Arctic Sea Ice Topography Information From RADARSAT Constellation Mission (RCM) Synthetic Aperture Radar (SAR) Backscatter

Sea ice topography information can be obtained from altimetry but these data are spatially and temporally limited compared to recent synthetic aperture radar (SAR) missions such as the RADARSAT Constellation Mission (RCM). We analyze the relationship between sea ice roughness and height obtained from three Ice, Cloud, and Land Elevation Satellite (ICESat)‐2 tracks on two dates in March 2022, with RCM backscatter from 17 images in the McClintock Channel, Canadian Arctic. We analyze how this relationship varies with ice type, polarization, and incidence angle. We find particularly notable relationships between sea ice roughness and horizontal‐transmit/vertical‐receive backscatter for first‐year ice, and sea ice height and backscatter for multi‐year ice. We develop a preliminary model for winter sea ice roughness retrieval using RCM. In comparison with independent ICESat‐2 data in our study region, we find the model performs effectively at estimating a roughness distribution and key roughness statistics, and characterizes spatial variations in roughness at a sub‐kilometer scale.


Introduction
Warming Arctic air temperatures have caused Arctic sea ice thickness, volume, and age to decline in recent decades (Meier et al., 2023).Between 2003 and 2021 Arctic sea ice volume declined by one third (Kacimi & Kwok, 2022).This sea ice is key to the global climate system, through the ice-albedo effect (e.g., Curry et al., 1995) and its influence on global atmospheric and oceanic circulation (e.g., Francis et al., 2017).
In order to better understand and model variations in the state of Arctic sea ice, accurate, extensive and highresolution sea ice topography information is needed.Geometric sea ice roughness information is necessary to parametrize momentum transfer between the atmosphere and ocean (Martin et al., 2016;Tsamados et al., 2014), reduce uncertainty in satellite altimetry retrievals (Landy et al., 2020), and to predict the evolution of summer melt pond coverage (Eicken et al., 2004;Perovich & Polashenski, 2012).Sea ice roughness information is also valuable to Arctic communities for planning safe and efficient sea ice travel (Segal et al., 2020).Additionally, sea ice height information is of great importance as it is typically a component of sea ice freeboard and thickness calculations, and it is valuable in assessing variations in mass balance.
However, sea ice topography remains poorly constrained and existing methods of detection are spatially and/or temporally limited in resolution.The Ice, Cloud, and Land Elevation Satellite (ICESat) laser altimetry mission, and its successor ICESat-2 (2019-present) have provided estimates of sea ice height, freeboard, and roughness information at a high spatial resolution and a 90-day return time.These measurements have enabled better detection and understanding of complex sea ice surface topography (e.g., Duncan & Farrell, 2022;Farrell et al., 2020).However, these estimates are restricted to the narrow track of the ICESat-2 Advanced Topographic Lidar Altimeter System (ATLAS).Radar altimetry missions, such as CRYOSAT-2 recorded topography over a large spatial extent, but at a relatively low spatial resolution.C-band Synthetic aperture radar (SAR) missions, in contrast, collect data at a wide swath of tens to hundreds of kilometers, while maintaining a high spatial resolution (∼≤100 m).Other techniques and instruments, such as Interferometric SAR and the Multi-Angle Imaging SpectroRadiometer, have also been applied for lower resolution, seasonal or local studies (e.g., Dierking et al., 2017;Johnson et al., 2022).
Previous studies have found relationships between backscatter and roughness for L-band (e.g., Toyota et al., 2021), X-band (e.g., Guo et al., 2023) and C-band SAR instruments.Previous analyses of recent C-band satellite missions, such as RADARSAT-2 (Cafarella et al., 2019) and Sentinel-1 (Segal et al., 2020), however, have been limited to comparisons with validation data from individual airborne campaigns, and have focused only on the horizontal transmit-receive (HH) channel, due to the insufficiently-low noise floor in the horizontaltransmit/vertical-receive (HV) channel of those instruments.The recent and upcoming launches of several SAR missions with open data policies, such as C-band Sentinel-1 and RADARSAT Constellation Mission (RCM) and L-band NASA-ISRO SAR (NISAR), along with the collection of regular, extensive altimetry data from ICESat-2, increases the opportunity and motivation to better understand the relationship between SAR backscatter and sea ice topography, and develop SAR-based topography-retrieval models.
In this study, we first examine the relationship of C-band RCM backscatter and measurements of sea ice surface roughness with height from the University of Maryland-Ridge Detection Algorithm (UMD-RDA), derived from ICESat-2 data.We do this at a case study site in the Canadian Arctic Archipelago (CAA), focusing on the latewinter of (2021/2022) and assess how the relationship varies depending on ice type, polarization channel, and incidence angle.We then leverage these relationships with machine learning to develop a preliminary model for retrieving sea ice roughness from RCM backscatter, which we validate and test on data from the CAA, as well as sites in the Beaufort Sea and Baffin Bay.

