The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part Bl. Beijing 2008 
114 
a) 10 GCPs applied b) No GCPs applied 
Figure 4 Subsets of InSAR DEM overlaying Indiana counties 
Both InSAR DEMs (with and without GCPs applied) were then 
registered with reference DEM (SRTM). An example of the 
cross-correlation is illustrated in Figure 5. 
Those conjugate points are listed in Table 1. The seven- 
parameter transformation equations were derived from those 
conjugate points through least squares approach. 
Match Points 
X 
7 
h (m) 
X 
Y 
H(m) 
1 
318 
306 
173.777 
301 
301 
175.494 
2 
919 
303 
142.618 
901 
301 
154.377 
Table 1 Conjugate points • 
The InSAR DEM was then transformed through the seven- 
parameter transformation equations and resampled into regular 
posts. RMSE of the new InSAR DEM was computed against 
the “truth DEM”. 
4. RESULTS AND ANALYSIS 
CroM-Ccirifcfcon 30t-301-Ps»fc <5.17} 
X tO* 
*022 
600 0 
Figure 5 Cross-correlation of InSAR DEM and reference DEM 
The results of RMSE are in Table 2. 
Location 
(Average 
Slope) 
Number of 
GCPs 
RMSE Before 
Alignment (meter) 
RMSE After 
Alignment 
(meter) 
North (1.62°) 
0 
236.054 
4.779 
10 
12.477 
4.085 
South (4.13°) 
0 
117.954 
12.421 
10 
18.370 
12.192 
Table 2 RMSE of InSAR DEM against “Truth DEM” 
GCPs applied INSAR processing yields much more accurate 
InSAR DEM than no GCPs applied, before any registration and 
alignment (12.477 m « 236.054 m and 18.370 m « 117.954 
m). 
The peak location (Ay, Ax) = (5, 17) is the horizontal offsets 
between one patch of InSAR DEM and one patch of reference 
DEM. The conjugate points are determined from the offsets: (x, 
y) = (318, 306) and (X, Y) = (301, 301), as (x,y) - (X, Y) = (17, 
5). Elevations h and H are then acquired through the location of 
conjugate points on InSAR DEM and reference DEM. The 
pattern of conjugate points is demonstrated in Figure 6. 
Match Points and Offsets (Exaggerated) 
X Points on Reference DEM 
O Points on InSAR DEM 
200 400 600 800 1000 1200 1400 1600 1800 2000 2200 2400 
X 
Figure 6 Pattern of conjugate points 
After registration and alignment, the accuracy of refined InSAR 
DEM without GCPs applied improved tremendously, from 
236.054 m to 4.779 m and from 117.954 m to 12.421 m. They 
are also much better than GCPs applied InSAR DEM without 
refinement, as 4.779 m < 12.477 m and 12.421 m < 18.370 m. 
If applying both GCPs in the InSAR DEM processing and 
refinement in the post-InSAR DEM processing, the accuracy 
gets more improved, but not much over refinement alone: 4.085 
m < 4.779 m and 12.192 m < 12.421 m. 
The orbit data and terrain variation have different effects on 
InSAR DEM accuracy. Without GCPs applied or DEM 
refinement, the accuracy is mostly decided by the precision of 
orbit data, as both Indiana north and Indiana south have the 
huge InSAR DEM error (236.054 m and 117.954 m). After 
applying GCPs or refining InSAR DEM, the accuracy is related 
to the terrain variation. Indiana south has the larger error 
(18.370 m > 12.477 m and 12.421 m > 4.779 m), asit has the 
higher average slope (4.13 degrees >1.62 degrees) 
5. CONCLUSION 
Integrating GCPs into InSAR processing produces a final DEM 
with acceptable accuracy. InSAR DEM refinement substitutes 
DEM registration and alignment for applying imprecise orbit