digital surface model generation from corona satellite images
TRANSCRIPT
Digital surface model generation from CORONA satellite images
Angela Altmaier *, Christoph Kany
Department of Geography, University of Bonn, Meckenheimer Allee 166, 53115 Bonn, Germany
Received September 2001; accepted March 2002
Abstract
Digital surface models (DSMs) are used for various analyses in environmental science, e.g. for erosion and water studies.
Aerial photos and maps, which are necessary for the extraction of DSMs, often do not exist due to financial or political reasons.
This situation can be also encountered in Morocco and, in particular, a test area of the international research project IMPETUS
was used in this study. Therefore, stereo satellite images of CORONA have been used, as they allow DSM generation, have a
ground resolution of 1.83 m, reasonable price (US$12–18 per filmstrip of 188� 14 km) and large coverage (especially of Asia
and eastern Europe). The software program ERDAS IMAGINE OrthoBASE Pro was used to generate DSMs automatically
from CORONA satellite images with best vertical accuracy of about 10 m and planimetric accuracy of about 3 m. These DSMs
could afterwards be used to generate orthoimages, e.g. for mapping change detection and generating thematic maps or land use
classifications. D 2002 Elsevier Science B.V. All rights reserved.
Keywords: CORONA; Digital surface models; DTM; Orthoimage; ERDAS IMAGINE OrthoBASE Pro; DGPS; Morocco
1. Introduction
For many tasks of environmental analysis and
environmental science, such as analysis of erosion
and runoff dynamics or vegetation and infrastructure
changes, digital surface models (DSMs) are an impor-
tant basis. The 3D view of a DSM can be used in
different software programs, e.g. generate anaglyph
images in ERDAS Stereo Analyst, to derive plani-
metric and 3D information for further use in change
detection analysis. An example might be the digitising
of slopes in the anaglyph image and comparing them
to older ones. On the other hand, orthoimages can be
generated from DSMs and once again be the basis for
2D change detection analysis or mapping. Facing this
challenge, temporal changes may be one of the most
important factors: information of DSMs and orthoim-
ages of different years can be compared, e.g. from the
actual IKONOS satellite images to the 30 years older
CORONA satellite images.
1.1. Research project and the need of DSMs
The mentioned possibilities of DSM use and their
integration into the actual research project IMPETUS
have been technical and conceptual motivations for
the following study. IMPETUS is the abbreviation for
‘‘Integratives Management-Projekt fur einen Effi-
zienten und Tragfahigen Umgang mit Sußwasser in
West-Afrika (an integrated approach to the efficient
management of scarce water resources in West
Africa)’’ (IMPETUS, 2001b). It is an interdisciplinary
and application-oriented project, with the final objec-
0924-2716/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved.
PII: S0924 -2716 (02 )00046 -1
* Corresponding author. Tel.: +49-0228-2422008 and +49-089-
848151.
E-mail addresses: [email protected],
[email protected] (A. Altmaier).
www.elsevier.com/locate/isprsjprs
ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235
tive of establishing a sustainable water management in
western Africa. Researches on conditions, influences
and realisation of this management are ongoing in two
different catchments in western Africa. One of the two
catchments is drained by the river Draa and situated in
southern Morocco. It extends from the southern slope
of the Atlas mountain chain to increasingly desertifi-
cated areas nearby the Moroccan–Algerian frontier
(Fig. 1). The objectives for initialising the IMPETUS
research project were the hypothesised interchanges
between the climates of Africa and Europe via tele-
connection processes as well as relations between the
long-lasting drought periods in the northern and
southern parts of the Sahara Desert. Therefore, the
hydrological cycle and the quantification of fresh
water resources are at the center of all research
studies. Especially, there is a great interest in differ-
entiating between climatic and socioeconomic influ-
ences on the hydrological cycle and the origins of
such influences.
For the research of almost all fields of this
IMPETUS project, DSMs will serve as a basis for
analysis, as mentioned in the Introduction. Hydro-
logical as well as geological, meteorological and
urban development researchers need DSMs as basis
for studies on dynamics of, e.g. erosion, runoff and
infrastructure. Depending on the special application
in the different research fields, DSMs or either
digital terrain models (DTMs) are more suitable.
