50
zd
tanbul 2004
ing is lower
cy (roughly
ly the Dem
O bias with
1 meters.
ig 34) :
(North)
1 (South)
2 (lower) :
St.
No
ev. match.
8,7 | 0%
29 17,1%
09 11,2%
3,4 | 0%
5,7. | 096
4,4 | 0%
5,1 19,9%
6,6 | 595
10 | 096
,9 | 0%
25 areas
Kaczynski
declared no
‘e complete
hing areas,
the filling
tion, it can
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
be pertinent or not depending on the use of the DEM. All our
statistics are therefore computed from valid dots so if we
don't have any pieces of information about those areas we
can't get rid of them in our statistics.
Those results are worse than the first one, on one hand
because of the kind of landscape (this area has more
significant relief) and on the other hand because we've got no
declared invalid areas for some DEM. D. Poli doesn't declare
any invalid areas but has good results in spite of this fact. P.
Reinartz whose DEM has also a 50m sampling, which
probably is not thin enough for this kind of relief shape.
Even if there are some recognizing holes in K. Jacobsen and
in R. Kaczynski DEM, we can see on profiles that the bad
correlation has a bigger impact on DEM than only the size of
the holes.
Example: altimetric restitution of relief
- Fig 36: Profile on P. Reinartz DEM
This DEM is not too bad but obviously too smooth (filtering
and 50m spacing) for the relief shape.
— pee 1 |
Fig 37: Profile on K. Jacobsen DEM
K. Jacobsen DEM is quite good when the matching
processing is efficient but when it's not the case, the results
are disappointing in areas biggest than the rigorous no
matching area.
E n — vise
Fig 38 : Profile in R. Kaczynski DEM
The 203 m bias have been taken of the profile. This DEM has
the same defaults than K. Jacobsen near bad correlation areas
and the relief is a little bit less well drawn.
Fig 39: Profile in D. Poli DEM
This DEM is close to the reference, though a little bit
smoother.
In conclusion, in this area with a significant relief, the results
are worse than thus were in flat area, as expected, but
nevertheless D. Poli's obtained good results. Even with two
kinds of modelisations, she's got results better than 7m
without any invalid areas.
3.5.2 Four reference DEM S 5 1/2/3/4 are available with a
5 m sampling (Fig 40)
(400 « height « 600)
« heigh
Sas
S 5 3(460« height « 530) S 5 2(44
Fig 40: Reference DEMs with 5m sampling
All those areas are roughly flat, more particularly the third
one. The references have a 5m sampling and a really good
accuracy (better than 0,5m) but a small size only 5Km x 5
Km.
St. No
min max |mean| Dev. [match.
P. Reinartz 0 31 10,0 | 5,9 =
K. Jacobsen] -9 34 80 | 79 | 376
S 5 1 [R. Kaczynski] -10 26 |2044| 7,0 |1.1%
D.Polil | -6 14 | 40 | 35 | 14%
D.Poli2 | -7 14 | 26 | 37 | 9%
P. Reinartz -1 40 10.4 |: 7,5 =
K. Jacobsen] -15 31 sa | 82 | 3%
S_5 2 |R. Kaczynski} -12 30 120382] 78 7
D. Poli 1 -6 19 3,4 4,3 >
D. Poli 2 -7 18 1,5 4,5 T
P. Reinartz 0 24 7 0 4,3 =
K. Jacobsen | -15 27 | 3,1 | 6,6 19,2%
S 5 3|R. Kaczynski]. -12 28 1203,4| 57 T
D. Poli 1 -4 11 29 128 | -
D. Poli 2 -6 i0 10571 2971| *