A terrestrial radar interferometer (TRI) was deployed twice for a two-week period in August 2023 and July 2024, respectively, to continuously measure ice flow velocity. The instrument was positioned on solid bedrock atop an opposing hill, 496 m above sea level and three kilometres from the calving front (Fig. ). The TRI developed by GAMMA Remote Sensing, is a real-aperture radar interferometer with one transmitting antenna and two receiving antennas, operating at a wavelength of λ=17.4 mm (Ku-band, 17.2 GHz). The antennas are rotating along the vertical axis on a precision astronomical mount. The range resolution is about 0.75 m, while the azimuth resolution is 0.4°, which corresponds to 6.9 m at a slant range of 1 km and in our case about 21 m at the calving front . Whereas the TRI scan has a maximum range of 16 km, our specific location of the TRI system and the glacier topography enable seamless, comprehensive coverage within the first 6 km of the terminus area. In this study, the data acquisition, which is daylight- and weather-independent, was repeated at 1-minute intervals, allowing to receive almost continuous line-of-sight (LOS) velocity and DEM values.

In 2023, we installed a basic automatic weather station (AWS) near the shore east of the terminus of EKaS (called “Fjord”, data used in Fig. b). One year later, a weather station was deployed next to the TRI, on the hill on the opposite side of the glacier terminus, 500 m above sea level (called “Hill”, data used in Fig. b). Both weather stations were measuring air temperature, relative humidity, incoming solar radiation, and precipitation at a 30 min interval. Wind speed and wind direction were taken from the Mittafik Airport weather station in Narsarsuaq . This weather station, operated by the Danish Meteorological Institute (DMI), is located about 20 km from the front of EKaS, in a valley that is aligned parallel to the flow direction of EKaS. Since different weather stations were used to derive the air-temperature records for the two years, we compared both datasets with the AWS at Narsarsuaq to ensure that they reflect regional temperature variability rather than site-specific effects (e.g. temperature inversions). As shown in Figs.  and , the regional temperature signal is well captured in both cases. The only exception is the initial phase of the foehn event on 20 July 2024, which was suppressed at the “Fjord” station by a local inversion. Therefore, data from the “Hill” station were used in 2024.

A time-lapse camera was installed on the hill opposite the calving front (Fig. ) and was running year-round from July 2022 onwards. The camera took pictures of the calving front at intervals between two minutes during the summer field campaigns and 20 min to one hour for the rest of the year. This allowed us to classify the evolution of the ice mélange extents and subglacial plume extents using a visual assessment of the images. Plume extents were classified based on their estimated relative coverage in the observed area: none (0 %), small (1 %–25 %), medium (26 %–50 %), and large (51 %–100 %). Ice mélange was similarly classified, but with a slightly adjusted percentage range: none (1 %–25 %), small (26 %–50 %), medium (51 %–75 %), and large (76 %–100 %). Both variables were extracted twice a day in the time-series of Figs. d and d, with the four classes determined by visually estimating the relative extents. A second camera on the hill east of the terminus, running at the same intervals, allowed to approximate the timing, area and surface elevation change of a major subglacial lake drainage event L1 from the western tributary of EKaS.

Besides the time-lapse data, we used satellite imagery from Sentinel-1 and Sentinel-2 as well as data from the Arctic-DEM to constrain the timing, extent and the approximate volume of the subglacial lake drainage event L1 mentioned above and an additional ice-marginal lake drainage event L2 on the orographic right side of EKaS, 20 km upstream of the terminus. For the latter, discharge volumes were provided by , based on satellite and airborne altimetry data. The locations of the two lakes are shown in Fig. .

A pressure sensor was installed in the north-eastern end of the fjord about 4 km from the calving front (Fig. ) to monitor the tides. We installed a RBR-duet³ pressure-temperature sensor in a steal-pipe drilled onto bedrock at the shore in July 2022. It continuously sampled the water level at a 4 s interval until end of July 2024.

Ice flow velocities were derived from the TRI phase records using the GAMMA software stack and following an established workflow . The TRI transmits from a single antenna and measures radar backscatter using two receiver antennas. Temporal interferometry was obtained by analysing the phase signal recorded by a single receiver antenna (in this case, the upper antenna) at consecutive acquisition time intervals of 1 min , resulting in a 1 min single-look interferogram time-series. A multilook factor of 5 was applied to spatially average the TRI data in range direction e.g., as a compromise between noise reduction and high spatial resolution along the flowline, resulting in a resolution of 3.75 m (range) × 21 m (azimuth) at the glacier front. To minimize atmospheric noise in measurements caused by variations in air turbulence and humidity , interferograms collected within 30 min were stacked, leading to a final temporal resolution of 30 min. To support this choice, several temporal baselines between 10 min and 1 h were tested, with the aim to minimize noise while maximising temporal resolution. While the main velocity variation patterns could already be detected on a 10 min resolution, a resolution of 30 min allowed us to get rid of most of the atmospheric noise and prevent misinterpreting noise as a physical signal.