Whereas DSMs include 3D objects, like vegetation
and buildings, DTMs describe only the earth surface.
The technical method of this study creates DSMs by
matching. However, due to the rare vegetation
coverage existing in Morocco (see below), the
DSM can be used as DTM in almost all parts of
the catchment area (except, e.g. the oases).
Fig. 1. River Draa catchment of the IMPETUS project in Morocco (in black) (IMPETUS, 2001a, p. 8).
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235222
1.2. New method of DSM generation
Maps and aerial photos have always been used as
basis for the generation of DSMs. Unfortunately, only
few such materials could be found in Morocco. Maps
exist only for a part of the research area in 1:100000
scale and dated from the 1960s. The main reasons for
this scarcity of material might be financial problems
and security restrictions. Nevertheless, more and more
activities concerning the geographical information
system (GIS) and remote sensing sector are currently
being developed in different Moroccan institutions
and organisations.
Against this background of low data availability, a
new method of DSM generation based on the high-
resolution CORONA satellite images offers a lot of
advantages. Their best ground resolution—as used in
this study—is 1.83 m. For other CORONA missions,
the ground resolution may vary between 1.83 and
140.3 m (see below). Since 1995, CORONA satellite
images are available at the US Geological Survey
(USGS, 2002) for the reasonable price of US$12–18
per filmstrip (188� 14 km), i.e. less than 1 cent per
square kilometre, though without digital preparation.
Comparing this price to the IKONOS satellite images
(about US$29 per square kilometre) or SPIN-2
(US$35 per square kilometre), it is evident that
CORONA gives a favourable financial possibility
for DSM generation in countries like Morocco.
However, it should be taken into consideration that
only few software programs like ERDAS IMAGINE
OrthoBASE Pro are able to generate DSMs based on
CORONA satellite images (see Section 3).
In the following study, the software program
OrthoBASE Pro was tested for DSM generation for
a part of the Draa river catchment in southern
Morocco. The results have then been evaluated con-
cerning accuracy and effort.
Fig. 2. Test area in Morocco (outlined) (RV-Verlag, 2000).
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 223
The method of this case study should be applicable
for the whole catchment of the river Draa. Because of
the area of this catchment (estimated at ca. 30000
km2) and its extension up to the Atlas mountain chain,
the procedure of DSM generation must take into
account a large area with complicated geomorpholog-
ical features. Therefore, the driving goal was to
generate a DSM for an area as large as possible, with
the lowest possible financial and temporal effort (e.g.
only one DGPS field campaign, only a short but high
precision DGPS surveying, software processing time
of only some weeks).
1.3. Test area
For the first tests of DSM generation by Ortho-
BASE Pro, only a small part of the Draa catchment
was selected as a test area in this study. This test area
was situated at the southern slope of the Atlas moun-
tain chain, few kilometres north of the small village
Skoura. The test area’s size is about 100 km2 (Figs.
2–4). For DSM generation, the test area had to be
separated into two parts and a DSM of the northern
and southern part was generated. That was necessary
because the aft CORONA (see Section 2.1 for tech-
Fig. 3. View of the test area in the northern direction.
Fig. 4. View of the northwestern part of the test area.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235224
nical specifications of CORONA) image of the test
area was also separated into these two parts on two
different adjacent filmstrips. After generation of these
two DSMs, a mosaicking of the two DSMs and
orthoimages in ERDAS was performed.
From a hydrological perspective, the test area
belongs to a region with water resource levels below
the average. Therefore, the typical Mediterranean
vegetation in the test area is only existent in very
few areas or even absent. The oases of the river Draa
and its small adjacent rivers are sometimes the only
major vegetation patches. Along these rivers, often
only some metres wide, typical small Moroccan vil-
lages of clay construction can be found. This low
density of settlement and the scarcity of vegetation
became a great advantage for the DGPS surveying. In
addition, these two aspects and a lack of erosion led to
almost no negative influences on the accuracy of DSM
generation in most parts of the test area, i.e. the DSM
was very close to a DTM and multitemporal differ-
ences due to erosion were small (IMPETUS, 2001a).
