Novaya Zemlya (NVZ) has experienced rapid ice loss and accelerated marine-terminating glacier retreat during the past 2 decades. However, it is unknown whether this retreat is exceptional longer term and/or whether it has persisted since 2010. Investigating this is vital, as dynamic thinning may contribute substantially to ice loss from NVZ, but is not currently included in sea level rise predictions. Here, we use remotely sensed data to assess controls on NVZ glacier retreat between 1973/76 and 2015. Glaciers that terminate into lakes or the ocean receded 3.5 times faster than those that terminate on land. Between 2000 and 2013, retreat rates were significantly higher on marine-terminating outlet glaciers than during the previous 27 years, and we observe widespread slowdown in retreat, and even advance, between 2013 and 2015. There were some common patterns in the timing of glacier retreat, but the magnitude varied between individual glaciers. Rapid retreat between 2000 and 2013 corresponds to a period of significantly warmer air temperatures and reduced sea ice concentrations, and to changes in the North Atlantic Oscillation (NAO) and Atlantic Multidecadal Oscillation (AMO). We need to assess the impact of this accelerated retreat on dynamic ice losses from NVZ to accurately quantify its future sea level rise contribution.
Glaciers and ice caps are the main cryospheric source of global sea level
rise and contributed approximately
Arctic ice loss occurs via two main mechanisms: a net increase in surface
melting, relative to surface accumulation, and accelerated discharge from
marine-terminating outlet glaciers (e.g. Enderlin et al., 2014; van den
Broeke et al., 2009). These marine-terminating outlets allow ice caps to
respond rapidly to climatic change, both immediately through calving and
frontal retreat (e.g. Blaszczyk et al., 2009; Carr et al., 2014; McNabb
and Hock, 2014; Moon and Joughin, 2008) and also through long-term drawdown
of inland ice, often referred to as “dynamic thinning”
(e.g. Price et al., 2011; Pritchard et al., 2009).
During the 2000s, widespread marine-terminating glacier retreat was observed
across the Arctic (e.g. Blaszczyk et al., 2009; Howat et al., 2008;
McNabb and Hock, 2014; Moon and Joughin, 2008; Nuth et al., 2007), and
substantial retreat occurred on Novaya Zemlya between 2000 and 2010
(Carr et al., 2014): retreat rates increased markedly from around
2000 on the Barents Sea coast and from 2003 on the Kara Sea (Carr et
al., 2014). Between 1992 and 2010, retreat rates on NVZ were an order of
magnitude higher on marine-terminating glaciers (
Location map, showing the study area and outlet glaciers.
Initially, surface elevation change data from NVZ suggested that there was no
significant difference in thinning rates between marine- and land-terminating
outlet glacier catchments between 2003 and 2009 (Moholdt et al., 2012). This
contrasted markedly with results from Greenland (e.g. Price et al., 2011;
Sole et al., 2008) but was similar to the Canadian Arctic, where the vast
majority of recent ice loss occurred via increased surface melting
(
In this paper, we use remotely sensed data to assess glacier frontal-position
change for all major ( At multi-decadal timescales, is there a significant difference in glacier
retreat rates according to (i) terminus type (land-, lake- or
marine-terminating); (ii) coast (Barents Sea versus Kara Sea coast); (iii) ice
mass (northern ice mass, Sub 1, or Sub 2); and (iv) latitude? Are outlet glacier retreat rates observed between 2000 and 2010 on NVZ
exceptional during the past Is glacier retreat accelerating, decelerating, or persisting at the same
rate? Can we link observed retreat to changes in external forcing (air
temperatures, sea ice, and/or ocean temperatures)?
Number of outlet glaciers contained within each category used to assess spatial variations in retreat rate, specifically coast, ice mass, and terminus type.