At a next step, the phase signal was unwrapped using a stable location on bedrock as a control point. The phase differences between consecutive unwrapped interferograms, can be converted into LOS displacement δ by: δ=-λϕ4π, where λ=17.4 mm is the wavelength, ϕ is the differential phase difference, and where the displacement measurement sensitivity is better than 1 mm . Note, pixels with a stacked coherence <0.8 were masked out before unwrapping. However, since we originally measured the 1 min displacement, the coherence on the glacier was always given. Finally, the displacement data were used to calculate LOS velocity maps that allow for investigation of spatial and temporal flow speed variability . For visualization purposes, the radar image pixels were transformed into Cartesian coordinates. Since resampling may introduce errors, all computations were conducted in the original radar geometry, with georeferencing applied only to the final outputs.

The temporal variability of the velocity during both two-week campaigns was mainly analysed using the values along the centreline of the mapped area (Fig. ). Averaged velocity estimates close to the front can be affected by boundary effects such as from missing ice due to calving events happening within the 30 min between two consecutive velocity maps. Additionally, our data quality assessment deriving velocity estimates over exposed stable bedrock along the glacier showed that atmospheric noise increases with distance from the calving front due to longer travel times of the radar beams through the atmosphere (Fig. b). Therefore, the averaged centreline velocities were calculated using data for the first two kilometres along the centreline, where data quality is highest (Fig. b), but excluding the initial 100 m. The according stretch is labelled as “centreline part” in Fig. .

Acceleration maps were derived from smoothed velocity maps as follows: For each pixel in the velocity map, a temporal smoothing was applied. Therefore, a 30 min interval time-series covering the entire two-week period was processed for each pixel using a Butterworth low-pass filter with a cut-off period of 12 h (Fig. a and a). The Butterworth low-pass filter and the according cut-off period was chosen because this configuration was found to perform best to effectively suppress outliers while still preserving the full diurnal velocity variability. The filtered value for each time-step was subsequently used to update the corresponding pixel in the velocity map. Finally, the gradient between two consecutive smoothed velocity data points was calculated and divided by the time difference, yielding acceleration estimates for every time-step and pixel.

The TRI DEMs used to identify calving events were generated using a workflow similar to that applied for deriving TRI velocities. However, spatial interferometry was obtained by analysing the phase signal difference between the two receiving antennas. The resulting interferograms, that have an original acquisition time interval of 1 min , were multilooked and unwrapped using a stable location on bedrock as a reference and subsequently converted into topography following the workflow of . To correct systematic errors such as reference height inaccuracies, baseline errors and antenna misalignment , a correction factor was derived by comparing the generated DEMs with the ArcticDEM using stable control points at different radar distances. This factor was applied to the computed topography to reduce elevation uncertainty. To reduce atmospheric noise, 10 consecutive elevation maps were stacked, leading to a final temporal resolution of 10 min . Then, elevation changes between sequential stacked elevation models were calculated and negative changes at the front attributed to a calving event. Because of the stacking procedure, events occurring within a 10 min interval were combined. Elevation changes of less than 5 m are considered as noise and filtered out , leading to a detection of calving events exceeding a volume of 5000 m³.

Both line-of-sight (LOS) ice velocity time series from the 2023 and 2024 summer campaigns show clear diurnal fluctuations, which are characterized by an increase in velocity during the day followed by a decrease during night (Figs. a and a). While this diurnal signal is consistently observed in the 2024 dataset, it is slightly less clear in 2023, particularly during the period from 10–11 August. In both years, the diurnal accelerations generally start in the morning around 08:00–09:00 local Greenlandic time (UTC−2 in 2023, UTC−1 in 2024) and the velocities peak in the evening around 19:00–20:00 local Greenlandic time. However, the exact timing can vary up to six hours between individual days. An average diurnal velocity fluctuation, excluding periods that are heavily influenced by multi-day speed-up events, shows a peak-to-peak amplitude of 0.4 m d⁻¹ (2023) and 0.5 m d⁻¹ (2024), which corresponds to 7 % (2023) to 8 % (2024) of the average speed (Fig. ). These diurnal variations of the flow velocity are clearly correlated with the air temperature signal measured at a nearby weather station, which acts as a proxy for surface ice melt (Figs. b and b).

