Assessment of ICESat-2 ice surface elevations over the CHINARE route, East Antarctica, based on coordinated multi-sensor observations

This paper presents the results of the assessment of ICESat-2 ice surface elevations along the CHINARE (CHINese Antarctic Research Expedition) route in East Antarctica. The validation campaign was designed and implemented in 10 cooperation with the 36 CHINARE Antarctic expedition from December 2019 to February 2020. The assessment of the ICESat-2 ATL03 and ATL06 data was performed based on coordinated multi-sensor observations using two roof-mounted kinematic GNSS receivers, two line arrays of corner cube retroreflectors (CCRs), two sets of retroreflective target sheets (RTSs), and two unmanned aerial vehicles (UAVs) with cameras. This systematic validation of the ICESat-2 data covered a variety of Antarctic ice surface conditions along the 520 km traverse from the coastal Zhongshan Station to the inland Taishan 15 Station. This comprehensive investigation is complementary to the 750 km traverse validation of flat inland Antarctica containing a 300 km latitude traverse of 88°S by the mission team in a previous study. Overall, the validation results show that the elevation of the ATL06 ice surface points is accurate to 1.1 cm with a precision of 9.7 cm along the 520 km CHINARE route. The elevation of the ATL03 photons has an offset of 2.4 cm from a GNSS-surveyed CCR, and is accurate to 2.5 cm with a precision of 2.7 cm as estimated by using RTSs. The validation results demonstrate that the estimated ICESat-2 20 elevations are accurate to 1.1–2.5 cm in this East Antarctic region, which is important for overcoming the uncertainties in the estimation of mass balance in East Antarctica. The developed validation methodology and sensor system can be improved for continuous assessment of ICESat-2 data, especially during the later operation period.


Introduction
The Ice, Cloud, and Land Elevation Satellite-2 (ICESat-2) using the new photon-counting technology was successfully 25 launched by the National Aeronautics and Space Administration (NASA) on September 15, 2018(National Research Council, 2007Markus et al., 2017;Neumann et al., 2019). It is a follow-up to the previous ICESat laser altimetry mission, which is based on the full waveform ranging (Zwally et al., 2002;Schutz et al., 2008). The primary scientific objective of the ICESat-2 mission is to determine ice sheet height changes through continuous measurements at scales from the outlet glaciers to the https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. entire ice sheet (Markus et al., 2017). As the single instrument onboard ICESat-2, the Advanced Topographic Laser Altimeter 30 System (ATLAS) is a photon-counting laser altimeter using 532 nm wavelength laser pulses, which is designed to conduct surface-elevation observations at centimeter-level accuracy and is expected to reduce the uncertainty of the estimated sea level rise contribution from Antarctica (Markus et al., 2017;Neumann et al., 2018Neumann et al., , 2019. More specifically, ICESat-2 aims to improve the accuracy of the elevation change rate to 0.4 cm a -1 (Markus et al., 2017;Smith et al., 2020). The ICESat-2 data collected during its initial and nominal mission periods up to November 2020 (at the time of this submission), including ice 35 surface elevations of returned photons (ATL03) and those of estimated surface points (ATL06) on the Antarctic Ice Sheet (AIS), have been released to the public (Release 003; Neumann et al., 2019;Smith et al., 2019).
Calibration and validation campaigns are important for satellite altimetry missions, especially for the assessment of ICESat-2 performance with new photon-counting technology. Prior to launch, several airborne lidar experiments were carried out to validate this new satellite altimetry technology (Brunt et al., 2017;Magruder and Brunt, 2018). The results showed that surface-40 elevation biases for the tested altimeters, including the Multiple Altimeter Beam Experimental lidar (MABEL), over the flat ice sheet interior are less than 0.12 m with a precision of 0.09 m or better (Brunt et al., 2017). After launch, the mission team conducted a 750 km traverse validation of flat inland Antarctica containing a 300 km latitude traverse of 88°S from December 31, 2018 to January 11, 2019 using kinematic Global Navigation Satellite Systems (GNSSs) to assess the elevation and horizontal accuracy of the ICESat-2 data products (Release 001) on AIS (Brunt et al., 2019b). These results indicated that 45 ATL03 data (Release 001) are accurate to 5 cm of surface elevation with a precision of 13 cm, while ATL06 data (Release 001) are accurate to 3 cm with a precision of 9 cm. The team also used a corner cube retroreflector (CCR) array to collect ICESat-2 signatures and a horizontal geolocation accuracy of 2-5 m was achieved; additionally, the average beam diameter was estimated as ~11 m (Neumann et al., 2020b). Although this ICESat-2 validation campaign covered a long traverse of the flat Antarctic interior along the latitude of 88º S, additional coverage containing the lower-latitude interior and coastal regions 50 in AIS should make the validation of ICESat-2 data complete with an ample and comprehensive understanding of elevation of the whole AIS. More specifically, such a validation in East Antarctica should help confirm the regional ICESat-2 surface elevation accuracy with which it is expected to eliminate or reduce the uncertainty of the mass balance and change rate in East Antarctica (Zwally et al., 2015;Scambos and Shuman, 2016;Richter et al., 2016).
To assess the accuracy of the Antarctic surface elevations provided in the ICESat-2 ATL03 and ATL06 data products and their 55 capability for the estimation of the mass balance of AIS, complementary to the current validation efforts by the mission team, we designed and implemented an independent validation campaign based on a set of coordinated multi-sensor experiments along the 520 km 36 th CHINARE (CHINese Antarctic Research Expedition) route in East Antarctica from December 2019 to February 2020 (Fig. 1). The ground sensors used in this coordinated validation campaign include five GNSS master receivers on the ice surface serving as base stations, two roving GNSS receivers on a snowcat, two sets of CCRs, two sets of 60 retroreflective target sheets (RTSs), and two unmanned aerial vehicles (UAVs) with cameras. Five GNSS base stations with CHC i70 receivers from CHC Navigation Technology LTD (http://www.huace.cn/product/product_show/291) were deployed every ~100 km along the CHINARE route from the coastal Zhongshan Station to the inland Taishan Station to support the two https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