2. CORONA satellite images and DGPS surveying
2.1. CORONA satellite images
As the data basis for DSM generation, CORONA
satellite images were used. The CORONA satellite was
the first generation of the US reconnaissance satellites
built by the US Air Force and the US Central Intelli-
gence Agency (CIA). The CORONA satellite took
stereo photos from 1960 to 1972 using filmstrips of
7� 90 cm produced by a panoramic camera with
panchromatic channel (Day et al., 1998; Ruffner,
1995; Mc Donald, 1995a,b; Tappan et al., 2000) (Fig.
5, Table 1). The best ground resolution is 1.83 m for the
KH-4B mission at a flight height of 150 km (USGS,
2002). Two oblique viewing cameras realised the stereo
view, a forward- and an aft-looking camera each with a
15j tilt, i.e. an angle of 30j between them (Fig. 6).
Because of the panoramic camera, the filmstrips
have typical panoramic distortion (Slama, 1980;
Goossens et al., 2001; Fig. 5), which has to be
corrected after scanning by a mathematical algorithm.
Only few image and camera data specifications are
known for CORONA, as shown in Table 1 below. The
parameters ‘fiducial marks’ and ‘principal point’,
which are typically used for aerial photo triangulation,
do not exist. Therefore, triangulation and geometric
correction become very difficult (see Section 3).
Due to the military intention to spy on the former
Soviet Union and other sensitive areas, the principal
coverage areas of CORONA images are Asia, eastern
Europe and northern Africa. In 1995, the available
CORONA images of 2 billion square kilometres were
released to the public by the US government and can
be ordered from the USGS (USGS, 2002).
Depending on the CORONA mission, image qual-
ity, e.g. film and ground resolution or cloud coverage,
differs. The best film resolution varies between 160 L/
mm for KH-4B and 30 L/mm for KH-5, and the best
ground resolution is between 1.83 and 140.3 m.
However, cloud coverage prevents 40% of the images
from the full use (USGS, 2002).
In this study, CORONA satellite images of the very
successful last KH-4B mission with the ground reso-
lution of 1.83 m were used. The adjacent filmstrips
have an overlap of 10% in north–southern flight
direction. Therefore, a complete cover of the over-
flown area and a technical processing of the adjacent
images are made possible. In this study, the aft image
consists of two parts from two adjacent aft filmstrips
(about 30% for the northern part and 70% for theFig. 5. Film distortion of a panoramic camera (Slama, 1980, p. 201).
Table 1
Parameters of CORONA satellite for KH-4B system (USGS, 2002)
System, mission CORONA KH-4B, mission 1117
Date May 1972
Camera type panoramic, panchromatic
Frame format 5.54� 75.69 cm
Size of the imaged area 14� 188 km
Focal length 60.69 cm
Best film resolution 160 L/mm
Best ground resolution 1.83 m
Flight height 150 km
Image scale 1:247500
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 225
southern part of the test area). The forward image can
be found on one filmstrip.
Parts of the forward and aft CORONA filmstrips
showing the test area were scanned by a desktop
publishing (DTP) scanner, the DIN A3 flat bed colour
scanner EPSON 1640 XL with double CCD linear
sensors. The optical geometric resolution was 1600
dpi (ca. 16 Am), the radiometric resolution is 14 bits
per pixel and colour channel (internal and external)
and the optical density is 3.6. Leachtenauer et al.
(1998) mention that to get the CORONA film quality,
a scanning pixel size of 4 Am should be used, but this
was not possible with the used scanner and would lead
to very large data sets. The scanning pixel size
corresponds to a pixel footprint of about 4 m. Because
of the mentioned separation of the test area on two aft
filmstrips (north and south), two scans were necessary
for the aft images.
2.2. DGPS surveying
The second database for DSM generation was a
GPS surveying using a Differential GPS (Leica 300
System) in March 2001 in the test area in Morocco.
