| of the
\VHRR
reopair
> a spot
320-m.
ccuracy
regions
3sed.
SING
IRR
needed.
in angle
range of
as a off-
spot size
ly large
| terrain
of high
sed for
used for
racy of
image
).7 pixel
icient for
ain. For
1 size is
‚uracy of
r.
howed a
sat MSS
istration.
cy was
; not still
ted that
adaptation of this method in a small area
corresponding to MSS scene will improve the accuracy
because they used many MSS scenes rectified
separately for rectifying a whole AVHRR scene.
For extraction of terrain, MSS image providing a nadir
looking view and a part of a AVHRR image providing
the off-nadir view were used as a stereopair image
(Figure 1).
2.2 Discussion on AVHRR image as a off-nadir view
The specialty of using AVHRR image as off-nadir view
is that a spot size on the ground becomes larger along
the scan angle from the nadir. Figure 2 shows long
radiuses, short radiuses and interval distance from the
adjacent pixel along the scan direction. Upper axis
shows elevation angle from the ground point. Because
of curvature of the earth, a scan angle from the satellite
is not equal to the elevation angle from the ground
point. Conditions for detecting parallax with high
sensitivity are (a) a pixel to be high resolution on the
ground, (b) high B/H ratio between a stereopair and (c)
less distortion between two images for high correlation.
Near nadir image of AVHRR meets (a) and (c), and off-
nadir image of that meets (b).
Figure 3 shows the relation between sensitivity of
detecting parallax and pixel location from the nadir. If
a parallax is detected with 0.1 pixel accuracy of AVHRR
at 500 pixel from the nadir, the sensitivity of elevation
is approximately 260-m. Though sensitivity does not
change drastically at off-nadir pixels, low correlation is
anticipated because of distortion of image.
Range approximately from 350 pixel to 500 pixel from
the nadir across the track would be suitable for
generating DEM. In this range, B/H ratio is from 0.40
to 0.60.
2.3 Important factors for precise DEM generation
(1) Relative registration of stereopair images
Prior to the parallax detecting process, precise relative
registration of stereopair must be carried out. If
registration was done separately, it would be
impossible to detect parallaxes within sub-pixel
accuracy. In order to carry out the most accurate
registration between two images of a stereopair, using
rectified MSS image, image-to-image registration
method is applied to AVHRR image.
(2) Similarity of stereopair images
NOAA and Landsat are sun-synchronous orbit
satellites. Landsat passes the equator in the morning.
There are two types of NOAA satellites. One passes the
equator and the other passes the equator in the
afternoon. For reducing the shadow effect against
image correlation, AVHRR scenes acquired in the
morning such as NOAA-10, NOAA-12 must be selected.
As for wavelength of stereopair images, near infrared is
suitable because of less noise effect. Thus NOAA
AVHRR Channel 2 (0.725-1.10 x m) be selected in
AVHRR channels. The range of wavelength of channel
2 covers approximately both wavelength of Landsat
MSS band3 (0.7-0.8 x m) and band4 (0.9-1.1 4m). The
spectral response curve of channel 2 indicates a peek in
the range of band 3 (National Oceanic and Atmospheric
37
Administration, 1991). Therefore band 3 is selected.
The near infrared has much sensitivity of vegetation
changes. Thus stereopair should be selected in the
same season. Selecting procedure is as follows. At first
cloud free MSS scene is selected from archives.
Secondly AVHRR scene which does not have cloud in
the same part corresponding to the MSS scene is
selected from archives. For searching the similar
vegetation condition, AVHRR scene is sought day by
day from the same date of MSS scene. AVHRR scene
can be obtained every day; consequently the probability
of obtaining cloud free AVHRR scene is very high
because time difference between two scenes is small in
the same day.
2.4 Sub-pixel Correlation
As mentioned in 2.2, a pixel spot size on the ground
changes depending on scan angle. A interval of pixel
distance is smaller than the pixel spot radius along the
scan line. This mean that each spot overlaps the
adjacent pixel spot. To avoid resampling effect of
AVHRR pixels, AVHRR pixels treat as a coordinates
file and both systematic correction and rectification of
AVHRR scene is applied to coordinates. Coordinates file
include information of pixel value, satellite zenith angle
and spot size.
MSS image are degraded to AVHRR resolution by
simulating AVHRR spot with MSS pixels. Image-to-
image correlation between a reference window on
AVHRR image and a search window on MSS image are
accomplished. A maximum correlation can be identified
to accuracy of MSS pixel size (80m) (Figure 4). Since
sensor are different, normalized correlation (correlation
coefficient) should be computed for correlation.
This method is applied to both rectification of AVHRR
and detection of parallax between AVHRR image and
MSS image.
2.5 Flow of Experiment
As a test area, a scene including Mt. Olympus
(elevation 2428-m) in Washington State, USA was
selected. MSS scene is from Landsat 5 and AVHRR is
from NOAA 12. Acquisition date of both two scenes is
September 16, 1992. The range of AVHRR scene
corresponding to MSS scene is approximately from 330
pixel to 500 pixel from the nadir along the scan line.
The flow of experiment is following (Figure 5).
(1) Rectification of MSS
Using the points from Digital Line Graph (DLG)
produced by US Geological Survey (USGS) as ground
control points (GCPs), systematically corrected MSS
scene was rectified.
(2) Systematic Correction of AVHRR
Raw AVHRR image was systematically corrected as a
coordinates file using orbit model. This was done using
the AVHRR Data Acquisition and Processing System
(ADAPS) support routine developed by EROS Data
Center, USGS (Eidenshink, 1993).
(3) Rectification of AVHRR
Points along a coastal line were selected from the
AVHRR pixels as provisional GCPs. Corresponding
points in MSS images were selected by above described
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996