model, and ACTOR2/3.In this paper, we choose 6S model to 
correct the atmospheric effect of the CHRIS images. Fig 3 
shows the histogram of reflectivity distribution of the original 
image and the resultant images respectively. 
As we all show, the wave peak of the histogram moves to the 
left and the values of reflectivity are diminished after the 
atmospheric correction. Before the atmospheric correction, the 
pixel number and the reflectivity concentrate together, while 
after the processing, the peak of wave moves to the left .The 
values on the image are the real reflectivity of the similar 
objects, which is smaller than the original ones. In addition, the 
range of the reflectivity of the result image is wider than the 
original ones. That is because the reflectivity of water is 
reduced obviously. 
2.2.4 Geometric Correction: In this study, the Quick bird data 
of which the spatial resolution is about 0.6 meter is chosen as 
the warp image, the CHRIS/PROBA images are georeferenced 
using polynomial model and bilinear interpolation. The error is 
in half pixel. Images are resampled in 18 meter* 18 meter. 
3. REMOTE SENSING MODELING RESEARCH 
3.1 Correlation Analysis 
As there are more suspended substance and DOC in the Three 
Gorges Dam, the contribution of chlorophyll to spectrum 
characteristics of water is weaker than the contribution of 
suspended substance. In order to obtain the concentration 
information of chlorophyll effectively, the influence of 
suspended substance to chlorophyll should be restrained or 
eliminated first. AHN (AHN, 2001) simulated the spectrum of 
water and pointed that the existence of suspended substance 
may overrate the concentration of chlorophyll retrieved from 
images; the value of retrieved will be overrated by 20~30% in 
eutrophication water and 2~3 times in undernourishment water. 
Ma Chao-fei (Ma Chao-fei, 2005) analyzed the influences of 
suspended substance to the retrieval of chlorophyll contribution 
and pointed that the ratio of red and blue band (R652/R566) a can 
reduce the influences of suspended substance to the ratio of 
blue and green band (R46\/R566) effectively, moreover, he 
defined the heavy turbidity water, light turbidity water and the 
value of A respectively which respects the grade of turbidity 
water. Based on those definitions, we classify the water of the 
Three Gorges Dam as light turbidity water and the value of A 
is -0.62. 
Figure 4.The correlation of chlorophyll contribution and light 
spectrum band of water 
The ratio of bands can weaken the difficulty of data processing 
greatly. Generally, the correlation of band reflectance and 
chlorophyll contribution is investigated in the remote sensing 
of chlorophyll (Shu Xiao-Zhou, 2000). Figure 4 shows the 
correlation of band reflectance and chlorophyll contribution. In 
this paper, the ratio of the absorption peak and the reflection 
peak R617/R566 are used. The correlation of band 617nm and 
band 566nm is highness. As a matter of fact, the reflection peak 
between 550-570nm, which is the indication of chlorophyll 
existence, is due to the weak absorption of chlorophyll and 
scatter of plant cells. Since the absorption peak of the 
phycocyanins is around 624nm, the reflectance around it 
appears minimum value or shows shoulder shape (Li Su-ju, 
2002). Tests show that if a of the (R652/R566) A is used in 
the retrieval model, the correlation will be better ( R 2 = 0.75 ). 
3.2 Model building and estimation 
Based on discusses as we have done, the retrieval model of 
chlorophyll contribution is formulated as follows. 
CHL = 2.5659x 2 +0.1 8744jc+ 0.98624 (6) 
Where * = fan/>566X r 652/>*566) ( "°' 62) ,{R 2 = 0.75 ) 
r 566 » r ei7 > >652 = the reflectance of 566nm,617,652nm 
respectively. 
Figure5. The regression of chlorophyll contribution and 
combination of band 
The model is validated with the remnant four samples. Table 6 
shows the results. The maximum error is 11%, the minimum 
error is 3% and the average error is 8.68%. 
Samples 
Calculation 
(mg/m3) 
Measurement 
(mg/m3) 
Error 
% 
Sample 15 
3.04 
2.94 
-3 
Sample 16 
3.14 
3.25 
3 
Sample 17 
3.17 
3.37 
6 
Sample 18 
3.19 
3.58 
11 
Table 6. Results of the model error-tested 
3.3 Application