International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
P. Reinartz -] 28 8,6 5,9 =
K. Jacobsen | -12 32 4,6 8,4 7
S 5 4 RR. Kaczynski] -9 25 1204,71 7.5 =
D. Poli 1 -4 14 4,1 3,3 >
D. Poli 2 -5 14 2,8 3,6 T
Table 41: Accuracy and matching quality on S. 5 areas
Min and max values are relative values for R. Kaczynski
results.
The statistical results are nearly the same that we had on
north area with a low resolution and accuracy reference.
The P. Reinartz DEM has always an altimetric bias (7 / 10m)
but quite low standard deviations.
K. Jacobsen has also a variable altimetric bias (3 / 8m) and
higher standard deviation.
R. Kaczynski has always an impressive altimetric bias around
203 m but better standard deviation tha n K. Jacobsen.
D. Poli still has results very close to the reference, means are
a little bit higher and standard deviation lower for the Polil
modelisation and vice versa. We calculated the difference
between the two D. Poli DEM As we can see on the
following picture, the difference is not that big apart in the
relief in south where a small delocalization between DEM
gives a large altimetric difference.
min max mean |Stand. dev.
Polil — Poli2 -10 12 1.7 2,4
Table 42: Statistical differences between Polil and Poli2
/ mum
/
/
F
/
/
/
/
Tl i = jx "S
AA ame. aix Da Za
Ad ARTS % A ES
= ae Et
EA LIS a 3 T" zy
WEM. GO Hew
A ms
ie otro d JL
Fig 43: Visualized differences between Polil and. Poli2
We can't make any difference with a global reference but if
we compare the D. Polil DEM and D. Poli2 DEM with P.
Reinartz DEM we can see that it's the Polil DEM which
seems linearly closer to the P. Reinartz DEM. We can
mention an east-west tilt between those two DEM.
The differences between P. Reinartz DEM and K. Jacobsen
DEM or between K. Jacobsen DEM and D. Polil DEM show
an important parabola.
Fig. 45: P.Reinartz — K.Jacobsen / K. Jacobsen — Polil DEM
If we draw an east-west profile , considering that the D. Poli
DEM as a reference (because we’ve got no global reference
and D. Poli DEM are closer to the extract of reference), the
result is more evident (Fig 46):
Red =D. Poli
+8
— 1] Blue = P. Reinartz
AS Pink =K. Jacobsen
Fig 46: Profile differences versus D. Poli DEM
The differences of those profiles could be explained by some
modelisation defects in particular, P. Reinartz didn't use any
ground control points and K. Jacobsen didn't have any
ground control points on the east part of the stereopair where
the default is maximal(Bouillon, 2004),
Unfortunately we've got no global R. Kaczynski DEM but if
we took off the 203 meters bias, the mean of differences
seems quite regular.
The qualitative analysis of DEM on those areas are the same
that the ones we saw on North area. :
- P. Reinartz DEM is quite smooth but with no apparent
big defects, the 50M resolution of this DEM is not thin
enough to allow a definitive appreciation in hilly area.
- . R. Kaezynski DEM are smooth as well but artifacts has
not been removed.
- K. Jacobsen DEM shows the highest standard deviation
(ie the highest noise).
- The two D. Poli DEM are close to the reference (under 5
meters).
Standard deviation by area and investigator:
9,0 EE EEE
6,0 :
1 : x 0 oo : [—e—p. Reinartz -
5,0 f
T 7 -#—K, Jacobsen
ance s No Gil R. Kaczynski
4,0 : | D. Poli1
— — D. Poli2
3,0 eyes
2,0 1
|
1,0 1
|
0,0 |
areal area2 area3 area4
Fig 47 Standard deviations
4. CONCLUSION
Unfortunately, we are not able to present in this paper results
of all studies done in the 9 selected sites but most of the them
will be presented during the ISPRS Congress.
Intt
Ba
IG]
pro
bee
pro
of
tha
dor
lan
Sor
dist
pro
rest
The
HR
acc
are:
obt:
Uni
sam
reli
whi
Fin:
asse
of t
thar
Aire
ofa
Bau
asse
boai
Istai
Ber
Bull
Ber
trei
200:
Bou
IGA
Bou
and
Che
obta
Con
Cou
Ster«
ISPI