This paper focuses on the ice masses located on Severny Island, which is
the northern island of the Novaya Zemlya archipelago (Fig. 1). The northern
island ice cap contains the vast majority of ice (19 841 km
Outlet glacier frontal positions were acquired predominantly from Landsat
imagery. These data have a spatial resolution of 30 m and were obtained
freely via the United States Geological Survey (USGS) Global Visualization
Viewer (GloVis) (
Due to the lack of Landsat imagery during the 1990s, we use synthetic
aperture radar (SAR) image mode precision data during this period. The data
were provided by the European Space Agency, and we use European Remote-sensing
Satellite-1 (ERS-1) and ERS-2 products
(
Air temperature data were obtained from meteorological stations located on,
and proximal to, Novaya Zemlya (Fig. 1). Directly measured meteorological
data are very sparse on NVZ, and there are large gaps in the time series for
many stations. We use data from two stations, Malye Karmakuly (WMO ID: 20744)
and E. K. Fedorova (WMO ID: 20946), as these are the closest stations to the
study glaciers that have a comprehensive (although still not complete) record
during the study period (Supplement Table S2). The data were obtained from
the Hydrometeorological Information – World Data Centre Baseline
Climatological Data Sets (
Sea ice data were acquired from the Nimbus-7 SMMR and DMSP SSM/I-SSMIS
Passive Microwave data set (
Data on the North Atlantic Oscillation (NAO) were obtained from the Climatic
Research Unit (
We use ocean temperature data from the “Climatological Atlas of the Nordic
Seas and Northern North Atlantic” (Hurrell, 1995; Korablev et
al., 2014) (
We used a Kruskal–Wallis test to investigate statistical differences in
total retreat rate (1986–2015) for the different categories of outlet
glacier within our study population, i.e. terminus type (marine-, land-, and
lake-terminating), coast (Barents Sea and Kara Sea), and ice mass (northern
island ice cap, Sub 1, and Sub 2). The Kruskal–Wallis test is a
non-parametric version of the one-way ANOVA (analysis of variance) test and
analyses the variance using the ranks of the data values, as opposed to the
actual data. Consequently, it does not assume normality in the data, which
is required here, as Kolmogorov–Smirnov tests indicate that total retreat
rate (1986–2015) is not normally distributed for any of the glacier
categories (e.g. terminus type). This is also the case when we test for
normality at each of the four time intervals discussed below (1973/76–1986,
1986–2000, 2000–2013, and 2013–2015). The Kruskal–Wallis test gives a
We assessed the influence of glacier latitude on total retreat rate
(1986–2015), using simple linear regression. This fits a line to the data
points and gives an
Wilcoxon tests were used to assess significant differences in mean glacier
retreat rates between four time intervals: 1973/76–1986, 1986–2000,
2000–2013, and 2013–2015. These intervals were chosen through manual assessment of
apparent breaks in the data. For each interval, data were split according to
terminus type (marine, land, and lake), and marine-terminating glaciers were
further sub-divided by coast (Barents Sea and Kara Sea). For each category, we
then used the Wilcoxon test to determine whether mean retreat rates for all
of the glaciers during one time period (e.g. 1986–2000) were significantly
different from those for another time period (e.g. 2000–2013). The Wilcoxon
test was selected as it is non-parametric and our retreat data are not
normally distributed, and it is suitable for testing statistical difference
between data from two time periods (Miles et al., 2013).
As with the Kruskal–Wallis test, a
In order to further investigate the temporal pattern of retreat on Novaya Zemlya, we use statistical change-point analysis (Eckley et al., 2011). We applied this to our frontal-position data for marine- and lake-terminating glaciers, and to the sea ice and air temperature data. Land-terminating glaciers are not included, due to the much higher error margins compared to any trends, which could lead to erroneous change-points being identified. Change-point analysis allows us to automatically identify significant changes in the time series data and whether there has been a shift from one mode of behaviour to another (e.g. from slower to more rapid retreat) (Eckley et al., 2011). Formally, a change-point is a point in time where the statistical properties of prior data are different from the statistical properties of subsequent data; the data between two change-points are a segment. There are various ways that one can determine when a change-point should occur, but the most appropriate approach for our data is to consider changes in regression.
Box plots and Kruskal–Wallis test results for different glacier
terminus settings for
In order to automate the process, we use the cpt.reg function in the R EnvCpt package (Killick et al., 2016) with a minimum number of four data points between changes. This function uses the pruned exact linear time (PELT) algorithm (Killick et al., 2012) from the change-point package (Killick and Eckley, 2015) for fast and exact detection of multiple changes. The function returns change-point locations and estimates of the intercept and slope of the regression lines between changes. We give the algorithm no information on when we might be expecting a change or how large it may be, allowing it to automatically determine statistically different parts of the data. In this way, we use the analysis to determine whether, and when, retreat rates change significantly on each of the marine- and lake-terminating glaciers on NVZ, and whether there are any significant breaks in our sea ice and air temperature data. We also apply the change-point analysis to the number of sea-ice-free months, but as the data do not contain a trend, we identify breaks using significant changes in the mean, rather than a change in regression. Thus, we can identify any common behaviour between glaciers, determine the timing of any common changes, and compare this to any significant changes in atmospheric temperatures and sea ice concentrations.
The Kruskal–Wallis test was used to identify significant differences in
total retreat rate (1986–2015) for glaciers located in different settings.