Study Site
Data were collected in late winter 2021/2022 from the McClintock Channel, located in the southeast of the Canadian Arctic Archipelago, at ∼68-75°N and ∼99-103°W (Figure 1a).The channel hosts a range of ice conditions that have undergone different levels of deformation, including mixture of smooth and deformed firstyear ice (FYI), and multi-year ice (MYI, ice that has endured at least one summer's melt).Following ice break-up in the summer of 2021, some ice remained in the channel, becoming MYI and drifting slightly southward before freezing into place, primarily in the north of the study area.By mid-January 2022 the whole channel was fast with negligible drift taking place (Movie S1).The ice in this area remains fast through the winter until breaking up in late summer or early fall (Haas & Howell, 2015;Maxwell, 1981;Melling, 2002).Winter snow accumulation has been observed to be relatively low in the region, for example, Howell et al. (2016) report mean snow depths at nearby Cambridge Bay as 8.4 cm for October-May (1960-2014).The fast nature of the ice means that, in contrast to drifting ice sites, we are not limited to rare coincident ICESat-2/RCM overpasses.We assume that without drift, minimal deformation or change in roughness takes place, and we analyze a period of low and relatively stable temperature, when the effect of variations in dielectric constant on backscatter is minimal.This enables us to obtain numerous RCM images at a range of incidence angles for comparison with UMD-RDA ICESat-2 data.et al. (2020).Each ICESat-2 ATLAS track has six beams arranged in three pairs of strong and weak beams spaced 3.3 km apart and has a nominal along-track sampling interval of 0.7 m with a footprint of ∼11-12 m in diameter (Kwok et al., 2019;Magruder et al., 2020).The instrument uses a 532 nm (green) laser that measures returns from the snow surface.Comparisons with airborne Lidar (Duncan & Farrell, 2022) and airborne-laser scanner (Ricker et al., 2023) data show that data processed using the UMD-RDA detects ridges, sails and other obstacles at higher precision than ICESat-2's standard ATL07 height product, and is therefore more appropriate for determining roughness.This analysis also showed that ATL07 tends to overestimate the roughness of smooth ice and underestimate the roughness of rougher ice.The UMD-RDA algorithm performs better in this regard and is able to resolve features as narrow as 5.6 m wide and achieve a vertical height precision of 0.01 m.UMD-RDA surface finding is also applied on a per-shot basis to retain height at a consistent along-track resolution of ∼0.7 m, in contrast to ATL07 which provides data in variable segment lengths (typically 10-200 m).Atmospheric, tide, and mean sea surface (MSS) geophysical height corrections are applied to height information to obtain corrected heights relative to the MSS.Here we use the Technical University of Denmark 2018 Mean Sea Surface (DTU18 MSS) model (Andersen et al., 2018a).Surface roughness is then calculated as the standard deviation of height in 40 m along-track segments, to compare with the 40 m pixel size of the RCM data (see Supporting Information S1).

RCM Backscatter
SAR backscatter was obtained from RCM, a three-satellite constellation of C-band SAR instruments launched by the Canadian Space Agency in 2019, providing more than daily coverage of the Canadian Arctic (Thompson, 2015).For this study we obtained images in the Low Noise (SCLN) mode, because of (a) the lack of scalloping in the azimuth direction in the HV channel compared to the ScanSAR mode and (b) its low noise floor compared to other modes and instruments.RCM's SCLN has a pixel size of 40 m and nominal noise floor of 28 dB, but in reality we find it is typically 34 dB or less, compared to a nominal noise floor of 26 dB in RCM's ScanSAR mode and 22 dB for Sentinel-1's Extra Wide and Interferometric Wide swath modes.Both HH and HV bands for each image were radiometrically calibrated, speckle-filtered and map-projected and converted to decibels for analysis (see Supporting Information S1).We focus on the late-winter period because we expect the effect of meltwater on SAR backscatter after melt onset to alter the relationship between sea ice topography and backscatter (e.g., Scharien et al., 2014).

Comparing Roughness, Height, and Backscatter
In order to compare sea ice roughness and height from the UMD-RDA with RCM backscatter, RCM SCLN images over the McClintock study site were collected over eight dates through the period 27 February to 27 March, alongside three sets of ICEsat-2/UMD-RDA tracks, two dated 11 March and one 23 March 2022 (Figure 1a, Table S1 in Supporting Information S1).Due to the size of the study area, up to three images were collected in some days.Although on some days the entire study area is covered by RCM in a single day, images were collected on multiple days on different paths to obtain backscatter from the same locations at a range of incidence angles (25°-55°).
Pixels were identified as FYI and MYI manually, then HV backscatter, HH backscatter, incidence angle, sea ice height, sea ice roughness, and ice type were recorded for 40 × 40 m segments along each of the tracks for all images (see Supporting Information S1; Figure S1 in Supporting Information S1).In total there is 1,820 km of track covering FYI segments and 1,028 km covering MYI segments across the nine beams.This amounts to 305,766 FYI segments and 170,493 MYI segments with a roughness, height and backscatter value across all the images.

Relationship Between RCM Backscatter and Sea Ice Roughness
Surface roughness, as obtained from the UMD-RDA data, has a mean of 0.07 m and standard deviation of 0.08 m for FYI, and a mean of 0.11 m and standard deviation of 0.09 m for MYI.For the RCM HH channel in winter, overall backscatter has a Spearman's rank correlation of 0.31 with surface roughness for FYI.When broken down into near (<30°), mid (30°-45°), and far (>45°) incidence angle ranges, the correlations are 0.45, 0.35, and 0.41, respectively (Figure 1b).These values are substantially lower for MYI, with correlations of 0.24 (near), 0.25 (mid), and 0.27 (far).
Backscatter from the HV band has a stronger relationship with FYI surface roughness than HH, exhibiting an overall correlation of 0.42.The relationship is again strongest (0.50) at near-range incidence angles, with a correlation of 0.42 at mid-and 0.44 at far-ranges.When comparing the mean HV backscatter for each 0.01 m bin of roughness, there is a clear, largely consistent, non-linear relationship between the two variables, although there is substantial noise (Figure 1d; Figure S1 in Supporting Information S1).The relationship is approximately linear from around 0.01 m of roughness to 0.06 m.As roughness further increases, the rate of the increase in backscatter is lower, with the mean value not surpassing 26 dB.Beyond around a roughness of 0.20 m, there is a moderate, less-consistent increase in backscatter with roughness.However, 94% of FYI roughness values lie in the range of 0-0.20 m in this data set.It should be noted that both the correlation values (Figure 1b) and considerable spread in values shown in Figure 1d indicate that, although significant and moderately strong, the relationship between backscatter and roughness is modulated by other factors.For example, the influence of variations in temperature, grain size, and brine in the snowpack on backscatter, and discrepancies between roughness at the snow surface measured by ICESat-2, and roughness at the ice surface.As with HH, the relationship between roughness and backscatter is again substantially lower for MYI than FYI for the HV channel (near-: 0.27, mid-: 0.26 and farrange: 0.28).
The stronger relationship between roughness and backscatter exhibited by the HV channel highlights the value of RCM SCLN's low noise floor.While studies using RADARSAT-2 and Sentinel-1 did not use the HV channel to investigate roughness due to backscatter in this channel being below the noise floor in low-backscatter FYI (Cafarella et al., 2019;Segal et al., 2020), our study suggests that using the RCM HV channel is necessary to utilize the full potential of C-band SAR to detect roughness.
The stronger relationship between backscatter and roughness for FYI than MYI has been observed for other Cband instruments RADARSAT-2 (Cafarella et al., 2019) and Sentinel-1 (Segal et al., 2020).The differing relationship between ice types can be explained by the contrasting geophysical characteristics of the two ice types.The dielectric properties of the brine-wetted snow layer at the surface of FYI significantly influence microwave emission (Drinkwater & Crocker, 1988;Howell et al., 2006) and restricts C-band SAR penetration into the ice, causing it to scatter at the ice surface.Therefore, the backscatter returned to the sensor from FYI is largely sensitive to the characteristics of the ice surface, including its roughness (Figure 2a).In MYI, however, C-band SAR penetrates the freshened upper ice, and air bubbles within the ice cause substantial volume scattering (Geldsetzer & Yackel, 2009;Hallikainen & Winebrenner, 1992).This volume scattering causes the overall backscatter received by the sensor to be typically higher for MYI, and dampens the relationship between backscatter and surface roughness.
The stronger relationship between backscatter and FYI roughness at near-range incidence angles is in agreement with previous studies of other C-band SAR instruments (Cafarella et al., 2019;Segal et al., 2020).Incidence angle is known to play a significant role in the backscatter returned from a sea ice surface (Fors et al., 2016;Mäkynen et al., 2002).In FYI, although some deformation features, such as ridges, have brighter returns at higher angles (Melling, 1998), backscatter typically decreases with increased incidence angle (Onstott, 1992), as does its relationship with roughness (Cafarella et al., 2019;Mäkynen et al., 2002).
The influence of incidence angle on backscatter also explains part of the relationship between backscatter and roughness.Rough features on the surface of the ice, such as undulations, create a higher local incidence angle, increasing backscatter for both HV and HH channels (Figure 2a).Another reason for the relationship, for both channels, is the interaction of the SAR signal with roughness variations smaller than the instrument's 5.5 cm wavelength (Paterson et al., 1991;Richards, 2009).Additionally, we propose that the reason for a stronger relationship with roughness for the HV than HH channel is because as the signal "double-bounces" off surface roughness features, depolarization occurs, meaning that rough ice causes a substantial amount of horizontallypolarized signal to be returned with a vertical polarization.
The relationships and mechanisms analyzed here suggest that RCM backscatter has potential to be instructive for estimating sea ice roughness, particularly in areas dominated by FYI, where the relationship is stronger.The analysis suggests that the HV channel should be prioritized over the HH channel, but both channels may be useful.Our analysis also suggests that using backscatter to estimate roughness will be most effective at near-range incidence angles, but all incidence angles may be used.