Data
The ICESat-2 Release 003 data used for ground validation in this study are L2A Global Geolocated Photon Data (ATL03) and 80 L3A Land Ice Height data (ATL06; Neumann et al., 2019;Smith et al., 2019) collected along 60 ascending and 78 descending tracks between November 10, 2019 and February 21, 2020 (Table 1). We collected a total of ~1008 km (~516 km inbound and ~492 km outbound) of kinematic GNSS survey data during the 36 th CHINARE inland expedition between Zhongshan Station and Taishan Station (Fig. 1), using two roving GNSS receivers mounted on roof of the snowcat (Fig. 4a). were deployed with ~100 km spacing. Base 1 was located on the roof of a container near the runway of the Russian Progress Station and the other four stations were installed on the ice surface during the traverse. During the outbound traverse (February 8-14, 2020), we only used Base 1 (AC powered) since the other four stations had run out of battery after the inbound traverse trip. The sampling rate was set to 1 Hz to conserve the power and storage space, and the elevation angle mask was set to 7° to reduce the multipath effect. The same settings were also applied for the roving GNSS receivers. The RTK processing resulted 95 in a vertical accuracy of 0.1 ± 5.5 cm for the roving GNSS receivers.
Ten CCRs of 6 cm diameter (Fig. A1a) were placed linearly at a 10 m interval across a nominal ICESat-2 ground track to capture photons from ATLAS (Fig. 2a). To mitigate the error induced by the possible subsidence of the aluminum pole supporting each CCR, a wood base (40 cm × 40 cm × 1 cm) was assembled at the bottom of each pole. The elevations of the nominal centers of the CCR lens near Zhongshan Station were measured by an RTK survey with a vertical accuracy of 0.4 cm. 100 Due to logistic difficulties, the CCRs near Taishan Station were surveyed using the single-point positioning technique with a vertical accuracy of ~1 m. RTSs were designed to assess the reflectivity and elevation accuracy of the ICESat-2 photons from a known artificial surface in the Antarctic environment. We deployed an RTS of 5 m × 150 m at each site, which was oriented perpendicular to the two ICESat-2 tracks (weak and strong), that is ~90 m apart. An RTK GNSS survey of 137 randomly distributed points on the RTS near Zhongshan Station was performed. The coatings of the RTSs were selected from among 28 candidates through a pre-110 expedition experiment. For each coating, an Avafield-3 spectrometer (https://www.avantes.com/products/spectrometers/compactline) was used to measure the reflectivity ( ) of each coated RTS at the ICESat-2 wavelength of 532 nm; we also used a lidar ranger of Riegl BDF-1 (http://www.riegl.com/uploads/tx_pxpriegldownloads/RIEGL_BDF-1_Datasheet_2019-05-31.pdf) with a laser wavelength of 532 nm to measure the corresponding reflectance ( ). We selected the silver-gray coating with = 0.235 which is closest to 115 R = 0.28 which has reportedly the highest estimated probability (EP) of photon detection (Hartzell et al., 2018). We added two other coatings, i.e., yellow ( = 0.532) for high reflectivity and dark green ( = 0.060) for low reflectivity. Thus, the RTS near Zhongshan Station had yellow and silver-gray coatings, while the one near Taishan Station had yellow and dark green coatings.
A UAV equipped with an HD camera (1″ CMOS, 20 megapixel) was flown over an area of 750 m × 500 m near Zhongshan Station. A total of 154 images were collected with an along-track overlapping of 80% and a side overlapping of 75%. At a 120 flight height of ~250 m the ground resolution was 4.6 cm. The CCRs and RTS were inside the mapping area. A high-precision GNSS survey was carried out for photogrammetric processing and DEM reconstruction.

