PUBLICATIONS

Journal of Geophysical Research: Earth Surface

RESEARCH ARTICLE Spatial extent and temporal variability
10.1002/2016JF003869
of Greenland firn aquifers detected
Key Points:
• Firn aquifers represent a total area of
by ground and airborne radars
2
21,900 km when inferred from Clément Miège1, Richard R. Forster1, Ludovic Brucker2,3, Lora S. Koenig4, D. Kip Solomon5,
airborne radar data collected between
2010 and 2014 John D. Paden6, Jason E. Box7, Evan W. Burgess1, Julie Z. Miller1,8, Laura McNerney1, Noah Brautigam1,
• Water table interannual variations Robert S. Fausto7, and Sivaprasad Gogineni6
imply filling and drainage
1
• Water flows in the firn aquifer under Department of Geography, University of Utah, Salt Lake City, Utah, USA, 2Cryospheric Sciences Laboratory, Code 615, NASA
low hydraulic gradients following local Goddard Space Flight Center, Greenbelt, Maryland, USA, 3Universities Space Research Association, Goddard Earth Sciences
surface topography
Technology and Research Studies and Investigations, Columbia, Maryland, USA, 4National Snow and Ice Data Center,
Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA,
5
Supporting Information: Department of Geology and Geophysics, University of Utah, Salt Lake City, Utah, USA, 6Center for Remote Sensing of Ice
• Figures S1 and S2 Sheets, University of Kansas, Lawrence, Kansas, USA, 7Geological Survey of Denmark and Greenland (GEUS), Copenhagen,
Denmark, 8Byrd Climate and Research Center, Ohio State University, Columbus, Ohio, USA
Correspondence to:
C. Miège,
[email protected]
Abstract We document the existence of widespread firn aquifers in an elevation range of ~1200–2000 m,
in the high snow-accumulation regions of the Greenland ice sheet. We use NASA Operation IceBridge
Citation: accumulation radar data from five campaigns (2010–2014) to estimate a firn-aquifer total extent of
Miège, C., et al. (2016), Spatial extent
21,900 km2. We investigate two locations in Southeast Greenland, where repeated radar profiles allow
and temporal variability of Greenland
firn aquifers detected by ground and mapping of aquifer-extent and water table variations. In the upper part of Helheim Glacier the water table
airborne radars, J. Geophys. Res. Earth rises in spring following above-average summer melt, showing the direct firn-aquifer response to surface
Surf., 121, 2381–2398, doi:10.1002/
meltwater production changes. After spring 2012, a drainage of the firn-aquifer lower margin (5 km) is
2016JF003869.
inferred from both 750 MHz accumulation radar and 195 MHz multicoherent radar depth sounder data. For
Received 1 MAR 2016 2011–2014, we use a ground-penetrating radar profile located at our Ridgeline field site and find a spatially
Accepted 14 NOV 2016 stable aquifer with a water table fluctuating less than 2.5 m vertically. When combining radar data with
Accepted article online 15 NOV 2016
surface topography, we find that the upper elevation edge of firn aquifers is located directly downstream of
Published online 20 DEC 2016
Corrected 16 JAN 2020 locally high surface slopes. Using a steady state 2-D groundwater flow model, water is simulated to flow
laterally in an unconfined aquifer, topographically driven by ice sheet surface undulations until the water
This article was corrected on 16 JAN encounters crevasses. Simulations suggest that local flow cells form within the Helheim aquifer, allowing
2020. See the end of the full text for
water to discharge in the firn at the steep-to-flat transitions of surface topography. Supported by visible
details.
imagery, we infer that water drains into crevasses, but its volume and rate remain unconstrained.

1. Introduction
Mass loss of the Greenland ice sheet has been accelerating since the 1990s [Rignot et al., 2011], and in recent
years (2009–2012), about 85% of its rate of increase were due to increased surface melt and subsequent
runoff [Enderlin et al., 2014]. Greenland surface melt extent and duration have increased over the last decade,
in response to increased advection of warm air from the south during summers [e.g., Tedesco et al., 2013].
Surface melt is not only occurring in the lower elevation margins of the ice sheet but also migrating toward
the ice sheet interior, causing a reduction of the dry-snow zone and an expansion of the percolation zone
[McGrath et al., 2013]. Widespread firn aquifers have been detected in high accumulation regions
(1.2 m w.e. yr
1, water equivalent, in average) of the ice sheet percolation zone [Forster et al., 2014]. In these
regions, firn aquifers retain water for at least several years, storing a substantial amount of water respon-
sible of heating the firn and delaying runoff [Forster et al., 2014; Kuipers Munneke et al., 2014]. Their impacts
on the ice sheet are numerous, including firn densification, alteration of the ice thermal state, and water
input from the aquifer into the englacial hydrology system, which has the potential to affect ice dynamics
and Greenland’s contribution to sea level rise [Forster et al., 2014; Koenig et al., 2014].
This work extends previous firn-aquifer studies within the Greenland ice sheet, outlined in section 2 below
[Forster et al., 2014; Koenig et al., 2014; Kuipers Munneke et al., 2014], focusing on mapping firn-aquifer extent
©2016. American Geophysical Union. and investigating short-term temporal evolution between 2010 and 2014 using a combination of airborne
All Rights Reserved. radar and lidar data, together with ground observations (radar and firn cores). We first provide background

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2381

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

information on water storage systems on glaciers and ice sheets including water percolation and lateral flow
within the firn. We then present a map of the Greenland firn-aquifer extent and depth to water using 5 years
of airborne radar data. We investigate annual temporal evolution of the water table over repeated airborne
and ground tracks between 2010 and 2014. Finally, we simulate steady state 2-D lateral flow within the firn
aquifer and reveal local flow cells due to local surface undulation in the topography.

2. Background
The downward percolation of meltwater into the snow and firn is a nonuniform process in space and time
that accelerates snow metamorphism [e.g., Colbeck, 1975; Raymond and Tusima, 1979]. Water generally flows
downward due to gravity, although some water can be retained either in a liquid form in the temperate (0°C)
snow/firn microstructure due to surface tension (capillary forces) or by refreezing into a snow/firn layer below
0°C, generating ice lenses. Cold-room experiments by Coléou and Lesaffre [1998] suggested that capillary
forces hold less than 10% of water in the pore volume of 0°C seasonal snow and that holding capacity
decreases with depth and increasing snow/firn density. In their experiment, a water-saturated layer was
found at the bottom of each snow sample, with a bulk density of 900–950 kg m
3; this suggests that substan-
tial water retention is possible in natural snowpacks.

The presence of meltwater storage in the firn is well documented for mountain glaciers [e.g., Fountain and
Walder, 1998; Jansson et al., 2003], and a few examples include Vernagtferner; Oetztal Alps [Oerter and
Moser, 1982]; South Cascade Glacier, USA [Fountain, 1989]; and Storglaciären in northern Sweden
[Schneider, 1999; Schneider and Jansson, 2004]. For these last two glaciers, the water in the aquifer seems
to be present mainly in the melt season. Drainage from the aquifer is reported for South Cascade by late
November [Fountain, 1996], and only a small amount of water (irreducible water saturation) is observed to
persist by late winter in Storglaciären [Jansson et al., 2003]. In Arctic settings, perennial (>1 year) firn water
storage is observed within the Holtedahlfonna ice field, Svalbard [Christianson et al., 2015]. The depth to
the water table depends on elevation, surface slope, and glacier topography and ranges between 0 m and
40 m [Fountain and Walder, 1998; Christianson et al., 2015]. Despite some variations in depth to the water
table and storage time scale, firn aquifers show relatively similar hydraulic characteristics, reflecting a similar
seasonal snow grain metamorphism rate in response to water input [Fountain and Walder, 1998]. Firn hydrau-
lic conductivity values range from 1 to 5 × 10
5 m s
1 with an average porosity of 0.15 [Fountain and Walder,
1998]. Aquifer responds to surface meltwater input, and percolation vertical velocities are found to range
from 0.12 to 0.35 m h
1 for an isothermal firn column, equivalent to ~1–3.5 days to transfer water from the
surface to the water table at 10 m depth [Schneider, 1999]. In the absence of local sinks, water would flow
laterally under hydraulic gradients, following the glacier slope. Jansson et al. [2003] found that the firn-aquifer
thickness is related to the hydraulic gradient (water table slope) and the efficiency of the englacial network in
allowing water to exit the glacier system. The water can eventually drain into nearby crevasses or moulins
[Fountain and Walder, 1998].

These previous studies provide important observations for understanding the larger firn aquifers observed
within the Greenland ice sheet. In the ice sheet firn, stored water exists in regions with moderate to strong
summer surface melt and high snow accumulation [Forster et al., 2014; Kuipers Munneke et al., 2014].
Sufficient early autumn snowfall insulates the firn from the winter surface temperatures, preserving pore
space at depth to allow retention and storage of additional meltwater [Kuipers Munneke et al., 2014]. In the
ice sheet percolation zone, the firn stratigraphy is heterogeneous and characterized by ice lenses (horizontal)
and ice columns/pipes (vertical) [e.g., Benson, 1962]. Ice inclusions offer preferential pathways for meltwater
to travel, changing firn physical properties through a complicated meltwater routing system [Pfeffer and
Humphrey, 1996; Humphrey et al., 2012; Machguth et al., 2016]. At greater depths (>10 m), water-saturated
layers can persist in the firn over the winter [Humphrey et al., 2012], similar to groundwater aquifers
[Forster et al., 2014]. The Greenland firn aquifer is usually present in a 10–20 km wide band that rings the
southeast and south ice sheet perimeter and has smaller average slopes (less than 1°) [Forster et al., 2014].
Assuming a similar hydraulic conductivity value as glaciers and weak hydraulic gradients, it would take years
for water at the high-elevation end of the water-saturated firn layer to exit the firn via the englacial hydrology
system [Humphrey et al., 2012].

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2382

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

3. Study Site
Surface-based measurements were made on the southeast part of the ice sheet (66.18°N, 39.04°W; ~1560 m),
50 km west from Sermilik Fjord on a ridgeline south of Helheim Glacier, named “Ridgeline Site” in the following
(Figure 1). Fifteen kilometers west of our field site the ice thickness is 1000 m, according to observations from
the Operation IceBridge (OIB) 2012 Multichannel Coherent Radar Depth Sounder (MCoRDS) [Leuschen et al.,
2014b]. Our field site was initially chosen for firn-core extraction to verify the simulated high accumulation
rates (1.44  0.30 m w.e. yr
1 [Burgess et al., 2010] and 1.67  0.29 m w.e. yr
1 [Fettweis et al., 2013]) and large
interannual variations in the region [Miège et al., 2013]. At this location, melt averages 0.73  0.17 m w.e. yr
1
over the 1979–2014 period, are simulated by the Model Atmospheric Regional (MAR) with a substantial
increase over the last decade [Fettweis et al., 2013]. For the 2010–2014 period, summer melt in 2011
(0.73 m w.e.) was within the 1979–2014 average, whereas the summers of 2010 and 2012 were characterized
by important melting with 1.04 and 1.18 m w.e. yr
1, respectively (Figure 1c) [Fettweis et al., 2013]. Our field
work at this site was conducted in April to early May 2013–2014, prior to melt onset, when the firn aquifer
is assumed to be at the seasonal minimum capacity.
To complement surface-based measurements, we use airborne radar data collected by NASA OIB campaign
each year between April–May 2010 and April–May 2014. OIB data set is spatially extensive and covers land ice
over the ice sheet with an average of 60,000 km of airborne data collected yearly (flight line statistics given in
Table 1). On board the NASA OIB airplane, we primarily use the accumulation radar data at a similar frequency
range to our ground radar observations. This large airborne radar data set allows us to complement and
extend our field observations and provide an ice sheet wide estimate of the firn-aquifer extent and investi-
gate temporal changes in the water table elevation over repeated flight lines.

4. Data
4.1. Airborne Data
Two different radar systems designed and operated by the Center for Remote Sensing of Ice Sheets (CReSIS) on
board a NASA OIB aircraft were used in this study [Rodriguez-Morales et al., 2013; Leuschen et al., 2014a, 2014b].
To monitor extent and depth to the water table (section 5), ultrahigh frequency radar, referred to as the accu-
mulation radar (AR), is used, which operated at a center frequency of 750 MHz and had a bandwidth of 300 MHz.
The ultrawide bandwidth enables imaging of internal features, including the water table, with a 65 cm vertical
resolution [Rodriguez-Morales et al., 2013]. In the along-track dimension, the radar-trace spacing is 35.8 m for a
platform velocity of 140 m s
1. For smooth, quasi-specular targets (i.e., internal layers), the imaged area is
equal to the first Fresnel zone, which is 20 m at 750 MHz and 500 m height [Leuschen et al., 2014a].
In addition to the firn-aquifer water table being identified by AR, the Multichannel Coherent Radar Depth
Sounder (MCoRDS), which operates at 195 MHz during the OIB campaigns, is also used to infer aquifer
presence [Leuschen et al., 2014b].
Finally, we retrieve the along-track surface elevation profile using the NASA Airborne Topographic Mapper
(ATM) lidar [Krabill, 2013], on board the OIB airplane with the CReSIS radars. The accuracy of the measured
elevation is 8.5 cm [Krabill et al., 2002]. At nadir, the elevation profile is interpolated to meet the time stamp
of each individual traces from the radar data.

4.2. Ground-Based Data

Ground-based radar measurements were made using a 400 MHz ground-penetrating radar (GPR) from
Geophysical Survey Systems Inc. The vertical resolution of this system is about ~35 cm for dry firn in West
Antarctica [Spikes et al., 2004]. We collected 2048 samples per trace in a 500 ns window, for an ~0.24 ns
sample interval. Two different types of surveys were performed: one using a snowmobile (2011), moving
at ~3 m s
1, and the other using skis (2013 and 2014), moving between 1 and 2 m s
1. Average trace spacing
was 0.5 m for the snowmobile surveys and 0.3 m for the ski surveys. Time-dependent gain was used to
compensate for radar wave attenuation within the firn. In postprocessing, additional stacking (8 times)
increased the signal-to-noise ratio, bringing the average trace spacing to 4.0 m (snowmobile) and 2.4 m (ski).
GPR data are geolocated using a dual-frequency Trimble© R7 GPS receiver in kinematic mode with a
sampling interval of ~15 m using snowmobile and ~7.5 m while traveling by skis. We processed the GPS data

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2383

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 1. (a) The inset shows a Landsat 8 true color image mosaic (21 August 2014) over the study region. Firn-aquifer locations are color coded by depth to the
water table. OIB flight lines are in black. Firn core locations are represented by blue dots, with a two-digit number corresponding to year of collection. The blue
dashed line corresponds to the polygon generated from the aquifer linear detections. PROMICE weather stations (TAS_U and TAS_L) are represented with green dots.
The green lines indicate the last upstream crevasse visible from L-band SAR imagery [Moon and Joughin, 2008]. Elevation contours (orange) are from the 1 km
resolution Cryosat-2 DEM [Helm et al., 2014]. (b) The inset details inside the red box with diamonds for airborne depth to water table and lines for the 2014 GPR. The
gray elevation contours are obtained from linear interpolation of the GPS grid. (c) The inset shows both the simulated melt (red) and snowfall (black) outputs from the
MAR regional climate model (version 3.5.2; at 5 km resolution) at the FA-13 firn-core location [Fettweis et al., 2013].

in 2014 using the precise point positioning web-based processor hosted by the Canadian Spatial Reference
Service (CSRS) at https://webapp.geod.nrcan.gc.ca/geod/tools-outils/ppp.php. Formal errors generated by
CSRS are averaged for the three field seasons and are estimated at 3.5 cm for latitude and longitude and
7.0 cm for elevation (Figure S1 in the supporting information).
Five density profiles were obtained from our firn cores between 2010 and 2013. Each core section was
weighed to retrieve a depth-density profile. The vertical resolution of density measurements was ~1 m in
2010 and 2011 [Miège et al., 2013] and ~20 cm in 2013 [Koenig et al., 2014]. Seasonal snow depth (snow/firn
transition) is estimated from the extracted cores, where small (<0.8 mm) dry-snow grains transition to larger
(>1.5 mm) grains with rounded angles, showing evidence of previous melting (Figure S2).
Finally, we use temperature, ablation, and snowfall measurements collected from two Southeast Greenland
automatic weather stations (AWSs) in the Programme for Monitoring of the Greenland Ice Sheet (PROMICE)
network [Ahlstrøm and the PROMICE project team, 2008, www.promice.org]. TAS_L station (65.64°N, 38.90°W;
~270 m above sea level (asl)) and TAS_U station (65.70°N, 38.87°W; ~570 m asl) are located in the ablation
area of the ice sheet, with TAS_U being 55 km south of our field site (Figure 1a).

Table 1. AR OIB Flight Campaign Characteristics for 2010–2014 Over Aquifer Regions
OIB Flight Total Flight Lines over Greenland Land Total Linear Distance with Aquifer-Repeat Tracks (km)
Campaign Ice (km) Aquifer (km) 2010|2011|2012|2013|2014

2010 30,580 750 +270|210|50|130

2011 74,900 1,040 270 + 345|80|145
2012 86,220 1,130 210|345 + 85|245
2013 42,290 350 50|80|85 + 30
2014 69,790 660 130|145|245|30+
Total 303,780 3,930 660|840|885|245|550

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2384

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

5. Methods
5.1. Radar-Derived Depths to Water
The presence of water in the firn slows electromagnetic wave propagation as it increases the medium’s
dielectric constant [Smith and Evans, 1972]. For soaked-firn conditions, Smith and Evans [1972] calculated a
dielectric constant of εf′ = 15.2, about 6 times higher than dry firn (εf′ = 2.4) with a density of 650 kg m
3.
For AR and GPR data, this dielectric contrast produces an interface with high-reflectivity and high-amplitude
return, substantially brighter than the dielectric contrast due to dry density changes in the firn column only
(see section 6.1 and Figure 3 for details). For the airborne radar, the snow surface was identified and picked
using an amplitude threshold at the start of the surface return. This threshold had to be adjusted for each OIB
campaign, but no other changes were made in data processing.
The depth to the aquifer surface (water table) is estimated from the two-way travel time (TWTT) of the
electromagnetic wave between the snow surface and the water table surface. For nonmagnetic and low-loss
dielectric media like snow and firn, the velocity (v) of an electromagnetic wave can be approximated by
pffiffiffi
v ¼ c= ε′ ; (1)

where c = ~0.3 m ns
1 is the electromagnetic velocity in a vacuum and ε′ is the dielectric permittivity.
We estimate the variations of the dielectric permittivity at depth (above the aquifer) using an empirical
relationship linking it to the snow and firn density (ρ) [Kovacs et al., 1995]:

ε′ ¼ ð1 þ 0:845ρðz ÞÞ2 : (2)

