dentata Thunb.). Site B is a typically flat region of reed 
vegetation (Phragmites communis Trim.) in the Nakdong 
River estuary . The estuary is well-Known as winter habitat for 
migratory birds and is characterized largely by sandbars and 
sand dunes. 
JERS-1 OPS data for the study area were obtained on 26 
December 1992, 11 : 23 A.M. with a solar elevation of 28.9? 
and a sun azimuth of 160.8?. The system correction which has 
performed by RESTEC (Remote Sensing Technology Center) 
of Japan was made up to BSQ Level 2. The OPS VNIR sub- 
system has two bands in the visible region (band 1 : 0.52 - 
0.60pm, band 2 : 0.63-0.69um), and two bands in the near 
infrared region (band 3,4 : 0.76 - 0.86pm) including one band 
for stereoscopic view (forward looking) with a ground 
resolution of 18.3m x 24.2m. 
3. METHODOLOGY 
A total of 137 sample polygons of mountainous forest 
were selected within the study area image for four different 
aspects : N(316° - 45°), E(46° - 135°), S(136° - 225°), and 
W(226° - 315°). The digital numbers of three bands (green, 
red and near infrared) were measured for each of these 
polygons. In order to minimize the heterogeneity of the 
samples and to explore bidirectional effects on different 
canopy types resulting from vegetation and topography the 
normalized difference vegetation index (NDVI) and the 
transformed vegetation index (TVI) were investigated. 
In this study the NDVI was calculated as (OPS 3 - OPS 2) 
/ (OPS 3 + OPS 2) and the TVI was calculated as [(OPS 3 - 
OPS 2) / (OPS 3 + OPS 2) + 0.5 ] "”, where OPS 2 and OPS 3 
are the DN values in OPS band 2 and 3 , respectively. 
For most vegetated surfaces with variation in reflectance 
directionality the obtained vegetation indices were unable to 
discriminate the detailed classes for land cover when the 
change rates of vegetation indices against for different aspects 
were larger than the change rates of reflectance values for 
different aspects caused by the effect of shadows due to steep 
slopes and low solar elevation. The maximum-likelihood 
classification algorithm was also unable to classifying the 
mountainous areas accurately. 
On the other hand, water surfaces were flat so that those 
satellite images could be classified for water properties and 
qualities with maximum likelihood method. 
During the course of this study I found that the effects of 
shadows on the classification of JERS-1 OPS images could be 
reduced through the SAM algorithm. The SAM algorithm 
determines the spectral signature similarity between the 
representative spectral mean DN values calculated from the 
training field pixels and the spectral DN values derived from 
each pixel in the image through the spectral angle differences 
(angluar distance in radians) between their vector directions in 
n-dimensional (band) space. A more detailed mathematical 
description of the algorithm, concepts and applications of the 
program are described in CSES and CIRES(1992), Research 
Systems, Inc.(1995), and Yuhas and Goetz(1993). 
The OPS VNIR data were processed and analyzed using 
the Environment for Visualizing Image (ENVI) image analysis 
and the Arc/Info and ArcView GIS software on a Sun sparc 
10/51 workstation in the Department of Forest Resources 
Remote Sensing Laboratory at Kookmin University, Seoul, 
Korea. 
342 
4. RESULTS AND DISCUSSION 
4.] Directionality of Vegetation Indexes and Spectral 
Reflectances 
The weighted mean values of the spectral digital numbers 
and vegetation indices for four slope orientations obtained 
through 137 sample polygons are given in Table 1. 
In the mountainous forests at a sun elevation of 28.9? and a 
sun azimuth of 160.8°, the spectral reflectance(DN) values 
decrease in the order of south, east, west and north(Figuire 1) 
But the magnitude order of NDVI is neither consistent 
with that of TVI nor of the DNs, and the change rates of NDVI 
values for different directions are larger than those of TV] 
values. 
These phenomena can be explained by the heterogeneity 
of the samples and the variation of bidirectional effects. Table 
2 presents the variation of vegetation indexes due to 
bidirectional effects in the homogeneous forest canopy 
structure of the mountainous site A when compared to the 
vegetation indexes of reeds vegetation on the flat site B. 
Since the both VI values of site B are similar for all four 
reflected directions these normalized relationship methods can 
be used as a winter cover monitoring parameter on flat terrain, 
regardless of illumination under low sun elevation condition. 
Due to the strong illumination variation in mountain area 
with steep slopes the obtained vegetation indexes were also 
stratified by aspect against the four reflective directions. 
4.2 Spectral Signatures Analysis 
As point out in the above result, the spectral reflectance 
characteristics of land covers of the study area could be limited 
to case of flat terrains because it is difficult to refine the 
correction model for topographic effects. Figure 2 shows 
different spectral responses from nadir views of OPS-VNIR 
data (except band 4) for the ground in flat terrains. 
When compared to the other spectral DN values, land cover 
classes of reeds vegetation and sand dune bordered on the river 
and estuary low reflectance values in the green, red, and near- 
infrared band. This phenomenon was influenced by high water 
absorption. The spectral signature distinctiveness between 
cover type sand dune and rock & sands were also related to 
water content which altered the scatter directions of the 
reflectances. 
Typically spectral signature similarity for spectral 
reflectance curves and separability between vegetated and nor 
vegetated cover types on the images can be determined 
through the use of NDVI. This phenomenon is evident in the 
NDVI values for the classes reeds vegetation and fallow & 
arable farming shown in Table 3. 
For water surfaces the VNIR data acquired by an off-nadir 
optical system with a Charge Coupled Device(CCD) are 
relatively insensitive to geometric illumination conditions 
compared with whiskbroom scanning. 
Therefore, the spectral reflectance responses and derived 
vegetation indices could be used to estimate the quality of 
water, monitor the water pollution sources and the turbid wate 
induced by tide in Nakdong River estuary. 
Figure 3 shows the spectral digital numbers and 
vegetation indexes obtained from VNIR values over water 
bodies in Nakdong River and the coastal zone. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
sig 
sh 
Fi 
av 
sh 
ve 
on 
Sy 
co 
ve 
Ta 
pa 
evi 
sel 
rel 
me 
va 
cla 
in- 
ch 
Wi 
un 
ani