Fig. 6. Camera model of CORONA satellite (NRO, 2002; Campbell, 1996).
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235226
These DGPS-measured data were used afterwards
to evaluate the height accuracy of the software-gen-
erated DSM. Therefore, foot and upper slope positions
were measured, mainly on foot in a point distance of
10–20 m. On the other hand, points were measured
by DGPS to be used as ground control points (GCPs)
in the software OrthoBASE Pro during exterior ori-
entation. These GCPs were mostly measured by
vehicle in a point distance of about 50 m (Fig. 7).
Due to the DGPS, the accuracy of all points (in x, y
and z direction) is less than 10 cm. In this context, the
rare vegetation distribution was an advantage for
DGPS measuring. Even by vehicle, the basis for
surveying was always the ground, guaranteeing the
small deviation of DGPS measurements from the
ground. This precise point accuracy was needed to
improve the interior orientation, as shown below. The
time needed for the field campaign was 14 days.
For the measurement of the points, well-defined
positions corresponding to the image and the terrain
had to be found. However, the point collection was not
always easy due to the changing infrastructure (roads
and buildings) since the acquisition of the 30-year-old
CORONA images. These points had to be measured
evenly in all parts and height levels of the test area to
get precise results from OrthoBASE Pro triangulation
afterwards. The best positions could be found on the
crossings of roads and rivers. The last one is approx-
imately stable as well because of the arid climate in
Morocco. Still, this way of point collection was diffi-
Fig. 7. Points of DGPS surveying in the test area.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 227
cult because of the few existing roads, the complex
terrain structure and the limited DGPS surveying
period. Therefore, it was impossible to measure GCPs
and slope positions in all parts of the test area. In total,
about 5000 points were collected by DGPS. Out of
these GCPs, again, only a few were valid and good
enough for precise triangulation results in OrthoBASE
Pro (see below). Therefore, just a few GCPs were
selected concerning their quality of corresponding
image and ground position as well as their regular
vertical and horizontal distribution. Due to lack of
experience with using Pro, the time needed for selec-
tion and measurement of good GCPs was about 4
weeks, but now, this task could be completed in 2–3
days.
3. Generation and applications of DSM
3.1. Triangulation and DSM generation
Usually, the interior orientation is defined by the
help of—among others—the image parameters ‘prin-
cipal point’ and ‘fiducial marks.’ However, from the
CORONA camera, these parameters do not exist.
Furthermore, CORONA has a panoramic camera,
which produces images with high geometric deforma-
tions (Fig. 5). Therefore, usually many steps are
required to correct the image and calculate the interior
orientation by removing the panoramic deformation by
a mathematical algorithm, and then allocating a pseudo
principal point and/or fiducial marks (relative to the
ones of the whole filmstrip) to each forward- and aft-
scanned image part. In this study, this way of calcu-
lation was not applied because (1) the two-step proce-
dure for CORONA is very complicated, time-intensive
and insecure (not scientifically verified), especially
considering the mentioned aim ‘large area coverage,
the least financial and temporal efforts possible’ and,
therefore, (2) the possibilities of the software Ortho-
BASE Pro regarding orientation should be evaluated.
OrthoBASE Pro demands as minimum input the
focal length, flight height and the pixel size of the
scanned image. The principal point is automatically
set to zero (i.e. at the image center) by the software.
The program treats the scanned image part as the
whole filmstrip, thus shifting the principal point from
its original position in the whole filmstrip. That would
not be problematic if the scanned forward and aft
image parts were exactly of same size and at the same
position on the filmstrips. However, this is almost
impossible, as the test area is separated on two aft
filmstrips. Scanning of the whole filmstrip would be
impossible as well due to the resulting large file size.
