International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
reaches about 1,000 km, the time difference of harvest would be 
as small as 3 weeks. 
harvesting in June. In consideration with this feature the value 
of NDVI in the middle of May was allocated in X axis and the 
AREA 
which could 
'at sown area 
ain in China, 
ined a linear 
in proportion 
late period to 
ca. Secondly, 
uld show the 
tion with the 
equire a high 
d, the author 
) estimate the 
ixel of coarse 
ze the spatial 
rop estimated 
examined the 
| applied it to 
major crop 
site NDVI of 
| provided by 
s operated by 
Agriculture, 
pan. He also 
h: 123, Row: 
stern part of 
[he dominant 
ind maize in 
ted except in 
inter wheat is 
| harvested in 
north of plain 
  
Figure 1. Location of Huang-Huai-Hai plain in China 
2.3 Method 
2.3.1 Linear Unmixing: The basic assumption of the method 
lies on linear mixture modelling of NDVI as a function of time / 
expressed in the followings (Uchida, 2001). 
n 
NDVI(1)=S" p,NDVI,(1) (D 
i=l 
where p, — probability density of land use item i in the pixel 
in the condition of > p, «1 
i=l 
This formula can be solved if (n-1) temporal data are given. 
This method is supposed to be applied to NOAA/AVHRR data. 
From the theoretical point of view, temporal changes of NDVI 
for cach land use cannot be uniquely defined due to not only 
atmospheric effects on radiometric characteristics but also 
growing conditions affected by various environmental factors. 
The end-member of NDVI(t), which is the value at the pixel 
wholly occupied by land use item ; when time is /, can be 
estimated by combining with land use information extracted 
from LANDSAT-ETM- data. In this study it is assumed that 
the end-member value, which is obtained at specific site, can be 
used for other sites where physical and land use conditions are 
similar. 
In order to obtain end-member value of NDVI for each land use, 
first LANDSAT-ETM- data was classified by maximum 
likelihood method and converted to probability density value of 
objective land use within 33 by 33 pixels window. The author 
drew a linear regression line in the figure of probability density 
against NDVI of NOAA/AVHRR at the same location, and 
extrapolated it to the value of one of probability density. When 
à linear mixture modeling formula is solved, negative value of 
probability density may come to appear. This is treated by 
addition of values so as to be zero for the minimum probability 
density of land use items and thereafter by scaling to become 
one as summation of total probability density. 
23.2 2-temporal Scattergram: This method was based on 
the feature that NDVI of winter wheat showed a maximum at its 
flowering stage in May and considerably low value after 
value in the middle of June in Y axis. This temporal feature of 
NDVI represented in the scattergram could be discriminative 
from the patterns of other land use types and also might bring a 
formula of estimation of winter wheat sown area. The 
advantageous point of this method is that no higher spatial 
resolution data would be required, if the formula is once set up 
and applied commonly to the case of different years. 
3. RESULTS AND DISCUSSION 
Figure 2 shows the color composite image of 10-day maximum 
composite NDVI, which is assigned the value in the mid-April 
in 2001 as blue, mid-May as green and mid-June as red. This 
figure indicates that the parts represented by resembling color 
tone should have a similar temporal feature of NDVI during the 
period from April to June. It is possible, therefore, to classify 
the Huang-Huai-Hai plain into 2 areas, i.e. the northern part and 
the southern part, in terms of temporal changes of NDVI. This 
suggests that the common parameters would be employed in the 
estimation method for either classified area, respectively. In this 
study the author picked up the northern part, where the 
cropping pattern would be less complicated, for the examination 
of adoptability of estimation methods. 7 counties represented by 
the capital in the figure were the sites used to estimate end- 
member values as well as to verify the estimation results. These 
counties were Shunyi (S) in Beijing Capital Area, Xushui (X), 
Dacheng (D), Wuyi (W) and Feixing (F) in Hebei Province, and 
Yandxin (Y) and Gaotang (G) in Shanding Province. 
*April 1-10 
May 1-10 
June 1-10 
  
Figure 2. Color composite image of 3-temporal NDVI overlaid 
by indication of location of 7 counties used in the analysis 
The author classified LANDSAT-ETM+ data covering the 
northern part of Huang-Huai-Hai plain, and identified 4 major 
categories, which were winter wheat, mixed vegetation, bare 
land and forest, in consideration with the temporal