51
5-3. Discussion on correlation
When we examined the coincidence of tidal fronts extracted by single band and multi-band
clustering of NOAA image data with the tidal fronts confirmed by sea truth, it has clear that the
extent of coincidence was almost the same. Table 4 shows the distance average between the
tidal fronts obtained from sea truth and NOAA image.
Table 4: Average distance between the tidal fronts obtained from sea truth and NOAA
image
Observation date of sea
truth
Observation date of
satellite NOAA
Average distance by
single band data
Average distance by
multi-band data
1992:May 21
1992:May 21
4.49
4.48
1992:Aug 28_1
1992:Aug 28
6.54
6.60
1992: Aug 28_2
1992:Aug 28
5.59
5.78
1991: Sep 3_1
1991:Sep 3
0.89
0.87
1991: Sep 3_2
1991 : Sep 3
1.06
1.01
1991: Sep 3_3
1991:Sep 3
1.78
1.75
1991: Sep 5
1991:Sep 5
5.31
5.31
1991: Sep 6_1
1991:Sep 6
1.07
0.59
1991 : Sep 6 2
1991 : Sep 6
1.61
1.92
Average distance is number of pixel
It is considered from the above results that the multi-band clustering is not necessarily more
effective than the single band clustering as the method to classify the water blocks. The tidal
fronts obtained from NOAA image coincided relatively well with the tidal fronts confirmed by
sea truth as a whole. However, there are some cases where these tidal fronts were largely
different. This is considered to be due to the difference of observation date and the influence of
climate and sea current, etc. It should be necessary to consider the time serial changes including
those due to climate and sea current to improve the accuracy of tidal front estimation.
6. Conclusion and problems in the future
It has become clear from this study that there is a good possibility to estimate the location
of tidal fronts using relatively simple methods such as preliminary processing of NOAA data
and applying single band clustering method to temperature distribution data (MCSST).
However, there are still some problems to be solved in constructing practical method based on
automatic processing because there is a lack of stability in extracting the tidal fronts as the
coincidence with the tidal fronts of sea truth is largely deteriorated in some cases and also
because some methods used in this study necessitate the judgment of specialists. There are
following problems which should be solved in the future.
(1) Discussion on the automatic processing method
In this study, the judgments of specialists were jointly used for establishment of threshold
for eliminating cloud area and land area and extraction of tidal front on the water block
classification image. It is necessary to discuss the automatic judgment method which does not
necessitate the specialist's judgment in developing the practical method.