First, terminus type was investigated. Results demonstrated that total
retreat rates (1986–2015) were significantly higher on lake- and
marine-terminating glaciers than those terminating on land, at a very high
confidence interval (
Linear regression of total retreat rate (1986–2015) versus glacier
latitude. Latitude was regressed against total glacier retreat rate for
Mean retreat rates for Novaya Zemlya outlet glaciers, and mean air
temperatures at E. K. Fedorova (WMO ID: 20946) and Malye Karmaku (WMO ID:
20744) (Fig. 1). Data are split into four time periods, based on manually
identified breaks in the glacier retreat data: 1973/76–1986, 1986–2000,
2000–2013, and 2013–2015.
We used simple linear regression to assess the relationship between total
retreat rate (1986–2015) and latitude, as there is a strong north–south
gradient in climatic conditions on NVZ, but no significant linear
relationship was apparent (
Relative glacier frontal position over time, from 1973 to 2015, for
Wilcoxon test results, used to assess significant differences in
retreat rates between each manually identified time interval (1976–1986,
1986–2000, 2000–2013, 2013, 2015). Retreat rate data were tested separately
for each terminus type, and marine-terminating glaciers were further
sub-divided by coast. Following convention,
Based on an initial assessment of the temporal pattern of retreat for
individual glaciers, we manually identified major break points in the data
and divided glacier retreat rates into four time intervals: 1973/76 to 1986,
1986 to 2000, 2000 to 2013, and 2013 to 2015 (Fig. 4). Data were separated
according to terminus type and, in the case of marine-terminating glaciers,
according to coast. We then used the Wilcoxon test to evaluate the
statistical difference between these time periods for each category (Table 2). For land- and lake-terminating glaciers, there were no significant
differences in retreat rates between any of the time periods (Fig. 4; Table 2). Indeed, retreat rates on lake-terminating glaciers were remarkably
consistent between 1986 and 2015, both over time and between glaciers (Figs. 4 and 5). For marine-terminating glaciers on the Barents Sea coast, the
periods 1973/76–1986 and 1986–2000 were not significantly different from
each other, and mean retreat rates were comparatively low (
Results of the change-point analysis for glacier retreat rates and
climatic controls. Red dots indicate the start of a significantly different
period in the time series data, and grey dots represent the end of the
previous period, with grey dashed lines connecting the two. This is done to
account for missing data: we know that the change-point occurred between the
grey and the red dot, and that the new phase of behaviour occurred from the
red dot onwards, but not the exact timing of the change. Blue dots show the
start of a second significant change in the time series. Frontal-position
data were analysed separately for marine-terminating outlets on the Barents
Sea coast
Following our initial analysis, we used change-point analysis to further assess the temporal patterns of glacier retreat, by identifying the timing of significant breaks in the data. On the Barents Sea coast, five glaciers underwent a significant change in retreat rate from the early 1990s onwards (Fig. 6). Of these, retreat rates on four glaciers (MAK, TAI2, VEL, and VIZ; see Fig. 1 for glacier locations and names) subsequently increased, whereas retreat was slower on INO between 1989 and 2006. The most widespread step change on the Barents Sea coast occurred in the early 2000s, after which nine glaciers retreated more rapidly (Fig. 6). A second widespread change in glacier retreat rates occurred in the mid-2000s, which was also the second change-point for four glaciers (Fig. 6). Of these eight glaciers, only VOE retreated more slowly after the mid-2000s change-point. On the Kara Sea coast, we see a broadly similar temporal pattern, with two glaciers showing a significant change in retreat rate from the early 1990s and again in 2005 and 2007 (Fig. 6). In the case of MG, retreat rates were higher after each breakpoint, whereas for SHU1 retreat rates were lower between the 1990s and mid-2000s. Four glaciers began to retreat more rapidly from 2000 onwards, and five other glaciers showed a significant change in retreat rates beginning between 2005 and 2010 (Fig. 6), with VER being the only glacier to show a reduction in retreat rates after this change (Fig. 6). Focusing on lake-terminating glaciers, a significant change in retreat rates began between 2006 and 2008 on all but one glacier, which began to retreat more rapidly from 2004 onwards (Fig. 6).
At E. K. Fedorova, mean annual air temperatures were significantly warmer
in 2000–2012 (
Mean retreat rates for Novaya Zemlya outlet glaciers, and seasonal
mean sea ice concentrations and number of ice-free months for the Barents
Sea and Kara Sea coasts. Data are split into four time periods, based on manually
identified breaks in the glacier retreat data: 1973/76–1986, 1986–2000,
2000–2013, and 2013–2015.