Method
As shown in Fig. 3, this Antarctic validation campaign was designed to achieve three goals: a) assessment of ICESat-2 surface elevations (ATL03 and 06) using a kinematic GNSS survey along the 520 km CHINARE route (69.46-73.86 o S); b) assessment 125 of elevations of the targeted ICESat-2 photons (ATL03) from the CCRs and RTSs; and c) assessment of ICESat-2 elevations (ATL06) using a UAV survey. The major steps of processing are illustrated in Fig. 3. https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

GNSS data processing
Each GNSS base station was deployed before the snowcat started its ~100 km survey along the inbound traverse (Fig. 1). The master receivers collected data from GPS, GLONASS, Galileo, and BDS systems during the batteries' lifetime (~3 days), 135 except Base 1 (all days, powered by AC). The double-frequency data were post processed using the precise point positioning (PPP) technique implemented in the software system of multi-frequency and multi-system instantaneous PPP (MUSIP) developed at Tongji University (Li et al., 2019), with the precise ephemeris and clock products provided by GFZ (ftp://ftp.gfzpotsdam.de/GNSS/products/mgex). Two roving receivers, Antenna 1 and Antenna 2, were mounted on the roof of the snowcat University of Marine Science and Technology (Takasu et al., 2009;Takasu, 2013). The snowcat positions within a segment of up to ~90 km from each base station were estimated.
In addition to the precision of the estimated snowcat positions given by the software system, we used the accuracy computed 145 from elevation differences at crossovers where the GNSS surveyed track intersected itself, as shown in Fig. 2c. These crossovers are the intersections of tracks by the snowcat during instrument installations, observations, and overnight breaks.
Within a neighborhood of the intersection we fit two lines to compute the crossover location and elevation difference (Kohler et al., 2013).

Figure 4. (a) GNSS roving receivers mounted on the roof of a Pisten Bully snowcat and boresight parameters for ice surface elevation estimation from the GNSS observations, and (b) CCR installed in the field with the pole top elevation (hPole top) surveyed by GNSS and the CCR center height (dhCCR) measured using a steel tape.
https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

Derivation of ice surface elevations from GNSS observations
Given the antenna phase-center elevation of a roof-mounted GNSS roving receiver (Fig. 4a), the elevation of the ice 155 surface can be computed as where ℎ 0 is the antenna phase-center height above the mounting plane and ℎ 1 is the vertical distance between the mounting plane and a reference point marked by a yellow dot on the snowcat in Fig. 4a, from which the vertical distance to the ice surface ℎ 2 can be measured. ℎ 0 was given by the manufacturer as 10.1 cm. Prior to the GNSS traverse survey, a boresight 160 calibration was performed to estimate the fixed parameter ℎ 1 . We used a SOKKIA CX-102LN total station to measure ℎ 1 three times and estimated an average of 191.1 cm. During the breaks along the traverse we used a steel tape to measure ℎ 2 . To reduce measurement error due to the uneven ice surface, we measured it three times across a section with an interval of 10 cm. Thus, wherever ℎ 2 is measured, the ice surface elevation can be computed from the roving receiver observation and the boresight parameters ℎ 0 and ℎ 1 . 165

