Michele Crosetto 
  
  
  
—#-— Atmo. Corr. 
—iüi-— No Corr. 
  
  
Correlation 
  
CO OC 0 OS O00 00 D0 CD 
NO NO O00 C QV t Qoo Co C t$ NO 0. 
-— Co o ND D 00 QOO C € t wv) o o0 
w— 00H Áo 0 — — — 
Distance [m] 
  
  
  
Figure 4: Descending DEM. Autocovariance function of the height differences (InSAR versus reference DEM) before 
correction of the atmospheric distortions (“No Corr.”) and after correction using optical data (“ Atmo. Corr.”). 
  
  
  
  
Terrain Mean Error Standard Deviation 
Type [m] [m] 
hilly/flat 0.11 9.34 
mountainous -0.28 16.56 
entire area - 0.01 12.11 
  
  
  
  
  
Table 7: Ascending and descending data fusion DEM results. 
Compared with the 30 m ascending grid, the 60 m descending one is less affected by atmospheric artefacts (the 
correlation length of the height differences is about 118 m, see Figure 4). We compensated for the atmospheric 
distortions in the descending grid using the 250 m DEM coming from optical images described in section 2.1.3. The 
new InSAR DEM was compared with the 60 m reference one (the relative statistics are reported in Table 6). The 
increase of the DEM precision (the global standard deviation drops from 20.9 m to 15.4 m) and the reduction of the 
correlation length from 118 m to 65 m indicate the effectiveness of the atmospheric effect correction. 
2.3 Ascending and descending data fusion 
In order to perform the data fusion, the ascending and descending grids have to be accurately geocoded with respect to 
the same reference system. The accurate relative geolocation was obtained through the joint InSAR geometry 
calibration based on GCPs and tie points. The ascending and descending grids were fused weighting each grid point 
according to its relative coherence (high weights are associated to points with high coherence). The DEM obtained by 
data fusion was compared with the reference one (see statistics in Table 7). The statistics refer to the entire area covered 
by the ascending DEM. In this area the fusion with descending data gives sensibly worse results than those obtained 
with the ascending grid alone. This is due to the very low quality of the descending data (low spatial resolution and 
much worse precision, compare the statistics in Tables 2 and 6) which is caused by the low coherence of the SAR 
images. 
However, analysing more locally the data fusion DEM, its quality is higher than the ascending one in the slopes facing 
the ascending SAR antenna. In these areas, the ascending and descending grids have very different characteristics. For 
instance, considering a profile along one of such slopes (see Figure 5), the original descending irregular grid has an 
average sampling step of 23 m and a coherence of 0.52, while the ascending one has an average sampling step of 32 m 
and a coherence of 0.18. Along this profile the ascending grid shows huge height errors (due to aliasing errors and to 
low coherence), while the descending and data fusion profiles cope very well with the reference one. Despite the low 
quality of the employed descending InSAR data, this example confirms the effectiveness of the data fusion for areas 
affected by SAR image distortions (foreshortening and layover). 
  
52 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part Bl. Amsterdam 2000. 
E CS m " FAO NN 
eS 9 med YY NO