In the ERA-Interim reanalysis data, mean annual air temperatures increased
significantly between 1986–1999 and 2000–2012 at both the surface and
850 hPa pressure level (Table 3). Winter (surface) and autumn (850 hPa)
temperatures also warmed significantly between these time intervals (Table 3). Surface air temperatures were significantly warmer in 2013–2015
than in 1986–1999, in winter and annually (Table 3). No significant
differences in air temperatures were observed at either height between
2000–2012 and 2013–2015 for any season (Table 3). Surface air temperatures
were comparable between 2000–2012 and 2013–2015 in winter and autumn, and
somewhat warmer in spring (
On the Barents Sea coast, sea ice concentrations during all seasons were significantly lower in 2000–2012 than in 1976–1985 or 1986–1999, as was the number of ice-free months (Fig. 7; Table 4). Between 1976–1985 and 2000–2012, mean winter sea ice concentrations reduced from 68 to 35 %, mean spring values declined from 59 to 28 %, and mean autumn averages fell from 27 to 7 % (Fig. 7). Mean summer sea ice concentrations reduced slightly, from 12 to 5 % (Fig. 7). Over the same time interval, the number of ice-free months increased from 3.0 to 6.9 (Fig. 7). Summer sea ice concentrations on the Barents Sea coast reduced significantly between 2000–2012 and 2013–2015, but no significant change was observed in any other month, nor in the number of ice-free months (Fig. 7; Table 4). With exception of winter, sea ice concentrations were significantly lower in 2013–2015 than in 1976–1985 or 1986–1999 (Fig.4; Table 4). As on the Barents Sea coast, sea ice concentrations on the Kara Sea were significantly lower in all seasons in 2000–2012 than in 1976–1985 or 1986–1999 (Fig. 7; Table 4). Summer mean sea ice concentrations declined from 25 % in 1976–1985 to 13 % in 2000–2012 (Fig. 7). Over the same time interval, autumn mean concentrations reduced from 56 to 33 %, spring values declined from 87 to 73 %, and winter values decreased from 87 to 79 % (Fig. 7). The number of ice-free months also reduced from 1.6 (1976–1985) to 3.0 (2000–2012) (Fig. 7). No significant differences were apparent between seasonal sea ice concentrations and the number of ice-free months in 2013–2015 and any other time period, with the exception of summer sea ice concentrations between 1976–1985 and 2013–2015 (Table 4).
Time series of
Focusing on the change-point analysis, we see a significant change in air temperatures at E. K. Fedorova from 2008 onwards, after which air temperatures increased markedly (Fig. 6). On the Barents Sea coast, we observe significant breaks in summer sea ice concentrations at 2000 and 2008: before 2000, summer sea ice showed a downward trend but large interannual variability; between 2000 and 2008, there was a slight upward trend and much lower variability; and from 2008 onwards, summer sea ice concentrations were much lower, showing both a downward trend and limited interannual variability (Supplement Fig. S2). From 2005 onwards, we observed much lower interannual variability in spring, summer, and autumn sea ice concentrations (Supplement Fig. S2). After 2005, summer sea ice concentrations on the Kara Sea coast showed much smaller interannual variability and had lower values (Supplement Fig. S3). The number of ice-free months increased significantly on both the Kara Sea (from 2003) and Barents Sea (from 2005) (Fig. 6).
Ocean temperatures from the “Climatological Atlas of the Nordic
Seas and Northern North Atlantic” (Korablev et al., 2014) at
Between 1970 and 1989, the summer and annual NAO index were largely
positive, with a few years of negative values (Fig. 8a). From 1989 to 1994,
values were all positive, followed by strongly negative values in 1995 (Fig. 8a). Subsequently, the summer and annual NAO index remained weakly negative
between 1999 and 2012, with values becoming increasingly negative in the
final 5 years of this period (Fig. 8a). In 2013, the NAO index became
strongly positive, particularly during summer, and values were also positive
in 2015 and 2016 (Fig. 8a). The AO index follows an overall similar pattern
to the NAO until
Glaciers identified as surging during the study period, based on the
surge criteria compiled by Grant et al. (2009).
