across-track capability. Two antennas displaced across-
track acquire two images observing all image points from
two slightly different angles of incidence. By overlaying
the two complex SAR images during processing and
determining the phase differences (interference genera-
tion), slant range differences can be determined with
fractional wavelength accuracy. From slant range, dif-
ferential slant range measurements, and knowledge of the
baseline, i.e. the displacement from one antenna to the
other, the 3-D target location can be determined in a fully
automated process.
The system presently uses a combination of an inertial
navigation system and a P-code GPS system to provide
the required navigational information. A more accurate
navigation and attitude measurement system is presently
under procurement.
The initial test of the EMISAR single pass mode showed
that the limiting performance factor was multi-path on the
radar platform and limited isolation between the two
interferometric antennas which can be seen to be equi-
valent to a multi-path reflection of the other antenna. For-
tunately, the calibration phase screen has been found to
be basically identical from mission to mission.
Significant amounts of data were acquired during 1995 in
both Denmark and Greenland indicating height accuracies
before multi-path calibration in the order of 5 to 10
meter. By applying the above described phase screen
calibration technique, systematic errors have been dra-
stically reduced. Analyses of data acquired in Denmark
where the topographic relief is moderate (heights vary
from 0 to 137 m in the data analysed) indicates that the
height error after removing a tilted plane is from 1 m in
the near range to 3 m in the far range (at 10 m horizontal
pixel spacing) when operating from 41,000 ft. Data
acquired in the double baseline mode on a 25,000 ft. pass
over the same area indicates stochastic RMS height errors
of 0.6 m in the far range. Studies are presently on-going
to evaluate the height error performance in more detail,
however, sufficiently accurate reference data are not
readily available. A shaded relief image of a subsection
of a 25,000 ft. pass is shown in Fig. 1.
Fig. 1. Shaded relief map of a 5.2 x 3.8 km subsection of
a radar generated height map. Data were acquired from
an altitude of 25,000 ft. Note the clear signature of the
buildings near the centre of the image. The area in the
lower right hand corner is water.
EMISAR repeat pass SAR interferometry. In case of
repeat pass interferometry the two images are acquired
from two successive tracks. The mutual displacement of
16
the two tracks defines the baseline, and hence the base-
line size and orientation are not fixed. They have to be
measured/estimated through the aircraft position (as
distinct from attitude). That is why the repeat pass base-
line is less well known.
The fact that the two images are not acquired simulta-
neously introduces a 'temporal' baseline. For DEM gene-
ration this temporal baseline is undesirable as it make
non-stationary targets decorrelate. Still, repeat pass inter-
ferometry is justified, one reason being that it allows a
larger baseline to be formed, thereby giving a higher
sensitivity to terrain elevation.
To obtain a sufficient navigational accuracy for repeat
pass applications the radar has been enabled to steer the
aircraft via the instrument landing system (ILS). A de-
sired reference track is input into the radar control com-
puter, as well as programmable across-track and height
off-sets (typically in the order of 20 m). The control
computer receives continuous inputs from the real-time P-
code GPS, and calculates position off-sets and across-
track velocities, which are used to emulate ILS receiver
signals for the flight director computer. When the auto-
pilot is set up appropriately the emulated ILS signals
control the horizontal and vertical manoeuvres of the
aircraft. The track control system has proved capable of
synthesising baselines deviating no more than 5 to 10 m
from the desired baselines. This is surprisingly good
considering that the GPS position is specified at 15 m.
The DEMs generated with repeat pass interferometry
typically have a height noise in the order of 10-20 cm in
areas with little vegetation. In forested areas the technique
fails due to decorrelation. The systematic height errors are
believed to be in the order of 1-3 m, typically with a
correlation lenght of one to several kilometres. A contour
map of a parabolic sand dune in Northern Jutland, Den-
mark, is shown in Fig. 2. Major contour lines are sepa-
rated by 10 m, minor lines are separated by 2 m.
A ok ia)
X
Q s
Fig. 2. Contour map of Raabjerg Mile generated from
EMISAR repeat pass interferometry data.
Jgrgen Dall and Sgren Ngrvang Madsen
TECHNICAL UNIVERSITY OF DENMARK
Department of Planning - Surveying
The Institute of Surveying and Photogrammetry at the
Technical University of Denmark in Lyngby is merged
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B6. Vienna 1996
wit
De
Ph
yez
