The document summarizes research analyzing differences between elevations measured by the ICESat satellite laser altimeter and the Shuttle Radar Topography Mission (SRTM) digital elevation model over Australia. The analysis found a mean elevation difference of approximately -2 meters, with SRTM biased high relative to ICESat. It also revealed undulating elevation errors in SRTM at the 100s of km scale. The differences were reproducible across ICESat observation periods, indicating they were characteristic of SRTM rather than ICESat.

1. Claudia C. Carabajal1,
David J. Harding2,
and Vijay P. Suchdeo1
1 Sigma Space Corp. @ NASA/GSFC – Planetary Geodynamics Laboratory
2NASA/GSFC - Planetary Geodynamics Laboratory
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010

2. Globally-distributed Repeated Profiles
Geoscience Laser Altimeter System (GLAS)
Footprint: ~70 m (lasers 1 & 2), ~50 m (laser 3)
Along-track spacing: 170 m
Vertical Precision: 3 cm (flat surfaces)
Vertical Accuracy: ~10 cm (flat surfaces)
Horizontal Accuracy: < 6 m
Primary Objectives
Ice sheet elevation change
Sea ice thickness change
Secondary Objectives
Cloud and aerosol profiles
Geodetic land topography profiles
Forest canopy height sampling
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
2

3. • The high accuracy of the ICESat elevation measurements in
a consistent reference frame provides a unique, globally
distributed Ground Control Point (GCP) data set
Vertical Accuracy: 10 cm (flat surface) Horizontal Accuracy: < 6 m
• Three main applications of ICESat geodetic control are:
Independent assessment of the accuracy of DEMs
defining their random and systematic error characteristics.
Correction of systematic errors in DEMs
improving their utility scientific and applied purposes
including detection of elevation change
Use as ground control points in the production of DEMs
either by stereo photogrammetric or interferometric SAR techniques
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010

4. 1A 2A 2A
2B 2C 3A
3B 3C 3D
3E 3F 3G
3H 3I
3J 3K 2D
2E
L1 & L2 8-day Laser 2 – 91 day Laser 3 – 91 day
Laser Energy Corrected for ICESat was in a precisely
Mean per Pulse Energy (mJ)
FOV Shadowing Effects
repeated orbit (±86°),
acquiring data along the
same 491 orbit tracks in
Laser 3 ~33-day long periods.
Laser 2
Laser energy dropped
significantly during the
Observation Period course of the mission
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
4

5. • ICESat Land/Canopy Product (GLA14), Release 31
GLAS waveform-derived elevations
highest detected signal
signal centroid (average)
inferred ground peak
lowest detected signal
Each Laser 2 and 3 month-long observation periods used separately
to assess reproducibility of the results
• SRTM Finished Product
DEM elevation interpolated to laser footprint location, provided on GLA14
geoid corrected to be in ICESat reference frame
Elevation standard deviation (relief) from 3 x 3 cells at footprint location
• ESA’s MERIS Globcover
Global land cover at 300 m resolution (Regional products)
51 land cover classes possible
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
5

6. After Harding & Carabajal, 2005.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
6

7. Stringent editing applied to identify appropriate returns:
Low within-footprint slope and roughness
Vegetation absent or very low stature
Not impacted by measurement artifacts
• Surface returns not from cloud tops
ICESat - SRTM DEM elevations < 50 m
• Non-saturated returns
Saturation index ≤ 2
• Data acquired near nadir
Incidence angle ≤ 1°
• No potential range delay due to atmospheric forward scattering
When correction available, in the mm range
• No broadened returns from high relief or vegetation cover
Width ≥ 0.5 m and ≤ 5 m
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
7

8. • Negative elevation differences: SRTM biased high relative to ICESat
absolute datum by several meters, on average, across western Australia.
• The along-profile variations reveal undulating elevation errors in the
SRTM DEM at the 100s of kilometer length scale and ~5 m amplitude.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
8

9. • We quantify differences between ICESat GCP’s and SRTM
along the ICESat ground tracks using a sliding 1 degree box-
car filter.
• We compute average 1 degree gridded ICESat-SRTM
elevation differences.
• We evaluate spatial patterns of mean elevation differences
(biases) and standard deviations (noise component).
• We do this using each ICESat observation period separately,
testing the reproducibility of ICESat elevation
measurements with different laser energies.
• We include topographic relief and land cover information to
establish empirical relationships between ICESat - SRTM
elevation differences with respect terrain characteristics.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
9

10. • Difference histograms for ICESat’s highest, centroid,
inferred ground and lowest elevations show well-defined
normal distributions.
• ICESat centroid and inferred ground are essentially
equivalent for the narrow waveforms selected by editing
• SRTM elevation bias ~ 2 m above ICESat’s centroid.
15
Frequency (%)
10
Highest
Centroid
5 Ground
Lowest
0
-10 0 10
ICESat – SRTM Elevation (m)
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
10