Ice surface elevation comparison along the GNSS traverse
Each ICESat-2 orbit has three pairs of ground tracks (1, 2 and 3 in Fig. 5a) which is 3 km apart (ICESat-2 Technical Specs, 2020). Each track pair corresponds to two laser beams, L (left) for weak beam and R (right) for strong beam. This correspondence may change as the ATLAS system changes its orientation. For example, an ascending ICESat-2 orbit (six tracks) intersected the CHINARE route (red line) at ~420 km from Zhongshan Station (Fig. 5b). In the enlarged area of the 170 intersection ( Fig. 5d), we select the ICESat-2 ice surface points (ATL06) and photons (ATL03) along track 2R (~80 m for ATL06 and ~40 m for ATL03). Specifically, we select the ATL06 points with their quality tag "atl06_quality_summary" equal to 0 (no likely problems) and the ATL03 photons with their confidence tag "signal_conf_ph" equal to 3 (medium) and 4 (high) to ensure high data quality; we also consider the ATL03 photons with "signal_conf_ph" equal to 1 (buffer) and 2 (low) if additional qualified photons are needed (Brunt et al., 2019b). Furthermore, these should be collected within 30 days of the 175 GNSS campaign period to avoid significant changes of the ice surface and other environmental conditions between the two data sets. On the other hand, in order to estimate the ice surface elevations (HIce surface) at the GNSS traverse points using Equation (1), the height from the snowcat to the ice surface (ℎ 2 ) is measured at the locations when the snowcat stops during campaign breaks. At other traverse points between these direct ice surface measurements, we interpolate the values of ℎ 2 using inverse distance weighting (IDW). 180 https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

185
Within an intersection area (Fig. 5d), the elevations of the GNSS traverse points are compared with the ICESat-2 ATL06 and ATL03 data. First, at each ATL06 ice surface point we open a 20 m radius circle inside which there should be at least five GNSS points. We then calculate the difference between the ATL06 point elevation and the median ice surface elevation of the GNSS points. Secondly, for each ATL03 photon we find its nearest GNSS point within a 4 m wide search window to form a pair and calculate their elevation difference. Finally, if the number of pairs is sufficient ( 30), we use the median among them 190 to be the representative elevation difference between the ATL03 and GNSS data at this intersection.

Validation of ICESat-2 photons using CCR and RTS elevations
Design of a photon-capturing CCR array requires a number of considerations. The spacecraft orbiting and sensor pointing capability has been improved after the initial mission (Brunt et al., 2019a). Our preliminary analysis of Releases 001 and 002 data indicated that the offsets between the actual and reference ground tracks in the study area were reduced from up to ~3000 195 m during the initial mission to 1-6 m before our expedition. Given a footprint of ~13 m diameter every ∼0.7 m along the track https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. (Markus et al., 2017;ICESat-2 Technical Specs, 2020), ICESat-2 measured ~4 photons for weak beams and ~10 photons for strong beams per pulse in our study area; with a CCR field of view (FOV) of ±35° (see Appendix A), a single line CCR array across a track ensures along-track photon capture. In the cross-track direction we installed 10 CCRs spaced every ~10 m across one weak (or strong) beam track (Fig. 5c); this ensures that at least one CCR is placed inside a footprint of ~13 m along one 200 track. Each CCR has a unique elevation for distinguishing the target CCR from other CCRs (Fig. 4b). The elevation of the CCR pole top (hPole top) is surveyed by using the precision RTK GNSS positioning method; then, the height from the CCR center, marked by the manufacturer as the theoretical photon bounce point, to the CCR pole top ( ℎ ) is measured using a steel tape. Thus, the elevation of the CCR is estimated as hCCR = hPole top + dhCCR. The field deployment and survey are carried out a few hours before the expected satellite pass to avoid severe sinkage of the CCR. 205 Once the photons are reflected from a CCR and received by ATLAS, a streak of photons along a track with the elevation close to the GNSS-surveyed elevation can be detected in the ICESat-2 ATL03 data (Figs. 6a). We use the photons in the central section of the streak (one footprint long; red rectangles in Fig. 6a and 6c) where the photons are generally of high quality and have the confidence tag "signal_conf_ph" equal to 3 or 4 used to calculate an average elevation in each pulse to reduce noises and potential atmospheric effect (blue dots in Fig. 6b). The average elevations are then further fitted to a symmetric curve, e.g., 210 a Gaussian curve. The central point (or peak) of the curve is treated as the representative CCR photon position. An offset is then calculated between the positions estimated from the ATL03 photons and the GNSS survey.

