4.1. Offset Tracking Velocity Mapping of Southern Ellesmere’s Ice Caps
In general, our results show that the performance of the amplitude matching algorithm is good, and the method is able to resolve the glacier velocities in most areas. There is a large variability in the observed velocities, due to the different dynamic behaviour of the various glaciers. Some glaciers present a surge-type behaviour [42
] or high seasonal and interannual variability of their velocities [43
]. We can broadly group our studied glacier into three groups: (1) fast flowing with winter velocities greater than 600 m year
; (2) medium flowing with winter velocities within 50–200 m year
; and (3) slow flowing with speeds less than 50 m year
. Trinity and Wykeham glaciers (Figure 2
) belong to the fast flowing group with speeds up to 1200 m year
for Trinity and 600 m year
for Wykeham. To the medium flowing group belong glaciers such as Ekblaw Glacier (Figure 1
c) or neighbouring glaciers such as Stygge or Cadogan (north and south of Ekblaw, respectively; not shown in figure), reaching winter velocities of 80 and 60 m year
, or the southern glacier of Manson Icefield, with speeds up to 75 m year
. This is currently the fastest glacier of the Manson Icefield because Mittie Glacier, which can have velocities greater than 1 000 m year
when in full surge [42
], is currently in its quiescent phase [41
]. Most of the slow flowing glaciers are land-terminating glaciers located on the western side of the POW Icefield (Figure 3
) and on Sydcap Ice Cap. However, some of the land-terminating glaciers in the western POW Icefield have larger velocities, as happens with unnamed West 3, 4 and 5 (Supplementary Materials Figure S1
), with velocities up to 75, 150 and 100 m year
, respectively. These high velocities are due to the narrowing of the valleys confining the glacier flow. In the case of unnamed West 4, the velocity increase is more noteworthy because of its larger accumulation area.
However, the feature tracking algorithm is not able to resolve the surface velocity in slowest-moving zones of the ice cap’s accumulation areas (Figures S1
) and in some zones of the Trinity glacier (Figure 2
). In the case of the accumulation areas, this can be attributed to the absence of visible features due to the snow cover. In the case of the Trinity Glacier, the underlying reason is the strong surface velocity gradients that deform the surface features (e.g., at the junction between the main glacier trunk and its principal tributaries, at the corners where the glacier changes direction or at the glacier margins.
The feature tracking algorithm is best suited for resolving areas of fast glacier movement. This is clear when comparing the feature tracking results with those from differential interferometry (Figures S1 and S2
). However, there is also a limit on the amount of movement that the feature tracking algorithms are able to resolve. Nagler et al. [21
], using an in-house developed algorithm, were able to resolve velocities up to 12 m d
when using Sentinel-1 IW data (the same sensor that we are using). With increasing image resolution (smaller pixel size), larger velocities could be resolved. Anyway, this restriction is not relevant to us; because 12 m d
is equivalent to 4380 m year
, a velocity well above the largest glacier velocities observed in the Canadian Arctic.
We turn now our attention to the performance of offset tracking using ascending and descending tracks without using the azimuth offsets. As expected, the areas of stable ground do not present the typical azimuth streaks that one would expect when using azimuth offsets [28
]. In the scenes that we used, the orientation of the ascending and descending across-track vectors was close to the horizontal (e.g., 12
below the west-east direction for the ascending case). This makes the retrieval of surface velocities for glaciers oriented in the north-south direction more challenging. Nevertheless, glaciers oriented in this direction (e.g., most of the tributaries of Ekblaw Glacier) still present homogeneous surface velocities (Figure 1
c). Furthermore, when analysing the surface velocity gradients of Ekblaw Glacier in the northern direction (Figure 1
c), the better performance of our approach is noticeable, which avoids the use of azimuth offsets. We can observe how the northern component of the velocity increases and decreases depending on the direction of glacier flow. By contrast, when using the standard approach that uses azimuth offsets, we were unable to resolve the velocity gradients for this glacier. Please also note that the standard approach requires the correction of the ionospheric effect, which increases the complexity when considering Sentinel-1 TOPS acquisition mode. This is because each sub-swath presents its own ionospheric-induced azimuth pattern. When no information on the total electron content of the atmosphere is available, we need to have stable ground available in different sub-swaths, a condition not easily fulfilled in many glacierized areas. Furthermore, when using a modelled trend of the ionospheric disturbance for subtracting the ionospheric effect, residues that degrade the azimuth offset results always remain.
