36 
cluster, pixel data of each cluster were plotted in 
the three dimensional principal component(PC) space 
of the 1st to the 3rd PC. Figure 8(1) shows the 
projected distribution of each cluster on to the 
lst-2nd PC plane by looking from the positive side 
of the 3rd PC. Figure 8(2) is the same image 
except on to the 2nd-3rd PC from the negative side 
of the 1st PC. As shown in Figure 8, each cluster 
distributes continuously and this means that there 
may be another possibility to get different 
clusters by using different algorithm or parameters 
of cluster analysis. To cope with this uncertainty 
of clusters, 16 clusters were merged into 7 groups 
by comparing seasonal vegetation dynamics of each 
cluster. Figure 9 shows mean value curves of 
monthly GVI data for each cluster in which similar 
curves are drew in the same graph. Vertical scale 
is not SNVI and is changed to NVI for better 
understanding of vegetation dynamics. Figure 10 
shows 7 groups of land cover in Asia. 
Classified results (Figures 7 and 10) were 
compared with DEM image (Figure 11), snow cover 
data (for example Figure 12) and Wilson's land 
cover data (for example Figure 13). Table 3 shows 
land cover types and its distribution of each 
cluster which were put in order with reference to 
the above images and some documents (Justice 1985; 
Willett 1987). 
Classified results have the following 
characteristics besides the ones described in Table 
3. 
Some part of boundaries of different land 
cover clusters quite coincide with the elevation 
pattern in Figure 11, for example Himalaya Range, 
Plateau of Tibet, the Irrawaddy valley in Myanma, 
plain in Indo-China Peninsula, Sulaiman Range in 
Pakistan and Tien Shan Range. 
Figure 12 is monthly snow cover images in 
February 1983 which is one of the 12 monthly 
images. It explains number of weeks with snow 
cover in a specific month. By snow cover 
data(Figure 12), it was found out that the low NVI 
values from October to next April of land cover 
group 3 in northern Eurasia are due to snow covers. 
7. CHANGE DETECTION 
One of the objectives of land cover monitoring in 
this study is to detect areas which have land cover 
changes. If different land cover has different 
seasonal vegetation dynamics, GVI data can be used 
to detect land cover change. 
Seasonal vegetation dynamics in a year by 12 
monthly GVI data can be represented by a point in 
the 12th dimensional space. Therefore change of 
seasonal vegetation dynamics between two different 
years can be measured by Euclid distance between 
two points in the 12th dimensional space. 
Figure 14 shows level-sliced Euclid distance 
image between 1983 and 1987. Dark color of the 
figure means smaller Euclid distance which explains 
that there is no land cover change within this time 
period. Red color means larger Euclid distance 
which does not directly mean a land cover change in 
this area. Because an apparent land cover change 
may occur by 'multi-temporal sampling problem’ or 
by cloud contamination mentioned in the section 
2.1. In the area where land cover is not uniform 
in GVI pixel size, that is approximately 20km by 
20km, 'multi-temporal sampling problem' causes 
wrong vegetation seasonal dynamics and results in 
apparent land cover change between two different 
years. If some areas are covered by cloud during 
the whole month, this causes wrong seasonal 
vegetation dynamics and leads to an apparent land 
cover change. 
8. CONCLUSIONS 
Monthly GVI data was found as a strong tool for 
global land cover monitoring by their information 
of seasonal vegetation dynamics which is 
effectively displayed by color composite of 
arbitrary three monthly GVI images or principal 
component images. 
Cluster analysis is a useful technique for 
global land cover classification. Asia was 
classified into typical seven different land cover 
types by cluster analysis of 12 monthly GVI data. 
It was found out that simple difference of 
monthly GVI data between two different years does 
not work for land cover change detection because of 
'multi-temporal sampling problem' and cloud 
contamination. One solution to cope with 
'multi-temporal sampling problem’ is to use GAC 
data which are the original data of GVI data. For 
the problem of cloud contamination, there are two 
approaches, one of which is to detect cloud using 
other data and the other is the use of microwave 
remote sensing. 
References 
Goward.S.N.,Dye,D..Kerber,A. and Kalb,V, 1987, 
"Comparison of North and South American biomes 
from AVHRR observations".Geocarto 
International, No.l, 27-39 
Holben.B.N., 1986, "Characteristics of maximum- 
value composite images from temporal AVHRR 
data", Int.J.Remote Sensing, Vol.7, No.11, 
1417-1434 
Johnson,G.E.,Dijk,A.V. and Sakamoto,C.M., 1987, 
"The use of AVHRR data in operational 
agricultural assessment in Africa", Geocarto 
International, No.l, 41-60 
Justice,C.O..Townshend,J.R.G.,Holben.B.N. and 
Tucker,C.J., 1985, "Analysis of the phenology 
of global vegetation using meteorological 
satellite data", Int.J.Remote Sensing, Vol.6, 
No.8, 1271-1318 
Justice,C.O..Townshend,J.R.G. and Choudhury.B.J., 
1989, "Comparison of AVHRR and SMMR data for 
monitoring vegetation phenology on a 
continental scale", Int.J.Remote Sensing, 
Vol.10, No.10, 1607-1632 
Kennedy,P.J., 1989, "Monitoring the phenology of 
Tunisian grazing lands", Int.J.Remote Sensing, 
Vol.10, Nos.4 and 5, 835-845 
Kerr.Y.H.,Imbernon,J. ,Dedieu,G.,Hautecoeur,0., 
Lagouarde,J.P. and Seguin.B., 1989, "N0AA AVHRR 
and its uses for rainfall and evapotran 
spiration monitoring", Int.J.Remote Sensing, 
Vol.10, Nos.4 and 5, 847-854 
Malingreau,J.P..Tucker,C.J. and Laporte.N., 1989, 
"AVHRR for monitoring global tropical 
deforestation", Int.J.Remote Sensing, Vol.10, 
Nos.4 and 5, 855-867 
SDSD,1986,"Global vegetation index users' guide", 
Satellite Data Services Division(SDSD), 
National Climatic Data Center,NESDIS,NOAA,U.S. 
Department of Commerce, Washington,D.C., USA 
Singh,S.M., 1988, "Lowest order correction for 
solar zenith angle to Global Vegetation 
Index(GVI) data", Int.J.Remote Sensing, 
Vol.9,Nos.10 and 11, 1565-1572 
Townshend , J. R. G. .Justice, C.O., Choudhury.B. J. ,Tuck 
er.C.J..Kalb.V.T. and Goff,T.E., 1989, "A 
comparison of SMMR and AVHRR data for 
continental land cover characterization", 
Int.J.Remote Sensing, Vol.10, No.10, 1633-1642 
Tucker,C.J..Justice,C.O. and Prince,S.D., 1986, 
"Monitoring the grasslands of the Sahel 1984 
-1985", Int.J.Remote Sensing, Vol.7, No.11, 
1571-1581