364
(6),
X = 127 ( 1
R • ) / ( R - R ■ )
min v max min
Where I is the mean theoretical intensity , R m g X and R m £ n is
a gain and offset factor, respectively. These factors are
known quantities , namely, R ma x = 2*63 and R m in = ^*^^ for band
4 of Landsat II.
We apply these curves in Figure 3 to the band 4 data
covering the Kanazawa area taken on May 23, 1979 by the
Landsat II. The solar elevation was 58°at the time of
observation, we need also to have information on ground
albedos for a few uniformly extended areas. For such test
areas, we choose the sea and coastal sand dune in the image.
The observed average CCT levels and the measured mean albedo
values for these test areas are tabulated in Table 1 at the
wavelength X =0.55>um. We may find a curve with an _
appropriate value of t which passes throuth the points (X,A)
having both the observed CCT level and the measured albedo
value of test areas. It is found from Figure 3 that a curve
with t =0.4 passes through_such two points, more
specifically, a point of ( x S ea = ^* 2i *» A sea : ^’^ ) and that
°f ^ X sand = ^”' - ^’ A sand = P* 1, ( / * Similarly, we can estimate
the value of tfor other band by making another curves of X
and A.
Once the value of t at the time of observation is
obtained,then the AECS can give a ground albedo data set.
In Figure 4 we show the 11 class-classification map of the
Kanazawa area ( 256 x 256 pixels ) computed from the
original CCT level data set taken by the Landsat II on May
23, 1979. We also show the same classification map in Figure
5 , except that the ground albedo data set made from the May
23,1979 Landsat II data set is used in the computation. In
both cases the cassification is performed by a
maximumlikelihood dicision rule and a supervised method
choosing the same test sites in four band space. We can
recognize that the quality of the image based on the ground
albedo data set is much improved, compared with that in
Figure 4. For example, the river systems,water way
connecting the port to the lake and several valleys entering
the mountain regions are more visible in Figure 5 than in
Figure 4. This example shows clearly the importance of the
correction for the atmospheric effect in the remote sensing
data.
TEMPORAL CHANGES IN CLASS STATISTICS
Let us consider the temporal changes in the CCT level
between two Landsat data sets covering one particular
geographical location ,namely, the Kanazawa area, taken at
different time. The statistical regions for four typical
pattern classes are plotted in the feature space of band 5
and 6 in Figure 6. The solid line elipses represent the
reliability range of certain pattern classes based on the
May 23, 1979 Landsat data set. The broken line elipses are
the same pattern class ranges, except that they are based on
the Oct.23, 1979 Landsat data set. The center of the elipse
is given by the mean vector point of a certain pattern class
and its boundary is drawn with 95 % reliability limit. The