daily GVI images are registered exactly, the
probability that a real area of a daily GVI pixel
in one image overlaps the one in another image is
low. Because the ratio of a real area of GVI pixel
to an area represented by the pixel is very low and
the location of a real area of GVI pixel in the
area represented by the pixel depends on images.
In other words, pixels of so-called multi-temporal
GVI data do not actually coincide geographically.
Therefore, in a strict sense, seasonal vegetation
dynamics by 12 monthly GVI data are an apparent one
produced from GVI data of twelve different small
areas, that is approximately 1.1km by 4.4km( in the
case of nadir observation), within approximately
20km by 20km. This problem is called as
'multi-temporal sampling problem’ in this study.
Though clouds contamination in daily GVI data
is reduced considerably by the above processing
(2), weekly GVI data still have clouds in some
degree.
2.2 Other Data
(1) Wilson's global land cover data
Wilson et al produced data sets which archive 53
categories of land cover types with a resolution of
1° latitude by 1° longitude from a number of
existing maps. The data set represents a land
based survey of surface cover roughly appropriate
to the period of 1950-1970. In their paper(Wilson
et al 1985), they described a method to derive 24
vegetation types (listed in Table 2) from 53
categories. Authors produced images of each
vegetation type in Polar Stereographic projection
where a pixel value represents a percentage of
coverage of a particular vegetation type in a unit
area of 1° by 1°.
(2) Global DEM
This is a product of NOAA/NGDC(National Geophysical
Data Center) which is called 'ET0P05'. It includes
global altimetric and bathymetric data at grid
points of 5 minutes latitude by 5 minutes
longitude.
(3) Snow cover data
The official name of this data is 'Satellite
Observation of Variations in Northern Hemisphere
Seasonal Snow Cover data’ which is a product of
NOAA/NESDIS (National Environmental Satellite,
Data, and Information Service). The data indicate
the existence of snow in a unit area by binary
value (0 or 1). 89 by 89 unit areas cover Northern
hemisphere in Polar Stereographic projection. This
data were produced from NO A A AVHRR visible images
by human interpretation.
3. PREPROCESSING
(1) Extraction of Asian region from GVI image
The size of GVI images in Polar Stereographic
projection is 1024 pixels by 1024 lines on Northern
or Southern hemisphere. A part of the image with
512 pixels by 400 lines which covers almost whole
Asia were extracted from weekly GVI images of
Northern hemisphere.
(2) Production of monthly GVI image
In order to reduce remaining clouds and eliminate
noises in weekly GVI images, a monthly GVI image
were produced from four or five weekly GVI images.
If a pixel has a remarkably low SNVI value among
the other values in a month, it is regarded as a
noise and ignored in the next processing. Since
pixels of clouds have high SNVI(low NVI) values,
lowest SNVI(highest NVI) value is selected to
produce monthly GVI image.
Almost all noises can be removed by the above
processing but it is impossible to produce a
completely cloud free image.
(3) Extraction of land area
Global DEM were used to extract land area from
monthly GVI images in order to eliminate remaining
clouds over the sea.
4. COLOR COMPOSITE
12 monthly GVI images show the seasonal vegetation
dynamics. One method to display the change of the
seasonal vegetation dynamics is to produce color
composite images from selected three monthly GVI
image.
Figure 2 is a color composite image of
April(red), August(green) and December(blue) in
1987. Green region in Siberia shows less
vegetation in April and December. Red regions from
Mediterranean Sea to Persian Gulf and along Hindu
Kush Mountains and Tien Shan Mountains show high
vegetation in April. These areas correspond to the
type C s (Warm temperate rain climates with dry
summer season) of Koppen's classification of
climate. Yellow region in Great plain of China has
high vegetation in April and August. This region
is a typical two-crop area in China. There is
wheat in April and rice in August.
Figure 3 is another color composite image of
consecutive three months, May(red), June(green) and
July(blue) in 1987. Red region in Myanma shows
high vegetation in May and purple region along
Western Gharts in India and In Great Plain in China
has less vegetation in June compared with May and
July. This figure explains the monthly vegetation
changes in three months.
5. PRINCIPAL COMPONENT ANALYSIS
Principal component analysis was applied to 12
monthly GVI data of land area in 1987 in order to
visualize seasonal vegetation dynamics throughout
a year.
Figure 4 shows eigen vectors for the 1st
principal component(PC) to the 6th PC. The meaning
of the 1st PC is the sum of NVI in a year and the
2nd PC corresponds to the difference between
summer(July and August) and winter(November to
March). The 3rd PC explains the difference between
April and July to January. The 4th PC shows the
difference between May, October, November and
February, July. The 5th PC is the difference
between July, December and April, September. The
6th PC is the difference between April, July and
May, June.
Figure 5 and 6 are color composite images of
the 1st to the 3rd PC and the 4th to the 6th PC
respectively. Each color shows the degree of the
meaning of each principal component. For example,
red color of Kalimantan in Figure 5 shows high
vegetation throughout the year. Blue region in
northern India in Figure 6 has high values of the
6th PC which means that there is more vegetation in
May and June than April and July.
6. CLUSTER ANALYSIS
Since land covers with different vegetation types
have different seasonal vegetation dynamics, land
cover can be classified using 12 monthly GVI data.
In this study land cover in Asia were classified by
cluster analysis using 12 monthly GVI data in 1983.
Applied algorithm of cluster analysis is K-means
method with merger of clusters.
Figure 7 shows classified 16 clusters in Asia.
In order to investigate the similarity of each
