International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
RMSE of 19 m and a non-zero mean error, indicative of a
systematic bias in the DEM. They did not use ground control
points to control the error, but constructed another DEM from
conventional maps with 5 m contour intervals.
The more recent literature reports similar findings, so that there
seems to have been no influence of the steady improvement of
the sterco-matching algorithms or the development of specific
SPOT-image modules designed to take into account the
information of orbital parameters, image capture angles, etc.
Hae-Yeoun et al. (2000) obtained RMSE values of 25.5 m and
33.6 m according to the study area: their controls were made by
comparison with other DEMs of 100 m (DTED, USA) and 60 m
(Korean National Geographic Institute) resolution.
The best results are those reported by Al-Rousan & Petrie
(1998). They generated DEMs in a single desert zone of north-
eastern Jordan with different commercial program suites, and
obtained variable error values depending on the accuracy of the
images and on the programs used. The RMSE ranged in value
from 3.3 to 6.7 m. The number of ground control points was
also variable, ranging from 10 to 47. It was not possible in their
work to deduce the type of relief of the terrain, possibly because
of the size of the error.
Table I shows a set of significant examples about accuracy in
SPOT-DEM. We can see that RMSE values are not comparable,
varying from 3.2 m (Toutin, 2002) to 33 m (Hae-Yeoun, 2000).
The number of declared check points is also very different, from
6 to 40, but many authors do not provide information about this
issue. Also, other aspects that may be crucial, such as the terrain
topography, remain unknown.
There are some common problems in using digital stereoscopic
pairs, e.g, the difficulty of identifying the ground control
points, and the frequent radiometric differences between the two
images. These differences arise because the left and right
components are usually taken on different days, and the light
conditions may change (Baltsavias & Stallmann, 1993).
Nonetheless, there are many works which lend support to the
quality of the results of using this type of data, due to their wide
coverage and good temporal resolution.
Date First RMSE* ;
Author (m) Method
1988 Priebbenow 5.4
: ; 40 CPs extracted from
$990: Mukal 26 cartography 1/25000
Comparison with DEM
1992 Sasowki 19 generated from cartography
1/5000
3.3 and
1998 Al-Rousan 6.7
A Comparison with DEM of
2000 Hae-Yeoun 25:5 amd 100 m and 60 m of
33.6 :
resolution
2002 Toutin 32 6 CPs from DGPS
^ Root Mean Square Error.
b Method which has been calculated the root mean square error
(RMSE) in elevation of DEM. When method is hole paper no
specify it.
Table 1. Some works about SPOT-DEM accuracy
determination
Once these problems that are inherent to the photogrammetric
process have been overcome, the resulting DEMs have to
integrated into a GIS. In a literature search, we could find no
references to possible optimization strategies for this phase of
the process, in which one would like to guarantee the
elimination of redundant and, in so far as possible, erroneous
data. In earlier work, we have studied this latter part in some
detail (Felicísimo, 1994; López, 2000), so that the present study
will focus on the selection and control of the points which are to
be kept in the data structure that will be integrated into the GIS.
3. OBJECTIVES
This paper aims to:
. To analyse the DEM accuracy genercted by automatic
stereomatching techniques from SPOT-HRV images by
different photogrammetric softwares (OrthoBase PRO,
Socet Set) and to compare results each other and whith a
DEM generated from cartographic data. It is emphasized
that, to guarantee reliable error control, it is necessary to
have sufficient well distributed, and highly accurate, ground
control points available.
. Propose a method of improvement for the structure of
DEM without a loss in accuracy. This process of
simplification enables the data structure to be better adapted
for integration in a GIS.
The working hypothesis is that the correlation coefficient
associated with each elevation datum may be used to determine
whether that datum should be kept or discarded, thereby
simplifying the TIN structure without significant degradation of
its quality. The actual loss of quality of the DEM will be
checked against a large number of ground control points
measured with high-precision GPS techniques.
4. MATERIAL AND METHODS
4.1 Area under study
Work area is a 23 km x 28 km rectangle in the province of
Granada (Southern Spain) (Fig. 1). It is an area with a
complex topography: Steep slopes in the South and flat
surfaces in the North. Elevations are in the range 300-2800 m
with an average of 1060 m.
4.2 Data
We have used two panchromatic SPOT-HRV images with 10
m pixel size. The images were taken on the days 2-11-1991
and 2-01-1992, and cover a total area of 60 km x 60 km.
Error estimation was performed using 315 check points to
calculate all SPOT-DEM errors and 7071 check points t0
stydy the method of improvement. These check points have
taken with differential GPS techniques.
256
Interne
43 S
SPO
aplic
(Erd.
using
and |
The
char:
" Si
S
data
rast
data
moc
DE!
Tria
44 D
In the
basical
to the
attaine
they ai
relativ
one to
the im
orienta
parame
errors.
of a ste
Triang
be perf
The ele
begins