At the broad spatial scale, data indicate that surface ocean temperatures
have warmed in the Barents Sea over time (Fig. 9). Warming was particularly
marked in the area extending approximately 100 km offshore of the Barents
Sea coast and south of 76
During the study period, we observed three glaciers surging: ANU, MAS, and SER (Fig. 1). These were excluded from the analysis of glacier retreat rates and are discussed separately here. ANU has previously been identified as possibly surge type, based on the presence of looped moraine (Grant et al., 2009). Here, we identify an active surge phase, on the basis of a number of characteristics identified from satellite imagery and following the classification of Grant et al. (2009): rapid frontal advance, heavy crevassing, and digitate terminus. High flow speeds are also evident close to the terminus (Melkonian et al., 2016), which is consistent with the active phase of surging. Our results show that advance began in 2008 and was ongoing in 2015, with the glacier advancing 683 m during this period (Fig. 10a). MAS was previously confirmed as surge type (Grant et al., 2009), and our data suggest that its active phase persisted between 1989 and 2007 (Fig. 10a). The imagery indicates that surging on MAS originated from the eastern limb of the glacier, which may be partially fed by the neighbouring glacier (Fig. 10b–f). The exact timing of this tributary surge is uncertain, but imagery from 1985 (Fig. 10c) shows limited evidence of surging, whereas a number of surge indicators are clearly visible by 1988, including looped moraines and rapid advance (Fig. 10d), suggesting it began in the late 1980s. The tributary glacier then advanced into the eastern margin of the main outlet of MAS, causing it to advance, and produced heavy crevassing on the eastern portion of its terminus (Fig. 10d and e). The main terminus of MAS reached its maximum extent for the study period in 2007, and the tributary continued advancing from the 1980s until 2007 (Fig. 10f). The role of the tributary glacier in triggering the surge is consistent with the lack of signs of surge type behaviour on the western margin of MAS and considerable visible displacement of ice and surface features on the eastern tributary (Fig. 10b–f). SER was also confirmed as a surge-type glacier by Grant et al. (2009), who suggested that glacier advance occurred between 1976/77 and 2001. Our results indicate that advance began somewhat later, sometime between July 1983 and July 1986, and ended before August 2000 (Fig. 10a).
Our results demonstrate that retreat rates on marine-terminating outlet
glaciers (
Our data showed no significant difference in total retreat rates for
marine-terminating (
One potential explanation for the common behaviour of the lake-terminating outlet glaciers on NVZ is that retreat may be dynamically controlled and sustained by a series of feedbacks once it has begun. As observed on large Greenlandic tidewater glaciers, initial retreat may bring the terminus close to floatation, leading to faster flow and thinning, which promote further increases in calving and retreat (e.g. Howat et al., 2007; Hughes, 1986; Joughin et al., 2004; Meier and Post, 1987; Nick et al., 2009). This has been suggested as a potential mechanism for the rapid recession for Upsala Glacier in Patagonia (Sakakibara and Sugiyama, 2014) and Yakutat Glacier, Alaska (Trüssel et al., 2013). However, rapid retreat was not observed on all lake-terminating glaciers in Patagonia (Sakakibara and Sugiyama, 2014), and the potential for these feedbacks to develop depends on basal topography (e.g. Carr et al., 2015; Porter et al., 2014; Rignot et al., 2016). Consequently, the basal topography would need to be similar for each of the NVZ glaciers to explain the very similar retreat patterns, which is not implausible but perhaps unlikely. Alternatively, it may be that the proglacial lakes act as a buffer for atmospheric warming, due the greater thermal conductivity of water relative to air, and so reduce variability in retreat rates. Furthermore, lake-terminating glaciers are not subject to variations in sea ice and ocean temperatures, which may account for their more consistent retreat rates, compared to marine-terminating glaciers (Figs. 4 and 5). In order to differentiate between these two explanations, data on lake temperature changes during the study period and lake bathymetry would be required. However, neither are currently available, and we highlight this as an important area for further research, given the rapid recession observed on these lake-terminating glaciers.
For the period between 1986 and 2015, we find no significant difference in retreat rates between the Barents Sea and Kara Sea coasts (Fig. 2). This is contrary to the results of a previous shorter-term study, which showed that retreat rates on the Barents Sea coast were significantly higher than on the Kara Sea between 1992 and 2010 (Carr et al., 2014) and the higher thinning rates observed on marine outlets on the Barents Sea coast (Melkonian et al., 2016). Furthermore, there are substantial differences in climatic and oceanic conditions on the two coasts (Figs. 4 and 7) (Pfirman et al., 1994; Politova et al., 2012; Przybylak and Wyszyński, 2016; Zeeberg and Forman, 2001), so we would expect to see significant differences in outlet glacier retreat rates. This indicates that longer-term glacier retreat rates on NVZ may relate to much broader, regional-scale climatic change, which is supported by the widespread recession of glaciers across the Arctic during the past 2 decades (e.g. Blaszczyk et al., 2009; Carr et al., 2014; Howat and Eddy, 2011; Jensen et al., 2016; Moon and Joughin, 2008). One potential overarching control on NVZ frontal positions is fluctuations in the NAO, which covaries with Northern Hemisphere air temperatures, Arctic sea ice, and North Atlantic Ocean temperatures (Hurrell, 1995; Hurrell et al., 2003; IPCC, 2013). More recent work has also recognized the influence of the AMO on oceanic and atmospheric conditions in the Barents Sea and broader North Atlantic (Drinkwater et al., 2013; Oziel et al., 2016). Our data suggest that the major phases of frontal-position change on NVZ correspond to changes in the NAO and AMO (Fig. 8; Sect. 4.2): rapid retreat between 2000 and 2013 coincides with a weakly negative NAO and positive AMO, following almost 3 decades characterized by a generally positive NAO and negative AMO (Fig. 8). As such, these large-scale changes may overwhelm smaller-scale spatial variations between the two coasts of NVZ when retreat is considered on multi-decadal timescales.
