We conducted a 750 km kinematic GPS survey, referred to
as the 88S Traverse, based out of South Pole Station, Antarctica, between
December 2017 and January 2018. This ground-based survey was designed to
validate spaceborne altimetry and airborne altimetry developed at NASA. The
88S Traverse intersects 20 % of the ICESat-2 satellite orbits on a route
that has been flown by two different Operation IceBridge airborne laser
altimeters: the Airborne Topographic Mapper (ATM; 26 October 2014) and the
University of Alaska Fairbanks (UAF) Lidar (30 November and 3 December 2017). Here we present an overview of the ground-based GPS data quality and
a quantitative assessment of the airborne laser altimetry over a flat
section of the ice sheet interior. Results indicate that the GPS data are
internally consistent (
The Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) is a
next-generation laser altimeter developed by the National Aeronautics and
Space Administration (NASA) and launched on 15 September 2018 (Markus et al.,
2017). ICESat-2 will carry a single instrument, the Advanced Topographic
Laser Altimeter System (ATLAS), a six-beam photon-counting system using
Plans for the post-launch validation of ICESat-2 elevation data products include utilizing both ground-based and airborne elevation datasets. The relatively short ground-based datasets, such as presented here, will provide error assessments for airborne surveys, such that longer airborne surveys can then be designed with sufficient length scales to provide the data volume required for meaningful statistics of satellite data validation. The ground-based activities include the kinematic GPS validation efforts at Summit Station, Greenland (Brunt et al., 2017), and airborne activities, such as those associated with NASA's Operation IceBridge (OIB; Koenig et al., 2010), which includes a lidar as part of the instrument suite.
In support of the ground-based component of ICESat-2 data validation, we conducted a 750 km traverse based out of South Pole Station (Fig. 1), referred to as the 88S Traverse (28 December 2017–10 January 2018). Kinematic GPS data collected along this traverse were used to validate airborne data and will ultimately be used for validation of ICESat-2's spaceborne datasets.
Map of the 88S Traverse route, color coded based on elevation. Locations for Figs. 4–9 are also shown. The South Pole Operational Traverse (SPoT) route is indicated in orange. Background is the Landsat image mosaic of Antarctica (LIMA; Bindschadler et al., 2008).
ICESat-2 will have 1387 unique orbits over a 91-day orbital cycle (i.e., all
1387 unique tracks are sampled every 91 days, or four times per year). The
orbit has an inclination of 92
The GPS antenna configuration on a PistenBully. GPS
The design of the 88S Traverse was based on validation studies associated
with ICESat and OIB research. Brunt et al. (2017) used data from the 11 km
ground-based kinematic GPS Summit Station Traverse (Siegfried et al., 2011),
in the center of the Greenland Ice Sheet, to assess the elevation bias and
surface measurement precision of OIB laser altimeters, including the
Airborne Topographic Mapper (ATM) and Land, Vegetation, and Ice Sensor
(LVIS). Using precise point positioning (PPP) post-processing methods for six
ground-based GPS surveys, elevation biases for the associated six ATM airborne
surveys (conducted between 2009 and 2016) ranged from
Here we present results from the first 88S Traverse and show that (1) this part of Antarctica is ideal for this type of airborne and spaceborne data validation and (2) the surface elevation is probably changing minimally, with respect to ice flow, snow accumulation, and surface melt, making it an ideal absolute elevation validation surface, but there is some level of snow redistribution (sastrugi migration) necessitating near-coincident airborne surveys in space and in time to improve estimates of surface measurement precision.
We conducted a 750 km kinematic GPS survey near Amundsen–Scott South Pole
Station, Antarctica, using two tracked vehicles (PistenBullys) provided by the
US Antarctic Program. The 88S Traverse departed from South Pole Station on
28 December 2017 and traveled for 4 days to the 88
The height of each roof-mounted GPS antenna was measured twice along the 88S
Traverse; specifically, the measurement made was the distance between the
antenna base plane and the bottom of the indentation of the tracks of the
PistenBully into the snow (Fig. 2). The average antenna heights for the two
vehicles were 281.3 cm (vehicle A, 280.7 and 281.9 cm) and 282.3 cm
(vehicle B, 282.6 and 281.9 cm). The depths of the tracks of each of the
vehicles into the snow surface were measured 30 times along the traverse.