Relationship Between RCM Backscatter and Sea Ice Height
Ice surface height, as obtained from the UMD-RDA data, has a mean of 0.23 m and standard deviation of 0.19 m relative to mean sea level for FYI, and a mean of 0.03 and standard deviation of 0.23 m for MYI.
For the RCM HH channel in winter, backscatter has an overall correlation of 0.41 with surface height for MYI.For near, mid, and far incidence angle ranges, the correlations are 0.49, 0.48, and 0.52, respectively.Backscatter from the HV band in winter has a stronger relationship with MYI surface height than HH across all incidence angles (0.49), but similar or marginally lower relationships when broken down by incidence angle range (Figure 1c).For HV, the correlation is 0.49 at each incidence angle range.When comparing the mean HV backscatter for each 0.01 m bin of height (Figure 1e; Figure S2 in Supporting Information S1), there is a clear, non-linear relationship for the range in which the majority of values lie.At the very lowest height values, between 0.5 and 0.35 m, backscatter in fact decreases with increasing height.However, as height increases from 0.35 m to around 0.05 m, backscatter consistently increases until the relationship plateaus.As with roughness, there is considerable noise and spread in values, which along with the moderate strength of the correlation values demonstrates that there are other factors contributing to the backscatter.For both polarization channels, the relationship between height and backscatter is found to be substantially lower for FYI than for MYI.For the HH channel, the correlations are 0.39 (near), 0.37 (mid), and 0.42 (far), while for HV they are 0.39 (near), 0.40 (mid) and 0.41 (far).
The fact that a stronger relationship exists between sea ice height and backscatter for MYI than FYI (Figure S3 in Supporting Information S1), despite the fact the relationship between backscatter and sea ice roughness is weak for MYI, suggests that this relationship is not simply a function of the roughness relationship with height (Figure S4 in Supporting Information S1).Instead, we suggest that this relationship is related to the salinity of MYI.As MYI ages and grows it becomes fresher that is, less saline (Figure 2b; Cox & Weeks, 1974).As it becomes fresher, the SAR signal is able to penetrate further into the ice, causing more volume scattering and greater returns to the sensor.This older, fresher ice has undergone more growth and is therefore thicker and has a greater height.In younger MYI, the ice is more saline and the signal penetrates less far, returning less backscatter to the sensor.In this case the younger ice has typically undergone less growth and is therefore less thick and has a lower height.In FYI the signal is primarily returned at the snow-ice interface, and thus this mechanism has less influence.However, as roughness has a strong relationship with height, and FYI backscatter has a notable relationship with roughness, the still-notable relationship with height for FYI is likely a function of this relationship.
These results suggest that RCM backscatter, particularly in the HV channel, should be instructive for estimating sea ice height, particularly within a height range of approximately 0.40 m ( 0.35 to 0.05 m relative to DTU18 MSS).Although this range is limited, in this study that represents 67% of height values.Our results also suggest that the relationship can be instructive across all incidence angles.

Estimating Sea Ice Roughness From RCM Backscatter
Leveraging the relationship we observed between RCM backscatter and surface roughness, we developed a preliminary model for the estimation of surface roughness from FYI in RCM imagery.Using the data discussed above as training data, we trained a decision tree regression model to estimate sea ice roughness based on both the HV and HH backscatter of a pixel, while also considering the relationship between incidence angle and backscatter.Given the relationships we observed (Figure 1), we only trained the model on/for FYI.A decision tree model is a non-parametric supervised machine learning method that predicts the value of a target variable by learning simple decision rules inferred from the data features, using a piecewise constant approximation approach (Pedregosa et al., 2011; see Supporting Information S1).
We applied the data to three sample sites within the previously discussed McClintock Channel study region, as well as one each at additional sites in the Beaufort Sea and Baffin Bay (Figure 3).For the three McClintock Channel sites, the model was applied to the HV, HH, and incidence angle bands of an independent RCM SCLN Geophysical Research Letters 10.1029/2023GL107261 mode image, dated 26 February 2022, and the results compared to two sets of independent ICESat-2/UMD-RDA tracks in areas dominated by FYI (Figures 3a and 3b).Additionally, we test it on one beam of Track A that was included in earlier analysis but excluded from training, in an area that contains substantial MYI as well as FYI (Figures 3c and 3d; Figure S1 in Supporting Information S1).In the case of the latter, the ice type of pixels was labeled FYI or MYI and the results of the model are considered when only applied to FYI (Figure 3c) or both FYI and MYI (Figure 3d).There were 4,759 FYI data points and 6,623 MYI points along this beam, meaning that the combined data set is majority MYI for that sample.For the Beaufort Sea (11 March 2022) and Baffin Bay (17 March 2022) sites, near-coincident (<3 hr) images and tracks from RCM and ICESat-2 overpasses were obtained, because of drift taking place in those regions.
Comparing the frequency distributions of roughness produced by the model and measured by ICESat-2/UMD-RDA shows the model performing well at capturing the overall distributions and key statistics in areas dominated by FYI.In all but one case, the modal roughness is captured by the model (0.03 m for both the measured and modeled data), and in the exception there is only 0.01 m difference between the modeled and measured modes.For sites where FYI dominates, the modeled and measured means for the distributions are within a maximum of 0.02 m for all the sites dominated by FYI, and is within <0.001 m in one case 3e,and 3f).When including all data in the site with substantial MYI, the model performs a little less well at estimating the mean, with a difference of 0.025 m between the modeled and measured values.However, it should be noted the model still captures the mode and much of the overall shape of the distribution in this case.In all cases the model estimate of the mean is higher than that measured by ICESat-2/ UMD-RDA, suggesting that one could potentially apply a systematic offset to the modeled value.Additionally, the modeled standard deviation is within 0.01-0.02m of the measured value in all instances apart from the site including substantial MYI, where it differs by 0.06 m.
In addition to performing well at capturing the overall distribution of the roughness, we find that it is also effective at capturing spatial variations.Though the model does not perform well when comparing modeled and measured roughness at the 40 m pixel scale, when downsampling the data we find it performs well at a sub-kilometer resolution (Figure 3f; Figure S5 in Supporting Information S1).For example, Figure 3f shows the model effectively capturing key variations in roughness along an individual track beam for data that has a moving-mean of 800 m.Notable spikes in roughness between around 50 and 90 m, 100 and 130 m, and 210 and 250 m along track are all visible in both sets of data.When comparing this measured and modeled data, we record a root-mean-squared error of 0.03 m and a Pearson correlation of 0.78.Additionally, application of the model to an RCM SCLN image in April 2022 allows visualization of key roughness features.For example, an area of rough FYI can be seen in Victoria Strait (Figure S6 in Supporting Information S1), where ice deforms between the two land masses.

