International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, V. ol XXXV, Part B7. Istanbul 2004 Internationa
Experimental Forest to island-wide forestland classification in
Taiwan.
2. MATERIALS AND METHODS
2.1 Study area and materials
The Liukuei Experimental Forest of the TFRI located in
Kaohsiung County of Taiwan (Figure 1). is divided into 25
forest compartments with a total of about 9616 ha. The major
forest type is natural forest intermixed with a proportion of
plantations. particularly Taiwania plantations. Elevations in
this area range from 330 to 2400 m. The average annual
temperature. rainfall. and relative humidity are 16-23 °C. 2150-
3748 mm. and 71% -86%. respectively.
A digital terrain model (DTM) of the Liukuei Experimental
Forest. located in the TERI of Taiwan (Figure 1). was used as
the material of this study. The resolution of each pixel is 40 m
^
x 40 m. The DTM was used to derive 3 terrain data layers of
elevation. slope. and aspect. and also to automatically extract
watersheds. which are treated as basic ecosystem units. All 3
data layers were then incorporated into the GIS. which was
applied to establish a hierarchical classification system. In
addition, the EMDS svstem (Reynolds 1998) was used to
integrate the database and the knowledge base in order to
generate a forestland classification DSS. As for Taiwania site
selection. soil data and a 1/5000-scale forest type map
generated by digital photogrammetric techniques were also
needed. In addition. the hardware used in this study included a
PC computer. plotter. and color laser printer. and the software
included ARC/INFO. ArcView. SAS. and EMDS.
Figure 1. Study area and material of digital terrain model
2.2 Methods
To achieve the above objectives. the processes focused on the
automatic delineation of ecosystem units using DTM. the
development of a hierarchical ecosystem classification scheme
using GIS and. multivariate — statistical analysis. the
establishment of a forestland-classification DSS using the
EMDS. and its application for Taiwania site selection.
using DTM:
ecosystem
2241 Delineation of ecosystem units
Ecosystem delineation is a prerequisite for
classification. This study treated watersheds as the basic
ecosystems and used the DTM to automatically extract
watersheds as ecosystem units. Meanwhile. the watersheds
extracted from different stream orders were regarded as
ecosystems of different sizes. A computer program written in
ARC/INFO AML (ARC Macro Language) was then developed
for delineating watersheds. Two major steps were emphasized
as follows.
2.2.1.1 Extraction of stream networks: In practice.
the accuracy of extracted stream networks is correlated with the
which refers to the [flow
There are several
setting of a threshold value.
accumulation calculated from the DTM.
approaches for setting the threshold value (Montgomery and
Foufoula-Georgiou. 1993). This study was based on previous
studies (Cheng. 1993: Cheng et al.. 2000) apd used 400 as the
threshold value to extract stream networks. Furthermore. cach
stream in the stream network was assigned a distinct number,
and a numeric order using the Strahler method was given to
assign the correct order for the watersheds alter the watershed
boundaries were delineated.
automatic
22.1.2 Delincation of watersheds: An
approach was used to extract watersheds in order to map
ecosystem boundaries. However, one problem faced in this step
was the delineation of different-sized watersheds for different
ecosystem units. To overcome this. a modified automatic
approach that differs from traditional automatic approaches was
developed and implemented in this study. The approach is
unique in its capability to automatically identify streams of
different orders and to search for the outlets to watersheds using
the stream networks extracted from Ihe DTM.
2.2.2 Development of a hierarchical — ecosystem
classification using GIS and multivariate statistical analysis:
To develop a hierarchical ecosystem classification, watersheds
delineated based on streams of different orders were used to
investigate spatial differences of the watersheds. Multivariate
statistical analysis was then applied for grouping ecosystems
according to their spatial similarities using SAS software.
During the grouping
nonhierarchical | clustering
grouping smaller ecosystems into larger ecosystems, and cubic
clustering criterion (CCC) was used for determining the optimal
number of clusters. Meanwhile. only 3 data layers (i.c..
elevation. slope, and aspect) were used as input variables for
clustering analysis according to Cheng's (1995) research in the
Liukuei Experimental Forest.
process. the
analysis was implemented for
2.2.3 Establishment of a forestland classification DSS and
its application on Taiwania site selection: The EMDS was
used as a framework to establish a forestland classification DSS
by integrating the above results of ecosystem classification with
a GIS environment. During the process of establishing DSS. all
GIS themes including different hierarchical ecosystems were
first constructed as a database. Then the knowledge bases that
describe relations among. ecosystem states and processes of
interest (e.g.. forestland productivity) were also constructed. In
this step. the truth value representing the basic state variable
was applied to express an observation's degree of membership.
and evaluations of degree of set membership were quantified in
the semantics of fuzzy logic.
182
k-means approach of
A case «
implemen
DSS. In tl
and soil
assumed t
according
Forest (T
texture fo
loam. sanc
other data
to -1. whe
suitable [o
the open
this case s
site selecti
manageme
AI Deli
l'igure 2 «
threshold
different ©
shows the
different s
that 4 dil
response
The result
orders cott
network wi
the numbe:
and the nu
number inc
obvious th
order numl
disappear.
their spati
ecosystem «
This study a
unit. Howey
dealing witl
and algorith
same area,
affecting the
