Ice Surface Velocity in the Eastern Arctic from Historical Satellite SAR Data

Knowledge on ice surface velocity of glaciers and ice caps contributes to a better understanding of a wide range of processes related to glacier dynamics, mass change and response to climate. Based on the recent release of historical SAR data from various space agencies we compiled nearly complete mosaics of winter ice surface velocities for the 1990’s over the Eastern Arctic (Novaya Zemlya, Franz-Josef-Land, Severnaya Zemlya and Svalbard), a region with sparse optical velocity records from these years. We mainly applied offset-tracking to JERS-1 SAR data and filled data gaps using SAR interferometry and offset-tracking from ERS-1/2 SAR data. We studied the long-term variability of winter ice surface velocity by comparing our 1990’s results to 2008-2011 velocity maps from ALOS-1 PALSAR-1 and 2020-2021 maps from Sentinel-1. A general increase of winter velocities from the 1990’s to present along with a retreat of glacier fronts is obsverved. Exceptions to this general pattern are surges, which are widespread over Svalbard but rarely found in the other three regions. The dense time series of ice surface velocity from Sentinel-1 since 2015 were also considered to infer the representativeness of winter data with respect to mean annual values. We found that for non-surging glaciers short-term seasonal fluctuations are relatively small and winter ice surface velocities are a good representative of mean annual velocities with an underestimation of less than 10%. Together with consistent datasets of glacier ice thickness and terminus position, the ice surface velocities in the Eastern Arctic provide the basis to quantify the regional decadal average calving flux during the 1990’s. The ice surface velocity data set for the 1990’s over the Eastern Arctic from satellite SAR data can be downloaded from https://doi.pangaea.de/10.1594/PANGAEA.938381 (Strozzi et al., 2021).


Introduction
Glaciers and ice caps are retreating and thinning nearly everywhere in the world. Mass loss in recent decades was ascertained in various studies from a range of techniques and sensors, including glaciological observations (Zemp et al., 2019), satellite gravimetry observations (Wouters et al., 2019), differencing surface elevations from satellite and airborne observations (Hugonnet et al., 2021), and analysis of satellite interferometric altimetry (Tepes et al., 2021). In order to understand the mechanisms behind glacier mass loss and the discrepancies between the various studies, it is worthy to investigate the contributions of various components of mass loss. The calving flux of a glacier can be determined by multiplying its flow 1 5 The archive of JERS-1 data at ESA was systematically mined and all image pairs in series from the same track and frame were analysed. Only results obtained for 44-day winter image pairs were further considered, because in summer and over longer time periods the ice surface velocity maps contained much noise and reduced spatial coverage. Where no 44-day winter JERS-1 image pairs were found in the ESA archive, the JERS-1 data archive at JAXA was searched to complement the results. Further data gaps existed for the far east of Franz-Josef-Land, entire Severnaya Zemlya and the central part of Svalbard. In these regions we tracked ERS-1 data with time intervals of 9 to 18 days. Step

SAR interferometry
The use of differential SAR interferometry to map surface displacements at centimetre-resolution is well established in research and operational projects Dowdeswell et al., 2008;McMillan et al., 2014). The interferometric phase is sensitive to both surface topography and coherent displacement along the look vector occurring between the two acquisitions of the interferometric image pair. The differential use of a high-resolution DEM (2-pass InSAR) or of two SAR image pairs acquired within short time periods (3/4-pass InSAR) allow the removal of the topographic phase from the interferogram to derive a displacement map. In the case of 2-pass InSAR, the acquisition date of the DEM has to match that of the SAR dataset close enough to ensure that no major topographic signal is left on the differential interferograms. In the case of 3/4-pass InSAR, displacement within the two image pairs is assumed constant. In addition, short perpendicular baselines of the SAR image pairs are preferred, in order to minimize the effect of the residual topographic phase so that phase signals can be interpreted as ice surface displacement in the satellite line-of-sight direction, with possible atmospheric disturbances.
ERS-1/2 InSAR results from previous studies (Dowdeswell et al., 2008;Nuth et al., 2019) were considered to improve the quality and completeness of the results over Nordaustlandet, south Spitsbergen and north-west Spitsbergen (Svalbard). The ERS-1/2 InSAR ice velocity map of Nordaustlandet combines interferometric phases from 1-day ERS-1/2 image pairs and is produced for the winter 1995/1996 at 100 m resolution. Method and uncertainties are described in Dowdeswell et al. (2008).
In most cases errors are assumed to be smaller than 7 m/a, while for unfavourable combinations of image pairs this value is slightly larger. The ERS-1/2 InSAR ice velocity map of south Spitsbergen is produced from winter 1996 and 1997 data at 20 m resolution. Method and uncertainties are described in Nuth et al. (2019). The accuracy was measured by extracting displacements across a point grid of 1000 m spacing excluding points which were not on stable terrain (glaciers, fjords, etc.). Nuth et al. (2019) found a median value of 2.3 m/a and a standard deviation of 2.6 m/a. The same processing and accuracy assessment procedure was also applied to 1-day ERS-1/2 image pairs of winter 1995/1996 over north-west Spitsbergen . In this case, a median value of 4.0 m/a and a standard deviation of 3.7 m/a were found.

