DEM GENERATION FROM MULTISENSOR STEREOPAIRS
- AVHRR AND MSS -
Kazuhiko AKENO
Geographical Survey Institute
Kitasato-1, Tsukuba-shi, 305, JAPAN
email: akeno@graph.gsi-mc.go.jp
Commission IV
KEY WORDS:
ABSTRACT
Stereoscopic, Global DEM Generation, AVHRR, MSS.
There are several studies on digital elevation model (DEM) generation from multisensor stereopairs. One of the
critical condition for generating global DEM is the availability of cloud free images. By combining NOAA AVHRR
and Landsat MSS, which can be widely obtained from huge
archives, it is possible to create DEM using a stereopair
with the MSS providing the nadir looking view and a part of the AVHRR providing the off-nadir view. Since a spot
size of AVHRR is more than one-kilometer, stereo matching must be accomplished by sub-pixel accuracy.
Test project conducted around Mt. Olympus, USA, demonstrated a vertical accuracy of approximately 320-m.
Though the accuracy is low, this result shows that this method has potential for creating DEM within the accuracy
of the Digital Chart of the World, which is the only digital source covering all over the world except some regions
where data is missing, and for filling elevation data in blank regions.
1. INTRDUCTION
Geographic information is indispensable for
understanding global environment. In particular,
elevation data is one of the most basic geographic
information. Thus projects of development of global
one-kilometer resolution digital elevation model (DEM)
are ongoing by compiling existing digital elevation data
such as the Digital Chart of the World (DCW) and the
Digital Terrain Elevation Data (DTED). Yet in a few
regions , there are no existing digital elevation data. In
order to fill elevation data in these blank regions, it is
feasible to take applications of remote sensing
technology.
In case of using optical sensors, availability of cloud
free data is critical, because probability of existing
cloud is very high in some regions. Satellites which can
be obtained stereopairs are limited. As an alternative
way, there are some studies of DEM generation from
stereopairs by combining with different sensors
(Raggam and Almer, 1991 and Storey, 1993).
NOAA and Landsat have long histories and many
scenes are available in the world. However the
resolution of NOAA Advanced Very High Resolution
Radiometer (AVHRR) is low and AVHRR is not seemed
to be suitable for digital elevation data extraction. By
the adaptation of the basic idea of Cracknell and
Paithoonwattanakij (1989) method to stereopairs of
AVHRR and MSS, correlations between AVHRR image
and MSS image are performed with sub-pixel accuracy
of AVHRR resolution, and approximately one-kilometer
grid DEM are generated from the detected parallaxes.
The accuracy of obtained DEM are compared with the
accuracy of DCW and possibility of filling digital
elevation data in blank regions of DCW is discussed.
2. APPROACH TO GENERATING DEM USING
AVHRR AND MSS
2.1 Basic Idea of DEM generation using AVHRR
For elevation data extraction, stereopairs are needed.
AVHRR has wide field of view (FOV) and the scan angle
is +55.4 degree from the nadir. Using some range of
AVHRR scene from the nadir, we can use it as a off-
nadir view image for one of stereopair. Since a spot size
of the AVHRR on the ground is relatively large
compared with terrain of the earth, even rough terrain
can not be obtained without measurement of high
precision of parallax.
Image-to-image correlation technique is used for
parallax detection. This technique is also used for
registration between two images. Accuracy of
correlation between AVHRR scenes for image
registration are approximately 0.6 pixel - 0.7 pixel
(Kelly and Hood, 1991). This accuracy is insufficient for
detecting parallax against the earth terrain. For
example, if B/H ratio is 0.5 and AVHRR pixel size is
one-kilometer on the ground, the vertical accuracy of
detectable elevation is more than one-kilometer.
Cracknell and Paithoonwattanakij (1989) showed a
more accurate registration method that Landsat MSS
image are used for AVHRR image registration.
According to their result, the accuracy was
approximately 260-m. Though this accuracy is not still
sufficient for this study, it is anticipated that
36
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
adapt
corres
becau
separ:
For e:
lookin
the o
(Figui
22 5]
The s
is tha
the s
radiu:
adjac
shows
of cur
is noi
point.
sensit
groun
less d
Near
nadir
Figur
detect
a par:
at 50¢
is ap]
chang
antici
Rang
the 1
gener
to 0.6
23 |
(1) Re
Prior
regist
regist
impos
accur
regist
rectif
meth
(2) Si
NOA,
satell
There
equat
afterı
imag:
morn
As foi
suital
AVHI
AVH]
2 cov
MSS
spect
the r:
