A, 9-11 Nov. 1999
in areas of overlap. The
were estimated from their
ons used in deriving the
in the online documentation
rd Space Flight Center, was
t to evaluate engineering and
high-resolution, orbital laser
surfaces. The first flight of
ary 1996 aboard the space
S-72 mission. Of the
ots transmitted during the
roximately 475,000 yielded
surfaces. Due to the shuttle
ions are distributed between
Details on the SLA-01
vided in Bufton et al. (1995,
1e geolocation processing of
ssentially the same methods
Isewhere in this volume by
sets and documentation are
ov/lapf.
scheme yielding geolocated
ghest detected surface within
tection of a surface requires
ter energy exceeding the
e backscatter return depends
laser-illuminated surface, its
vavelength, and atmospheric
hreshold is varied as the
nges. The background noise
olar illumination (e.g., day
‘the surface observed by the
oud-free locations where
ed elevation will depend on
of the vegetation. For areas
cover the reported elevation
egetation canopy. Similarly,
:evation will depend on the
orresponding to the building
tance to cause the detection
free areas lacking vegetation
onds to the highest ground
flectance. Where optically
ields a cloud-top elevation.
A elevation data has been
arison to Mean Sea Surface
n TOPEX/Poseidon radar
plied for ocean tides but not
). For nearly 728,000 SLA-
Iting residuals show a near
difference of 0.26 m and a
3arvin et al, 1998). The
in surface are thought to be
orbit errors (e.g., once or
twice per revolution). A procedure was developed to correct
these errors using smoothed ocean residuals and to extrapolate
the correction over land, as described in Carabajal et al. (2000).
The horizontal accuracy has not been as well quantified but the
reported SLA footprint locations are thought to be within
several hundred meters of their actual location based on inferred
instrument pointing uncertainty and by matching SLA
topography profiles to 90 m resolution DTED.
2. SLA VERSUS GTOPO30 DIFFERENCES
SLA-01 elevations correspond to the highest detected surface
within a 100 meter diameter footprint whereas GTOPO30
elevations are 'representative' of elevations in 30 arc second
grid cells (approximately a 1 km x 1 km area). Therefore,
elevations in the two data sets are not equivalent; SLA
elevations refer to a specific location covering only
approximately 0.8% of the area represented by a GTOPO30 grid
cell. Also, the manner in which GTOPO30 is representative of
the topography varies as a function of source and continent, as
described above. Because of these differences in the way
topography is sampled, differences between SLA and
GTOPO30 elevations may be large for an individual footprint,
particularly in areas of high relief. However, because of the
high absolute vertical accuracy of the SLA data and the large
number of observations, the GTOPO30 elevation accuracy can
be assessed in a statistical sense using the SLA data.
GTOPO30 elevations are referenced to a mean sea level vertical
datum, an approximation of the geoid, whereas SLA-01
elevations refer to the TOPEX/Poseidon ellipsoid reference
frame. SLA elevations were therefore converted to orthometric
heights with respect to the geoid by subtracting the geoid height
at the footprint as defined by the Earth Geoid Model 96
(EGM96) (Lemoine et al, 1998) SLA to GTOPO30
differences were then computed by subtracting an interpolated
GTOPO30 elevation from the SLA orthometric elevation. The
interpolated GTOPO30 elevation was computed for the SLA
footprint location by bilinear interpolation using the four
nearest GTOPO30 grid cells. Elevation differences were
computed for the four regions having the greatest density of
SLA-01 ground tracks. The regions are Africa including Saudi
Arabia, southern Asia between 60? and 120? E longitude, South
America south of 10? S latitude, and Australia. Refer to
Carabajal et al. (2000) for a global map of SLA-01 ground
tracks.
The elevation differences (SLA orthometric minus interpolated
GTOP30) are summarised in Table 2 as a function of region and
in Table 3 as a function of region and GTOPO30 data source.
Only elevation differences less than or equal to 200 m are
included in an effort to exclude SLA returns from clouds above
the land surface. Histograms of differences for each region
show distributions that decrease to near zero at elevation
differences well less than 200 m, indicating no significant
number of land surface returns are excluded.
International Archives of Photogrammetry and Hemote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
Region Number of Mean St. Dev.
Differences (m) (m)
Africa 244,640 -1.40 44.75
Southern Asia 109,286 14.22 49.03
South America 50,602 6.78 53.32
Australia 29,139 -21.72 48.92
Table 2. Mean and standard deviation of elevation
differences (SLA-01 orthometric minus interpolated GTOPO30)
as a function of geographic region.
Region and Number of Mean St. Dev.
ource Differences (m) (m)
Africa DTED 134,062 2.00 28.88
Africa DCW 109,240 -5.64 58.46
S. Asia DTED 97,804 15.30 45.73
S. Asia DCW 10,654 4.98 72.64
S. Am. DTED 27,583 16.22 32.12
S. Am. DCW 21.112 -4.31 69.09
S. Am. AMS 498 -39.88 86.71
S. Am. IMW 1147 3.39 S3.75
Australia DCW 29,139 -21.72 48.92
Table 3. Mean and standard deviation of elevation differences
(SLA-01 orthometric minus interpolated GTOPO30) as a
function of region and source.
3. DISCUSSION
The mean differences in Table 2 are indicative of systematic
biases between the two data sets on continental scales, with the
southern Asia and South America regions being systematically
higher in the SLA data, Africa showing little systematic
difference, and Australia being systematically lower. For flat
surfaces SLA elevations show little bias, based on the near-zero
mean difference with respect to the ocean surface. However,
one might expect SLA land elevations to be systematically
biased high in the presence of vegetation cover, buildings or
high local relief at the footprint scale due to the first return
ranging. This might account for the mean SLA to GTOPO30
differences observed in southern Asia and South America.
However, if that were the case one would also expect the
extensive forested landscapes of Africa to yield a positive mean
difference that is not observed. The large negative mean for
Australia is also contrary to an expected high SLA bias.
A bias could also be introduced in converting SLA from
ellipsoid to orthometric elevations. However, the formal root
mean square error at long wavelengths for EGMO6 is less than
50 cm (Lemoine et al., 1998), and the ellipsoid references used
for SLA and EGM96 agree at the centimetre level. Therefore,
the continental variations in mean differences must at least in
part be due to systematic errors in the GTOPO30 data set.
These systematic errors may well be due to deviations from
mean sea level of the vertical datums in the GTOPO30 source
materials. The variations in mean differences between data
sources for individual regions (Table 3) indicate that there are
