n), (b) 
correlation 
to TIN, R? 
ctral Index 
(1) 
band and 
Our results 
3and 4 was 
NDSI has 
is the input 
    
  
3.0 Fusion of RADARSAT-2 image and HJ-1 CCD image 
In order to accurately estimate the TIN in continuous sea 
surface on a large scale, the fusion of RADARSAT-2 image and 
HJ-1 CCD image was adopted. In this study, the RADARSAT-2 
image in the study area was converted into 30 x 30 m grids by 
nearest neighbor algorithm to keep consistency with HJ-1 CCD 
image in the spatial resolution. The operation was conducted 
using a routine written in MATLAB in combination with 
ENVI 4.8. Two types of image data sources were corrected to 
WGS-84 (World Geodetic System 1984) Geographic datum 
and Geographic Lat/Lon projection system. 
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 
3.3 Spatial distribution of input variables from image 
Sea surface backscattering coefficients were converted to 30 x 
30 m grids for RADARSAT-2 images. Figure 3 shows the 
distribution VH, HV, Band4 and NDSI input variables in the 
same spatial resolution. From Figure3, the backscattering 
coefficient of HV ranged from —45 to 9 dB, whereas VH 
ranged from —44 to 9 dB. HV and VH had the similar range, 
and the former had a little higher range than the latter. Band4 
reflectance had a low value with ranging from 0 to 0.012, while 
NDSI ranged from -0.8 to 0.4. Detail statistics of four input 
variables were calculated in Table 2. 
  
(b) 
  
0. 012 
  
  
  
  
  
Figure 3. Distribution of four input variables in the same spatial resolution 
(a) HV, (b) VH, (c) Band4, (d) NDSI 
  
  
  
  
  
  
  
  
Input variables Min Max Mean Stdev 
HV -44.08 7.98 -32.62 4.72 
VH -43.86 8.07 -32.53 4.71 
Band4 0 0.012 0.0011 0.0008 
NDSI -0.79 0.36 -0.66 0.092 
  
  
  
  
Table 2. Statistics of input variables in study area 
3.4 Establishment of model 
À series of HJ-1 CCD reflectance and SAR backscattering 
coefficient were chose in the study area, in addition, they kept 
the same geographical coordinates of situ measured TIN data. 
The statistical model was created to establish the relationship 
between the above four input parameters and TIN. Of all 
statistical models, multiple regression analyses method was 
most popular (Singh et al., 2009). Regression analysis is widely 
used for predicting and forecasting, and it is also used to 
understand which among the independent variables are related 
to the dependent variable. Therefore, in this study, multiple 
regression analysis was used to establish the relationship 
between four input parameters and TIN. 
3.5 Evaluation of model's performance