|
Using ASTER DEMS to Produce IKONOS Orthophotos
In many areas of the world, it is difficult to obtain or create
an accurate digital elevation model (DEM). In other areas, obtaining
an accurate DEM is prohibitively expensive. Lack of an accurate
or affordable DEM over certain areas of the earth means that
producing accurate orthophotos over these areas is difficult,
cost-ineffective, or even impossible. ASTER, a sensor aboard
NASA’s Terra satellite, provides low-priced Visual and Near-Infrared
(VNIR), Thermal Infrared (TIR), and Short Wave Infrared (SWIR)
data. The VNIR segment of the data contains stereo bands that
can be used to produce DEMs. The DEMs derived from ASTER data
using control points extracted from IKONOS imagery are of sufficient
accuracy to allow the production of Reference Product (25m CE90)
orthophotos from IKONOS imagery anywhere in the world where
ASTER data is available. This article outlines the process for
creating DEMs from ASTER data, including obtaining the data,
necessary software, and an overview of the actual production
steps. Included are case study accuracy assessments in which
DEMs produced from ASTER data were used to orthorectify IKONOS
imagery, and the resulting orthophotos compared to those produced
using DEMs of known accuracy derived from IKONOS stereo data.
Peter Aniello
ASTER is a sensor aboard the Terra satellite, which was launched
in December 1999. ASTER as two telescopes, one nadir-looking
and one back-looking, which collect stereo data at 15-meter
resolution. A single ASTER data granule containing stereo data
and covering 3600 square kilometers costs $55 US, plus a $5
US handling charge. The two stereo bands (3n and 3b) contained
in an ASTER Level 1A (L1A) data granule can be used to produce
DEMs using a Space Imaging proprietary sensor model. These DEMs
can be used to orthorectify IKONOS imagery to Space Imaging
Reference Product accuracy standards (25-meter CE90). This is
significant because, for some areas of the world, ASTER provides
the only means to produce an accurate DEM at low cost. The ASTER
sensor will eventually image the entire Earth between latitude
80 north and 80 south; most of the Earth has already been imaged,
although an exact percentage is unavailable at this time.
Space Imaging’s IKONOS satellite was launched in September 1999.
IKONOS is the world’s first commercial satellite to collect
panchromatic imagery at 1-meter resolution and multi-spectral
imagery at 4-meter resolution.
To achieve good accuracy using the Space Imaging ASTER sensor
model, ground control points (GCPs) are read from IKONOS imagery
and applied to the ASTER imagery. The advantage of using Space
Imaging’s proprietary ASTER sensor model is that few GCPs are
required, and GCP distribution is not critical. As noted in
previous studies, ASTER DEM accuracy may deteriorate in steeper
terrain (A. Kaab et al, 2002). However, in many instances, it
is not unreasonable to expect ASTER DEM accuracies to be comparable
to USGS 1-arc second DEM accuracies (T. Toutin and P. Cheng,
2001).
Accurate control points are not available for many areas of
the world, but they can be derived from overlapping IKONOS monoscopic
imagery with Rational Polynomial Coefficients (RPCs), viewed
stereoscopically. The advantage of using overlap areas of IKONOS
mono imagery as opposed to stereo IKONOS imagery to derive point
locations is that the cost is lower. Stereo IKONOS overlaps
by nearly 100 percent, and so requires that a given area be
imaged twice; mono imagery may overlap by a very small amount
and still be useable for control point extraction when viewed
stereoscopically, so that fewer images are needed. The cost
savings means that the derived product can be offered at a lower
price.
Obtaining ASTER Data
ASTER data can be searched and ordered from the USGS EROS Data
Center’s EDG (EOS Data Gateway) website (http://edcimswww.cr.usgs.gov/pub/imswelcome/).
Browse images are available for search results, and the results
may be displayed in a text list format for importing to a spreadsheet.
ASTER data granules can be staged to an FTP site, or delivered
on CD-ROM or 8mm tape.
Production Process
Space Imaging uses a proprietary software to reformat the ASTER
L1A stereo bands. After reformatting, any software that is capable
of block-adjusting IKONOS and auto-correlating stereo pairs
can be used to produce ASTER DEMs.
A project is created, and the ASTER L1A stereo bands are imported
as separate image files. GCPs are read from the IKONOS imagery
in the overlap area, and applied to the ASTER images.
Once the GCPs have been applied, a triangulation, or block-adjustment,
is performed. Auto-correlation of the ASTER stereo pair(s) produces
a DEM (for this study, post spacing was set to 30 meters). Editing
may be necessary, as areas of clouds, shadows, and water will
miscorrelate.
Once the DEM is edited, it is either output to a format useable
by Space Imaging’s IKONOS production area for orthophoto production,
or an orthophoto is produced using the same software as
was used for the block-adjustment and DEM production (see Figures
1 and 2 for an overview of the production process).
Test Areas
Three areas of the world were chosen as test areas: southwestern
Australia, the southern tip of the north island of New Zealand,
and western Puerto Rico (see Figure 3).