4.2. D-InSAR Velocity Mapping of Southern Ellesmere’s Ice Caps
Glacier surface velocity results using ascending and descending passes show good performance on all land-terminating glaciers, such as the western glaciers of POW Icefield (Figure 3
a and Figure S2
) and the eastern glaciers of Sydcap Ice Cap. Moreover, there are two tidewater glaciers that also show good results when applying this technique, namely the South Margin glacier (POW Icefield) and Mittie East and West arms (Manson Icefield), as well as their tributary glaciers. All of them present surface velocities below 50 m year
and smooth surface velocity fields.
D-InSAR performs better than offset tracking in areas with low glacier velocities. Furthermore, it is able to resolve small surface velocity gradients. The low amount of precipitation in this region [32
] allows for a successful application of D-InSAR in the region’s ice mass accumulation areas. This is the case of POW Icefield, where we see a fine and continuous ice velocity field. It is precisely this fine resolution that allows us to discern the ice divide between the eastern and western flowing glaciers (Figure S2
). We note that in some cases the Randolph Glacier Inventory (RGI) [36
] ice divides do not match well with our surface velocity field estimates. This fact is noticeable e.g., in Taggart Lake glacier and in the unnamed West 4 and 5 glaciers (Figure S2
). This fine resolution also allows us to recognize different glacier flows (tributaries) in some western land-terminating basins of the POW Icefield. Examples are Taggart Lake glacier, which shows three distinctive flows (one in its northern side and two in the southern one), Unnamed West 2 glacier and Unnamed West 4 glacier (Figure 3
Regarding surface velocities for individual glaciers, Unnamed West 1 reaches 24 m year and Taggart Lake glacier reaches 33 m year in their main branches. None of the three different ice flows visible in Unnamed West 2 glacier exceeds 30 m year. The southernmost one shows the slowest velocities, of up to 15 m year, its intermediate branch 12 m year and the northern branch ∼21 m year.
4.3. Comparison of Offset Tracking and D-InSAR Results
If we compare the results from offset tracking and differential interferometry (Figure 3
), the more irregular velocity patterns produced by the offset tracking technique are noticeable (e.g., there is some degree of variability on stable ground). Additionally, we perceive a lower resolution of the technique, either due to the size of the matching window or to the quality of the amplitude features. This lower resolution prevents the offset tracking technique from producing the continuous and clearly defined glacier surface flow fields produced when the differential interferometry technique is used. However, the offset-tracking technique allows us to derive region-wide results with less constraints as compared with other methodologies (e.g., resilience to de-correlation or fast surface velocities).
Differential interferometry produces a high-quality, continuous and smooth velocity field. The main shortcomings of this technique include the inability of the minimum cost flow algorithm to resolve strong surface gradients and its deterioration with longer time spans between acquisitions (e.g., increased de-correlation). The former situation would be exemplified by the case of the glacier tongues of glaciers Unnamed West 3, Unnamed West 4 and Taggart Lake north, whose velocities cannot be retrieved with the D-InSAR algorithm (Figure 3
). Offset tracking, on the contrary, is capable of capturing these type of velocity gradients, making this methodology especially useful to retrieve region-wide surface velocity fields.
The areas where the differential interferometry performs best are those with subtle transitions between parallel or adjacent surface flows or in accumulation areas with lower velocities. In such areas, we notice the difference between offset tracking and D-InSAR. While the former hardly resolves the velocity gradients (and, when it does, it produces a nonsmooth field; see the accumulation area of Taggart Lake glacier), D-InSAR results illustrate the capabilities of this technique in accumulation areas with low amount of precipitation, which helps to preserve the correlations between subsequent acquisitions.
