ss
tential in
ecology
; not been
hodology
mapping
)bjectives
e natural
ross land
ng Kong,
of forests
>xpansion
(Hau and
ausal and
ides and
re, 1974:
portant to
vithin Pat
tern New
1). In this
'egetation
ary rocks
s a fault-
east. A
he Range
. Tai Mo
ell et. al.,
field and
es of the
r reaches
upper hill
nds. The
of Hong
. Country
n on fire
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
distribution within any given park. Hong Kong citizens, however,
remember all too well a particularly tragic hill fire that swept
across Pat Sin Leng Country Park in 1996 killing members of a
school party.
4. METHODOLOGY
Two types of remotely sensed data arc used in this study.
Georeferenced LANDSAT TM data (dated 10-01-03) and
IKONOS data (dated 4-11-02).
LANDSAT data has visual, near-infrared and mid-infrared bands.
Mid-infrared bands (TMS and TM7) are sensitive to canopy
moisture content and may be linked to vegetation type and canopy
structure. They are therefore are useful to establishing vegetation
indices that identify burned or stressed vegetation; Fraser et. al.,
2000). Two indices are determined in this study. They are NDVI
normalized difference vegetation index), (Bannari et. al., 1995)
and NDMIDIR (normalized difference mid-infrared). High-
resolution IKONOS data has visual and near-infrared bands and is
used to create a false colour composite image to detect shallow
landslides. The spectral resolution of LANDSAT TM and
IKONOS bands used is 30m and 4m, respectively.
The following components were used to develop a methodology
(this study) to identify disturbed vegetation (including fire scars)
and detect associated shallow landslides:
« Produce NDVI image using LANDSAT TM data
e Produce NDMIDIR image using LANDSAT TM data
e Identify scars on NDMIDIR image and compare the indices
values with the NDVI image
* Produce False Color Composite (FCC) using IKONOS
multispectral data and identify landslides
* Locate fire scars on NDMIDIR image
e Ground check for the verification of shallow landslides and
fire scars
The equations for producing NDVI and NDMIDIRIR using
LANDSAT TM bands are: as follow:
—— (TM4- TM3)
NDVI S T CINA + TM3) a
-— (TMA—-TM7) (2)
NDMIDIR (TM4 + TM7)
5. RESULTS AND DISCUSSION
5.1 Vegetation Indices and Disturbed Vegetation
NDVI and NDMIDIR images generated using equations (1) and (2)
above are presented as Figures 2 and 3 below. Relevant statistics
for these indices are given in Table 1.
Healthy vegetation shows higher values in the NDVI image, but
from the index values, it is difficult to discriminate between
healthy vegetation and the disturbed vegetation types (Fig. 2). The
values from the NDMIDIR image, however, do very clearly
discriminate vegetation vigor (Fig. 3). Disturbed (burned) or
49]
stressed vegetation is indicated by values between zero and 0.235
(mean value) and healthy vegetation by values greater than the
mean. In the NDMIDIR image (Fig. 3), disturbed vegetation is
seen as dark pixels. These areas are not depicted in the NDVI
image within vegetated areas.
Indices Minimum Mean Maximum
NDVI -0395 0.235 0.512
NDMIDIR -0.135 0.395 0.635
Table 1. Statistics of NDVI and NDMIDIR indices.
Figure 2. NDVI image showing healthy vegetation (bright white
tonal areas).
Figure 3. NDMIDIR image identifying possible fire scars
principally along the south and west Pat Sin Leng Range (arrowed;
black pixels). Small areas of disturbed vegetation are also seen to
the north. Fire scars and landslides were identified in ground
thruthing studies, particularly along the western Range.
IKONOS multispectral near-infrared, green and blue bands were
used to produce a false colour composite (FCC) image (Fig. 4).
Red areas depict healthy vegetation. Pale colours (yellow, blue),
prominent to the south of the Pat Sin Leng escarpment (Figs. 4-6).
identify exposed denuded hill slopes.