11. L2A L2B L2C
L3A L3D L3G
The along-profile smoothed differences show long wavelength
undulations in the SRTM DEM, of several meters magnitude, that are
consistent for all observation periods and lasers.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
11

12. Along-track differences show large
wavelength undulations (100s of
kilometers) for the various periods, not
correlated with relief.
The along-track differences are
independent of the ICESat observation
period, and are therefore characteristic of
SRTM.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
12

13. Along-track differences show large
wavelength undulations (100s of
kilometers) for the various periods, not
correlated with relief.
The along-track differences are
independent of the ICESat observation
period, and are therefore characteristic of
SRTM.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
13

14. Along-track differences show large
wavelength undulations (100s of
kilometers) for the various periods, not
correlated with relief.
The along-track differences are
independent of the ICESat observation
period, and are therefore characteristic of
SRTM.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
14

15. Residual height error of the
SRTM X-band DEM.
(a) Error along a particular
data take acquired over the
pacific for calibration
purposes.
Shown is the band of the
relative and absolute vertical
accuracy requirement.
(b) Schematic distribution of
SRTM error sources across
spatial scales in azimuth
direction.
The largest error contribution
comes from roll angle firings
used to counteract the torque
exerted on the mast by the
earth gravity field gradient.
Rabus et al., 2003
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
15

16. points/cell mean st. dev.
0 1000 -20 20 0 10
rmse minimum maximum
0 10 -20 20 -20 20
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
16

17. Laser 2 Laser 3 2m
-2 m
-6 m
• Centroid differences for all laser periods show very consistent means
of ~ -2m, a demonstration of ICESat’s highly accurate and
reproducible absolute elevations.
• There is a slightly decreasing trend with laser energy decay,
especially for Laser 2. It is not related to editing of saturated
returns during high energy periods.
• The origin of this ICESat L2 drift and the associated increase in
standard deviation requires further investigation.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
17

18. sparse
vegetation
Water
grassland/
cropland/ short
grass/ stature
shrubs vegetation bare
areas
grassland/
Cropland/ short
grass/ stature
shrubs vegetation
Bare areas
Sparse Vegetation
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
18

19. 5m
0m
-5m
Histograms of differences between Mean: -1.91 m
ICESat and SRTM 90 m elevations at the St. Dev.: 2.12 m
ICESat footprint locations for bare ground
land cover. The Mean and Standard
Deviation of the distribution are -1.91 m
and 2.12 m, respectively, for a population
of 46271 laser returns.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
19

20. Narrow ICESat Waveforms
L2B (Feb.-Mar., 2003)
Waveforms with narrow
pulse-widths (0 to 5 m), are
consistent with low relief
surfaces having no or only
short-stature vegetation
cover, and are suitable for
use as ground elevation
Waveform Pulse Width (m)
5.0
control points.
4.5
4.0
Approximately 30%-35% of
the data acquired in North
3.5
America fits this criteria
≤3.0
(however, a large fraction
are at higher latitudes
where the ground track
spacing is smaller).
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
20

21. Identify narrow last peaks in broad waveforms that are likely
to be returns from the ground beneath the vegetation to
increase the number of global GCPs.
Use of last peaks as GCPs in vegetated terrain must be restricted to areas of
low topographic relief due to the complex merging of ground and canopy returns
in waveforms from areas moderate to steep relief. (Harding & Carabajal, 2005)
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010
21

22. • Using careful editing of ICESat elevation data, we are developing a
Global Geodetic Control database for a variety of Solid Earth
applications.
• Edited data apply to locations of low relief and absent to short
stature vegetation cover (< a few meters).
• As an application of ICESat for Ground Control, we have performed a
comprehensive analysis of the spatial distribution and magnitude of
the ICESat - SRTM differences for Australia.
• A negative mean difference of ~ 2 m (SRTM on average higher than
ICESat) is observed for Australia, but there are regionally correlated
mean differences that vary from about -10m to 5m. These might be
associated with differences in land cover type.
• We have investigated the repeatability of the results for all ICESat
observation periods, exploring possible intra-period instrument/
pointing biases remaining in the ICESat elevation data.
• Identification of ground peaks in broadened waveforms will expand
the number of GCPs for vegetated regions.
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010

23. • Methodologies developed to use ICESat data for global geodetic
control purposes are a pathfinder for similar use of the data to be
produced by the Lidar component of the DESDynI mission.
• With substantially improved sampling as compared to ICESat
DESDynI will provide a more comprehensive set of global GCPs
- Multiple beams spaced across track by ~ 1 km
- Smaller footprints (25 m) that are contiguous along track
- Continuous, rather than episodic, operation
• Differencing the densely sampled DESDynI Lidar data through
time with respect to a common DEM should reveal surface elevation
changes at the decimeter level during the course of the mission on
a local to regional (TBD) scales, including for surfaces that are
decorrelated at radar wavelengths
E.G. seasonal snow accumulation; soil loss in agricultural regions
Carabajal et al. - IGARSS 2010 - Honolulu, HI, July 25-31, 2010