Validation of ICESat-2 ice surface elevations using UAV-DEM
To georeference the UAV images, GNSS observations are taken using the D-RTK GNSS mobile station 220 (https://www.dji.com/cn/d-rtk-2?site=brandsite&from=nav) near Zhongshan Station during the image acquisition. This mobile station and other three ground control points (GCPs) are used in GNSS RTK processing to achieve an accuracy of better than 1.0 cm. We generate a UAV-DEM and an orthophoto from the UAV data by using the structure-from-motion multi-view stereo ( At Taishan Station, however, owing to the lack of GCPs and the D-RTK GNSS mobile station, the accuracy of the UAV-230 derived products is estimated to be several meters and cannot be used for ICESat-2 ice surface elevation validation.

Kinematic GNSS -ICESat-2 ice surface elevation validation
The elevation precision values given by the GNSS PPP processing software system for the five base stations, from Base 1 to Base 5, are at the sub-centimeter (0.5 cm, 0.9 cm, 0.8 cm, 0.7 cm, and 0.9 cm) level. For the positions of the roof-mounted 235 GNSS receivers along the inbound and outbound traverses, a threshold of the precision values given by the kinematic GNSS RTK positioning software system was used to filter out traverse points with large errors originating from rough terrain features and other noises, i.e., 3σ for the undulated topography near the coast (0-67 km) and 2σ for the relatively flat inland topography (67-520 km). As a result, a total of 625,358 GNSS traverse points were obtained, which have an average internal elevation precision of 1.6 ± 0.6 cm given by the software system. Finally, the elevation accuracy of the GNSS traverse was assessed as 240 0.1 ± 5.5 cm by using 27 crossovers of the traverse itself (Fig. 2c) where the GNSS surveyed elevations from two intersecting traverse segments were compared.
There were 20 locations along the traverse (Fig. 1) where ℎ 2 was measured (Fig. 4a). Along with the other two bore sight parameters, ℎ 0 and ℎ 1 , these direct measurements were used to derive the ice surface elevations from the roofmounted kinematic GNSS observations . The average of the measured ℎ 2 is 93.0 cm with a standard deviation of 3.2 245 cm. This variation is mainly attributed to the microtopography and firn density changes at different locations along the 520 https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. km traverse from the coast to the highland interior. Out of the 134 intersections between the GNSS traverse and ICESat-2 tracks, we selected 25 intersections that are within 2 km of the ℎ 2 direct measurement locations to enhance the comparability between the elevations observed by the ICESat-2 satellite and the kinematic GNSS receivers along the traverse. The average distance between the intersections and measurements is ~816 m. We validated the elevations of the ICESat-2 ATL06 ice 250 surface points and ATL03 photons using the GNSS-surveyed elevations that are summarized according to six ICESat-2 tracks separately ( Table 2).  Compared to the kinematic GNSS elevation observations, the ATL06 ice surface points have median elevation differences (bias) for the six ICESat-2 tracks ranging from -5.0 cm to 8.1 cm and precision values (1σ) ranging from 6.8 cm to 10.4 cm, resulting in an overall bias of 1.1 cm and precision of 9.7 cm. Similarly, the ATL03 photons have an overall bias of 3.7 cm 260 and precision of 9.1 cm. Note that due to the 2 km limit from the direct ice surface measurements h2 for the selection of intersections, there are no comparison results for the ATL03 tracks of GT2L and GT2R. No significant elevation differences were found between the tracks of the weak and strong beams. Thus, the processed ATL06 application product (L3A Land Ice Height data) has a smaller bias of 1.1 cm than that of 3.7 cm for the unprocessed ATL03 product (L2A Global Geolocated Photon Data), although their precision values are almost the same (9.7 cm and 9.1 cm, respectively). 265 We further extended our assessment to all intersections of the ICESat-2 tracks and the GNSS traverse without the above 2 km selection limit. At each intersection, the ℎ 2 value was calculated between two measurement locations using the IDW interpolation method. As shown in Table B1, the ATL06 data also present a bias of 0.5 cm, smaller than 3.7 cm for the ATL03 data, which are comparable to these in Table 2. However, the overall precision values of 15.7 cm (ATL06) and 14.2 cm (ATL03) in Table B1 are significantly larger than 9.7 cm (ATL06) and 9.1 cm (ATL03) in Table 1, mainly due to the uncertainty caused 270 by the interpolated ℎ 2 value. https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

