, Giant
golden
Idlife's
ccurate
ld take
s of its
nature
n NRs,
nber 8
29 and
onifer-
d bare-
than in
eflects
clear
ildlife,
nagers
strong
pplied
results
cation
detect
while
es can
ch has
yield
Jue to
F giant
China,
]scape
ge) in
red in
k and
rs, (2)
agers
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
2. STUDY AREA
The study area is located on the southern slope of the middle
Qinling Mountains, Shannxi, China (Figure 1), which includes
three giant panda NRs: Changqing, Foping and Guanyingshan.
Its geo-location is 107° 26'- 107° S58" E, 33° 26'- 33° 46'N.
The total area of the study site is about 709 km?. The elevation
ranges from 800 m to 3071 m. The annual mean temperature is
about 14.6°C and annual precipitation about 813.9 mm. This
region is the most northern refuge of the Giant Panda and a
crossing area of the southern and northern fauna and flora. It
has rich species and diverse habitats. The vegetation is well
developed with a more complete vertical distribution under the
sub-tropical monsoon climate. Bamboo has a wide understory
distribution (CNNRA 2001, FNNRA 2001, SDIFR 2001).
Foping, Changqing and Guanyingshan NRs were established in
1978, 1994 and 2002 respectively. Therefore, Foping NR has
been conserved for more than 20 years and its ecosystem and
wildlife habitat are well protected and developed. Foping's
ecosystem has reached a relatively stable stage. However, there
are still some local people resident in the center of Foping.
Changqing NR was originally belonging to the Changqing
Forest Bureau (constructed in 1967) and got 9 years’ protection.
Changqing’s strong-disturbed ecosystem and wildlife habitat
has been starting secondary succession and got certain
restoration. Changqing’s ecosystem is still keeping changing.
There are no people living inside Changqing. While
Guanyingshan NR was converted from the Longcaoping Forest
Bureau (constructed in 1986) and stopped forest cutting in 1998.
The human activities, such as farming and commercial logging
have remained the strong impacts on the ecosystem and wildlife
habitat until now. Guanyingshan's ecosystem is unstable.
Guanyinshan
Foping
Changging
Figure 1. Location of the study area: the southern slope of the middle Qinling Mountains in Shannxi, China.
Three giant panda nature reserves are included: Changqing, Foping and Guanyingshan.
3. RESEARCH METHODS
3.1 Data and preparing
This research used vector data (including boundaries of
research area and nature reserves; rivers; sample points), raster
data (including DEM; slope model; aspect model), and Landsat
TM images acquired on September 15 1988 and September 8
1997. All map data have been geo-referenced in order to have a
same coordinator system as well as spatial resolution of 30*30
m^. Due to the incompletion of the 1997-image scene and its
cloud cover, we derived the same-size area for LUC mapping
and NDVI calculating for both 1988 and 1997 images without
the cloud-cover area and the missed image part. Sample points
were from the field survey conducted in 1999 (Liu 2001) and
from the 3™ national panda and its habitat census.
3.2 Mapping LUC and assessing the mapping accuracy
In order to improve the LUC mapping accuracy for change
detecting, eight data layers were used, containing TMI,
TM2, TM3, TM4, TMS, TM7, DEM and aspect model
(Zhao and Li 2001). DEM can help identify the LUC type with
similar spectral characteristics but different elevation
information, and aspect model can correct the changed spectral
information caused by the mountain shadow (Mo and Zhou
2000). Eight LUC types were defined based on the previous
research works, which are (1) conifer forest, (2) mixed conifer
and broadleaf forest, (3) deciduous broadleaf forest, (4) bamboo,
(5) shrub-grass land, (6) farmland and settlement, (7) bare-land
and rock, (8) water (Liu 2001).
855
The stratified random sampling method (Table 1) and the
traditional maximum likelihood classification (Singh 1989,
Chen et al. 2004) were applied for image classification and
LUC mapping. The total samples used for classitying 1997-
image are 1578 points and they basically cover the whole study
area. Due to that the 1988-image was acquired 10 years ago, we
made some adjustment during classifying 1988-image based on
our knowledge on the spectrum, ground objects and experience
from 1997-image classification. Therefore, some more sample
points were taken into account in order to increase certainty.
The final number of samples to classify 1988-image is 1848.
The error matrix and Kappa methods were used to assess the
LUC mapping accuracy. The overall mapping accuracy only
considers the correction of diagonal elements in the matrix,
while the kappa method also takes the other elements in the
matrix into account, which can compensate the disadvantage of
the error matrix method (Liang et al. 2002, Ding et al. 2001). In
general, the mapping results are listed very good, better, good,
normal, bad, worse, very bad in turn when the kappa values fall
into the ranges of 0.8-1.0, 0.6-0.8, 0.4-0.6, 0.2-0.4, 0-0.2, and
«0 (Liu et al. 1998).
3.3 Detecting the LUC change and quantifying the change
The LUC change map was obtained by minus calculation of
two produced LUC maps from 1988 and 1997 images. The
research work detected not only whether changing happened or
not but also what kind of changing happed and how much
changing happened. The changing areas and the changing rates
for eight LUCs were statistically analyzed (Ye et al. 2002).
