ever,
when
ks on
evant
ainly
ation
eme,
data
1 this
1) are
ively
of the
pared
juires
ferent
may
ioned
rly at
rding
LU
tions,
| and
e past
have
oment
ige of
stored
]ue to
CL
owest
, each
| with
works
tween
th the
e LU
z up to
upport
yblem,
vances
spatial
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
2. STUDY OBJECTIVE
The objective of this study was to develop an image-based
multidimensional LU classification scheme as a part of the
development of versatile LU information system (VLUIS) for
local planning in Indonesia. Following the classification
scheme development, example of LC information extraction as
the first dimension of the VLUIS was given. Image datasets of
Landsat ETM+ and Quickbird covering Semarang area, Central
Java, Indonesia were used. This study is a part of a longer term
research aims to develop versatile LU classification scheme and
information extraction methods for each category within the
scheme, followed by a demonstration in applying the obtained
spatial data to support several local planning tasks.
3. PREVIOUS WORKS
Studies on the development of LC/LU classification systems
have been carried out by various authors. One of the most
eminent systems is the USGS LC/LU classification system
(Anderson ef al., 1976), which mixes up LC and LU terms in its
categorisation. The USGS LC/LU classification system is
widely used in various projects in the USA. For Indonesian
environment, Malingreau and Christiani (1982) and Sandy
(1982) also developed systems mixing up LC and CU concepts.
Van Gils ef al. (1991) proposed a two-level ‘ITC World LC and
LU Classification’, which tried to separate LC from LU
categories and simultaneously established relations between the
two. Recent development of LC/LU classification systems were
undertaken by Food and Agricultural Organisation (Jansen and
Di Gregorio,1998), Young (1998) and Cihlar and Jansen
(2003).
Similarity between all aforementioned classification systems is
the use of single attribute for each category on each level. The
single attribute of LC/LU categories may become problematic
at the subtler level, e.g. level III and IV of the USGS
classification system, since more detailed information in a
single attribute tends to be more specific. Thus, at a subtler
level, translation or conversion from a classification scheme to
another is inhibited. As a consequence, it is more difficult to
use similar categories under different schemes for practical
purposes, e.g. monitoring of LU change. That is why Young
(1998) emphasised the need for development of LU
classification system containing multiple attributes comparable
to soil properties found in the World Reference base for Soil
Resources.
By using digital satellite imagery, multispectral classification
can automatically derive LC-related spectral classes (Jensen,
1996; Mather, 1999). The tentative categories can then be
regrouped and relabelled into more meaningful LC classes. Liu
et al. (2002) suggested the combination of various automatic
image classification methods, i.e. maximum likelihood, expert
system, and artificial neural network for improving land cover
map accuracy. Derivation of subtler information on LC or LU
through per-pixel image classification can also be done with
contextual information (Stuckens et al, 2003), such as
landscape characteristics related to soil properties and slope
steepness (Folly, 1996; Danoedoro, 2001; Ehlers ef al., 2003).
4. METHODS
4.1. Development of Classification Scheme
The classification scheme development was started with the
distribution of questionnaires to 36 stakeholders related to
planning in the study area. Findings obtained from the
questionnaire data was analysed together with previous works
dealing with LU based environmental assessment and
modeling. In addition, several classification schemes widely
used such as USGS LC/LU classification systems (Anderson et
al, 1976), LC/LU classification system for Indonesia
(Malingreau and Christiani, 1982), ITC (van Gils ef al., 1991)
were taken into account. Moreover, various concepts related to
LC, LU as viewed from spectral, spatial, temporal, ecological,
and socio-economic aspects were also considered.
include spectral characteristics of various objects (Hoffer, 1¢
Curran, 1985; and Jensen, 1996); spatial pattern and geographic
position/ site (Lillesand and Kiefer, 2000); tempora! pattern of
LC and LU (van Gils e a/., 1990), tropical ecology (F
1990; Osborne, 2000); and socio-economie aspect of LC and
LU (Sutanto, 1986; Jensen. 2000).
l'hese
VUSSsiEe,
LU DIMENSION DESCRIPTION
Spectral Strongly related to, or may directly be identified
based on, spectral information of the objects. In
general, the spectral dimension is expressed by
cover types
Spatial Related to particular spatial pattern or
arrangement, position or site, which is normally
used as an additional key factor (besides spectral
dimension) to distinguish one feature from others,
e.g. river, lake, regularly spaced stands, interleave
planting, coastal mudflat
Temporal Related to temporal or seasonal changes, e.g.
length of indundation and crop rotation.
Information related to spectral and spatial aspects
is also required to determine temporal dimension.
LC and LU forms express interaction between
vegetation, animals and human activities with the
land they exist. Their existence also represent the
environmental characteristics of the area, e.g.
mangrove formation, upland agriculture, slum
areas
Ecological
Socio-economic
function
Basically, many LC types and LU functions have
economic or socio-economic functions too.
However, the socio-economic dimension needs to
be explicitly presented, if they have.
Legal Basically it is difficult to extract using remotely
sensed imagery.
Table 2. Description of each LU dimension used in this
study
A multilevel classification was considered more suitable for
local regions in Indonesia, which show a wide range of areal
coverage. Therefore, various satellite imagery with various
spatial resolutions were taken into consideration. Previous
works using various satellite data were reviewed with respect to
the level of details of the categories generated, methods of
processing used or developed, and accuracy levels reached.
The works of Phinn ef al. (2000) and Phinn ef al. (2002) were
also taken into account. Meanwhile, types of information to be
included in LU categories were also specified with respect to
the previous works in environmental applications.
4.2. Image Analysis and Classification
Image classification was run based on the classification
scheme. In this study, the first (spectral) dimension of the
versatile LU information was derived using image processing