In 2024, cross-correlation analysis of the velocity and air temperature time-series reveals a peak correlation of 0.6 at a lag of 4 h, indicating a moderately positive relationship throughout the entire 2-week field period. In other words, the ice flow speed reaches its daily maximum 4 h after the temperature peak. For the days from 16 July 2024, onward, the correlation becomes even stronger, reaching a cross-correlation of 0.8, again at a lag of 4 h. However, before 16 July, no correlation between flow speed and air temperature can be detected, with a cross-correlation value <0.1 at a lag of 4 h. Furthermore, the fjord was strongly covered by ice mélange, while the plume extent remained small for most of the time during the 2024 campaign (Fig. d).

In 2023, the cross-correlation between air temperature and flow velocity over the entire dataset is weak (cross-correlation <0.1), but most positive at a lag of 5 h. Nevertheless, for the last few days, between 12–15 August 2023, the cross-correlation is moderately positive, with a value of 0.7 at a lag of 5 h. In that year, the glacier terminus was barely covered with ice mélange and the plume extent fluctuated in size (Fig. d).

In both years, periods with poor correlation between air temperature and flow velocity are likely due to additional processes disrupting the diurnal signal (see below). Further, no clear link to multi-day scale velocity variations was found for either the ice mélange or the plume extent. The diurnal ice velocity variations are not directly correlated to the tidal signal, with cross-correlation values close to zero regardless of any lag (Figs.  e and e). At the study site, the tides are dominated by a shorter periodicity of 12.4 h.

During both field campaigns multi-day variations in velocity are superimposed onto the diurnal changes. In 2023, a distinct velocity peak is observed between 7–10 August (Fig. a). During this period, a continuous velocity increase of in total 1.5 m d⁻¹ (27 %) occurred, followed by a fast velocity decrease of 1.7 m d⁻¹ (30 %) within 13 h. After 12 August, another continuous increase of similar magnitude over a period of at least 4 d was observed. The weather, however, remained rather stable with no substantial additional water inputs such as rainfall, varying wind conditions, or foehn events (Fig. c). Thus, no direct relation to meteorological conditions is apparent for this multi-day speed-up event.

In the 2024 campaign, a very distinct speed-up event occurred between 19–21 July (Fig. a), starting with a sudden velocity increase of 2 m d⁻¹ (28 % of the average speed) within 13 h, followed by a large diurnal fluctuation peak-to-peak amplitude of 1 m d⁻¹ (14 %), before dropping back down to pre-event velocity levels. At the same time, the air temperature rose drastically to almost 20 °C, accompanied by a noticeable shift in the prevailing wind direction from south-west to north-east (Fig. c). This shift comes along with a low relative humidity of 30 %–50 %, which indicates a warm, foehn-like wind descending from the ice sheet. Later, another one-day speed-up event occurred on 25 July, with a large velocity increase of 1.2 m d⁻¹ (17 %), followed by a sudden speed drop of 1.8 m d⁻¹ (26 %), mostly driven by a large local velocity change signal at the front. During the initial phase of the 2024 time-series, a clear slowdown over 3–4 d can be observed on top of some diurnal variations, while air temperatures remain rather low or even slightly increase. This is also reflected in a weak cross-correlation between velocity and temperature.

Both years show a multi-day speed-up event (7–10 August 2023; 12–15 July 2024), which exhibits only weak correlation with air temperature and therefore cannot solely be explained by surface melt. Instead, these events align with the timing of subglacial and ice-marginal lake drainage episodes.

The slowdown in the first 3–4 d of the 2024 dataset coincides with the cessation of a major subglacial lake drainage event (L1, Fig. b), expected to release approximately 100–300 million cubic metres of freshwater. The drainage event from below the western tributary about 3 km upstream of the glacier front has been detected in the time-lapse imagery. The onset of the lake drainage event started already on 4 July, a few days before the start of our velocity record, and lasted until 15 July.

Having deployed our TRI from 12 July onwards, we only managed to capture the ending tail signal from this lake drainage event occurring on 4–15 July, but the timing corresponds well to the phase of slowdown in flow speed and weak cross-correlation between air temperature and ice flow speed. Note, the time-lapse cameras also recorded this large subglacial lake drainage L1 in summer 2023. But this discharge event occurred between 28 July and 3 August, stopping just as we began our field campaign, and hence we could not detect any influence in our TRI-velocity record.

