mapping ground co-ordinates into the image by
solving the equations relating the Doppler shift to the
sensor and image co-ordinates taking into account the
movement of the sensor and of the earth. The ground
will be represented by a digital elevation model
(DEM). This mapping is done exactly for a number
of points arranged on a three dimensional framework,
know as a supergrid, surrounding the area to be
mapped, all other points are mapped by interpolation.
The number of points rigorously mapped will depend
on the nature of the terrain and the order of the
polynomial chosen. The radiometric value to map
back to the output image will be determined by
resampling the image and a number of resampling
routines will be available.
The method was developed and tested with ERS-1
data and the results are described in Dowman et al,
[1993a] This paper will concentrate on results from
JERS-1 SAR data.
3. GEOCODING RESULTS AND PROBLEMS
FROM JERS-1 SAR
3.1 Test data
The region concentrated upon was the Marignane
region of France. We have a DEM of this region, and
have previously geocoded several ERS-1 images to it
[Dowman et al 1993]. The DEM was generated by
IGN France. The map projection is France Lambert
zone 3, the sample spacing is 80m, and the sample
quantization is 0.1m. The DEM covers an area
approximately 16km x 16km, and elevations span the
range [0m, 279m]. The geoid-ellipsoid separation for
this region is in the range [7m, 7.4m], but as the DEM
samples were assumed to be ellipsoid elevations, the
separation was not applied. The projection from map
to image was computed in the same map projection
and on the same grid as the DEM, so no preprocessing
was required.
The image was a JERS-1 SAR level 2.1 product
[NASDA, 1992, Shimada, 1993], with a nominal
pixel spacing of 12.5m. This is the standard SAR
product. The map projection of this product is stated
to be "geocoded" (as opposed to "ground range" or
"slant range"), and polynomial transformations
between GRS80 UTM zone 31 northern hemisphere
and image coordinates are given. The polynomials are
approximately rotations. Other documentation
describes the coordinate system of the level 2.1
product as being azimuth and range. We therefore
treat the image as being a ground range product. It
should be noted that the product is alread geocoded
only in the sense of ellipsoid correction; terrain
correction is still necessary.
The sensor state vectors are given in ECR (Earth
Centre Rotated), every 60 seconds. We have
approximated ECR with WGS84. In the projection
from global cartesian to slant range, the iteration
termination threshold was set to 0.0001 seconds.
438
3.2 Tie pointing
The headers do not contain the information necessary
to relate the azimuth coordinate of the image to
azimuth time (zero-doppler or not). As a
substitute, until a more satisfactory solution could be
found, tiepointing was performed, as follows. Two
points well separated in azimuth were chosen, and
their azimuth coordinates measured in the image.
Their coordinates were also measured in the map,
then transformed to slant range, giving their zero-
doppler azimuth times. A linear relation between the
azimuth coordinate and the zero-doppler azimuth time
was easily derived. This shows the time across one
pixel to be approximately 0.0024 seconds, which
justifies the choice of iteration termination threshold
in the projection from global cartesian to slant range.
Samples of the range are given for the first, centre and
last pixels in azimuth for each block of 1024 lines in
range. They are only approximately zero-doppler, but
were used as if they were exactly zero-doppler. The
linear relation between the azimuth coordinate and the
zero-doppler azimuth time, derived above, was used
to generate azimuth time values on the same sample
points. The resulting slant range samples were used to
form the projection from slant range to ground range.
The final tiepoint correction to the overall projection
requires that tiepoint coordinates be measured in
ground range and in the image. We chose to do
this step by reference to the DEM rather than
published maps, as follows. First, the overall
projection was derived without tiepoint correction.
This projection was then used to generate an energy
conservation map in image space, to be used as a
ground range simulated image. The required tiepoint
coordinates were then measured manually using an
interactive image display ‘tool in both the
ground range simulated image and the input image,
and used to form the projection from ground range to
the image. The errors, above, in the approximations
used to derive samples of the projection from ground
range to slant range arecorrected for in this tiepointing
step, along with any other errors.
It would have been just as possible to get the ground
range tiepoint coordinates by projecting
measurements from published maps. However,
the approach taken avoided possible errors between
the map and the DEM, and also allowed a greater
number of tiepoints to be generated. It is usually very
difficult to find a sufficient number of features that
both appear and are well defined in both published
maps and the image. A drawback of the approach
taken is that the features used result from terrain
effects, and so are less well defined in the azimuth
direction than in the range direction.
After the overall projection was computed, the image
was resampled to the map, completing the geocoding
process. The ancillary products were also generated:
In image space, layover and energy conservation; In
map space, shadow, layover and energy conservation.
3.3
In the
approxi
be gene!
and lay
The sc
nominal
the DEN
which w
An imp
compar
saturate
compar
against
Howeve
can be
conserv:
geometr
3.4
There w
from N/
in the h
late, son
our geoc
the stan
discuss t
The coo
"ECR".
context.
The act
systems
We have
were zer
this is ne
appropri
necessar
does not
correctio
We need
the imag
use can |
involvin
(describ
NASDA
informat
the infori
It may e
Amongst
standard
given in
product ,
ellipsoid
31 north:
slant rar
between
azimuth