Data
Two sets of data containing ice surface velocities over the Eastern Arctic can be downloaded from https://doi.pangaea.de/10.1594/PANGAEA.938381. The first set of data contains the velocities derived from offset-tracking of all image tracks from this study as original research data in vector format with metadata information. Nine collections of data wrapped up in single files for easy storage (tar packaging followed by a gzip compression) are available: the JERS-1 (1992-1998) results over Novaya Zemlya, Franz-Josef-Land, and Svalbard, the ERS-1 (1991-1992 results over Franz-Josef-Land, Severnaya Zemlya and Svalbard and the ALOS-1 PALSAR-1 (2006-2011) results over Novaya Zemlya, Franz-Josef-Land, and Svalbard. A comma-separated values file (.csv) provides the northing and easting coordinates of measurement points, the elevation from the above-mentioned DEMs, the displacement in metres in the x, y and z directions and the crosscorrelation coefficient for each measurement. A metadata file in extensible markup language format (.xml) provides information about the SAR images (<inputSatelliteData1> and <inputSatelliteData2>), the processing parameters (<processingParameters>) and quality aspects of the data such as the percent of valid information over ice (<QA-IV-2>) and statistical measures over ice-free regions (<QA-IV-3>). In addition, the original research data packages include for each image pair GeoTIFF files of the three-dimensional ice surface displacement maps (.tif in single-precision floating-point format and .300.tif as exemplary display of the colour-coded displacement map with saturation at 300 m/a), the two intensity images (.pwr1.tif and .pwr2.tif), the differential interferogram (.tflt.tif), the phase coherence image (.cc.tif), an RGB colour composite of the coherence, intensity and intensity difference between both images (.rgb.tif), and the layover and shadow map (.ls_map.tif. In all cases data are provided the Universal Transverse Mercator (UTM) projection (zones 33N for Svalbard, 40N for Novaya Zemlya and Franz-Josef Land, and 47N for Severnaya Zemlya) with a spatial resolution of 100 m.
The second set of data contains velocity mosaics of the best JERS-1 results over Novaya Zemlya, Franz-Josef-Land, and Svalbard, the best ERS-1 results over Franz-Josef-Land, Severnaya Zemlya and Svalbard and the best ALOS-1 PALSAR-1 results over Novaya Zemlya, Franz-Josef-Land, and Svalbard in GeoTIFF format. In addition, we provide in the same format also the winter ERS-1/2 InSAR ice velocity map, used to improve the quality of the results over Svalbard, and Sentinel-1 mosaics computed from winter 2020/2021 data, considered in the following section to study the long-term variability of winter ice surface velocity over the Eastern Arctic. Appendix A provides the lists of the satellite data considered over the four study regions along with some technical information and Figure 1 shows the SAR footprints.

Flow velocities for distinct periods
Mosaics of ice velocity maps of the 1990's for Novaya Zemlya, Franz-Josef-Land, Severnaya Zemlya and Svalbard are presented in Figures 2a, 4a, 5a, and 6a, respectively. In the Russian High Arctic, priority in the winter ice velocity mosaics was given to JERS-1 offset-tracking over ERS-1 offset-tracking. Glaciers were masked out from land and sea using glacier outlines from satellite imagery acquired between 2000 and 2010 during summer (Moholdt et al., 2012), after manual adjustment using the SAR backscattering intensity images of the front position of the glaciers that significantly retreated or advanced from the 1990's to the 2000's. In addition, for Severnaya Zemlya we included ice flow over the Matusevich Ice Shelf that collapsed in 2012 (Willis et al., 2015). Over Svalbard, priority in the winter ice velocity mosaic was given to ERS-1/2 InSAR over JERS-1 offset-tracking, with ERS-1 offset-tracking considered to fill minor data gaps over north-east Spitsbergen. Over this region we considered the glacier outlines from summer satellite imagery spanning the period 2000-2010 (Nuth et al., 2013) after correction from the SAR backscattering intensity images for the front position of the major retreating or advancing glaciers. Sentinel-1 mosaics computed from winter 2020/2021 data that showed the best spatial coverage are considered to study the long-term variability of winter ice surface velocity. The four Sentinel-1 velocity maps for Novaya Zemlya, Franz-Josef-Land, Severnaya Zemlya and Svalbard are shown in Figures 2c, 4b, 5b and 6b, respectively. Glaciers were masked out from land and sea using satellite imagery acquired between 2013 and 2016 over Novaya Zemlya (Rastner et al., 2017) Where available, we include in our discussion previously published velocity results from ALOS-1 PALSAR-1  in order to highlight the temporal consistency of the changes or point to possible trends and differences. In particular, we consider a nearly complete mosaic computed for Novaya Zemlya from winter ALOS-1 PALSAR-1 data acquired between 2008 and 2010. This mosaic is shown in Figure 2b with glaciers masked out from land and sea using glacier Glacier, we did not detect any sign of destabilisation for the glaciers on Novaya Zemlya. In this region the inter-annual changes of winter ice surface velocity between 1998 and 2021 exceed seasonal variability ( Figure 11) and can be considered a significant representation of the long-term variability of ice surface velocity over this region.

Franz-Josef-Land
Over Franz-Josef-Land, the general pattern of the differences between winter ice surface velocities from the 1990's to Over Franz-Josef-Land, we did not detect any clear sign of destabilisation and the inter-annual changes of winter ice surface velocity, which exceed seasonal variability, can be considered representative of the long-term variability of ice surface velocity.

Severnaya Zemlya
Changes in the ice surface velocity observed between the 1990's and 2020/2021 over Severnaya Zemlya ( (Dowdeswell et al., 2002) and summer ALOS-1 PALSAR-1 data . Over the Academy of Sciences Ice Cap, we observe a widespread increase of ice surface velocities from 1991 to 2020/2021 of more than 100 m/a for five large and two small glaciers. For two glaciers over this ice cap, the maximum speeds at the front were actually higher in 1991 than in 2020/2021. This is not reflected in Figure 5c Severnaya Zemlya, we detected two glaciers with clear signs of destabilisation. Also for these, however, the inter-annual changes of winter ice surface velocity, which exceed seasonal variability, again represent the long-term variability of ice surface velocity rather well.

Svalbard
The difference map between ice surface velocities in the 1990's and 2021 for areas with Sentinel-1 velocities larger than 50 m/a over Svalbard ( Figure 6c) shows a large increase of velocities for many glaciers. Along with glaciers having significantly higher frontal velocities (e.g., Monacobreen, Kronebreen and Osbornbreen on Spitsbergen and Schweigaardbreen, Leighbreen, and Basin 7 on Austfonna), there are also many prominent surges, e.g., Basin 3 (Austfonna), Stonebreen (Edgeyoa), Negribreen and Sonklarbreen (Spitsbergen) and a few other smaller ones. Over south Spitsbergen we observe the only glacier with significantly higher velocities in the 1990's, Mendelejevbreen, which surged apprixmately in 2000 (Blaszczyk et al., 2009). Other large glaciers that were surging during the 1990's, such as Monacobreen in 1994  and Fridtjovbreen in 1996 (Murray et al., 2003), are not captured in our mosaic shown in Figure  from several years is much less representative of yearly averages over Svalbard than over the three other study regions, even without taking into account that over this region a large number of glaciers (more than ten according to Leclercq et al. (2021)) underwent surging events in recent years.      Zemlya with Sentinel-1 velocities larger than 100 m/a, the velocities in the winter period differed between -12% and +4% (average -3%) from the average annual velocities. For the summer period, the related differences were between +33% and -12% (average +8%). Summer differences are larger than those in winter, because the summer speed-ups are short, intensive events and the summer period is shorter.

Severnaya Zemlya
On and +2% (average -5%) from the average annual velocities. The velocities in the summer period differed between +27% and -1% (average +11%) from the average annual velocities. summer 650 m/a (+20.1%), mean winter 488 m/a (-9.8%)) and (c) Academy of Sciences 6 (mean 1239 m/a, mean summer 1341 m/a (+8.2%), mean winter 1189 m/a (+4.0%)).  (Figure 12e), the decrease of velocity after the summer speed-up is much slower than for Kronebreen and Hornbreen, lasting a few months. For Idabreen (Figure 12f), maximum velocities are reached during the winter, slowly decreasing to reach the minima in summer. For the 30 glaciers analysed in our study over Svalbard with Sentinel-1 velocities larger than 100 m/a, the velocities in the winter period differed between -13% and +14% (average 1%) compared to the average annual velocities. The velocities in the summer period differed between -33% and +46% (average -2%) compared to the average annual velocities. The variability of ice surface velocities over Svalbard, also graphically summarised in Figure 14, is much larger than over the Russian High Arctic. It is worth noting that in our list of glaciers, we included only two surging glaciers, Basin 3 and Negribreen.  (mean 402 m/a, mean summer 461 m/a (+14.5%), mean winter 377 m/a (-6.3%)), (e) Hinlopenbreen (mean 305 m/a, mean summer 447 m/a (+46.2%), mean winter 266 m/a (-12.8%)) and (f) Idabreen   (mean 199 m/a, mean summer 133 m/a (-33.2%), mean winter 227 m/a (+14.2%)).

Figure 13: Differences between annual average ice surface velocity and winter (October-May, blue) and summer
(June-September, red) averages for the glaciers analysed in our four study regions.

Interpretation of the long-term trends
We observed a general increase of winter velocities from the 1990's in the Eastern Arctic, along with a retreat of glacier fronts. Notable exceptions to this general pattern are surges, which are widespread over Svalbard but rarely found in the other three study regions. Indeed, for a couple of glaciers on Novaya Zemlya (e.g. Figures 9c), three glaciers on Severnaya Zemlya (e.g. Figures 11a, 11b) and more than ten glaciers on Svalbard (e.g. Figures 12a and 12b, the others not shown in this contribution but listed e.g., in Leclercq et al. (2021)), intra-annual trends in the Sentinel-1 ice surface velocity time-series dominate over inter-annual variability. Surges were not detected with our 1990's data. Few examples of surging glaciers as found in the literature (e.g., Monacobreen  and Fridtjovbreen (Murray et al., 2003)) are missed, because they were asynchronous with our JERS-1 data coverage. At present, many more surging glaciers are observed over Svalbard than in the 1990's (Morris et al., 2020), making the interpretation of the long-term variability of ice surface velocity over this region challenging. It is expected that with the continued global temperature increase, the variability in the way glaciers respond to climate change will further increase. Some glaciers feature dynamic instabilities and discharge large amounts of ice, while others become dynamically less active and exhibit moderate rates of mass loss. As dynamic instabilities allow for larger, more rapid ice mass loss than surface melt, it is expected that ice discharge into the oceans will increase. Characteristic patterns of time series of ice surface velocities over dynamic instabilities will be analysed in future work. The dynamic response of land ice to climate forcing constitutes the main uncertainty in global sea level projections for the next century (Church et al., 2013;Martin and Adcroft, 2010). To address this knowledge gap, observations of the relative contribution of both surface mass balance and ice dynamics to the global mass losses are necessary. Using seven years of CryoSat-2 swath interferometric altimetry, Tepes et al. (2021) tracked changes in the volume of Arctic glaciers and ice caps and partitioned their losses into signals associated to atmospheric processes and glacier dynamics. They concluded that while surface ablation is responsible for 87% of losses across the Arctic, dynamic imbalance is increasing in the Barents and Kara Sea regions, where it now accounts for 43% of total ice loss. The persistent incursion of warm North Atlantic Ocean water into the Arctic Ocean, associated with a northward shift of Atlantic climate, is thought to have a direct impact on sea ice extent and tidewater glacier dynamics (McMillan et al., 2014;Polyakov et al., 2017;Barton et al., 2018). However, the respective role of internal processes, commonly ascribed to glacier surges, and external climatic forcing in driving these dynamic instabilities remains largely unclear (McMillan et al., 2014;Dunse et al., 2015). Quantifying calving fluxes to the ocean, composed of ice discharge to the ocean and marine-terminus retreat under the assumption of negligible melting and sublimation for grounded tidewater termini, is important to partition the causes of glacier mass loss. In order to quantify the regional decadal average calving flux, spatially and temporally consistent datasets of ice surface velocity are required together with consistent datasets of glacier ice thickness and terminus position (Kochtitzky et al., submitted). Our mosaics of winter ice surface velocities for the 1990's over the Eastern Arctic, with uncertainties of generally less than ±20 m/a, can support the assessment of the past regional decadal average calving flux for this region.

