! 2004 
"M: 
tem 
asic 
ract 
eds 
as 
1 in 
ped 
zed 
ice. 
the 
low 
eral 
and 
ous 
the 
ach 
Der, 
| (o 
hed 
atic 
nap 
(ep 
'ent 
atic 
Nas 
| ds 
of 
ing 
em 
eds 
to 
ate 
ms 
re. 
of 
for 
bic 
nal 
el 
for 
he 
nd 
‘as 
SS 
ith 
all 
C 
rat 
of 
In 
ale 
Ip. 
in 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
A case study concerning the T'aiwania site selection was 
implemented for a demonstration of a forestland classification 
DSS. In this part. topographic factors (e.g.. elevation and slope) 
and soil factors (e.g.. soil texture and soil suitability) were 
assumed to be the main factors influencing the l'aiwania habitat 
according to the management plan of the Liukuei Experimental 
Forest (TFRL 1992). As for membership representation, soil 
texture focuses on moderate texture (e.g. sandy loam. silty 
loam, sandy clay loam. clay loam. and silty clay loam). and the 
other data are based on fuzzy arguments and standardized as 1 
to -1. where | means the environmental condition is completely 
suitable for the l'aiwania: -1 means completely not suitable; and 
the open interval (-1.1) means partially suitable. In addition. 
this case study only applied Ist-order watersheds for Taiwania 
site selection because the site selection was regarded as a forest 
management practice on a small scale. 
3. RESULTS AND DISCUSSION 
3.1 Delineation of ecosystem units using DTM 
Figure 2 depicts stream networks extracted using 400 as the 
threshold value and the result of encoding streams with 
different orders according to the Strahler method. Figure 3 
shows the result of delineating watersheds by specilying 
different stream orders as the minimum order. It is apparent 
that 4 different Kinds of watersheds can be delineated in 
response to the need to consider watersheds of different sizes. 
The result shows that watersheds derived from different stream 
orders coincide very well with the distribution of the stream 
network when these two maps are overlaid together. Moreover. 
the number of watersheds varies with different stream orders. 
and the number of watersheds decreases as the stream order 
number increases. Irom the delineation of watersheds. it is 
obvious that the Experimental Forest cannot accept a stream 
order number that is higher than 2. because the upper part will 
disappear. Therefore. this study used 2" watersheds to study 
to develop the 
their spatial differences and hierarchical 
ecosystem classification. 
  
Figure 2. Extraction of stream networks. (a) Delineating stream 
networks. (b) Encoding stream networks. Red. green. 
blue. and yellow represent 1%, 27%, 3" 4™ order 
streams, respectively 
This study adopted the use of watersheds as the basic ecosystem 
unit. However. there are several problems encountered when 
dealing with watershed delineation because different criteria 
and algorithms will generate different watershed maps for the 
same area. In addition. the map scale is an important factor 
allecting the delineation of watershed boundaries because the 
patterns of ecosystem boundaries on various maps may differ 
(Bailey. 1996). This study did not have a problem with this 
because the DTM with 40 m x 40 m resolution for each pixel 
was used for watershed delineation. No matter who delineates 
the watershed boundaries, the result will be the same when 
applying the extraction. algorithm as proposed in this study. 
This is the reason why this study treats the watershed as a basic 
ecosystem unit and uses D'TM for ecosystem delineation. 
Watersheds of lust order and above | Wales sheds of secon order and above 
À 
| 
  
Watersheds of third order and above Watersbed of fourth order 
J 
3 o 3 6 1:3 " 3 $ 
km 
  
^ 
Figure 3. Delineation of watersheds based on different stream 
orders 
3.2 Development of a hierarchical ecosystem classification 
using GIS and multivariate statistical analysis 
hierarchical 
is clear that the 
Figure 4 shows the result of a ecosystem 
classification using 2" watersheds. It 
Experimental Forest was classified into 3 different clusters (or 
zones) based on 3 data layers. The result is quite satisfactory 
because the distribution of the clusters coincides with the 
terrain characteristic and varies along a continuum. For 
example. the 1*' cluster is located in the downstream area of 
lower elevation and slope. while the 3" cluster is located in the 
upper stream area of higher elevation and slope. The result 
indicates that the Experimental Forest can be geographically 
divided into 3 large ecosystems if' Miller's scheme of 3-scale 
perception (i.c.. site, landscape. and ecoregion) is applied in this 
study. 
As mentioned previously. ecosystems exist at multiple scales. 
Several countries have proposed and implemented schemes for 
recognizing such scale levels. In this study. a 3-scale scheme 
similar to Miller's approach (Miller. 1978) was implemented 
although more levels were suggested by previous literature. 
Therefore. the Liukuei Experimental Forest has 3 landscapes 
mosaics. and cach landscape mosaic is composed of different 
sites. This scheme. proposed in this study on the basis of a 
hierarchical ecosystem classification system. looks satisfactory. 
183