -classification 
results, the standard remote sensing accuracy method KHAT 
(Congalton and Mead, 1983) and four Landscape Pattern 
Metrics as applied in landscape ecology (Forman, 1995; 
McGarigal et al., 2002). The discrete multivariate analysis 
technique KHAT and the Zstatistic for significance testing are 
used to evaluate the overall classification results (Hudson and 
Ramm, 1987). The elementary objects are used as reference to 
calculate the similarity matrices. The Landscape Pattern 
Metrics were used to evaluate the LCM classification results on 
variability and arrangement of forest cover and forest cover 
pattern. The metrics Percentage of Landscape (PLAND) and 
Number of patches (NP) evaluate forest cover, whereas the 
metrics Simpson 's Diversity Index(SIDI) and Landscape Shape 
Index (LSI) evaluate forest cover pattern. PLAND and SIDI are 
spatially independent metrics, which refer to composition or 
variability of the landscape, or ‘how different things are’. NP 
and LSI are spatially dependent metrics, which refer to the 
configuration or arrangement of the landscape. or ‘how things 
are distributed’ (Forman, 1995; Gustafson, 1998). 
3. RESULTS 
3.1 LCM classification results 
Figure 4 shows the LCM classification results for two different 
values for minimum area, i.c., MA is 25 ha and 300 ha, for the 
p1990 image (with BN=0.55). For comparison, figure 5 shows 
a standard per-pixel classification result uing the maximum 
likelihood classifier. Figure 4 clearly shows that LCM 
classification results provide crisp maps. In addition, the LCM - 
classification with the larger minimum-area results in fewer 
small objects; it is more aggregated. 
Table 1 shows the classification results of elementary objects 
versus composite objects expressed in cover percentages (i.e., 
PLAND). Overall, both aggregation levels show similar 
ranking of major and min or classes. For the p1990 image, there 
792 
  
Figure 4: LCM-classification results for the p1990 image; MA 
is 25 ha (a), and MA is 300 ha (b), both with BN at 0.55. 
is almost no difference («196 cover) in PLAND values between 
the two aggregation levels for all thematic classes. However, 
for the 1996 image two striking differences (= 4% cover) exists 
for the thematic classes related to heavily logged forest and 
shrub. From 1990 to 1996, shrub vegetation and grass 
vegetation have enormously increased, reducing the vegetation 
of heavily logged forest. The increase of grass vegetation is 
depicted by both aggregation strategies meaning that grass 
tends to occur in (homogeneous) clusters. However, the 
increase of shrub is differently depicted by both aggregation 
strategies meaning that shrub tends to be (heterogeneously) 
distributed over the landscape. This finding supports the fact 
that shrub is an intermediate vegetation between the transition 
of forest to agriculture and vice versa. 
Intern 
  
cle 
mc 
val 
M; 
bet 
de 
bet 
Th 
sig 
tha 
Sp 
the 
bet 
M/ 
‘pal 
sig 
unt 
sin 
is 1 
Tb 
KHAT 
PP POPPE PES 
Fig 
(M: