The mass balance of the Greenland Ice Sheet (GrIS) in a warming climate is
of critical interest in the context of future sea level rise. Increased
melting in the GrIS percolation zone due to atmospheric warming over the
past several decades has led to increased mass loss at lower elevations.
Previous studies have hypothesized that this warming is accompanied by a
precipitation increase, as would be expected from the Clausius–Clapeyron
relationship, compensating for some of the melt-induced mass loss throughout
the western GrIS. This study tests that hypothesis by calculating snow
accumulation rates and trends across the western GrIS percolation zone,
providing new accumulation rate estimates in regions with sparse in situ data or
data that do not span the recent accelerating surface melt. We present
accumulation records from sixteen 22–32 m long firn cores and 4436 km of
ground-penetrating radar, covering the past 20–60 years of accumulation,
collected across the western GrIS percolation zone as part of the Greenland
Traverse for Accumulation and Climate Studies (GreenTrACS) project. Trends
from both radar and firn cores, as well as commonly used regional climate
models, show decreasing accumulation rates of
Greenland Ice Sheet (GrIS) mass loss has accelerated over the past few
decades, with modern mass loss rates more than double those from Antarctica
(van den Broeke et al., 2016). The 2010–2018
GrIS mass loss was calculated as
Enhanced GrIS surface melt is driven fundamentally by positive Greenland
summer temperature trends of
In addition to measuring snow accumulation rates with ice cores and automated snow depth sensors, several studies have used ground-based and airborne radar to calculate GrIS accumulation rates and trends (e.g., Medley et al., 2013; Spikes et al., 2004; Hawley et al., 2014; Koenig et al., 2016). We build upon these previous studies by collecting ground-penetrating radar (GPR) data across the lower percolation zone of western Greenland, where airborne radargrams are often obscured by refrozen melt percolation (Nghiem et al., 2005). The in situ GPR used in this study operates using a ultra-high-frequency (UHF) pulsed radar, while other systems such as frequency-modulated continuous-wave (FMCW) radars use phase-sensitive antennas that include both amplitude and phase information. By having our GPR antenna coupled with the snow, we avoid losing energy, and therefore penetration depth, from a strong reflection off the snow–air interface.
In addition to temperature–precipitation relationships through the Clausius–Clapeyron equation, previous studies have analyzed the dynamic climate controls on Greenland precipitation. Mernild et al. (2014), Auger et al. (2017), and Lewis et al. (2017) have hypothesized that a positive Atlantic Multidecadal Oscillation (AMO) index correlates with rising accumulation rates over most of the GrIS interior since higher sea surface temperatures increase moisture flux over the GrIS and induce greater snowfall. In addition, high-pressure (blocking) systems east of Greenland tend to deflect eastward-moving storms over central Greenland and increase precipitation, whereas blocking directly over Greenland or in Baffin Bay has the potential to decrease accumulation rates over the GrIS by displacing the polar jet stream and corresponding storm tracks equatorward (Auger et al., 2017). Over the 1991–2015 period there has been particularly strong summertime Greenland blocking (Hanna et al., 2016), but its effects on GrIS accumulation rates have not been determined with in situ data.
Here we develop new accumulation records across the western GrIS percolation zone using 16 firn cores and 4436 km of GPR data collected during an over-ice traverse spanning two field seasons. We evaluate the veracity of the accumulation data through comparisons of our firn core time series with previous measurements. We quantify multi-year trends in accumulation rates across western Greenland to test the hypothesis that precipitation has recently increased from the Clausius–Clapeyron relationship and higher GrIS temperatures. Further, we assess the ability of regional climate models (RCMs) to capture the year-to-year variability and multi-year trends in western GrIS accumulation rates. Finally, we evaluate relationships between recent accumulation rate trends and atmospheric circulation patterns, particularly changes in storm tracks.
This study uses data from the 2016–2017 Greenland Traverse for
Accumulation and Climate Studies (GreenTrACS), which measured accumulation
rates and melt across the western GrIS percolation zone over two summer
snowmobile traverses (closely following the 2150 m a.s.l. elevation
contour). The May–June 2016 season traversed 860 km from Raven–Dye-2
northward to Summit Station, while the May–June 2017 traverse made a 1230 km clockwise loop starting and ending at Summit Station
(Fig. 1). This paper focuses on accumulation
rates derived from 400 MHz GPR data collected along the entire traverse
path, as well as 16 shallow (22–32 m deep) firn cores spaced 40–100 km apart along the backbone of the traverse
(Fig. 1). Firn Cores 1–7 were collected in 2016
and Cores 8–16 were collected in 2017. We returned to the Core 7 location
at the beginning of the 2017 traverse to recover a weather station and to
connect the two season's GPR data. Additionally, we collected GPR data
Average accumulation rates across the GreenTrACS traverse for the length
of each record showing the location of each firn core, ACT11d, D4, D5, Katie
(K), Raven–Dye-2, and Sandy (S) ice cores, and Summit Station. Transect
During the 2016 traverse we collected GPS data using a Trimble NetR8
reference receiver with a Zephyr Geodetic antenna mounted to a Nansen sled
During the 2017 traverse we used GPS data from a Garmin 19x GPS receiver
wired directly to the GPR instrument, which recorded position data at every
radar sample with rms values of 3 m. During radar processing we average 75
adjacent traces, corresponding to a distance of
We develop a spatially continuous record of accumulation rates using GPR
profiles collected with Geophysical Survey Systems Inc. (GSSI) SIR-3000
(during 2016) and SIR-30 (during 2017) radar units with a 400 MHz antenna
(following Hawley et al., 2014). The antenna was
towed on the snow surface in a small plastic sled
Radargram showing the top 32 m of the transect along the main 2017 traverse from Core 13 to Core 15. Cores are indicated as red lines down to their final depth, with dates plotted every 5 years at corresponding depths. Traced internal reflecting horizons are shown as isochronous green lines. The depth scale on the vertical axis is calculated from the two-way travel time-to-depth conversion (see Sect. 2.4) for Core 13, although there is no visual difference in depth scale across this radargram.
Depending on signal attenuation within the firn column, IRHs can be traced
to a depth of 20–50 m (Fig. 2), providing
accumulation records over the past 20–60 years
(Fig. 3). For areas with high attenuation (i.e.,
shallow penetration of the radar signal), such as lower-elevation regions
with more refrozen melt layers, we calculate accumulation results for
shorter time periods. We are not able to trace as many IRHs to the west of
Cores 10–13 compared to the east due to higher signal attenuation (Fig. 3),
resulting in slightly different (less than 0.03 m w.e. a
We reduce the GPR data volume and signal noise by averaging 75 adjacent traces, which has the effect of suppressing random noise by the principle of trace stacking (Yilmaz, 2001). We apply a combination of median trace filtering, residual mean filtering (Gerlitz et al., 1993), and bandpass filtering using a butterworth design (Selesnick and Sidney Burrus, 1998) between 200 and 800 MHz. For data visualization, we apply an automatic gain control (Yilmaz, 2001) to give the interpreter more confidence when picking IRHs.
The amount of snow mass and the time span between IRHs are necessary to calculate accumulation rates from the GPR profiles. The accumulation rate is a function of the depth–age scale, travel time–depth conversion rate, and the firn density profile. We obtain the depth–age and depth–density scales from each of the shallow firn cores collected along the GreenTrACS traverse, and from density models based on temperature and accumulation rate data.
The 16 firn cores were drilled using an Ice Drilling Program hand auger
with a Kyne Sidewinder attachment (see Graeter et al.,
2018). We sampled the firn cores for chemical measurements using a continuous
ice core melter system with discrete sampling at Dartmouth College
(Osterberg et al., 2006). We used an
Abakus (Markus Klotz GmbH) laser particle detector to measure microparticle
concentrations and size distribution from the continuous ice core meltwater
stream, a Dionex model ICS5000 capillary ion chromatograph to measure major
ion (
We determined depth–age curves by identifying annual layers based on robust
seasonal oscillations in
We determine depth–age curves for each core by identifying annual layers
based on seasonal oscillations in
At each firn core and at the ends of each spur, we measured the density in
the top meter of snow using a 1000 cm
Date of oldest resolvable internal reflecting horizon throughout the entire GreenTrACS traverse route. Anomalously young ages from Core 7 to Summit are due to equipment malfunction.
After collecting each firn core, we measured borehole temperature for 24–48 h using a 20 m long thermistor string. We estimate mean annual
temperature from the deepest thermistor on the 20 m long thermistor string.
These measurements agree with MODIS satellite-derived mean annual surface
temperature (Hall et al., 2012) to within
Depth–density profile along the main 2016 traverse used for calculation of electromagnetic wave velocity and accumulation rates in this study. Densities are linearly interpolated between the two nearest cores and are modeled using Herron–Langway profiles below the depth of each core. The left and right boundary data come from the Raven–Dye-2 and Summit firn cores, respectively. Ice layers in Cores 1–5 are clearly visible as red lenses, but their extent is, in reality, likely more localized.
As shown in Fig. 4, ice layers within several firn cores are extrapolated laterally along the traverse, although these dense lenses are typically both localized and heterogeneous at these elevations (Brown et al., 2011; Rennermalm et al., 2013). Numerous studies have documented the heterogeneity of firn throughout the percolation zone and the complications of calculating SMB due to ice pipes and lenses (Brown et al., 2011, 2012; de La Peña et al., 2015). Here we attempt to accurately calculate accumulation rates using interpolated firn cores and in situ GPR throughout this complicated region. Our ice lens density interpolation is as accurate as possible between firn cores without additional in situ data, and this estimation does not significantly alter our results, as discussed in Sect. 2.6, since the ice layers represent a small fraction of the total depth to IRHs.
We convert the radar travel time to depth by iteratively multiplying the
velocity of the electromagnetic wave by the signal's one-way travel time to
each internal reflecting horizon (IRH). The electromagnetic speed of the radar wave,
We manually select 10 clear, strong IRHs spaced approximately 5 years apart
to consistently trace from Raven–Dye-2 to Summit Station and throughout the
2017 main traverse (Fig. 2). We trace each layer
manually by visually identifying strong amplitude peaks throughout the
radargram, starting with the 2016 layer and working downwards. We use a
spline interpolation between manual picks to trace each layer along large
amplitude reflections every
We trace layers between cores using a connect-the-dots approach using the depth–age scale at each firn core. After tracing layers from one firn core to the next, we check that layers intersect the core location at the proper depth for the age of our traced IRH. Note that the depths of several layers at Cores 2–16 are located below the bottom depth of those cores. Since these layers are isochronous, they are used to calculate accumulation rates over appropriate time epochs by using dates obtained from intersections with other cores (see Fig. 3).
Finally, we calculate snow accumulation rates using the firn core depth–age
scales, measured and interpolated depth–density profiles
(Fig. 4), and traced IRHs
(Fig. 2). We calculate the water-equivalent
accumulation rate
We perform a leave-one-out cross validation to calculate accumulation errors
at locations where we do not have firn core density profiles. Here we choose
one of the 16 firn cores, in addition to the Raven–Dye-2 and Summit
cores, to omit from our density interpolation
(Fig. 4), so that we interpolate density profiles
between adjacent firn cores and a Herron–Langway profile at the missing core
location. We find maximum single-epoch errors of 0.079 m w.e. a
Accumulation rates from GPR and collected firn cores (this study) compared with cores from the PARCA Campaign. Thin lines represent annual PARCA (blue) and GreenTrACS (black) firn core accumulation rates, while thick lines are 5-year averages over corresponding GPR epochs. Error bars represent one standard deviation over each epoch. GPR and PARCA accumulation rate averages and decadal trends are statistically indistinguishable.
Similarly, we perform a leave-one-out validation by omitting a firn core
density profile location entirely and interpolating density profiles over a
larger distance (e.g., between Core 1 and Core 3). In this case we find
maximum single-epoch errors of 0.057 m w.e. a
We conservatively take our accumulation error from missing density
measurements to be 0.079 m w.e. a
We assume uncertainty in dating the firn cores from annual variations in
chemistry to be
We calculate the total uncertainty from formal error propagation (following
Bevington and Robinson, 1992) from the average accumulation
rate
Due to the random and non-systematic nature of these errors, we can assume
that they are unlikely to contribute to a regional or temporal accumulation
rate bias. To calculate uncertainty for accumulation rates averaged over
multiple epochs (
We compare our GreenTrACS accumulation results with annual outputs from
Box et al. (2013;
hereafter “Box13”; 1840–1999), the Fifth Generation Mesoscale Model
(Polar MM5; 1958–2008;
Burgess et al., 2010), MAR
(1948–2015;
Fettweis et al., 2016), and the Regional Atmospheric Climate Model (RACMO2;
1958–2015; Noël et al.,
2018) over common time periods. Grid cell sizes for these model outputs are
5, 3, 5, and 1 km, respectively. For each radar trace we calculate
statistically significant differences (at
To investigate recent changes in GrIS accumulation rates, we calculate
trends in accumulation rates across our GPR and GreenTrACS firn core
dataset. We fit a linear model to the accumulation rate time series for each
radar trace and analyze the trend for both slope and statistical
significance. Likewise, we calculate trends and their statistical
significance for total precipitation (snowfall
To investigate the potential role of changing storm tracks in precipitation
changes over the western GrIS, we utilize the updated
Serreze (2009) storm track database. This database contains
6 h interval positions of extratropical cyclone storm centers on a
2.5
Difference between averaged (1966–2016) GreenTrACS accumulation and average (1962–2014) IceBridge Accumulation Radar rates from Lewis et al. (2017) across all 562.5 km of overlap. Spatially overlapping section of 2016 and 2017 traverses displayed as adjacent tracks. Also showing extent of GreenTrACS traverse (black) and IceBridge accumulation radar (grey). Inset shows map location with respect to GreenTrACS traverse (black).
Figure 1 displays the mean accumulation rates at each location along the
traverse route, with higher accumulation rates along the main traverse and
lower accumulation rates at higher elevations of the ice sheet interior,
broadly consistent with previously published accumulation rate compilations
(e.g., Bales et al., 2009) and RCM
output (Box et al., 2013; Burgess et al., 2010; Fettweis et al., 2016;
Noël et al., 2018). We analyze localized differences between GPR-derived
accumulation rates and these RCMs in Sect. 3.3. There is an especially
high accumulation rate zone near Core 11 (0.685 m w.e. a
Average GreenTrACS GPR accumulation rates (black) compared with
We validate our accumulation record with published core records from the PARCA campaign and accumulation data from the NASA IceBridge program. The locations of GreenTrACS core sites 2, 5, 9, 10, 11, 14, 15, and 16 were chosen to reoccupy PARCA core locations 6745, 6945, 7147, 7247, 7249, NASA-U, 7347, and 7345, respectively. These GreenTrACS cores overlap with the accumulation history of each PARCA core and extend the record from 1997/1998 to 2016/2017. Annual and epoch-averaged accumulation rates derived from GreenTrACS firn cores are within uncertainty ranges of those determined from corresponding PARCA cores during the period of overlap. Averaging accumulation rates over 5-year epochs reduces noise in year-to-year accumulation variability. Figure 5 compares the accumulation records from PARCA sites 6745, 6945, 7345, and NASA-U to their corresponding GreenTrACS cores, demonstrating that each pair of cores has similar long-term mean accumulation rates and nearly identical decadal variability. Thus, we have confidence in firn-core-derived accumulation rates that are used in subsequent GPR calculations of accumulation rates throughout the GreenTrACS traverse.
Accumulation record at GreenTrACS Core 8 and Core 15 (black) compared with RCM output from RACMO2 (red), Polar MM5 (cyan), MAR (green), and Box13 (blue). We find statistically significant Pearson correlation coefficients between GreenTrACS and RCM accumulation rates for these cores (see Table 2).
Average (1966–2016) GPR accumulation rates are statistically
indistinguishable from average (1962–2014) IceBridge Accumulation Radar
measurements analyzed by
Lewis et al. (2017), with
an rms difference of
Similarly, our 2011–2016 accumulation rates are statistically
indistinguishable from average 2009–2012 IceBridge snow radar
measurements analyzed by Koenig et al. (2016), with an rms difference of
We assess differences between RCM accumulation output and GreenTrACS
accumulation records at each firn core site, two of which are shown in
Fig. 8. In general, year-to-year correlations
between GreenTrACS firn core accumulation records and RCM output for the
corresponding grid cell are strong, positive, and statistically significant
(Table 2). On average, GreenTrACS firn cores'
correlation coefficient with MAR output is 0.718, with PolarMM5 is 0.701,
with Box13 is 0.607, and with RACMO2 is 0.763. Every correlation is
statistically significant at
Difference between accumulation rates at each GreenTrACS core site calculated using Herron–Langway profiles and firn core density information.
Pearson correlation coefficients between accumulation rate time series
from firn cores and co-located RCM output over their common time
period
We also assess spatial differences between GreenTrACS accumulation rates and
mean RCM accumulation rates averaged over several decades
(Table 2). Figure 9 shows
that differences between GreenTrACS accumulation rates and RCM output
increase in magnitude, become more spatially heterogeneous, and vary by
model at lower elevations of the ice sheet where topographic variations are
larger and surface melt increases. Averaged over all 4436 km of the
traverse, the rms difference (
Differences between GreenTrACS accumulation rates and
However, the high spatial resolution of our dataset shows significant
accumulation variability not captured in model output
(Fig. 9). For example, Polar MM5 and MAR both
underestimate accumulation rates between Core 4 and Summit, while
overestimating accumulation rates to the west of Cores 10–12. Likewise,
RACMO2 overestimates accumulation rates between Raven–Dye-2 and Core 5 by
0.03 to 0.08 m w.e. a
Our study is almost entirely contained within drainage basin E from
Vernon et al. (2013), who
note that basin E is the only major Greenland drainage basin with no
statistically significant differences in SMB between the four RCMs. However,
differences of 0.1 to 0.4 m w.e. a
In summary, the RCMs do an excellent job of calculating accumulation rates
averaged over this drainage basin, with rms values between 0.048 and 0.082 m w.e. a
In most locations, there are no statistically significant trends in the
GreenTrACS accumulation record from 1966 through the mid-1990s. However, a
change-point analysis (Lavielle, 2005) reveals
that accumulation rates in the western GrIS percolation zone changed
significantly after the 1995–1996 accumulation year. Since 1996, our
record indicates a statistically significant average accumulation rate
decrease of
In Fig. 10, we compare the negative accumulation
trend in the GreenTrACS record (1996–2016) to best-fit linear trends in
total precipitation (rain
Best-fit linear trends for each grid cell showing magnitude (left) and
percent (right) changes in total precipitation for
We find strong agreement between the accumulation rate decrease in the
GreenTrACS record and widespread precipitation decreases in the RCMs over
the study area (Fig. 10). On average, the RCMs
have a more negative precipitation trend than the GreenTrACS record by
Increased melt throughout the 1996–2016 period is a confounding variable when analyzing trends in accumulation rates. With increased melt over the past several decades in this region, meltwater percolates down through several years of firn (Benson, 1962; Graeter et al., 2018; Harper et al., 2012; Wong et al., 2013). This movement of mass into lower years can artificially increase the mass balance at depth and lower the mass balance during the most recent years, which have not experienced as much meltwater percolation from more recent annual layers. Therefore, it is necessary to evaluate the degree to which the recent accumulation rate decrease in the GreenTrACS record is biased by the recent increase in surface melt and percolation.
On average, we find larger negative accumulation trends (
Our analysis indicates that snow accumulation rates have been declining in western Greenland since 1996, despite significant warming and resulting increases in saturation vapor pressure from the Clausius–Clapeyron relationship. Instead, precipitation decreases over western Greenland likely result from changes in atmospheric and/or oceanic circulation. Mernild et al. (2014) and Auger et al. (2017) found that the positive phase of the Atlantic Multidecadal Oscillation (AMO) is associated with a precipitation increase over interior and southwestern Greenland based on ice core records and the Japanese Meteorological Agency 55 Year Reanalysis (JRA-55; Kobayashi et al., 2015), respectively. In direct contrast with these findings, the decline in western Greenland accumulation rates documented in the GreenTrACS record began in the mid-1990s, contemporaneous with a switch to the AMO positive phase.
We hypothesize that the differences between our results and those of
Auger et al. (2017) and
Mernild et al. (2014) stem from
different causes. Auger et al. (2017) validated the
reanalysis data by demonstrating that JRA-55 precipitation at Nuuk,
Greenland, is significantly correlated with Nuuk station data from 1958 to 2013. Furthermore, coastal precipitation in western Greenland is strongly
and significantly (
We find that the decrease in accumulation rates over the western GrIS is
associated with a significant decrease in the number of storm days since
1996. The GreenTrACS region experienced an average of
The decline in summer storm days indicates a relationship with
well-documented stronger summer blocking over Greenland over the past 2
decades (Hanna et al., 2013;
McLeod and Mote, 2016), with a positive Greenland Blocking Index (GBI)
during 17 out of 21 summers between 1996 and 2016
(Hanna et al., 2016). The June–August GBI
had a statistically significant positive trend of 1.87 (unitless; normalized
to 1951–2000) from 1991 to 2015 (Hanna et
al., 2016). The summertime 500 mbar geopotential height increased 50–70 m over the 1996–2016 period compared with the 1979–1996 baseline
(Fig. 11c), indicating stronger blocking that we
suggest likely reduced precipitation over the central GrIS by deflecting
storms poleward from the Greenland interior. This is consistent with an
observed
The effect of summertime Greenland blocking has previously been discussed primarily in the context of increasing surface melt (Hanna et al., 2013, 2018; Ballinger et al., 2017; Hofer et al., 2017), while the effect of blocking on precipitation has received less attention (Hanna et al., 2013; McLeod and Mote, 2016). Our results highlight that stronger summer blocking reduces GrIS SMB through both an increase in surface melting and a decrease in accumulation rates. Stronger summer blocking has been tied to an observed increase in surface melt since 1996 across the western GrIS percolation zone (Graeter et al., 2018), and to the July 2012 melt event, during which 98.6 % of the GrIS experienced melt (Nghiem et al., 2012). We show here with in situ data that snow accumulation rates have declined in this same region as strong blocking has decreased the number of summer storm days. Presently, none of the GBI outputs from the Coupled Model Intercomparison Project Phase 5 (CMIP5) suite of global climate models accurately capture the recent summer GBI increase (Hanna et al., 2018). Improved predictions of summertime Greenland blocking under future anthropogenic forcing scenarios are therefore critical for accurately predicting Greenland SMB and its contribution to sea level rise.