Conclusions
We computed nearly complete mosaics of winter ice surface velocities over the Eastern Arctic from the 1990's, presented long-term trends compared to results from ALOS-1 PALSAR-1 in winter 2008-2011 and Sentinel-1 in winter 2020/2021, and analysed the seasonal variability from Sentinel-1 data to infer the representativeness of winter velocities compared to annual averages. In most of the cases, winter velocities are a good representative of mean annual velocities, as seasonal fluctuations are relatively small for non-surging glaciers, with an underestimation of less than 10%, in particular over the Russian High Arctic. Summer velocities, on the other hand, can be significantly larger than the annual mean. Additionally, there are strong, short-time speed-up events during the summer period for many glaciers. We conclude that winter velocities give a better idea of long-term trends in speed, even if the spatial extend of the summer acceleration events, which can also be subject to long-term changes, is missed. Available ice surface velocity products based on Landsat-8 optical data such as GoLive, which also provide regularly updated scene-pair velocity fields, are limited to periods of solar illumination, therefore missing the winter season. As we showed, these data are more subject to summer speed-up events and thus less suited to analyse long-term trends in velocity. In addition, these products are less accurate in the very slow-moving accumulation areas, as these regions are difficult for optical feature tracking due to low-feature surfaces. Best approaches to integrate results derived from satellite SAR and optical missions still need to be further investigated. One of the crucial lessons learned from previous research on partitioning the causes of glacier mass loss is a severe problem with glacier velocity data and uncertainty of the calving flux estimation coming from this source. For an improved glacier monitoring strategy in the Eastern Arctic from Sentinel-1 SAR data, 6 days repeat is better suited than 12 days to retrieve high quality data, because time-series used to study dynamic instabilities are denser and the effects of ice and snow melting are less severe. In addition, acquisition gaps as occurred in 2018 and 2019 over Franz-Josef-Land should be avoided, as the Sentinel-1 time-series over this region (Figure 10) are of arduous interpretation. The spatial resolution of the IW Sentinel-1 SAR data (Table 1) is not optimal for studying ice surface velocity of Arctic glaciers and ice caps, but still acceptable as a compromise to the global acquisition strategy. Future L-band SAR missions such as NISAR (https://nisar.jpl.nasa.gov; last access: 7 September 2021), ALOS-4 (https://global.jaxa.jp/projects/sat/alos4; last access: 7 September 2021) or ROSE-L (https://www.esa.int/Our_Activities/Observing_the_Earth/Copernicus/Candidate_missions; last access: 7 September 2021) can complement the Sentinel-1 results.

Output data
The ice surface velocity data set for the 1990's over the Eastern Arctic from satellite SAR data can be downloaded from https://doi.pangaea.de/10.1594/PANGAEA.938381 (Strozzi et. al. 2021).