Validation of ICESat-2 photon elevations using CCRs and RTSs
ICESat-2 passed across the line array of 10 CCRs near Zhongshan Station along orbit 0416 at 23:49 UTC on January 22, 2020 ( Fig. 5c). Around 6 h before the ICESat-2 pass, the CCRs were deployed and GNSS was surveyed with an accuracy of 0.3 cm (horizontal) and 0.4 cm (vertical), respectively. A GNSS sinkage survey was performed over 2 days and estimated a negligible 275 sinkage of less than 1 cm. The ATL03 data showed a streak of 137 photons in 40 pulses, spanning ~35 m along the weak beam track (1L; green dots in Fig. 6a). We selected 68 photons of 18 pulses located in the central part of the streak (~13 m, inside the red rectangle). Both the track location and elevations of the photons matched those of CCR #5 (black square), suggesting that it was illuminated during the ICESat-2 pass. The elevations of the photons within each pulse were averaged and used to fit a Gaussian function curve with an 2 of 0.7043 and a fitting RMSE of 5.1 cm (Fig. 6b), whose peak point was used as the  Fig. 6c). To estimate a CCR elevation hCCR (Fig. 4b) that is more accurate than the meter-level GNSS result, we first used 10 ATL06 ice surface points (red dots in Fig. 6c) to fit a terrain surface plane with an 2 of 0.9960.
Using the approximate GNSS-surveyed CCR location and the fitted ice surface, the improved CCR elevation hCCR (black square) was calculated from the measured pole height ℎ . After that, the photons were averaged within each of pulse (blue dots in Fig. 6d) to fit another Gaussian function with an 2 of 0.8647 and an RMSE of 3.1 cm. The peak position of the 290 Gaussian function was used as the representative photon of the CCR that has an offset of 60.0 cm in the horizontal direction and 34.9 cm in the vertical direction from the estimated CCR location.
ICESat-2 passed across the RTS and the CCR line array along the same orbit at each site near Zhongshan Station and Taishan Station, respectively. The combinations of the strong and weak beams and different coatings are listed in the first column of Table 3. There are seven check points on the yellow sheet near Zhongshan Station, which were surveyed using the GNSS RTK 295 method with an accuracy of 0.4 cm. Pairing of the check points with the corresponding ATL03 photons was achieved with a maximum in-between distance of 1.22 m. The comparison between the two datasets resulted in an elevation uncertainty of 2.5  2.7 cm.
The three deployed RTS coatings were tested prior to the Antarctic expedition and their reflectivity was measured as 0.532 for yellow, 0.235 for silver-gray, and 0.060 for dark green coatings, respectively (Table 3). The 5 m wide, high-reflectivity yellow 300 coating reflected photons from the strong beam. There were 183 and 132 ATL03 photons reflected within the footprint containing the RTS near Zhongshan Station and Taishan Station, respectively. Furthermore, the full saturation fraction (FSF) tag "full_sat_fract", ranging from 0 to 1, represents the ratio of fully saturated pulses over a 20 m along-track segment (Neumann et al., 2020a). As shown in Table 3, FSF of the two yellow coatings at the two sites are 0. For comparison, the https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License. similar number averages of received photons (149 and 144) and FSF values of 0 were estimated over the 300 m bare ice surface 305 segments next to the RTS for the same strong beam tracks. Therefore, the footprints of the high-reflectivity yellow coating RTSs showed similar photon reflectivity characteristics to the bare ice surface. On the other hand, the two lower-reflectivity coatings, i.e., silver-gray and dark green coatings, reflected photons from the weak beam, but showed higher FSF values (0.321 and 0.429). We suggest that this abnormal phenomenon be attributed to contamination by the photons reflected from the illuminated CCRs, which were only ~9 m away from the RTSs (Fig. 5c). 310 In summary, one CCR of the precisely GNSS-surveyed line CCR array near Zhongshan Station was illuminated. The estimated representative CCR photon has an offset of 2.4 cm in elevation and 180.0 cm in horizontal direction from the GNSS-surveyed 315 CCR position. Seven precisely GNSS-surveyed check points that are evenly distributed on a highly reflective RTS near Zhongshan Station were used to validate the elevations of their corresponding ATL03 photons (strong beam) and resulted in a bias and precision of 2.5 ± 2.7 cm.

