41 
2. Developing database and model 
Fig. 1 shows the routine observation points covered by Osaka prefectural fisheries 
experiment station. The color of sea database and the model derived from the database are based 
on the data collected at 6 points aligned with the long axis of the bay. Fig. 2 illustrates the flow 
of the analysis. One of the new approaches adopted in this study is that the model is not based 
on the direct relationship between Landsat TM and in situ data, but the model is solely based on 
in situ data and accepts the atmospherically corrected Landsat TM data later. The estimated 
transparency from Landsat TM data is compared with coordinated in situ data and the model is 
verified. Fig.3 indicates the distribution of chromaticity in the chromaticity diagram. The 
chromaticity is scattered around the white point(w). In order to evaluate the chromaticity the 
concept of main wavelength and purity is adopted shown in Fig. 4. The main wavelength is 
defined as the wavelength(in unit of nano meter) on the curved spectral locus and corresponds to 
hue. The purity is defined as the index of closeness to the white point(w). Therefore the 
chromaticity with purity 100% is located on the spectral locus, while the chromaticity with 
purity 0% is located on the white point. Fig. 5 and 6 show the relationship between transparency 
and purity, transparency and main wavelength respectively. While the relationship between 
transparency and purity is not clear, the relationship between transparency and main wavelength 
show an clear negative correlation. Fig. 6 also shows a logarithmic curve fitted with this 
relationship, which is later used as an empirical model for estimating transparency from Landsat 
TM data. 
3. Application of Landsat TM data to the model 
Atmospheric correction is always the main issue in the area of ocean remote sensing 
especially in the visible range. This study adopted the so called “Dark-object-subtraction 
technique” presented by Chavez (4) to estimate the contribution of atmospheric scattering for each 
TM band. Fig. 7 shows an example of the atmospheric contribution to the Landsat TM visible 
bands. Dark portions of the bar graph represents the scattering effect and the rest of the graph 
represents the corrected digital counts. In this example “Very clear”model is adopted so that the 
scattering contribution is greater in the shorter wavelengh bands. Atmospherically corrected 
Landsat TM data is then used for calculating tri-stimulus values X,Y,Z, chromaticity and main 
wavelength with the equations shown below. The main wavelength is then input to the model 
developed above and the transparency is estimated and compared with in situ data.