The average track depths for the two vehicles were 6.2 cm (vehicle A,
Surveys were conducted at
Sample footprint spacing for the UAF lidar (dark blue) and ATM (cyan), and the 88S Traverse ground-based GPS data are in shades of red (GPS A is in light red while GPS B is in dark red). WorldView-2 imagery, copyright 2017, DigitalGlobe, Inc.
The University of Alaska Fairbanks (UAF) lidar is a line-scanner laser
altimeter that has typically been deployed during Alaska-based OIB campaigns
(Johnson et al., 2013). The UAF lidar surveyed the 88S Traverse on two
separate flights (30 November and 3 December 2017) while integrated in a
commercial (Airtec) BT-67 (Basler). The UAF system is a commercial RIEGL
LMS-Q240i scanning laser altimeter transmitting in the 905 nm wavelength
part of the spectrum. The system has a full scanning angle of 60
ATM (Krabill et al., 2002; Martin et al.,
2012) is a laser altimetry system used by many OIB campaigns in both the
Arctic and Antarctic. ATM collected data along the 88S Traverse on 26 October 2014,
while integrated on the NASA DC-8. For that deployment, ATM
(version T4) consisted of a dual instrument configuration, with both
wide-scan and narrow-scan lidar systems integrated simultaneously. The
wide-scan lidar system is more appropriate for ice sheet surveys and has a
full scanning angle of 30
For completeness, we note that ATM also conducted a mission that included
the 88S Traverse on 26 October and 15 November 2016 (also integrated on the
NASA DC-8 and flying at
Following the data processing methods of Brunt et al. (2017), we
post-processed 88S Traverse GPS data using PPP methods. PPP solutions use
precise GPS satellite orbit and clock information to determine the kinematic
GPS antenna position. Position solutions for each vehicle were determined
using NovAtel's Inertial Explorer (v.8.6); processing for each vehicle was
performed on nearly continuous stretches of GPS data, which typically represented
1 full day of driving, or approximately 50 km. Position solutions were
solved to the L1 phase center of each antenna; the elevations are given in
the ITRF08 reference frame and the geographic coordinates are referenced to
the WGS84 ellipsoid. We used a GPS satellite elevation mask, or a cutoff
angle, of 7.5
The elevation of the snow surface, relative to the position solutions of the
L1 phase center of each antenna (Fig. 2), was then determined using data
from the field and the appropriate National Geodetic Survey (NGS) antenna
model phase-center offset. The height of the snow surface (
We obtained the UAF Lidar Scanner L1B Geolocated Surface Elevation Triplets,
version 1 data (Larsen, 2010) through the National Snow and Ice Data Center
(NSIDC) OIB Data Portal (
We also obtained the ATM IceBridge ATM L1B Elevation and Return Strength with Waveforms, version 1 data (Studinger, 2018) through the NSIDC for the 26 October 2014 flight over the 88S Traverse area (17:04 to 19:45 UTC). The data files include geographic coordinates and elevations derived from an integrated on-board GPS (Javad) and inertial system (Applanix POS AV). Differential GPS (DGPS) post-processing methods use a base station installed at the departure airport for this deployment. DGPS was accomplished using a software package developed by the ATM team called GITAR (GPS Inferred Trajectories for Aircraft and Rockets; Martin, 1991). These data are distributed with the elevations given in the ITRF08 reference frame and the geographic coordinates referenced to the WGS84 ellipsoid.
We based our comparison strategy on Brunt et al. (2017). We compared the post-processed snow surface elevations from the 88S Traverse with the airborne surface elevation data, using a nearest-neighbor approach. In this method, we compared the closest lidar data point to every single ground-based GPS data point. We limited our statistical analysis based on a distance criterion, making elevation comparisons only where the lidar footprints and the GPS measurements were within a distance 1 m of one another. We then assessed the difference between the filtered GPS and ATM and UAF lidar surface elevation datasets.