Validation of ICESat-2 ice surface elevations using UAV-DEM
The 750 m × 500 m UAV-DEM mapping area near Zhongshan Station contains the RTS and CCR line array (Fig. 5c). ICESat-320 2 orbit 0416 passed through the area 8 h before the UAV image acquisition. First, 167 check points on the RTS were GNSS surveyed at an elevation accuracy of 0.4 cm and used for the UAV-DEM accuracy assessment. The generated UAV-DEM elevations were compared against those of the check points and achieved an elevation uncertainty of 0.2  6.3 cm. Second, the 1L (weak beam) and 1R (strong beam) tracks have 48 ATL06 ice surface points in the DEM area, among which three were affected by the photons from the CCR and excluded from the validation. The elevation differences between the ATL06 ice 325 surface points and the UAV-DEM were computed and resulted in an estimated ICESat-2 ice surface elevation uncertainty of 1.1  4.9 cm. https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

Discussions
The validation of ICESat-2 elevations at crossovers between the ground tracks and the CHINARE route using a kinematic GNSS positioning technique and direct ice surface measurements produced an elevation bias and precision of 1.1 ± 9.7 cm for 330 the ATL06 ice surface points, compared to 3.7 ± 9.1 cm for the ATL03 photons (Table 2). This indicated that the method used for estimating the ice surface elevations every 20 m along a track in ATL06 data (Smith et al., 2019) may be able to significantly reduce the bias, i.e., a reduction of bias from 3.7 cm to 1.1 cm as demonstrated in this study. The precision remained at an approximately same level (≤ 9.7 cm) and may be mainly caused by the variation of ℎ 2 in Equation (2), the height from the snowcat to ice surface. Although the average distance between the direct ℎ 2 measurements and the crossovers 335 is limited to ~816 m, the factors that influence the pairing of ℎ 2 with the kinematic GNSS positions may include the local microtopography and uneven firn properties from the coast to the interior of Antarctica along the GNSS traverse. This trend is further evidenced in the results by extending the assessment to all crossovers with both directly measured and interpolated ℎ 2 values (Table B1) where the bias can be similarly reduced from 3.7 cm for the ATL03 photons to 0.5 cm for the ATL06 ice surface points. However, the precision decreased (that is, the values increased) from less than 9.7 cm for the experiment using 340 only the directly measured ℎ 2 (Table 2) to less than 15.7 cm for the experiment with interpolated ℎ 2 . An improvement of the validation system using continuous measurements of ℎ 2 along the GNSS traverse is planned for experiments in future expeditions.
The accuracy of the ATL03 data was better validated by using targeted photons returned from CCRs and RTSs. One precisely GNSS-surveyed CCR was illuminated by a weak beam and showed an offset of 2.4 cm between the GNSS elevation and that 345 of the representative CCR photon. Furthermore, seven precisely GNSS-surveyed check points on an RTS were compared with the corresponding ATL03 photons and presented a bias and precision of 2.5 ± 2.7 cm.
The fact that the photons returned from the CCRs are saturated is indicated by the full saturation fraction (FSF) values of 0.429 and 1.000 for the segments containing the illuminated CCRs near Zhongshan Station and Taishan Station, respectively (Table   4). They are drastically higher than those of their ice surface areas nearby (0.009 and 0.043). It is also noted that the CCR FSF 350 at Zhongshan Station is more than 50% lower than that at Taishan Station (max. FSF). A similar trend exists between FSFs of the ice surface areas at the two sites despite the variation in the ratio. These consistent FSF differences between the two sites may be explained by their corresponding "total neutral atmospheric delay" (TNAD) in the ATL03 data (/gtx/geolocation/neutat_delay_total), which represents the totally neutral atmospheric delay correction (Neumann et al., 2020a); TNAD is dependent on the state of the atmosphere, which itself is dependent on the total pressure, partial pressure of 355 water vapor, and air temperature. The TNAD of 2.353 m at Zhongshan Station over that of 1.726 at Taishan Station suggested that atmospheric conditions may have had a severe impact on the CCR photons at Zhongshan Station, including the increased level of refraction and thus, prolonged the time delay of the returned photons. Therefore, we see less-saturated photons (lower FSF) near Zhongshan Station. And the elevations of the CCR photons at Zhongshan Station are also more dispersed (Fig. 6a) compared to the regularly distributed pattern at Taishan Station (Fig. 6c).

