ACCURACY ANALYSIS OF DEMS DERIVED FROM ASTER IMAGERY
J. A. Gonçalves, A. M. Oliveira
Science Faculty — University of Porto, Portugal
jagoncal@fc.up.pt, anamariap@@mail.telepac.pt
Commission III, WG III/2
KEY WORDS: Mapping, DEM/DTM, Orientation, Matching
ABSTRACT:
ASTER acquires along track stereoscopic imagery, with a spatial resolution of 15 meters. Automatic generation of Digital Elevation
Models (DEMs) from these images is a well established process, implemented in many commercial software packages. It can
provide relief information for areas with poor coverage of topographic mapping.
This paper presents a study of the accuracy achieved in DEMs extracted from ASTER, for an area in Portugal, using the PCI
OrthoEngine software. Images were orientated with ground control points (GCP) obtained from topographic maps. Experiments
were carried out in reducing the number of GCPs. A number of 5 or 6 GCPs was always required to orient the images, in order to
keep the accuracy achieved with larger numbers of points. It was possible to conclude that more use could have been done of the
approximate orientation provided in the image header. A grid of points derived from the sensor position and attitude, estimated by
onboard equipment, is given in the image header. The precision of these data was assessed as being better than 100 meters. A
significant reduction in GCP requirements would be possible if more use of this information would be done.
The vertical accuracy of the extracted DEMs, assessed with a 8 m DEM grid, produced by a local mapping agency, was of 9 meters
in areas of mountains with low vegetation. The percentage of coverage of the DEM is very large, however the automatic filling of
gaps, in areas where the stereo-matching failed, degraded the DEM detail and accuracy. A manual edition of gaps, according to user
perception of relief forms is more appropriate when deriving DEMs for production and update of topographic map products.
1. INTRODUCTION
1.1 DEMs from optical satellite data
Digital elevation models are important data sources in
topographic mapping production, such as contour generation
and image ortho-rectification. Terrain analysis in a GIS also
requires those types of data, which in general are available from
mapping agencies in developed countries. However, large parts
of the Earth are not properly mapped in medium scales, such as
1:50,000.
Satellites can provide DEMs for the map production and other
applications through different means. Among the optical
sensors, across-track stereoscopy has been widely used, for
example in the SPOT programme, with SPOT 1 to 4. The two
images of a stereopair are acquired pointing the sensor to the
same area, with different incidence angles, in different orbits.
Time separation between images can range from a few days to
months. Cloud coverage makes the acquisition of stereopairs
difficult. Different sun illumination conditions, as well as
different vegetation growth, produce significant image
differences, complicating the automatic stereo matching.
Along-tack stereoscopic image acquisition requires two sensors
with different inclinations, observing at the same time. SPOT-5,
the most recent satellite of the SPOT programme carries, among
other sensors, a system of this type, the High Resolution Stereo
(HRS) system (Spotimage, 2002). That is also the case of
ASTER which is equipped with two telescopes, both sensible in
the range of 0.78-0.86 um, one pointing in the nadir direction
and the other pointing backwards, with an offset angle of 26
degrees. This leads to a base-height ratio of 0.6 (ERSDAC,
2001), which is appropriate for automated extraction techniques
in various terrain conditions (Hirano et al.,, 2003). Time
separation is of only 60 seconds. Illumination conditions are
very similar allowing for an efficient automatic matching. The
two bands are identified as 3N (nadiral) and 3B (backwards).
Band 3N is associated with two others. The three bands (1, 2
and 3N) are acquired by Aster VNIR imaging system
(ERSDAC, 2001) and have an image size of 4100 by 4200
pixels. Band 3B has 5000 by 5400 pixels.
1.2 Orientation of linear array images
The extraction of DEMs in a well defined cartographic
reference system requires that exterior orientation is well
known. Ground control points are required in order to
determine a set of parameters of a sensor model that express the
mathematical relation between ground and image coordinates.
Sensor models for optical sensors, such as SPOT or ASTER,
comprise position and attitude parameters. Usually some orbital
parameters are considered, in order to describe the sensor
position, and 3 absolute attitude parameters (roll, pitch, yaw) at
the initial image instant. Together with the orbital perturbation
theory and attitude variations measured on board, position and
attitude can be calculated for any image line. The sensor model
is based on the colinearity equations. Different variations are
described by many authors. See for example Kratky (1987),
Gugan and Dowman (1988), Westin, (1990) or Toutin (1994).
Commercial software packages for mapping from satellite
images implement this type of physical sensor models.
Satellites carry navigation equipment that can estimate the
exterior orientation parameters. The image geo-location of
International /
optical satellit
example, in
Orbitography
positioning sy
meter or bette
uncertainty is
significant rec
only attitude p
Alternative o
Grodecki (20(
(also a linear
space.
Commercial
implement, in
of PCI Orthol
However, the
considered in
1.3 DEM gei
PCI Orthoen;
orientation a
straightforwar
It consists oft
1. Define the
such as UTM,
WGS84, can
by means of tl
2. Chose the s
different sens
preferably in |
the use of th
ancillary data.
3. Enter the €
report is outpt
4. Generate er
so that paralla
ASTER this s
that, due to th
direction.
5. Automatic
epipolar imag
options respe
filled automat
related to th:
acceptable cor
6. DEM geoc
space. This ste
There is an
resulting from
DEM generat
Results obtain
10 meters or
This study i
commercial sc
the vertical ac
the number of
2. EX
21 Data use
Some images
Porto, were p
these images +
