Quite a few criteria have been defined in the literature to 
evaluate the performance of a model (Wang and Elhag, 2007). 
These criteria include the sum of squared error (SSE), mean 
square error (MSE), mean absolute error (MAE), root mean 
squared error (RMSE), absolute percentage error (APE), root 
mean square percentage error (RMSPE), correlation coefficient 
(RY), and so on. Among of them, RMSE, APE and R? are the 
most widely used performance evaluation criteria and will be 
used in this study. They are defined as follows: 
2 
  
R= E 6 55) (9m 3) 
i=] i=1 : (2) 
2 2 
n n 
G7») 3o») 
=] i=l 
(3) 
  
18S bs x«l 
APE - 21009 4 
2 ni x : e 
  
Where yu; Ya: - y, are the predicated value, 
measured value, average measured value and average 
predicated value, respectively, n is the sample number. 
4. RESULTS 
4.1 TIN inversion model 
  
  
  
   
    
  
   
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012 
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia 
    
According to situ measured TIN data and matched four input 
parameters, a multiple regression model was established. 
Namely, 
y=0.685-0.004x, +0.018x, +1.930x, +0.137x, (5) 
Where y is TIN, X, , X , X4 , X, are VH,HV,Band4 and NDSI, 
respectively. The summary statistics for the above models of 
TIN (at the 9594 confidence level) are as following: R°=0.774, 
F-value-20.48, constant, X, , X,, X, and X,of T-value are 
5.74,1.12,4.37,1.40 and 2.22, respectively. In addition, 
According to the above three parameters for assessing the 
performance of model, RMSE=0.063, APE=8.651%, it 
indicated that the model had relative low RMSE, APE and high 
R° value. 
4.2 Spatial distribution of TIN 
According to the above multiple regression model based on situ 
measured data, the model established was applied to calculate 
TIN of area covered by two images in the sea. Figure 4 shows 
the spatial distribution of TIN in sea surface of study area. As 
seen in Figure 4, TIN of sea ranged from 0 to 0.30. The 
majority of study area had a low value of TIN with ranging 
from 0.05 to 0.10. However, the area adjoining to land had a 
high value of TIN above 0.20. It inferred that the area was 
nearly influenced by land matter. It still agrees well with the 
actual distribution of TIN in sea. 
Based on the above analysis, the multiple regression model for 
predicting TIN in sea surface performed well. It confirmed that 
the fusion of optical data and SAR data was effective. 
TIN Gus L) 
zo 
0. 29-0. 30 
9. 15-ü. 20 
9. 10-0. 03 
ü. 05-0. [0 
Figure 4. Distribution of TIN in sea surface of study area 
5. CONCLUSION 
RADARSAT-2 quad-polarization image and HJ-1 CCD image 
have been used to estimate TIN of sea surface. Based on the 
situ measured data, four parameters were selected as sensitive 
factors. Moreover, the multiple regression model, which 
interprets the variation of TIN as a function of sensitive factors. 
According to the models and correlation analysis, the results 
can be summarized in three points: 
    
1) Band4 reflectance and NDSI are relative sensitive to TIN of 
sea surface with R° above 0.3. 
2) Compared with HH and VV, VH and HV has a better 
correlation with the change of TIN of sea surface, implying the 
advantages of cross-polarization radar backscatter in sea 
biochemistry monitoring. 
3) Fusion of optical data and SAR data can improve the 
accuracy for estimating TIN in sea surface. It is because 
different image data sources can get more subtle information 
for oceanic biochemistry from different view.