Once the lidar elevation data (Lidar
The
The lidar biases and precisions reported here are determined relative to the GPS data, which we take to represent truth, with zero errors. In actuality, these errors are not zero and are a function of (1) formal GPS errors, which include factors such as ephemeris and clock errors; (2) ionosphere and troposphere errors; (3) multipath errors; and (4) errors due to geophysical effects, such as variable snow surface strength causing variable vehicle sinking or antenna motion due to short-scale surface undulations (sastrugi). We note that given the short distance between the two survey vehicles, our results are somewhat blind to the full magnitude of the error terms that can be correlated on short timescales, such as those associated with the ionosphere and troposphere.
We compared the GPS position solutions of each vehicle to assess consistency
of the ground-based data. After the 88S Traverse GPS data for each vehicle
were post-processed, the data were then filtered based on the 8 cm vertical
sigma; this reduced each GPS dataset by about a third (GPS unit A: 316 948 data
points were reduced to 203 603; GPS unit B: 321 689 data points were
reduced to 209 253). The mean vertical sigma values for the data used in
further analysis were 7.16 and 7.19 cm for ground-based GPS units A and B,
respectively. We then used a nearest-neighbor approach, limited based on a
0.5 m distance criterion, and calculated the mean elevation residual between
the elevation measured by the two vehicles. This residual was
PPP GPS post-processing methods are often used in regions where long-term base-station data are not available for DGPS methods, such as the center of ice sheets. Brunt et al. (2017) showed that PPP position solutions for their traverse outside of Summit Station, in the center of the Greenland Ice Sheet, were comparable to GPS position solutions using differential methods. Therefore, while we are limited with respect to the availability of permanent GPS base stations for post processing, we feel confident that our methods provide consistent and accurate results and are appropriate for this data analysis.
To assess the internal consistency of the UAF lidar, we compared the
processed elevation data from the 30 November 2017 flight to the 3 December 2017
flight, using a nearest-neighbor approach, limited based on a 1 m
distance criteria, and calculated the mean elevation residual. This residual
was
Table 1 lists the results for the nearest-neighbor analysis of the ground-based GPS and lidar elevation comparisons for both ATM and the UAF lidar. Both altimeters had elevation biases of less than 10 cm and surface measurement precisions of less than 15 cm; we note that these values are similar to those in Brunt et al. (2017), which is a similar study in a similar geophysical setting. Figure 4a shows the elevations of ground-based GPS unit A. Panel (b) shows the difference between GPS A elevations and the 30 November UAF lidar elevations, minus the mean difference. Panel (c) is similar to panel (b) but using the 3 December UAF lidar data, and panel (d) compares the GPS data to the 2014 ATM data. Figure 5 is the same as Fig. 4, but the results are relative to ground-based GPS unit B.
Elevation bias and surface measurement precision (cm), relative
to ground-based GPS survey data, for ATM and UAF airborne lidar elevation
data. Results are posted as GPS
Along-track elevation and elevation differences associated with
GPS A.
We examined the spatial correlation of the elevation differences calculated
between the ground-based GPS data and the airborne lidar data (Motyka et
al., 2010; Rolstad et al., 2009). When measurements are made within close
spatial proximity of one another, they are generally similar, and
measurement errors tend to be correlated; over increasing distances,
measurement errors become uncorrelated. Similar to Rolstad et al. (2009),
which is a detailed summary of semivariograms, we created semivariograms of
the elevation differences. This analysis is intended to provide an
assessment of the length scales at which measurement errors become independent
of one another, or uncorrelated. Figures 6 and 7 provide the semivariograms
for GPS units A and B, respectively, relative to the two UAF lidar flights (Figs. 6a, b and 7a, b) and the ATM flight (Figs. 6c, 7c). The
Along-track elevation and elevation differences associated with
GPS B.
Elevation bias and surface measurement precision (cm), between
ATM and the UAF lidar. Results are posted as ATM
The
While the residual between the 88S Traverse vehicles is low, it is not zero.
