Marc Honikel
Figure 6. Degraded fringe Figure 7. Restored borders
borders. after the Wiener filtering.
Figure 8. Random noise Figure 9. Noise removal by
Occurrence. filtering.
43 Improvement of InSAR topographic height measurement
After the successful application of the filter to the phase noise reduction process, its impact on the interferometric height
determination has been studied. A sub-region of the site of approximately l00knr, containing 1.5 million phase
measurements, has been selected from the ERS data sets for the DEM generation. The phase unwrapping has been
performed with the ghost-line method. For the evaluation of the filtering, two DEMs, one with and the other without
Wiener noise reduction (Fig. 10), have been produced and compared to the SPOT results and the reference DEM.
The SPOT and ERS DEMs have been sampled in a 30m grid and were compared to ground truth by bilinear
interpolation of each height value in the reference DEM. The resulting error statistics in terms of mean, rms and the
maximum errors are given in Tab. |.
Error [m] Stereo-Optical InSAR Wiener filtered
InSAR
Mean 1.4 -2.9 2.1
RMS 8.4 20.8 S.S
a -71.6 -139.4 -37.6
Gun 104.4 84.6 33:1
Table 1. Error statistics of the SPOT, ERS-1 and filtered SAR-optical DEM
In comparison to the original interferometric height measurements, the Wiener enhanced InSAR DEM shows improved
statistics in all categories. The relatively high mean error may be a result of the gcp determination. The difference of 5m
for the mean between the IDSAR DEMSs indicates the systematic changes due to the Wiener filtering. The Wiener
InSAR DEM is considerably improved with respect to the rms error, which decreased by 74%, and maximum errors,
both being fractions of the original ones.
InSAR and stereo-optical height measurements achieve an initial rms error, which is beneath the pixel resolution of the
sensor (SPOT: 10m, ERS-1: 23m). The initial pixel resolution is a fact that is hardly considered, when comparing the
InSAR with stereo-optical DEMs. In this context, it is most remarkable that the Wiener InSAR DEM is superior to the
SPOT height measurements in rms (-35%) and maximum errors, again considerably reduced. No errors higher than
37.6m occurred in the improved InSAR DEM, while in the original InNSAR DEM 27.2% and in the SPOT DEM 2.4% of
all values showed an error higher than 40m. None of the extreme corrupted measurements of the SPOT DEM entered
the fusion process due to the weighting procedure.
These results emphasize the interferometric height measurement capabilities in case of restored phases. It is important
to note that the INSAR measurements have been improved with the simulated phases to an extent that the quality of the
result is higher than the assumed ideal data source.
Finally, the amount of measurements has been increased by this method from 113.000 points, originating from the
SPOT measurements, to more than 118.000 points in the improved InSAR DEM.
5 CONCLUSIONS
A new method for the fusion of optical and SAR interferometric data by applying Wiener filtering for phase restoration
has been introduced and tested. It reduces the effect of noise on the interferometric height determination, namely the
degradation of fringe borders, typically caused by layover, and the occurrence of local phase noise, e.g. due to
vegetation. In this way, new interferometric height measurements, like from SRTM, can be related to and enhanced
with former photogrammetric measurements. The presented procedure helps overcoming the limitations of the
interferometric height determination, by taking advantage from the synergy between optical and SAR height
measurements in critical areas. The noise-adaptive data fusion process preserves valid interferometric measurements,
154 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B1. Amsterdam 2000.
REF
Andr
Gold
Sci,
Honi
Int. \
Honi
Phot
Leic:
Pratt
Patia
Cont
Prati
ISPR
Smal
EAR
Wert
Dete
Zebk
IEEE