Nevertheless, it may be assumed that the impact of
this definition of the principal point on the object
point accuracy can be almost neglected due to (1) the
great flight height and the narrow bundle of the
projection rays and mainly because (2) this effect is
absorbed by the exterior orientation parameters. Fur-
thermore, the accuracy of object coordinates is
improved by using more than the usually required
three GCPs in absolute orientation (see below). It
must be emphasised that the calculation of the interior
orientation of the CORONA images is a very critical
but interesting research topic and should be further
examined and improved in further studies. This study
is an empirical work based on certain data and tools
and the mentioned aim ‘large area coverage, the least
financial and temporal efforts possible.’ It shows that
the calculation of orientation is possible with some
good results but it also discovers weak points, which
should be further studied.
The interior orientation is followed by the exterior
orientation, divided into the relative orientation of the
forward and aft images and their absolute orientation
to the ground (Heipke, 1997; Mikhail et al., 2001; Fig.
8). The relative orientation establishes the relation
between the forward and aft image by using tie points.
OrthoBASE Pro automatically measures tie points in
both images. In the northern part of the test area,
OrthoBASE Pro generates about 200 tie points and in
the southern part about 400, in both cases almost
evenly distributed.
The absolute orientation in OrthoBASE Pro needs
either approximate exterior orientation parameters,
ephemeris information for the satellite images and
two measured tie points for each pair or three GCPs
for each image. To determine the approximate per-
spective center and rotation angles of each image,
either the similarity transformation, the simplified
space resection or the ephemeris transformation is
used (Wang et al., 2000).
The automatic tie point collection is realised in
OrthoBASE Pro by a structural image matching based
on the correspondence between two structural descrip-
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235228
tions. A structural description of an image consists of
a set of image features (points, lines and regions and
their attributes) and the topological and geometrical
relationships among the features. In order to improve
accuracy and speed of structural matching, procedures
like evaluation function, search method, rotation-
invariant cross-correlation method, correctness check
and extraction of structural descriptions are added. In
combining all these tools, the matching is fully
automated and realised without any a priori informa-
tion (Wang, 1998).
After the measurement of tie points, the triangu-
lation procedure was carried out. Three triangulation
procedures are available: self-calibration (if the cam-
era interior parameters are unknown) which was used
in this study (employing in this case the 12 additional
parameter model of Ebner), relative triangulation (if
there are no GCPs) or the triangulation based on
exterior orientation parameters and GPS/INS data
(Wang et al., 2000). The orientation model in the
form of self-calibrating direct linear transformation
(SDLT) does not require any sensor parameters like
the interior orientation and approximate exterior ori-
entation parameters (incident angle and ephemeris
information), and it also does not require any geo-
metric precorrection of the original image data (Wang,
1999).
The absolute orientation is usually calculated by
the input of at least two full GCPs and one height
GCP. As discussed above, more GCPs than usual were
Fig. 8. Measurement of GCPs and tie points in OrthoBASE Pro in three different zoom images.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 229
necessary for this study: seven for the northern part of
the test area and six for the southern part (Fig. 9).
They were empirically selected to get the best possible
triangulation results, considering the exact corre-
sponding ground and image position of the GCPs as
well as a regular horizontal and vertical distribution of
the GCPs in the test area. A higher number and good
distribution of GCPs were necessary to improve the
complicated and approximate interior orientation of
the CORONA images. If there was an optimal interior
orientation, perhaps three required GCPs would be
enough, maybe even for the very large area of the
whole Draa catchment.
The same matching as for the tie point generation
was used in OrthoBASE Pro—after triangulation—to
create about 50000–70000 irregularly distributed
points with an average point distance of 20m for further
DSM interpolation. The area covered by the DSM was
about 100 km2. The point file and its contours were
created in ERDAS as well as in ESRI shapefile and
ASCII compatible format. The DSM interpolation was
possible in ERDAS IMAGINEVisualGIS, DataPrep or
ArcView 3D Analyst (Figs. 10–12).
The evaluation of triangulation and DSM accuracy
gave the following results. First, the accuracy of
triangulation is shown in OrthoBASE Pro by the
Fig. 9. Position of used GCPs (marked with a flag) in the northern and southern part overlaid on the calculated CORONA orthoimages (dark
regions are palm oases).
