Istanbul 2004 International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Here, the author introduced a modified distance (D) from the
line connecting two vertices, where the probability density of
winter wheat was presumed as 0 %.
wheat
vegetation
and
= dv Q2)
dv + dw
The author examined the relationship between D and the
percentage of winter wheat sown area in a pixel. As he plotted
the data on graph, the distribution pattern tended to align along
the line of logistic function. Then, he discovered a relationship
expressed by a logistic formula shown in Equation (3). The
mend parameters used in Equation (3) were given in the case of 2001
0.6 and its coefficient of determination (R^) was 0.8006.
P(%) = 100 (3)
Land the 14- 89.1928exp(-9.024D)
Figure 7 shows the estimated distribution of winter wheat sown
area in and around Gaotang county for the case of 2001. Figure
7(a) represents the calculated probability density of winter
wheat sown area obtained from LANDSAT-ETM- data within
33 by 33 pixels window. Figure 7(b) represents the result of a
linear unmixing method, where the value indicates the average
of estimation for 4 combinations of NDVI data, i.e. May 1-10
counties. [t is
1 all the values
ability density.
ed in Figure 6.
O Shunyi and June 1-10, May 1-10 and June 11-20, May 11-20 and June
@Dacheng 1-10, and May 11-20 and June 11-20. This process was aimed
OFeixiang at the reduction of estimation error due to the spatial
O Gaotang heterogeneity of influence of atmospheric condition on 10-day
e Wy composite NDVI data. Figure 7(c) represents the result of
© che application of the logistic function shown in (3). This figure
eee describes that the patterns of distribution of high and low
percentage areas are generally reproduced by either method.
However, by examination of the result at pixel level, the
» estimated values would not be properly expressed at the parts of
medium level of percentage especially for the method using
logistic function. This could be mainly caused by the result of
ely geometric correction, which remained some tolerance of
0.6 registration of geographic location of pixel.
Figure 7. Comparison between methods to estimate winter
wheat sown area in and around Gaotang county
((a) LANDSAT-ETM+, (b) linear unmixing, (c) logistic on 2-
temporal scattergram)
Even though the estimated values involve small errors in terms
of percentage of acreage in a pixel, it would be useful to obtain
ility density of the distribution of winter wheat sown area at an appropriate
spatial unit in the wide range. Figure 8 shows the comparison of
percentage area of winter wheat sown area aggregated by
county as a unit. The estimated values by the method applying 70 r OLANDSAT-ETM+
linear unmixing indicates the higher correlation with the values HE linear unmixing
eat (0%) obtained from LANDSAT-ETM»- data. The values by the other [logistic on 2-temporal
method also shows considerably good correlation, so that the
estimated values by either method would be accepted to apply
to analyze the characteristics of temporal changes of the winter
wheat in the Huang-Huai-Hai plain.
Because the author could not collect the high spatial resolution
data for every year, the method using logistic function was
selected to analyze the change of winter wheat sown area.
Figure 9 shows the results of estimation. He noticed that the
fluctuations with large amplitude were figured for all the
Ratio of winter wheat sown area (%)
9 - - -—
heat (10096) counties before 1997. Contrastively the systematic patterns of g = 2 2 & 5
es change were recorded after 1998. This feature was presumably & = $ s £ E
id May) caused by the problem of pre-processing of the original dataset à > be
and not by the defect of the estimation method. Then, the
1 winter wheat estimated values after 1998 could be used to analyze temporal Figure 8. Comparison of percentage area of winter wheat by
characteristics of winter wheat sown area. county among different estimation methods
141
