query and retrieve data from the database and to perform
analysis on these data.
SATELLITE IMAGE ANALYSIS
Using satellite image to detect land use changes has the
advantages of lower cost and shorter processing time
over aerial photograph interpretation and field
investigation.
Another advantage of using satellite image over field
survey for Der-Chi area is that satellite images can be
received and processed without the interference of local
interest groups.
Image Source
The satellite receiving facilities in Center of Space and
Remote Sensing Research at Central University receives
SPOT, LANDSAT and ERS data regularly. This assured
the data availability for the research and for future
monitoring. SPOT images were preferred than
LANDSAT images because SPOT had better spatial
resolution. Two images were chosen for the study, one
was taken on December 30; 1993 and the other was
taken on January;7 1995.
Preprocessing
ERDAS's imagine software was used in this research to
process satellite images. The images were geocoded to
Transverse Mercator projection. The RMS error for the
registration was 1 to 2 pixels, or 20 to 30 meters. 40m
DTM was used to produce ortho Satellite images.
Land Use Classification
This research was interested in distinguishing between
natural vegetation and different types of framing and in
identifying landslides.
First, unsupervised classification were used to uncover
the nature categories in the images. The results indicated
that the rugged terrain of the research area caused large
area within shadow. The shadow regions' spectral
characteristics were very different from those of the
unshaded regions. Several research had proposed
methods to resolve this problem (Golby 1991; Hodgson
and Shelley 1993, Michael 1993). The band ratio method
was chosen for this research. To further increase the
accuracy of classification, two sets of training samples
are chosen for the shadow regions and unshaded regions
respectively for performing supervised classification.
In order to reduce the effect of shadow and distinguish
different types of vegetation, a variety of vegetation
Indices, such as IR-R, IR/R and (IR-R)(IR R) were
tested. The IR-R and IR/R indices were used to classified
the 1993 and 1995 image respectively. Because the
correlation between the G and the R channel was very
high (0.99), the R channel was not used. Vegetation
index, G and IR channels were used for supervised
classification.
The classification scheme must be reliable and met the
requirements for policy makers. After several trail
classifications, a classification scheme was derived for
the research (figure 2).
The 1993 image was taken during a serious drought
period. Large area of river beds and river banks was
SPOT
image
|
| | |
water shadow unshaded
region region
|
| | | | | |
vegetation buildup barrean land vegetation buildup barrean land cloud
area and area and
landslide forest
| |
p | | | | |
forest farm grass forest farm grass
land
dense sparse orchard vegetable tea dense sparse orchard vegetable tea
forest forest farm garden forest forest farm garden
Figure 2 Classifiction Scheme
132
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996