in documentation provided, along with the data set, at
http://edcwww.cr.usgs.gov/landdaac/dataproducts.htm.
The manner in which a GTOPO30 elevation value in a 30 arc
second grid cell represents the topography for that portion of
the Earth’s surface varies depending on the data source and on
processing methods which varied between continents.
Processing of the raster source data involved generalising the
higher resolution data to the 30 arc second horizontal grid
spacing. As the GTOPO30 project progressed, several methods
of generalisation were used. Selection of the representative 30-
arc second value was accomplished by systematic subsampling
for North and South America, by calculation of the median
value for Eurasia, and by the breakline emphasis approach
(Gesch and Larson, 1996) for Africa (essentially selecting either
the maximum or minimum value from the higher resolution
grid). The 500-meter New Zealand DEM was generalised to
30-arc seconds by reprojecting it from the New Zealand
National Grid projection to geographic co-ordinates using
bilinear resampling. The topographic information from the
vector cartographic sources was converted into elevation grids
through a vector-to-raster gridding approach. Contours, spot
heights, stream lines, lake shorelines, and ocean coastlines were
input to the ANUDEM surface gridding program developed at
the Australian National University (Hutchinson, 1989).
ANUDEM employs an approach known as drainage
enforcement to produce raster elevation models that represent
more closely the actual terrain surface and contain fewer
artefacts than those produced with more general purpose surface
interpolation routines.
The absolute vertical accuracy of GTOPO30 varies by location
according to the source data. Generally, the areas derived from
the raster source data have higher accuracy than those derived
from the vector source data. The vertical accuracy associated
with each source type is provided in Table 1, either from
product specifications, calculation or estimation. For DTED,
and the derived USGS DEM's, vertical accuracy specifications
are provided by NIMA. However, these are generalised
specifications and the method for establishing DTED accuracy
Source 90% LE RMSE
DTED 30 18
USGS DEM 30 18
N.Z. DEM 15 9
DCW 160 97
AMS maps 250 152
IMW maps 50 30
Peru map 500 304
ADD variable variable
Table 1. GTOPO30 absolute vertical accuracy by source, as
linear error at the 90% confidence level (90% LE) and as root
mean square error (RMSE).
is not documented. DTED accuracy is known to vary
geographically and with method of production. The DCW
accuracy reported in Table 1 was obtained by calculating
International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 3W14, La Jolla, CA, 9-11 Nov. 1999
differences with respect to DTED in areas of overlap. The
accuracies of the other sources were estimated from their
contour interval. The assumptions used in deriving the
accuracies in Table 1 are detailed in the online documentation
provided with the data set.
1.2; Shuttle Laser Altimeter
SLA, developed at NASA’s Goddard Space Flight Center, was
designed as a pathfinder experiment to evaluate engineering and
algorithm techniques for obtaining high-resolution, orbital laser
altimeter observations of terrestrial surfaces. The first flight of
the SLA instrument was in January 1996 aboard the space
shuttle Endeavour on the STS-72 mission. Of the
approximately 3 million laser shots transmitted during the
course of the 10 day mission, approximately 475,000 yielded
geolocated laser returns from land surfaces. Due to the shuttle
orbit inclination, the SLA observations are distributed between
28.45° N and S latitudes. Details on the SLA-01
instrumentation and results are provided in Bufton et al. (1995,
1999) and Garvin et al. (1998). The geolocation processing of
the SLA-01 laser footprints used essentially the same methods
as those for SLA-02, described elsewhere in this volume by
Carabajal et al. (2000). SLA data sets and documentation are
available at http://denali.gsfc.nasa.gov/lapf.
SLA utilises a first-return ranging scheme yielding geolocated
elevations that correspond to the highest detected surface within
the 100 m diameter footprint. Detection of a surface requires
reception of sufficient backscatter energy exceeding the
instrument detection threshold. The backscatter return depends
on the nadir-projected area of the laser-illuminated surface, its
reflectance at the 1064 nm laser wavelength, and atmospheric
transmissivity. The detection threshold is varied as the
background optical noise level changes. The background noise
is dependent on the amount of solar illumination (e.g., day
versus night) and the reflectance of the surface observed by the
receiver field-of-view. For cloud-free locations where
vegetation is present, the geolocated elevation will depend on
the density and spatial organisation of the vegetation. For areas
with sufficiently dense vegetation cover the reported elevation
will correspond to the top of the vegetation canopy. Similarly,
in urbanised areas the geolocated elevation will depend on the
spatial organisation of buildings, corresponding to the building
top with sufficient area and reflectance to cause the detection
threshold to be exceeded. In cloud-free areas lacking vegetation
or buildings, the elevation corresponds to the highest ground
surface of sufficient area and reflectance. Where optically
dense clouds are present, SLA-01 yields a cloud-top elevation.
The vertical accuracy of the SLA elevation data has been
assessed for flat surfaces by comparison to Mean Sea Surface
ocean topography, derived from TOPEX/Poseidon radar
altimeter data, with a correction applied for ocean tides but not
for sea state (Carabajal et al., 2000). For nearly 728,000 SLA-
01 ocean surface returns the resulting residuals show a near
Gaussian distribution with a mean difference of 0.26 m and a
standard deviation of 2.78 m (Garvin et al., 1998). The
observed deviations from the ocean surface are thought to be
primarily due to long-wavelength orbit errors (e.g., once or
Internatio
twice per revolutior
these errors using si
the correction over |
The horizontal accu
reported SLA foot
several hundred met
instrument pointin
topography profiles
2. SLA VE
SLA-01 elevations
within a 100 mete
elevations are 'repr
grid cells (approxir
elevations in the
elevations refer t
approximately 0.896
cell. Also, the mani
the topography vari
described above.
topography is sa
GTOPO30 elevatior
particularly in areas
high absolute vertic
number of observati
be assessed in a stati
GTOPO30 elevatior
datum, an approx
elevations refer to
frame. SLA elevati
heights with respect
at the footprint as
(EGM96) (Lemoin
differences were the
GTOPO30 elevatior
interpolated GTOP(
footprint location
nearest GTOPO30
computed for the f
SLA-01 ground trac
Arabia, southern As
America south of
Carabajal et al. (2X
tracks.
The elevation differ
GTOP30) are summ
in Table 3 as a fun
Only elevation diff
included in an effor
the land surface.
show distributions
differences well le
number of land surf