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235230
automatically calculated residuals (Table 2). The a
posteriori standard error from the triangulation adjust-
ment was 0.58 pixels in the northern part and 1.19
pixels in the southern part.
In addition, the absolute accuracy of elevation was
estimated by comparing DGPS-measured points with
the DSM points. These DGPS points belong to the
5000 DGPS points collected during the field cam-
paign in Morocco. In ArcView, the DSM points were
overlaid with the DGPS points, enabling the manual
calculation of the height difference of planimetrically
similarly situated points. For this comparison, it was
important to select points from almost all horizontal
and vertical parts of the image. As the DGPS survey-
ing was not as dense as the DSM points, a comparison
of 113 points for the northern part and of 70 for the
southern part of the test area was possible. The
following accuracy of elevation was calculated for
the northern and southern parts (Table 3).
The accuracy varies depending on the position of
the GCPs used for exterior orientation. Higher eleva-
tion differences between DGPS-measured points and
DSM points could be noticed in areas far from the
GCPs. The elevation accuracy of the southern part
DSM is a bit lower than that of the northern part
DSM. One reason for this is surely the less adequate
Fig. 10. DSM generated by OrthoBASE Pro interpolated in ERDAS VisualGIS: northern part above, mosaic of the two DSMs below.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 231
GCP distribution in the southern part. This influence
of number of GCPs and their distribution may be
partly due to errors and approximations in the interior
orientation calculation.
3.2. Dependence of triangulation and DSM accuracy
In total, it can be assumed that triangulation and
DSM accuracy depend mainly on the following fac-
tors: camera model, film deformations, geometric
accuracy and resolution of scanner as well as accuracy
of interior orientation. Furthermore, the number, accu-
racy and regular horizontal and vertical distribution of
the GCPs are important, as also influenced by the
approximative interior orientation in this study. There-
fore, a very high accuracy (only a few metres) of the
corresponding ground and image position of GCPs
was attempted during DGPS surveying and image
measurement. The image position of GCPs has to be
marked in the CORONA images (scan print or photo
print). Finding the exact corresponding ground and
image positions during the field campaign was some-
times quite difficult because of scarce roads, inacces-
sible terrain or structural changes since the CORONA
image acquisition. Achieving the above GCP accuracy
requirement, and the good triangulation and DSM
results shown above without a perfect interior orienta-
tion, was quite an effort in this study and would be a
greater challenge for large areas, like the whole Draa
catchment. If the GCP measurement and triangulation
require less effort, when an accurate interior orienta-
tion is calculated, has to be checked in another study.
3.3. Examples of DSM applications: orthoimage and
anaglyph image
The OrthoBASE Pro-generated DSMs can be the
basis of further applications. It can be useful for 2D
and 3D applications, e.g. anaglyph images and ortho-
images.
Fig. 11. Northern part of the DSM with contours generated in OrthoBASE Pro, interpolated and visualised in ERDAS IMAGINE VisualGIS.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235232
With the help of ERDAS Stereo Analyst, stereo
anaglyph images were created and the file with DSM
points was imported. A 3D visual control of the DSM
point positions was possible (especially in comparison
with roads, etc.). It is possible to continue processing
the DSM data in the anaglyph images by digitising,
e.g. slope or erosion features. These new data can be
used in a 2D and 3D comparison with older or recent
information to perform a change detection analysis.
Another application for DSM is the generation of
orthoimages. The DSM points created by OrthoBASE
Pro were used to generate the corresponding orthoim-
ages of the northern and southern parts. Afterwards,
the orthoimages of the test area were overlaid in
ArcView with the DGPS-measured points. The plani-
metric accuracy of the orthoimages (in x and y
direction) was calculated manually, as for the DSM
accuracy mentioned above. A comparison of 120
DGPS points for the northern and southern parts
relative to the orthoimages was made (Table 4).