We attribute the
Overall, the quality of the lidar data used in this survey was quite good.
While a quantitative assessment could be made for the UAF lidar, a similar
assessment of ATM could not be made in this region, as we were limited to
one flight. However, Brunt et al. (2017) analyzed ATM data from five different
airborne campaigns, which included five different versions of the ATM system
(including both narrow and wide scanning data) near Summit Station,
Greenland, on the relatively flat ice sheet interior, similar to this study.
Their results indicated an average ATM elevation bias and surface
measurement precision of
Semivariograms of elevation differences between GPS unit A and
elevations derived from ATM
Semivariograms of elevation differences between GPS unit B and
elevations derived from ATM
We note that there is a slight along-flight signature that is apparent in
the UAF lidar elevation data (Fig. 8). The signature is visible in the
southern side of the swaths of both the 30 November and 3 December 2017
datasets. Specifically, there appears to be a trough along the southern edge
of the swaths that has anomalously lower elevations, relative to the
surrounding edges. The magnitude is variable but based on a nearest-neighbor
assessment of the overlapping region in Fig. 8, where the flight line from
30 November 2017 intersected itself; the mean residual was
Elevation data from the UAF lidar (30 November 2017), where the flight line crossed itself. The along-track artifact in the data is visible in both passes; UAF lidar elevations are anomalously lower within the artifact and manifest as a narrow trough, parallel to the direction of flight. 88S Traverse ground-based GPS data are in shades of red. WorldView-2 imagery, copyright 2017, DigitalGlobe, Inc.
The elevation biases and surface measurement precisions of the two OIB lidars
presented here are comparable to those of the OIB lidars assessed in Brunt et
al. (2017); results based on PPP methods for both studies indicated biases
that are less than
We attribute the poorer surface measurement precision to the time difference
between the airborne ATM campaign (October 2014) and the ground-based GPS
survey (December 2017 to January 2018). Specifically, we hypothesize that
these differences were associated with the transient locations of sastrugi.
To assess this hypothesis, we used the same nearest-neighbor approach,
described in the methods section, to compare the 2014 ATM elevation data to
the 2017 UAF lidar elevation data (Table 2). Ultimately, the difference
between these two lidar datasets revealed a signature that was of a similar
magnitude (meters) and trend (
Ground-based GPS data (in shades of red) plotted on difference in elevation between ATM (26 October 2014) and the UAF lidar (30 November 2017). WorldView-2 imagery, copyright 2017, DigitalGlobe, Inc.
Sastrugi cause noise about the mean surface elevation from a measurement perspective. Sastrugi migration between the 2014 ATM campaign and the 2017/2018 ground-based traverse would not have an impact on the surface elevation bias, as the observed differences would be averaged out and lost in surface measurement noise. The migration of the sastrugi adds components of noise on the mean surface measurement. This effect is evident in the observed larger (poorer) ATM surface measurement precision assessment.
We note that our analysis does not attempt to account for elevation changes
due to the temperature- and accumulation-rate-driven effects of firn
compaction (Li and Zwally, 2015). In this region, we expect variation in
firn compaction rate to be driven by changes in firn temperature, which have
a large seasonal amplitude and a much smaller secular trend. As the firn
warms each austral spring, the surface elevation along our traverse should
decrease. Since the UAF lidar data and ground-based GPS data were collected
within a month, we expect firn compaction to have a negligible effect on our
results. Conversely, the
Overall, these results suggest that the 88S Traverse route is an ideal setting to assess airborne or satellite absolute elevation accuracy (Brunt et al., 2017), as the surface was relatively unchanged between 2014 and 2018 (i.e., no distinguishable change in bias). Further, our results based on the 2014 ATM elevation dataset suggest that airborne data collected along this route are applicable to absolute elevation validation for a few years. However, results based on our comparisons between our GPS measurements and ATM suggest that when a few years have passed between the datasets being evaluated, the surface elevation measurements become hard to reproduce; this manifests itself in a higher surface measurement precision assessment.