Conclusions
This paper presents the results of the assessment of ICESat-2 ice surface elevations along the CHINARE route in East 365 Antarctica. The validation campaign was designed and implemented in cooperation with the 36 th CHINARE Antarctic expedition. An assessment of the ICESat-2 ATL03 and ATL06 data was performed along the 520 km traverse using a kinematic GNSS positioning technique. Near Zhongshan Station and Taishan Station, additional coordinated multi-sensor observations of a CCR line array, an RTS, and a UAV were acquired for each site. Overall, this systematic validation of the ICESat-2 data covered a variety of the eastern AIS conditions from the coast to inland Antarctica along the 520 km traverse and is 370 complementary to the 750 km traverse validation of flat inland Antarctica along the latitude of 88 o S (Brunt et al., 2020b).
The following conclusions are drawn from this research.
1) The comparison of the ICESat-2 Release 003 data with the high-precision GNSS survey and direct snowcat-to-ice surface height measurements along the 520 km CHINARE route showed that the elevations of the ATL06 ice surface points are 375 accurate to 1.1 cm with 9.7 cm precision (1.1 ± 9.7 cm), and the elevations of the ATL03 photons are accurate to 3.7 cm with 9.1 cm precision (3.7 ± 9.1 cm). This is comparable to the similar result of the ICESat-2 Release 001 data validation, 3 ± 9 cm for the ATL06 data and 5 ± 13 cm for the ATL03 data along a traverse of the 88°S parallel (Brunt et al., 2019b).
2) The validation of the ATL06 ice surface points along two ICESat-2 tracks with a high-precision GNSS-controlled UAV-DEM (750 m × 500 m) near Zhongshan Station indicated that the ATL06 elevations are accurate to 1.1 cm with 4.9 cm 380 precision. Although the DEM coverage is relatively small, these validation results show the potentially achievable accuracy using the ICESat-2 ATL06 data in many applications in the AIS environment.
3) The validation of the ICESat-2 ATL03 data using a high-precision GNSS-surveyed CCR showed an accuracy of 2.4 cm.
An additional experiment using seven high-precision GNSS-surveyed check points on the RTS near Zhongshan Station indicated that the ATL03 data are accurate to 2.5 cm with 2.7 cm precision. This suggests that an assessment using targeted 385 and controlled (or partially) photons can result in high elevation accuracy. https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.

Appendix A. CCR lens
The mechanism of photons entering into and reflecting from a CCR is realized according to the optical total reflection principle (Born et al., 1999 originally designed for laser distance ranging (Fig. A1a). The lens system consists of three adjacent and mutually orthogonal plane-reflecting surfaces, which have an angle of 54.73° from the horizontal plane (Fig. A1b). An incident beam ① from the satellite comes out of the CCR as the main outgoing beam ② that is parallel to ① but in the opposite direction, allowing it to be received by the satellite. The system guarantees that any photon from the satellite entering the field of view (FOV) of the CCR, ±35° from the zenith, will be returned to the satellite. Ideally, the elevations of all photons returned from the CCR 490 would be placed at one location on the ice surface. However, the ATLAS places all photons collected inside a footprint (~13 m) at one location, which includes those from the CCR and the surrounding area. Therefore, bounded between the first footprint entering the FOV and the last footprint exiting the FOV, the elevations of photons returned from the CCR are recorded as a streak of photons in ATL03 data (green dots in Figs. 6a and 6c). https://doi.org/10.5194/tc-2020-330 Preprint. Discussion started: 25 November 2020 c Author(s) 2020. CC BY 4.0 License.