For the northern part, a very high accuracy could
be achieved. Based on a pixel footprint of the scanned
images and an orthoimage pixel size of 4� 4 m, the
RMS accuracy of the orthoimage in relation to the
DGPS measured points is 1.8 m in x direction and 2.8
m in y direction (Fig. 13). For the southern part, the
accuracy is lower because of the lower DSM and
triangulation accuracy (RMS is 13.9 m in x direction
Fig. 12. Northern part of the DSM with elevation ranges generated in OrthoBASE Pro, interpolated and visualised in ArcView 3D Analyst.
Table 2
Accuracy of triangulation shown as statistics of the absolute
residuals of tie points (in m)
Northern part Southern part
x y z x y z
Average 2.5 2.7 12.5 4.8 5.7 21.6
Minimum 2.0 2.1 8.4 3.8 4.0 14.2
Maximum 4.1 4.4 22.3 7.0 9.7 38.1
Table 3
Accuracy of DSM elevation relative to DGPS-measured points
(in m)
Statistic Northern part Southern part
Average with sign � 2.6 2.4
Average absolute 7.7 10.7
RMS 9.7 13.3
Minimum � 24.5 � 34.3
Maximum 21.5 33.3
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 233
and 13.7 m in y direction). In addition, there is a large
systematic error (bias) in the y direction, which the
authors could not explain. The planimetric (radial)
RMS for the northern and southern parts was 3.3 and
19.5 m, respectively, the former result being within
expectations (error is less than 1 orthoimage pixel).
These orthoimages can be used in further studies
for change detection analysis, e.g. a comparison of
urban and settlement development, changes of vege-
tation distribution or changes of water resources like
reservoirs. Therefore, the 30-year-old CORONAortho-
images can be compared to other satellite images (e.g.
Landsat, SPOT and IKONOS) or aerial photos, if
available.
4. Conclusions and outlook
This study has been one of the first tests to apply
methods of DSM generation on CORONA satellite
images.
The processing for generating DSMs and digital
orthoimages from CORONA satellite images was
developed in an empirical way. Results of DSM and
orthoimage evaluation show a best accuracy of about
10 m in height and 3 m in planimetry. Due to the
complicated characteristics of CORONA images, the
software program ERDAS IMAGINE OrthoBASE
Pro was used, requiring only few initial orientation
parameters and no geometric precorrection of the
image data. Thereby, errors and approximations made
in the interior orientation, as well as possible geo-
metric scanner errors and not ideal GCP image meas-
Table 4
Planimetric accuracy of CORONA orthoimages, relative to DGPS-
measured points (in m)
Difference statistic Northern part Southern part
x y x y
Average with sign 0.3 � 0.5 1.5 10.1
Average absolute 1.5 1.3 10.9 10.8
RMS 1.8 2.8 13.9 13.7
Minimum � 4 � 5 � 12 � 12
Maximum 4 4 36 35
Fig. 13. CORONA orthoimages generated in OrthoBASE Pro, overlaid by DGPS measured points: section of the northern part left, and of the
southern part right.
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235234
urement accuracy, and GCP number and distribution,
may have contributed in a deterioration of the object
point positioning accuracy.
Nevertheless, it would be an interesting effort to
verify these results in further studies, including pre-
processing of the CORONA satellite images concern-
ing panoramic distortion, scanning process and interior
orientation calculation, as well as possibly using
CORONA images over areas with good reference
DTM, and easy to find and measure GCPs in order
to examine the geometric accuracy potential of this
imagery and possibly compare various sensor models.
Regarding the aim ‘large area coverage, the least
financial and temporal efforts possible’ for countries
like Morocco, the work was no minor effort and it
would be a great challenge to apply this method to the
entire Draa catchment area of about 30000 km2. It
would be interesting to examine if financial and
temporal costs are reduced when applying alternative
methods for the processing of CORONA data.
Acknowledgements
The authors would like to thank all institutions,
persons and companies who participated in this study.
This project had been supported by the Federal
Ministry of Education and Research (project number
07GWK02) and the State Ministry of Education,
Science and Research (project number 514-21200200),
especially Prof. Gunter Menz and Dipl. Geogr.
