365 
pattern class numbers in Figure 6 represent the specific 
ground covers. The number 430 and 520 represents the shallow 
water and sand, respectively. The number 210 and 320 
corresponds to the rice field and the forest Respectively. 
We recognize immediately from Figure 6 that the 
temporal changes of the same pattern class in CCT level 
between May and October are too great. Although some 
temporal changes of the biological system are expected due 
to its natural variations with season, the temporal changes 
for the water or sand should be small.The result in Figure 6 
shows the opposite of what we expect. We believe that the 
discrepancy between the above result and the expectation is 
basically caused by the atmospheric effects, such as , the 
different haze thicknesses and solar elevations. Figure 6 
indicates that it is difficult for us to make a quantitative 
comparison among the data sets taken at different time. 
On the other hand , we can expect that the temporal 
changes in spectral responce become small if we use the 
ground albedo data sets because data are express in the same 
absolute unit. Daynamic range of the relative CCT level data 
depends on time (specifically , the solar elevation angle 
), whereas that of the absolute unit data is expected to be 
constant. In Figure 7 we plot the statistical regions for 
the same pattern classes based on the May and Oct.,1979 
Landsat data sets. Figure 7 shows that the temporal changes 
for the water and sand in albedo become very small, while 
those for the forest and rice field still exist, but they 
are not large. It can be said that the location of the 
elipse for a certain class in May is generally close to that 
in October, compared with the result in Figure 6. The 
spectral characteristics of the forest and rice field vary 
in the time interval between May and October. If we use two 
albedo data sets at the same season in different years, 
instead of using the May and Oct. data sets, the temporal 
changes for all classes are expected to be very small. This 
suggests that one Landsat data set could be classified 
according to the class statistics based on another Landsat 
data set taken at different time if these data set are 
converted in albedo unit. Such a signature extension method 
would be a powerful tool in the temporal data analysis. 
Let us now look at the real data, more specifically. In 
Figure 8 we show a 11 class classification map of the small 
area( 45x36 pixels, roughly correspond to 2.7km x 2.2km) 
including the central part of Kanazawa city. The boundary of 
this area is given in Figure 4 by a solid rectangle. This 
classification map is computed by using the ground albedo 
data set on May 23,1979. Our study site is composed of 
mostly the urban , residential area, except a big park(the 
center to the lower left in Figure 8), two shallow water 
rivers(one in the upper right, the other in the lower left), 
and the mountain edges(in the upper right). Since the width 
of these rivers is less than 60m, the rivers are recognized 
as the green banks along them in the classification. The 
classification result agrees fairly well with the land 
utilization map issued from the Geographical Survey 
Institute of Japan. The weather on May 23,1979 was very 
clear. 
In Figure 9 the same classification map is given, 
except using the ground albedo data set on Sept.2, 1980. The 
weather on this day was very bad and we have clouds and