sub-pixel correlation method. First order polynomial 
coefficients based on MSS pixel coordinates were 
calculated. The polynomial coefficients were applied to 
AVHRR coordinates. Systematically corrected AVHRR 
coordinates were rectified using the polynomial 
coefficients. 
(4) Detection of Parallax and Computation of Elevation 
Parallaxes were detected by sub-pixel correlation 
method. As indicated in the figure 6, an intersection of 
two looking vectors from the ground points is calculated. 
In practice two vectors does not intersect, thus a center 
of a minimum segment of a line between two vectors 
was calculated. 
(5) Evaluation 
Elevations were evaluated using 3-arc second DEM 
produced by USGS. 
3. TEST RESULTS 
3.1 Registration Accuracy of a Stereopair 
MSS scene was rectified with root mean square error 
(RMSE) of 64-m. AVHRR coordinates were rectified 
using the MSS scene as reference images. Thirteen 
AVHRR pixels were selected as GCPs and image-to- 
image correlation was achieved. Table 1 shows the 
result. RMSE changes slightly in case of more than 5 by 
5 reference window. It seems that RMSE 100-m would 
be reached a limit, considering conditions such as MSS 
IFOV, MSS rectification RMSE, incomplete way of 
AVHRR pixel simulation by MSS pixels etc. 
3.2 Accuracy of Generating DEM 
Elevations of all points in AVHRR scene corresponding 
to MSS scene (185-km by 170-km) except sea area were 
calculated. The number of points was 19380. Some 
points were rejected as miss-matching points. 
Conditions of miss-matching points are (1) correlation 
coefficient is less than 0.5; (2) distance between two 
looking vectors is more than 500-m; (3) the difference 
between an original elevation and an elevation after 
applied median filter is more than 500-m. Except these 
points, elevations were evaluated with USGS DEM. 
Table 2 shows the result. 
4. DISCUSSION AND CONCLUSION 
Elevation accuracy of a map is defined normally by the 
contour interval. According to National Map Accuracy 
Standards by USGS, 90% linear error (LE) is within 
half of contour interval. For example 1:1,000,000 scale 
map has usually contour line with intervals of 500-m or 
1000 feet (300-m); the RMSE is 152-m or 91-m 
respectively. RMSE 320-m, the result of this 
experiment, does not meet the criterion. In case of DCW, 
the source map scale is 1:1,000,000 but 90% LE is + 
610-m (Defense Mapping Agency, 1992). This criterion 
38 
is corresponding to RMSE 371-m. This fact indicates 
that the proposed method has potential to fill elevation 
information in contour blank regions within the 
accuracy of DCW; however the accuracy of DCW might 
be better than the criterion in most of regions where 
contours existing. 
Since RMSE of elevation is 320-m and the mean of B/H 
ratio is approximately 0.5, it means registration 
accuracy of a stereopair is less than 160-m. This shows 
that it is possible to rectify AVHRR using GCPs even 
from non coastal pixels. 
As future plans, one of a refinement way of DEM 
accuracy is using multi-direction's AVHRR scenes 
providing the off-nadir view and MSS scene providing 
nadir view like triplet matching. Another way is to use 
sensor with higher resolution and wide FOV. Satellites, 
ADEOS-II and EOS-AMI, which will be launched in 
1998, will have a sensor of 250-m resolution on the 
ground with wide FOV. The sensors names are GLI 
and MODIS respectively. By applying the same method 
to a stereopair of these sensor and Landsat TM, more 
accurate DEM generation is expected. 
ACKNOWLEGEMNT 
This research was done during the author's stay at 
EROS Data Center, US Geological Survey / UENP 
GRID Sioux Falls, supported by STA Long-Term 
Scholarship program of the Science and Technology 
Agency, Government of Japan. 
REFERENCES 
Cracknell, A.P. and Paithoonwattanakij, K., 1989. Pixel 
and Sub-pixel Accuracy in Geometrical Correction of 
AVHRR Imagery. INT.J.REMOTE SENSING, Vol.10, 
pp. 661-667. 
Defense Mapping Agency, 1992. Military Specification 
for Digital Chart of the World (DCW), Document No. 
MIL-D-89009. 
Eidenshink,J.C., Steinwand, D.R, Wivell,C.E., 
Hollaren, D.M. and Meyer, D.J., 1993. Processing 
Techniques for Global Land 1-km AVHRR Data. Proc. 
of PECORA 12 Symposium, Sioux Falls, South Dakota, 
pp. 214-221. 
Kelly, G. and Hood, J., 1991. AVHRR Contermious 
United States Referece Data Set. 1991 ACSM/ASPRS 
Annual Convention, Vol.3, pp.232-239. 
National Oceanic and Atmospheric Administration / 
National Environmental Satellite, 1991. Data and 
Information Services, NOAA Polar Orbiter Data Users 
Guide. 
Raggam, J. and Almer, A., 1991. A Multi-Sensor Stero 
Mapping Experiment. Proc. of ACSM/ASPRS Annual 
Convention, Baltimore, Maryland, Vol.4, pp. 173-182. 
Storey, J.C., 1993. Hybrid Steropairs from Space-Based 
Sensors. Proc. of PECORA 12 Symposium, Sioux Falls, 
South Dakota, pp. 55-60. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996 
w 
Length (km) 
oo 
Fig 
  
Figure