Marine-terminating outlet glacier retreat rates do not show a linear relationship with latitude, and there is considerable scatter when the two variables are regressed (Fig. 3). This may be due to the influence of fjord geometry on glacier response to climatic forcing (Carr et al., 2014) and the capacity for warmer ocean waters to access the calving fronts. In contrast, southerly land-terminating outlets retreat more rapidly than those in the north, which we attribute to the substantial latitudinal air temperature gradient on NVZ (Zeeberg and Forman, 2001). Conversely, lake-terminating glaciers retreat more rapidly at more northerly latitudes (Fig. 3), which we speculate may relate to the bathymetry and basal topography of individual glaciers, but data are not currently available to confirm this.
Our results show that retreat rates on marine-terminating outlet glaciers on NVZ were significantly higher between 2000 and 2013 than during the preceding 27 years (Fig. 4). At the same time, land-terminating outlets experienced much lower retreat rates and did not change significantly during the study period (Figs. 4 and 5). This is consistent with studies from elsewhere in the Arctic, which identified the 2000s as a period of elevated retreat on marine-terminating glaciers (e.g. Blaszczyk et al., 2009; Howat and Eddy, 2011; Jensen et al., 2016; Moon and Joughin, 2008) and increasing ice loss (e.g. Gardner et al., 2013; Lenaerts et al., 2013; Moholdt et al., 2012; Nuth et al., 2010; Shepherd et al., 2012). As discussed above, recent evidence suggests that glacier retreat in the Russian high Arctic can trigger substantial dynamic thinning and ice acceleration (Melkonian et al., 2016; Willis et al., 2015), but it not currently incorporated into predictions of 21st-century ice loss from the region (Radić and Hock, 2011). Consequently, the period of higher retreat rates during the 2000s may have a much longer-term impact on ice losses from NVZ, and this needs to be quantified and incorporated into forecasts of ice loss and sea level rise prediction.
Within the decadal patterns of glacier retreat, we observe clusters in the
timing of significant changes in marine-terminating glacier retreat rates
(Fig. 6). Specifically, we see breaks in the frontal-position time series on
both the Barents Sea and Kara Sea coasts in the early 1990s,
Between the 1950s and mid-1990s, positive phases of the NAO were associated
with the influx of warm Atlantic water into the Barents Sea (Hurrell,
1995; Loeng, 1991) and increased penetration of Atlantic cyclones and air
masses into the region, which lead to elevated air temperatures and
precipitation (Zeeberg and Forman, 2001). Conversely, negative NAO
phases were associated with cooler oceanic and atmospheric conditions in the
Barents Sea (Zeeberg and Forman, 2001). During this period, therefore,
the impact of the NAO was opposite in the Barents Sea and in western
portions of the Atlantic-influenced Arctic (e.g. the Labrador Sea)
(Drinkwater et al., 2013; Oziel et al., 2016). However,
since the mid-1990s, changes in the Barents Sea and the western Atlantic
Arctic have been in phase, and warming and sea ice reductions have been
widespread across both regions (Drinkwater et al., 2013;
Oziel et al., 2016). As such, increased glacier retreat rates on NVZ during
the 2000s (Figs. 4 and 5) may have resulted from the switch to a weaker, and
predominantly negative, NAO phase from the mid-1990s (Fig. 8), which would
promote warmer air and ocean temperatures, and reduced sea ice, as we
observe in our data (Figs. 4 and 7). Previous studies have suggested a 3–5-year lag between NAO shifts and changes in conditions on NVZ, due to the
time required for Atlantic water to transit into the Barents Sea
(Belkin et al., 1998; Zeeberg and Forman, 2001), which is consistent
with the onset of retreat in
Following higher retreat rates in the 2000s, our data indicate that marine-terminating glacier retreat slowed from 2013 onwards on both the Barents Sea and Kara Sea coasts, with several glaciers beginning to re-advance (Figs. 4 and 5). Our data demonstrate that marine-terminating glaciers on NVZ have previously undergone a step-like pattern of retreat, with short (1–2 year) pauses in retreat (Fig. 5). Thus, it is unclear whether this reduction in retreat rates is another temporary pause, before continued retreat, or the beginning of a new phase of reduced retreat rates. One possible explanation for reduced retreat rates on both coasts of NVZ is the stronger NAO values observed from the late 2000s onwards: winter 2009/10 had the most negative NAO for 200 years (Delworth et al., 2016; Osborn, 2011), and values were strongly positive in 2013 (Fig. 8a). This is consistent with the 3–5-year lag required for NAO-related changes in Atlantic water inflow to reach NVZ (Zeeberg and Forman, 2001), and so we speculate that reduced glacier retreat rates from 2013 onwards (Figs. 4 and 5) may relate to an increase in the influence of the NAO, relative to the AMO, from the late 2000s (Fig. 8). Evidence indicates that the impact of the NAO in the Barents Sea is now in phase with the western North Atlantic (Drinkwater et al., 2013; Oziel et al., 2016), and so a more positive NAO could lead to cooler conditions on NVZ and hence glacier advance. However, the relationship between large-scale features, such as the NAO and AMO; ocean conditions; and glacier behaviour is complex (Drinkwater et al., 2013; Oziel et al., 2016), and the period of glacier advance/reduced retreat on NVZ has lasted only 2 years. Consequently, further monitoring is required to determine whether this represents a longer-term trend or a short-term change and to confirm its relationship to synoptic climatic patterns.
