International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
off-nadir). The best available geographical sources (raster maps,
world or regional vector databases like VMAPO or VMAPI,
orthorectified space imagery) are used to get the most accurate
position of the water body to derive the most accurate height
from the altimeter data.
When many different water bodies are situated nearby the
satellite nadir and are potential candidates to match with
specular signal, an optimisation process is applied taking in
account all available altimetric data from different orbits and
keeping the most coherent solution among all the candidates
(the different orbits should give the same altitude for a single
water body)
3.3 Expected accuracy
When the satellite crosses vertically through the water body
(lake or river), the altitude of this water body can be derived
from altimeter data with about 2 meters accuracy. This result
was established after comparison with elevation reference
points extracted from best topographical sources along some
French rivers (Maheu, 2000).
When the water body is off-nadir, its altitude (derived from
altimetric data) will depend of the accuracy of its horizontal
position . The impact of this horizontal error determination on
the altitude determination can be approximated by the following
formula derived from the general formula given in 3.2. :
H.error = (L/Z)* L.error
when
{{<<Z,L<<l), L.error >> D'error, Lcrror >> Z.error)
Numerical example : for L = 15 km , Z = 700 km and
L.error = 250 m we get H.error = 5 m.
4. SELECTION OF ALTIMETRIC DATA IN FLAT
AREAS
The exploitation of radar altimetric data on water bodies was
very encouraging and made us extend its exploitation to land
areas.
The “threshold retracking” method developed by GRGS for
altimetric observation of continental ice sheets (Rémy, 1990 ,
Legresy 1995, 1997, 1998) and adopted by SPOT IMAGE to
support image rectification and DEM control, rejects altimetric
measurements in very rough terrain; then only flat to
moderately rough terrain measurements are kept after retracking
(about 80% of the on board memorised data).
We have to keep in mind that satellite radar altimeter was first
designed for oceanographic purposes dealing mainly with
specular data, that is why a severe selection process has to be
applied to keep adequate data matching with required accuracy
of the mapping project. To minimise unwanted or uncontrolled
errors due to “slope effects” or “smoothing effect” (which will
be studied in the next section of this paper), priority has been
given to very flat areas to collect very reliable height
measurements. The selection process designed to extract the
elevation of very flat terrain areas is described in following
subsection.
4.1 Criterions for selection of flat areas
Signal continuity is the first filter applied to available data after
retracking. It consists in keeping only the "ideal" sequences of
20 measurements per second, that means without any
discontinuity. A statistical analysis showed that about 50% of
data are rejected by this first filter due to terrain roughness, on
board tracking discontinuity or rejection by retracking .
Height variation within one sequence is the next filter applied to
remaining data. To minimise uncontrolled reflections due to
changing heights and slopes inside the impact zone of the radar
pulse, only data associated with very flat surface are selected. A
0,75 m threshold for standard deviation within a one-second
sequence (corresponding to a 8,3 km travelling of the satellite
on his orbit), equivalent to a maximum slope of 2 meters for 10
km was finally adopted .
Statistically about 8095 of the remaining data is rejected by this
test.
Inter-cycle height variation is the last filter applied; it consists
in computing for each height measurement the maximum height
difference with all other height measurements derived from
other passes of the satellite within a 2 km radius. This checks
the coherency and stability of ERS measurements along time.
The maximum height difference observed should be 5 m for a
1
minimum of 3 cycles available 2 km around the data point to
test.
4.2 Selectivity and accuracy
After passing through the different selection steps (retracking,
signal continuity, height coherency between several cycles)
about 5% of the total input data are kept for exploitation in
DEM control or ground control in photogrammetric block-
adjustment.
Comparison of this type of selected data with reference height
points extracted from reliable topographical maps showed
agreement better than 5 m in most cases (more than 95%)
5. ACCURACY EVALUATION OF ALTIMETER DATA
RELATIVE TO TERRAIN CARACTERISTICS
We have also tried to refine the modelisation of radar pulse
reflection on moderately rough and heterogeneous terrain . The
aim was to extend the domain of validity of ERS altimeter data
providing it keeps satisfying the required accuracy for mapping
projects .
For that purpose, we have used a special algorithm based on
radar simulation, developed by GRGS (Pace, 2003).
5.1 Principles of GRGS waveform simulation algorithm
For each radar pulse, the simulation algorithm builds a
simulated waveform taking into account the satellite position
and a refined physical model of propagation and reflection of
the radar pulse on the ground surface. The ground surface itself
is simulated by the best available DEM or the DEM to control.
The variation of reflectivity inside the total zone hit by the radar
pulse is modelised with the help of an existing vector database
like VMAP (only the water bodies have been considered in the
current version and were given a much bigger reflectivity
compared to land surfaces). The simulated height is then
computed from ramp mid-point of the simulated waveform.
Then the simulated height is compared to the height derived
from on-board data which is much more convenient than
comparing directly ERS observed height with DEM height, as
both the ERS observed height and simulated height carry the
same systematic errors like "slope effect", *smoothing effect"
and “lock on off-nadir water body”.
International Arcl
A local discrepa
heights is the sign
A systematic sh
systematic shift er
6. ERS ERROI
We investigated
terrain, using a 30
reference. Two
characteristics we
DEM. Errors bet
the simulated r
calculated. Finally
were analysed.
6.1 Study areas.
The first geograph
includes landform
relief foothills, ar
east. The other ai
contained larger :
response. The fielc
6.2 Elevation da
ERS elevation dai
retracking step ir
elevation points ov
30-m DEM : the I
photogrammetric 1
Spot, or by contou
A root mean squa
quality . The R!
coordinates, and 5
We extracted elev:
bilinear interpolati
ERS simulated ele
we use the DEM t
method described i
heigth (m)
Figure 6. Heig
6.3 Exploratory |
6.3.1 Errors sta
Fig 7 shows the dif
seem similar for tl
show a roughly nc