International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
DEM 
20m<a<40m 
40m<a<60m 
o>60m 
InSAR 
0% 
0% 
0% 
Stereo 
1.79% 
0.1% 
0.02% 
Fused 
0% 
0% 
0% 
Table 3. Percentage of points with height errors greater than 
20m in test site 1. All errors of the SPOT DEM are 
rejected by the fusion process. 
Much like in test site 1, the amount of extreme errors decreased 
drastically in test site 2. From the initial 4% (SPOT) and 17% 
(ERS), the percentage of errors fell to 1% in the fused DEM 
(Tab. 4). No errors greater than 40m occurred. The remarkable 
decrease of errors in the ERS DEM indicates that the fusion 
method also copes well with the InSAR error properties. 
DEM 
20m<a<40m 
40m<a<60m 
o>60m 
InSAR 
15.77% 
1.34% 
0.02% 
Stereo 
4.07% 
0.22% 
0.03% 
Fused 
1.04% 
0% 
0% 
Table 4. Percentage of points with height errors greater than 
20m in test site 2. Only few errors greater than 20m 
occur. 
The signed average (bias) is an indicator for the presence of 
systematic errors. For test site 1, the signed average is already 
low, and the fused DEM leads to an averaging of this statistic 
between ERS and SPOT results. For test site 2 where the signed 
average is larger, it is nicely shown that through the DEM 
fusion, the signed average is improving with respect to both 
ERS and SPOT DEMs and not simply averaged. The absolute 
averages of both sites before and after the fusion are given in 
Table 2. The absolute average after the fusion decreases to 3.2m 
in site 1, meaning an improvement of 14% (ERS) and 46% 
(SPOT) respectively. The absolute average of test site 2 
decreases by 57% (ERS) and 28% (SPOT). The results prove 
that the proposed data fusion method does not only reduce the 
statistical errors, but also is capable of reducing systematic 
errors and blunders, hence dealing with all occurring error types 
in DEMs. 
It is important to note that both results prove the ability of the 
method to fuse using only the less error-affected values. In both 
cases, the absolute average and RMS are reduced by fusing two 
DEMs of very different initial accuracy (in test site 1, the SPOT 
RMS is almost double of the ERS RMS). These results indicate 
that the assumption, that the correlation coefficient can be used 
as a quality indicator for both DEMs, is valid. Simple averaging 
would not lead to these results. 
Fig. 4. SPOT DEM of test site 1 (3D view from northwest, Z- 
axis scaled 5 times). The terrain is hidden behind 
spikes. 
Fig. 5. DEM of test site 1 after the fusion (3D view from 
northwest, Z-axis scaled 5 times). No spikes passed 
the fusion process. 
7. CONCLUSIONS 
A procedure for the fusion of InSAR and stereo DEMs has been 
introduced and tested. The proposed procedure represents a 
framework to achieve very accurate DEMs from spacebome 
remote sensing data. It takes advantage of the synergy between 
InSAR and stereo-optical DEM generation, by weighting the 
height values in both DEMs according to the estimated error. In 
this way, InSAR and stereo-optical DEMs, even of very 
different accuracy, can be fused to a DEM of higher accuracy. 
The resulting DEM improves in terms of 
• amount of valid measurements (reduction of holes), 
• mean and RMS Error, and 
• error distribution (extreme reduction of large errors), 
indicating that the effects of all types of errors, occurring in a 
DEM, are reduced.