Despite the changes in the NAO and AMO, our data show no significant change
in sea ice concentrations, nor the length of the ice-free season, between
2000–2012 and 2013–2015 on either the Barents Sea or Kara Sea coast
(Table 4; Fig. 7). Likewise, we see no significant change in winter
(January–March) air temperatures at E. K. Fedorova (Table 3; Fig. 4) nor in
the ERA-Interim data during any season (Table 3; Fig. 4). Although not
significant, we see summer warming of 0.7
Although we observe some common behaviour, in terms of the approximate timing and general trend in retreat, there is still substantial variability in the magnitude of retreat between individual marine-terminating glaciers (Figs. 4 and 5). Furthermore, not all glaciers shared common change-points, and certain outlets showed a different temporal pattern of retreat to the majority of the study population (Figs. 4–6). For example, INO retreated more slowly between 1989 and 2006 than during the 1970s and 1980s. We attribute these differences to glacier-specific factors and, in particular, the fjord bathymetry and basal topography of individual glaciers. Previous studies have highlighted the impact of fjord width on retreat rates on NVZ (Carr et al., 2014) and basal topography on marine-terminating glacier behaviour elsewhere (e.g. Carr et al., 2015; Porter et al., 2014; Rignot et al., 2016). This may result from the influence of fjord geometry on the stresses acting on the glacier once it begins to retreat: as a fjord widens, lateral resistive stresses will reduce, and the ice must thin to conserve mass, making it more vulnerable to calving (Echelmeyer et al., 1994; Raymond, 1996; van der Veen, 1998a, b), whilst retreat into progressively deeper water can cause feedbacks to develop between thinning, floatation, and retreat (e.g. Joughin and Alley, 2011; Joughin et al., 2008; Schoof, 2007). Thus, retreat into a deeper and/or wider fjord may promote higher retreat rates on a given glacier, even with common climatic forcing. In addition, differences in fjord bathymetry may determine whether warmer Atlantic water can access the glacier front (Porter et al., 2014; Rignot et al., 2016), which could promote further variations between glaciers. This highlights the need to collect basal topographic data for NVZ outlet glaciers, which is currently very limited but a potentially key control on ice loss rates.
Our data demonstrate that air temperatures were very substantially warmer between 2000 and 2012 than during the preceding decades and that sea ice concentrations were also much lower on both the Barents Sea and Kara Sea coasts during this period (Figs. 4 and 7). This is consistent with the atmospheric warming reported across the Arctic during the 2000s (e.g. Carr et al., 2013a; Hanna et al., 2013, 2012; Mernild et al., 2013) and the well-documented decline in Arctic sea ice (Comiso et al., 2008; Kwok and Rothrock, 2009; Park et al., 2015). As such, the decadal patterns of marine-terminating outlet glacier retreat correspond to decadal-scale climatic change on NVZ (Figs. 4 and 7), and exceptional retreat during the 2000s coincided with significantly warmer air temperatures and lower sea ice concentrations (Tables 2 and 3). Interestingly, step changes in the air temperature and sea ice data identified by the change-point analysis did not correspond to significant changes in outlet glacier retreat rates (Fig. 6), suggesting that such changes may not substantially influence retreat rates or that the relationship may be more complex, e.g. due to lags in glacier response.
