47
4-1. Preliminary processing of satellite image data
(1) Cloud area and land area
We excluded the clouds and lands to extract the study sea area. The identification of clouds
and lands area are easily distinguished among the visual, near infrared and temperature
distribution data.
(2) Discussion on smoothing method
For some cases, large difference was observed between the adjoining pixels in the
temperature distribution. It is considered that this is due to the influence of thin cloud and vapor
or the noise occurred during the data acquisition.Therefore, we discussed the method to
remove these noises by smoothing processing. As for smoothing processing, we employed the
smoothing matrix (low pass filter 2 x 2 - 7 x 7) to temperature distribution data. And,for some
cases,we applied ABIC minimization method to smoothed data.
4-2. Extraction of tidal front
We examined the three different methods for the water block classification.In the first
method,the temperature distribution data of NOAA image is clasified by applying single band
clustering and level slicing. In the second method,ISODATA method clustering is carried out to
the multi-band data (i.e. visual, near infrared and temperature distribution).In the last
method,the maximum likelihood method is applied to the multi-band data. The borderline of the
classified water blocks was regards as the tidal front and used for the discussion.
4-3. Discussion on correlation
For the discussion, we chose the estimated tidal front,which corresponds to the sea
truth,from the borderlines of classified water blocks. Comparison between the estimated tidal
front and sea truth was carried out by using the distance average.Distance average was
calculated by averaging the distance of both front from north to south.The distance in the
direction of east to west is used for the calculation.
5. Results and consideration
5-1. Preliminary processing of satellite image data
(1) Removal of clouds and lands area Table 2:The data and CCT values with distinction of clouds and lands
Clouds were extracted with relatively
high accuracy But,for some cases,
distinction of land and sea area was
difficult,because the both areas had the
same CCT values. In this study, we
established the threshold value from the
histogram of each data by human
judgment. There remained some issues
yet to be solved in objectivity and
processing time because it was
necessary to establish the threshold
value respectively for each data. Table
2 shows the data and CCT values with
distinction of clouds and lands.
Day of
observation
Lands
Clouds
Data*l
CCT
values
Data*l
CCT
values
1992 May 21
N
64- 115
V
84 - 255
22
N
57 - 108
V
90 - 255
24
V
54- 76
V
100 - 255
25
V
60- 90
V
120 - 255
1991 July 21
N
60- 130
V
100 -255
22
N
50- 130
V
100-255
23
V
0- 85
V
121 -255
T
207 - 255
N
120 -255
T
0- 35
24
V
0- 80
N
100 - 255
1992 Aug. 28
N
50- 70
V
61 -255
29
N
53 - 80
V
74 - 255
30
N
53 - 86
V
76 - 255
Sep 1
N
66- 79
V
81 -255
1991 Sep 1
N
53 - 105
V
90 - 255
2
N
67- 105
V
110-255
3
N
50- 100
V
83 - 255
5
N
61 - 100
V
77 - 255
6
N
57- 95
V
67 - 255
* 1 V: visual data
N near infrared data
T temperature data