4.4. Comparison with Previous Studies
We have produced regional velocity estimations for the Southern Ellesmere Island ice caps using intensity offset tracking and differential interferometry from Sentinel-1 acquisitions. What is special in our application of the intensity offset tracking technique is that we retrieved the surface velocity fields exclusively from range offsets using ascending and descending passes. We also applied differential interferometry for the same areas in order to establish a comparison on the performance of both methodologies in glaciers with different dynamic regimes [42
]. Sentinel-1 IW TOPS mode data was already tested with the offset tracking methodology by Nagler et al. [21
] giving successful region-wide results for the ice velocity field of Greenland. The application of the D-InSAR technique to TOPS mode acquisitions from ice caps is still a challenge due to the need for stable ground to achieve a fine co-registration with the spectral diversity method, aimed to avoid undesired azimuth phase ramps in the resulting products [8
The application of the intensity offset tracking approach to TOPS mode acquisitions was already dealt by Dall et al. [20
], who acknowledged the need for azimuth deramping and azimuth common band filtering when applying speckle tracking. These two latter steps are not required in the feature tracking that we apply. The application of the intensity offset tracking technique to ascending and descending passes was covered by Fallourd et al. [16
]. However, their approach considered the use of all azimuth and range velocity components solving by least-squares, while our approach discards the azimuth offsets from the estimation to avoid undesired ionospheric effects and, simultaneously, to avoid Sentinel-1 shortcomings such as its lower azimuth resolution. We did this by transforming a nonorthonormal coordinate system (e.g., ascending and descending across-track vectors) to a geographical coordinate system [12
The results of the application of the proposed method show that there is a noticeable improvement in the resolution of the components of the surface velocity together with an increase of the final precision. We avoid the ionospheric effects and therefore, considering the improved accuracy of the ephemeris, the remaining error sources are restricted to the co-registration, the matching procedure and the DEM related geocoding error. In addition, there are still some de-correlated areas in the accumulation zone devoid of results due to the lack of features for tracking. We would like to highlight that using azimuth offsets with Sentinel-1 TOPS data not only worsens the final velocity product but also increases the difficulty of correcting the azimuth streaks which present a different pattern for each sub-swath.
There are several articles devoted to the topic of interferometry using TOPS mode data [10
]. Furthermore, the use of ascending and descending passes for retrieving ice surface velocities is also well covered in the literature [12
]. The main obstacle that we faced when applying D-InSAR to TOPS mode data was to guarantee a successful fine co-registration using the spectral diversity method between contiguous bursts. The choice of Ellesmere Island as a test site was key for overcoming this limitation because of the wide areas of stable ground that are found next to the ice masses.
Our results show that differential interferometry is optimal for areas of low movement with no (little) surface change (e.g., accumulation areas with low amount of precipitation, and land terminating glaciers). The method produced smooth velocity fields for most land terminating glaciers of Ellesmere southern ice caps. On the other hand, the method failed to work in areas with strong velocity gradients and areas of fast glacier movement (e.g., the minimum cost flow (MCF) algorithm did not resolve a continuous velocity field [12
]). The comparison between methodologies illustrates the difficulties of MCF to resolve the velocity field of a few tongues from the western glaciers of the POW Icefield.
Regarding the comparison with velocities presented in previous studies in the region, focusing on those based on scenes closest in time to ours, we note that the velocities are very similar, especially for the fastest glaciers. Our maximum winter (Feb-March 2016) velocities of 1200 m year
near the terminus of Trinity Glacier or 600 m year
for Wykeham Glacier are very similar to those of 1250 and 500 m year
(respectively) reported by van Wychen et al. [41
] for winter 2014–2015 (from speckle tracking of RADARSAT-2 data), or those of 1200 and 650 m year
(respectively) given by Millan et al. [40
] for winter 2015-2016 (from speckle tracking of Sentinel-1a data). Similarly, for Ekblaw Glacier, both our results and those of van Wychen et al. [41
] show similar maximum velocities of ∼100 m year