Data collected from the 88S Traverse (and data collected on subsequent surveys of the same route) will provide 300 km of in situ data for direct comparison with ICESat-2 elevation data products. The GPS data collection strategies and post-processing methods presented here provide accurate and precise data for such an assessment. Further, the data analysis presented here provides guidance on how to make similar comparisons between ground-based and satellite elevations, given the satellite footprint size and associated rejection criteria. Approximately three to four ICESat-2 reference ground tracks will intersect this region daily to produce many statistical crossover points between the GPS and ICESat-2 datasets. While the crossover points represent only a small segment of along-track ICESat-2 data, the analysis will be based on data from many ICESat-2 reference ground tracks over the course of the entire satellite mission. Thus, the analysis of the derived ICESat-2 bias and surface measurement precision relative to these GPS data will provide an assessment ICESat-2 performance through time, independent of errors associated with single orbits or single points in time. Results of Brunt et al. (2017) and results presented here also provide an assessment of the accuracy and surface measurement precision of three airborne lidars that NASA has routinely deployed over the ice sheets (ATM, LVIS, and the UAF lidar). With a statistical understanding of how these instruments perform on the relatively flat ice sheet interiors, longer flight lines can be constructed over similarly flat ice sheet surfaces to create better statistics associated with comparisons using long length scales of along-track ICESat-2 data. In summary, the strategic location of the ground-based 88S Traverse provides a validation of ICESat-2 that is independent of the errors that are correlated with respect to most satellite timescales, and these ground-based data provide a better understanding of airborne lidars that will survey longer length scales of data, for better satellite error statistics.
Here we present a comparison of in situ GPS elevation data and laser altimetry in preparation for ground-based and airborne validation of ICESat-2. We show that the ground-based methods for GPS data collection and processing along the 88S Traverse provide internally consistent results, with accuracies and precisions appropriate for assessing airborne lidar data and ultimately satellite elevation data. Further, we have shown that airborne lidar data assessed here (ATM and the UAF lidar), relative to the GPS data, show elevation biases that are comparable to results from similar instruments in similar geophysical settings. However, discrepancies between the ATM surface measurement precisions observed here, and those observed in Brunt et al. (2017) under similar ice sheet interior conditions, suggest that the migration of sastrugi can have an adverse effect on assessments of surface measurement precision when significant time (on the order of a few years) has elapsed between surveys. Thus, absolute elevation bias can be determined with datasets from this surface that are a few seasons old, but for the best assessment of precision, comparisons need to be made with relatively coincident (spatial and temporal) datasets.
The ground-based GPS data associated with this study are
available online, as the Supplement related to this article. NASA ATM and the
UAF lidar data are publicly available on the NSIDC Operation IceBridge Data
Portal (
The supplement related to this article is available online at:
KMB conducted the traverse, processed the ground-based GPS data, and led the effort to make the comparisons between these data and the airborne lidar data. TAN conducted the traverse and was instrumental in the effort to make the comparisons with the airborne lidar. CFL conducted the UAF lidar campaign and processed those data. All authors contributed to the discussion of the results and shared the responsibility of writing this paper.
The authors declare that they have no conflict of interest.
We thank the NASA ICESat-2 Project Science Office for funding the field component and data analysis associated with this project. We thank the National Science Foundation, Office of Polar Programs for logistical support of the field component of this project. We thank Operation IceBridge for the data collection of the ATM and UAF lidar datasets. We thank our deep-field mechanic and mountaineer associated with the 88S Traverse (Chad Seay and Forrest McCarthy) for ensuring that we safely completed the full Antarctic ground survey. We thank the many science support staff of the US Antarctic Program that helped make the field component of this project possible. We thank the National Snow and Ice Data Center (NSIDC) for IceBridge data distribution. Finally, we thank our editor (Kenny Matsuoka) and the two anonymous reviewers for constructive comments on earlier drafts of this paper. WorldView-2 imagery was provided by the Polar Geospatial Center at the University of Minnesota, which is supported by grant ANT-1043681 from the National Science Foundation.Edited by: Kenichi Matsuoka Reviewed by: two anonymous referees