The criteria for choosing these areas included availability
of archived IKONOS imagery, availability of ASTER imagery, and
variability of terrain. The Puerto Rico test area consisted
of low-lying coastal flats and very rugged hills covered with
tropical vegetation. Since the Puerto Rico test area contained
many cloud areas, which resulted in non-correlation (failure)
of the DEM in these areas, cloud-free areas were subsetted out
for testing purposes. The Australia test area consisted of rolling
agricultural terrain with many field patterns. The New Zealand
test area consisted of coastal flats, rolling agricultural areas,
and steep mountains up to 2,000 meters in elevation.
Test Procedures
ASTER imagery was obtained and DEMs were produced from the imagery
at 30-meter post spacing using the procedure outlined above.
DEMs were also produced at 30-meter post spacing from stereo
IKONOS imagery for the Australia and New Zealand test cases.
For the Puerto Rico test case, the ASTER DEMs were compared
to DEMs of 6-meter post spacing obtained from Intermap. The
DEMs were then compared by “subtracting” the IKONOS or Intermap
DEM from the ASTER DEM. This produced a “difference surface”
for which statistics were calculated. Orthophotos were then
produced using the DEMs derived from both ASTER and IKONOS for
the Australia and New Zealand study areas.
Test Results
Using the generated statistics, the linear errors (LE90) of
the ASTER DEMs were calculated using Equation 1.
The LE90 of the three test area ASTER DEMs are shown in Table
1.
From the LE90, we can accurately calculate the amount of error
(CE90) that the DEM will contribute to an orthophoto produced
from IKONOS imagery taken at a given elevation angle and using
an ASTER DEM as the elevation source. The CE90 of the expected
error contribution would be described by Equation 2, where E
is the elevation angle of the IKONOS image (Figure 4).
Once we know the amount of error that the DEM will contribute
to an orthophoto, Equation 3 can be used to calculate the CE90
accuracy of an orthophoto produced using the DEM.
Table 2 shows calculated ASTER DEM error contributions and orthophoto
accuracies for an IKONOS image at various elevation angles and
assuming a line-of-sight geometric CE90 accuracy for the input
image of approximately 8.6 meters (J. Grodecki and G. Dial,
2002).
To determine the actual accuracy of the orthophotos produced
using the ASTER DEMs, a Space Imaging proprietary software was
used which compares pixel locations between two images and produces
a shapefile depicting pixel location differences (see Figure
5). This shapefile was then used to calculate statistics. Equation
4 was used to calculate the CE90 accuracy of the ASTER DEM orthophoto
as compared to the IKONOS DEM orthophoto.
For the Australia orthophoto accuracy test, the statistics are
shown in Table 3.
Solving for Equation 4, the observed accuracy of the Australia
orthophoto produced using the ASTER DEM compared to the orthophoto
produced using the IKONOS DEM is about 5.1 meters. The predicted
accuracy of the ASTER DEM orthophoto with respect to the IKONOS
DEM orthophoto (the ASTER DEM error contribution) would be approximately
4.6 meters, based on the actual elevation angle value of approximately
66 degrees and solving for Equation 3.
For the New Zealand orthophoto accuracy test, the statistics
are shown in Table 4.
Solving for Equation 4, the actual accuracy of the New Zealand
orthophoto produced using the ASTER DEM compared to the orthophoto
produced using the IKONOS DEM is about 8.5 meters. The predicted
accuracy of the ASTER DEM orthophoto with respect to the IKONOS
DEM orthophoto (the ASTER DEM error contribution) would
be approximately 11.2 meters, based on the actual elevation
angle value of approximately 75 degrees and solving for Equation
3.
The larger discrepancy between the observed accuracies and the
predicted accuracies for the New Zealand orthophotos can
most likely be attributed to the fact that the program used
to compare the pixel locations tends to throw out the higher
difference values, such as might be found in mountainous areas,
resulting in overly optimistic observed accuracies. The Australia
test area contained no mountainous areas.
Since no archived IKONOS imagery was available over the cloud-free
areas of the Puerto Rico ASTER DEM, no orthophotos were produced
for this test case, although it is expected that any orthophoto
generated from an IKONOS image collected at an elevation angle
of 52 degrees or greater would meet or exceed Reference Product
(25-meter CE90) specifications (see Table 2).
Conclusion
DEMs created from ASTER imagery using control points derived
from overlapping IKONOS imagery viewed in stereo are, given
any likely IKONOS imagery elevation angle, accurate enough to
be used for production of IKONOS Reference Product (25-meter
CE90) orthophotos anywhere in the world where ASTER L1A data
and archived IKONOS data are available. The fact that ASTER
is low-cost and that stereo IKONOS is not required to extract
control points means that derivative products can be made available
to the consumer at a lower price than would otherwise be possible.
References
A. Kaab et al (2002). Glacier Monitoring From ASTER Imagery.
EARSeL Proceedings, LIS-SIG Workshop, Berne, Switzerland.
J. Grodecki and G. Dial (2002). IKONOS Geometric Accuracy Evaluation.
ISPRS Proceedings, Denver, Colorado.
T. Toutin and P. Cheng (2001). DEM Generation With ASTER Stereo
Data. Earth Observation Magazine, June 2001.
About the Author
Peter Aniello is a photogrammetrist in the Photogrammetric Engineering
Department at Space Imaging in Thornton, Colorado.
Back
|