We have developed a new dataset of accumulation rates over the western interior of the Greenland Ice Sheet spanning the past 20–60 years, based on sixteen 22–32 m long firn cores and 4436 km of in situ GPR accumulation data. This accumulation record is internally consistent across the dataset and is validated by previous in situ field measurements and other radar-derived accumulation measurements (e.g., Lewis et al., 2017).
Overall, the Polar MM5
(Burgess et al., 2010), MAR
(Fettweis
et al., 2016), Box13 (Box et al., 2013), and RACMO2
(Noël et al., 2018) regional
climate models accurately capture large spatial patterns in accumulation
rates over the GrIS, but show statistically significant differences from GPR
accumulation rates on a regional basis. The average rms difference between
each model and GreenTrACS accumulation rates is
While global climate models predict a 21st-century increase in precipitation over the GrIS (e.g., Bintanja and Selten, 2014), we observe a decrease in precipitation across the western GrIS from 1996 to 2016 using records from firn cores, GPR, and published RCMs. We believe this study is the first to identify widespread negative GrIS precipitation trends during this period of enhanced surface melt, evident in these RCMs and our field observations (Graeter et al., 2018).
We attribute the decrease in accumulation rates over the western GrIS between 1996 and 2016 to more persistently positive Greenland blocking in the summer. We find a statistically significant 25 % reduction in the number of summer storms that precipitate over the GreenTrACS region since 1996. While increased temperatures from anthropogenic forcing and enhanced summer blocking have increased melt across the western percolation zone, here we show that summer blocking has also contributed to declining precipitation over the past 2 decades. This has led to a strongly negative SMB trend on both the input and output sides of the SMB equation that may not be accurately captured in global climate models that are currently unable to reproduce the recent increase in blocking. This highlights the importance of improving global climate models (GCMs) projections of future summer blocking to accurately forecast Greenland precipitation and melt rates under stronger greenhouse gas forcing.
We have uploaded all corresponding data from this project to the NSF Arctic Data Center (Lewis, 2019). This dataset includes the depth–age and depth–density scales, all associated chemistry data, final yearly accumulation, and yearly melt records for all 16 GreenTrACS firn cores. Additionally, the dataset includes all GPR accumulation measurements for each epoch and the traced IRH depths.
EO, RH, and HPM proposed the project and GL refined the methodology. GL, EO, RH, HPM, TM, KG, FM, and TO collected field measurements. GL, EO, KG, ZT, and DF analyzed firn cores and conducted chemistry analyses. GL traced radar layers, conducted data analysis, and wrote the paper, with substantial input from EO, RH, and HPM. All authors reviewed and approved the paper.
The authors declare that they have no conflict of interest.
This project was supported by the US National Science Foundation (NSF) under grants DGE-1313911 and ARC-1417640. We would like to thank Mary Albert for providing field validation measurements, as well as Jason Box, Xavier Fettweis, and Brice Noel for providing the most recent Box13, MAR, and RACMO regional climate model outputs. Our successful data collection would not have been possible without the support of Ch2M Hill Polar Field Services, Kangerlussuaq International Science Support, and the Air National Guard 109th Airlift Wing. We thank the U.S. Ice Drilling Program for support activities through NSF cooperative agreement 1836328. Special thanks go to Sean Birkel and the Danish Meteorological Institute for location-specific weather forecasts in Greenland. The authors would like to thank the two anonymous reviewers for greatly improving the paper.
This research has been supported by the National Science Foundation, Division of Arctic Sciences (grant no. NSF ARC-1417640).
This paper was edited by Michiel van den Broeke and reviewed by two anonymous referees.