Conclusion
By comparing RCM SAR imagery with sea ice topography data measured by ICESat-2 and processed with the UMD-RDA, we show that RCM data provides important information about sea ice topography in Arctic sea ice during winter.Most notably, sea ice roughness is correlated with RCM backscatter from FYI, particularly in the HV channel and at near-range incidence angles; sea ice height is correlated with RCM backscatter from MYI, particularly in the HV channel and evenly across incidence angles.Both relationships suggest that RCM has great potential for estimating key sea ice topography information at spatial and temporal resolutions and extents not possible with laser or InSAR altimetry missions.This is particularly the case in the western Arctic where RCM typically achieves sub-daily coverage in many regions.The relationship between HV backscatter and topography information on low-backscatter FYI also highlights the value of RCM's low noise floor, particularly in its low noise "SCLN" mode.
We leverage the relationship between roughness and backscatter, along with supervised machine learning, to develop a preliminary RCM backscatter roughness retrieval model for FYI.By comparing with validation data from ICESat-2/UMD-RDA, this model is shown to be effective at capturing the key features of a roughness frequency distribution, including mean, mode and standard deviation.The relationships we analyze, the training data used, and testing of the model suggest it is best applied to areas dominated by FYI, although it may also be of some use in areas with substantial MYI.The model also captures key spatial variations in roughness at a subkilometer scale.These results show the model has potential for use in validating sea ice model results, as well as for analyzing roughness and deformation.
Having developed a preliminary model for roughness retrieval in the Canadian Arctic based on the limited available RCM SCLN data for our study region so far, we aim to develop a more robust Arctic-wide model by extending our analysis and model development in a range of conditions across the Arctic and multiple years.We will also further research constraining the effects of incidence angle, as well as snow conditions, on the relationship.We also aim to take advantage of the relationship between height and backscatter on MYI, and the strong relationship between roughness and height, to develop a robust model for estimating thickness from RCM imagery.

Figure 1 .
Figure 1.(a) The McClintock study site and location of the Ice, Cloud, and Land Elevation Satellite (ICESat)-2 tracks shown in a RADARSAT Constellation Mission (RCM) horizontal transmit-receive (HH) image from 20 March 2022; The spearman rank correlation (P < 0.0001) between (b) roughness and backscatter on first-year ice (FYI) and (c) height and backscatter for the horizontal-transmit/vertical-receive (HV) and HH channels at near-(<30°), mid (30°-45°), and far-range (>45°) incidence angles.(d) Mean RCM HV backscatter for 0.01 m bins of sea ice roughness, as measured by ICESat-2/University of Maryland-Ridge Detection Algorithm (UMD-RDA), for FYI data in all images and tracks.(e) Mean RCM HV backscatter for 0.01 m bins of sea ice height relative to DTU18 mean sea surface (MSS), as measured by ICESat-2/UMD-RDA, for multi-year ice (MYI) data in all images and tracks.Error bars display the standard deviations of backscatter for each bin.Bins with fewer than 100 data points are excluded from the plot.

Figure 2 .
Figure 2. Conceptual figures showing how the RADARSAT Constellation Mission signal interacts with (a) roughness on first-year ice (FYI) and (b) height on multiyear ice (MYI) with lower and higher surfaces.Red arrows represent the Synthetic Aperture Radar signal, large white circles represent air bubbles and small circles represent salinity.

Figure 3 .
Figure 3. (a-f) Frequency distributions for roughness as measured by Ice, Cloud, and Land Elevation Satellite (ICESat)-2/University of Maryland-Ridge Detection Algorithm (UMD-RDA) and estimated by the RADARSAT Constellation Mission (RCM) roughness retrieval model for corresponding pixels in an RCM image.(a, b) Two independent tracks dated 24 March 2022 and an independent RCM image dated 26 February 2022 in the McClintock Channel study region.Both tracks are in areas dominated by first-year ice (FYI).(c) An individual beam dated 11 March 2022 and the same independent RCM image in the McClintock Channel study site, where the model has only been applied to areas identified as FYI.(d) The same beam and image as (c), but including both FYI and multi-year ice (MYI).(e) A near-coincident track and image dated 11 March 2022 in the Beaufort Sea (f) a near-coincident track and image dated 17 March 2022 in Baffin Bay.(g) Along-track roughness for an individual ICESat-2/UMD-RDA beam from the track that forms the distribution in (b), alongside roughness as modeled by the RCM roughness retrieval model for corresponding pixels.In both cases roughness is plotted as an 800 m running-mean.