Michael Schmidt (University of Bonn, Remote
Sensing Research Group). Special thanks to the
company GEOSYSTEMS, Germering for providing
the software ERDAS OrthoBASE Pro as well as the
TU Munich (Mr. Czaja), the company Leica, Munich
and Dusseldorf, Mr. Fuhlbrugge (University of Bonn,
Department of Geodesy) and the surveying company
Kany, Goldkronach for lending GPS devices and
technical advice.
References
Campbell, J.B., 1996. Introduction to Remote Sensing Taylor &
Francis, London.
Day, D.A., Logsdon, J.M., Latell, B., 1998. Eye in the Sky, The
Story of the Corona Spy Satellites Smithsonian, London.
Goossens, R., De Man, J., De Dapper, M., 2001. Research to pos-
sibilities of Corona-satellite-data to replace conventional aerial
photographs in geoarcheological studies, practised on Sai, Su-
dan. In: Buchroithner, M.F. (Ed.), A Decade of Trans-European
Remote Sensing Cooperation. Balkema, Lisse, Netherlands, pp.
257–262.
Heipke, C., 1997. Automation of interior, relative, and absolute
orientation. ISPRS Journal of Photogrammetry and Remote
Sensing 52 (3), 1–19.
IMPETUS (Ed.), 2001a. IMPETUS Westafrika. First intermediate
report 2000, Cologne.
IMPETUS (Ed.), 2001b. http://www.uni-koeln.de/globaler-wandel/
impetus (accessed January 2002).
Leachtenauer, J., Daniel, K., Vogl, T., 1998. Digitizing satellite
imagery: quality and cost considerations. Photogrammetric En-
gineering and Remote Sensing 64 (1), 29–34.
Mc Donald, R.A., 1995a. CORONA. Photogrammetric Engineering
and Remote Sensing 61 (6), 689–720.
Mc Donald, R.A., 1995b. Opening the cold war sky to the public—
declassifying satellite reconnaissance imagery. Photogrammetric
Engineering and Remote Sensing 61 (4), 385–391.
Mikhail, E.M., Bethel, J.S., Mc Glone, J.C., 2001. Introduction to
Modern Photogrammetry Wiley, New York.
NRO (Ed.), 2002. http://www.nro.gov/corona (accessed January
2002).
Ruffner, K.C., 1995. America’s First Satellite Program CIA History
Staff, Washington, DC.
RV-Verlag (Ed.), 2000. World-Landerkarte Marokko 1: 800.000.
Ostfildern.
Slama, C., 1980. Manual of Photogrammetry, 4th edn. The Amer-
ican Society for Photogrammetry and Remote Sensing, 5410
Grosvenor Lane, Suite 210, Bethesda, MD 20814-2160, USA.
Tappan, G.G., Hadj, A., Wood, E.C., Lietzow, R.W., 2000. Use of
Argon, Corona, and Landsat imagery to assess 30 years of land
resource changes in west-central Senegal. Photogrammetric En-
gineering and Remote Sensing 66 (6), 727–735.
USGS, (Ed.), 2002. http://edcwww.cr.usgs.gov/webglis (accessed
January 2002).
Wang, Y., 1998. Principles and applications of structural image
matching. ISPRS Journal of Photogrammetry and Remote Sens-
ing 53 (3), 154–165.
Wang, Y., 1999. Automated triangulation of linear scanner imagery.
Proc. Joint Workshop of ISPRS WG I/1, I/3 and IV/4 ‘‘Sensors
and Mapping from Space 1999’’, Hannover, September 27–30
Available at http://www.ipi.uni-hannover.de/html/publikationen/
1999/isprs-workshop/table_of_contents.htm (accessed 27 Feb-
ruary, 2002).
Wang, Y., Yang, X., Stojic, M., 2000. Automatic triangulation and
rectification of images from airborne and spaceborne sensors
Paper presented at the 19th ISPRS Congress, Amsterdam, July
16–23, 2000. (paper erroneously not included in the proceed-
ings).
A. Altmaier, C. Kany / ISPRS Journal of Photogrammetry & Remote Sensing 56 (2002) 221–235 235