The much lower retreat rates on land-terminating outlets (Fig. 4) may indicate an oceanic driver for retreat rates on marine-terminating glaciers. Previous studies have identified sea ice loss as a potentially important control on NVZ retreat rates (Carr et al., 2014), which fits with observed correspondence between sea ice loss and retreat, but it is unclear whether the two variables simply co-vary or whether sea ice can drive ice loss, by extending the duration of seasonally high calving rates (e.g. Amundson et al., 2010; Miles et al., 2013; Moon et al., 2015). The available ocean data indicate that temperatures were substantially warmer during the 2000s (Fig. 9), which would provide a plausible mechanism for widespread retreat on both coasts of NVZ (Fig. 4). However, oceanic data for the 2000s is sparse in the Barents and Kara seas, compared to previous decades, so it is difficult to ascertain the magnitude and spatial distribution of warming and to link it directly with glacier retreat patterns. Lake-terminating glaciers are not affected by changes in sea ice or ocean temperatures but could be influenced by air temperatures. However, despite much higher air temperatures in the 2000s, mean retreat rates on lake-terminating outlet glaciers were similar for each decade of the study (Fig. 4), suggesting that the relationship is not straightforward. Instead, the presence of lakes may at least partly disconnect these glaciers from climatic forcing, by buffering the effects of air temperatures changes and/or by sustaining dynamic changes, following initial retreat (Sakakibara and Sugiyama, 2014; Trüssel et al., 2013).
During the study period, we identify three actively surging glaciers, based
on various lines of glaciological and geomorphological evidence
(Copland et al., 2003; Grant et al., 2009), including
terminus advance (Fig. 10). Frontal advance persisted for 18 years on MAS
and 15 years on SER, whilst ANU began to advance in 2008, and
this continued until the end of the study period (Fig. 10a). This is
comparatively long for surge-type glaciers, which usually undergo short
active phases over time frames of months to years (Dowdeswell
et al., 1991; Raymond, 1987). For comparison, surges on Tunabreen,
Spitzbergen, last only
During the active phase of the NVZ surge glaciers, we observe large sediment plumes emanating from the glacier terminus (Fig. 10g), which indicates that at least part of the glacier bed is warm-based during the surge. Together with the comparatively long surge interval, this supports the idea that changes in thermal regime may drive glacier surging on NVZ, as hypothesized for certain Svalbard glaciers (Dunse et al., 2015; Murray et al., 2003; Sevestre et al., 2015). In addition, the surge of MAS appears to have been triggered by a tributary glacier surging into its lateral margin (Fig. 10b–f). This demonstrates an alternative mechanism for surging, aside from changes in the thermal regime and/or hydrology conditions of the glacier, which has not been widely observed but will depend strongly on the local glaciological and topographical setting of the glacier. The data presented here focus only on frontal advance and glaciological/geomorphological evidence, whereas information on ice velocities is also an important indicator of surging (Sevestre and Benn, 2015). Consequently, information on velocity and surface elevation changes are needed to further investigate the surge cycle and its possible controls on NVZ. This is important, as NVZ is thought to have conditions that are highly conducive to glacier surging (Sevestre and Benn, 2015) but has a long surge interval. We therefore want to ensure that we can disentangle surge behaviour and the impacts of climate change on NVZ.
At multi-decadal timescales, terminus type remains a major overarching
determinant of outlet glacier retreat rates on NVZ. As observed elsewhere in
the Arctic, land-terminating outlets retreated far more slowly than those
ending in the ocean. However, we see no significant difference in retreat
rates between ocean- and lake-terminating glaciers, which contrasts with
findings in Patagonia. Retreat rates on lake-terminating glaciers were
remarkably consistent between glaciers and over time, which may result from
the buffering effect of lake temperature and/or the impact of lake
bathymetry, which could facilitate rapid retreat that is largely independent
of climate forcing, after an initial trigger. We cannot differentiate
between these two scenarios with currently available data. Retreat rates on
marine-terminating glaciers were exceptional between 2000 and 2013, compared
to previous decades. However, retreat slowed on the vast majority of
ocean-terminating glaciers from 2013 onwards, and several glaciers advanced,
particularly on the Barents Sea coast. It is unclear whether this represents
a temporary pause or a longer-term change, but it should be monitored in the
future, given the potential for outlet glaciers to drive dynamic ice loss
from NVZ. The onset of higher retreat rates coincides with a more negative,
weaker phase of the NAO and a more positive AMO, whilst reduced retreat
rates follow stronger NAO years. This suggests that synoptic atmospheric and
oceanic patterns may influence NVZ glacier behaviour at decadal timescales.
Marine-terminating glaciers showed some common patterns in terms of the
onset of rapid retreat (1990s,
The primary dataset created by the paper is the glacier frontal position data, which are provided in the Supplement. Shapefiles of the data can be provided on request to the lead author. Other datasets (e.g. climate data) are available online, and the sources are given in the paper.
The authors declare that they have no conflict of interest.
The authors thank Xavier Fettweis, Robbert McKnabb, and one anonymous reviewer for their constructive comments that helped to improve the manuscript. Edited by: Xavier Fettweis Reviewed by: Robert McNabb and one anonymous referee