For the period of the observed main velocity speed-up event in the summer 2023 data, we observed an ice-marginal lake drainage event (L2, Fig. a), that may have provided a large volume of freshwater input to the glacier system. Satellite imagery from Sentinel-1 and Sentinel-2 showed that the lake, which is located 20 km upstream at the orographic right margin of the main glacier, drained between 7–9 August (Fig. a). Sentinel-1 data indicates a half empty lake on the morning of 8 August and thus confirms ongoing drainage . already observed this ice-marginal lake drainage in 2020 and 2022 and used satellite products to estimate an average water level drop of 12 m, which corresponds to about 2 million m³ of released water, considering a lake area of roughly 0.2 km². Given an expected delay due to the large distance to the terminus, the lake drainage date seems to align with the high velocity speed-up event that peaks at the start of 10 August.

The TRI allowed to detect calving events with subaerial volumes larger than 5000 m³ and relate them to our velocity variations. In 2023, a total of 134 events were captured during the 12 d field period, whereof the interquartile range (IQR) shows subaerial calving volumes from 10 000 to 30 000 m³, with a median calving volume of about 15 000 m³. In 2024 (total of 81 events in 14 d), the IQR of the subaerial calving volumes goes from 8000 to 60 000 m³, with a median calving size of about 30 000 m³. In both years, the largest captured calving events had a subaerial volume of about 200 000–300 000 m³.

Throughout the two summer campaigns, we observe that calving events are generally more prevalent during high-velocity periods (Figs. d and d). For example, the high-velocity day of 20 July 2024, recorded 9 events, compared to just 2 events on 14 July 2024, a day with relatively low flow velocity. The average flow velocity during calving events is significantly higher compared to periods without calving (p<0.05). However, no clear velocity response, such as slow-downs or speed-ups from the front travelling upstream, were found after an individual calving event. Additionally, no diurnal pattern for the appearance of calving events could be found.

The temporal patterns, such as the diurnal and the multi-day speed-up events, can generally be observed almost uniformly along the entire 6 km long centreline (Figs.  and ). The velocities are generally increasing towards the terminus (Figs. a and a). Subtracting the 2-week velocity average for each location indicates that the diurnal velocity fluctuations not only appear along the entire 6 km stretch, but also occur at a similar absolute magnitude along the centreline (Figs. b and b). This is further supported by comparing the average LOS velocity on three different flow and transverse lines, showing an almost constant velocity difference between these lines, regardless of any fluctuations, indicating a consistently detectable diurnal signal of similar magnitude at all distances from the terminus (Figs.  and ).

However, a close inspection of the velocity change and acceleration data (Figs.  and ) shows that the timing of the diurnal velocity peak and minima also varies spatially along the centreline and is further analysed below.

During high-velocity days, diurnal velocity changes initiate at the terminus and propagate upstream, whereas commonly diurnal accelerations start at locations further upstream and propagate downstream over time. The initiation of acceleration in the morning and deceleration in the evening generally occurs with a lag of about two hours between the terminus and the location 6 km upstream. This is visible in the inclined acceleration/deceleration signal with time on Fig. c and Fig. c. For better readability, a zoomed-in representation with supporting labels is provided in the Appendix (Fig. ). In nearly half of the observed initiations, the velocity change starts earlier at the upper part, propagating downstream with time (manually detected and marked as orange boxes in Fig. c), whereas upstream propagations, when velocity changes are initiated at the front, only occur in 20 % of the cases (green boxes in same figure). The remaining 30 % labelled as “unclear” show either no clear propagation direction or a combination of an upstream and a downstream signal. Table  in the Appendix provides a complete overview of the acceleration transition types for both years.

Representative 30 min interval time-series of velocity variation maps illustrate in more detail the spatial patterns during propagation of such flow transitions (Fig. ). These transitions occur either from acceleration to deceleration in the evening (a and c), or from deceleration to acceleration in the morning (b and d). The more common downstream propagation to the terminus appears spatially smooth and uniform across the entire width of the glacier, apart from the almost stagnant parts beyond the shear margins (Fig. ). In contrast, on days with high flow velocities, such as for example during the speed-up event around 20 July 2024 (Fig. c), the transition signal generally starts at the front and propagates upstream. In these cases, propagation typically exhibits a more “block-like” spatial pattern (Fig. a, b). The outlines of these blocks align with major crevasses or rifts, which are oriented in north-westerly direction due to generally faster flow velocities towards the western margin of the glacier (see black dashed polygons in Fig. ). The velocities are significantly higher (with p<0.02 in both years) during upstream propagating transitions compared to downstream transitions.