A regional steady state ensemble density profile is compiled from five dry firn cores, extracted either above
the water table surface or less than 20 km upstream of the firn aquifer. By using an average density profile, we
attempt to minimize the impact of small-scale spatial density heterogeneities in the firn of the lower percola-
tion zone. We also use a biexponential fit to smooth and interpolate the averaged density profile with depth.
The presence of water within the snow microstructure influences the TWTT to depth conversion within the
firn aquifer, since water slows the transmitted electromagnetic waves [e.g., Bradford et al., 2009]. Here in
focusing on the water table depth, we assume that firn above the water table is dry when measurements
are made in the spring (April–May), consistent firn temperature profiles collected at our field site [Koenig
et al., 2014].

5.2. Associated Uncertainties

There are two main uncertainties in mapping the spatial extent of the firn aquifer and its depth to the water
table with AR radar: (1) the uncertainty linked to the presence or absence of the firn aquifer and (2) the uncer-
tainty in determining the depth to the water table, if present. Picking of the water table was not a fully auto-
mated process, and thus, a subjective human bias is introduced. We did not use a constant brightness
threshold to detect the water table from the radar data as brightness levels varied substantially based on
the radar system used (GPR versus AR), the year of collection of the AR radar, the depth to the reflector,
and the location. A length of ~60,000 km of radar profiles was flown over the Greenland ice sheet during each
OIB campaign (2010–2014; Table 1). For each flight line, we manually examine radar data for bright and con-
tinuous reflectors, at a depth range of 5–50 m [Forster et al., 2014]. We identify a water table by the presence
of continuous reflector that can be followed for more than 0.5 km along profile. To quantify the human bias
introduced from the layer picking, the entire 2011 OIB data set was processed by two different operators,
using two different tracing methods: manual and semiautomatic. The semiautomatic routine is based on
an edge-detection method, looking at differences between lower and upper frequencies to accentuate the
peaks. The automatic part of the picker performed well in regions with clear water table signal, but manual
tracing was required for weaker signals. We find agreement (within uncertainties) over 75% of the total area
between the two operators. This difference between picks is structured and mainly occurred in locations
where internal reflectors were weak in the radar data. Discrepancies between layer picks occur on the edges
of the water table, where reflector intensity diminishes gradually.
Radar data quality and aquifer detections depend on airplane survey conditions, such as turbulence and
turning geometry. For example, aircraft roll greater than 10° prevents detection of internal layers and, as a

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2385

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

result, identification of the firn aquifer. Identifying deeper water tables is challenging due to the greater
attenuation of the higher-frequency radar wave of the AR. Usually, an aquifer deeper than 40 m in the
percolation zone is not well imaged in the AR profile, likely due to weak returns from internal layers that
are masked by surface clutter (off-vertical surface) at the edges of the percolation zone.
Layer depth uncertainty comes from two sources: TWTT-to-depth conversion (section 5.1) and layer
picking. The TWTT-to-depth conversion uncertainty is associated with the density profile used. For
example, within 1σ (standard deviation of the averaged density) of density, the uncertainty is 40 cm
for an average depth to the water table of 20 m. For the 2011 AR data, the layer picking was repeated
twice, using a semiautomatic approach and a manual approach. The averaged depth difference was
60 cm for the ~1000 km of radar data (~50,000 radar traces). Internal reflecting horizons imaged by the
GPR were brighter and more distinct than those imaged by the AR, probably due to the lack of surface
clutter that obscured the deeper returns in the AR. The repeated picking test over a 5 km segment shows
an averaged difference of only 7 cm (or three samples) between the two picking methods. Finally, we
calculate averaged error propagations based on the two uncorrelated uncertainties of 72 cm for the AR
and 41 cm for the GPR, which remain small compared to the range of water table depths investigated
(4–40 m; see Figure 6).

5.3. Estimation of Firn-Aquifer Extent

We use five consecutive years of AR radar data (2010–2014) to estimate the spatial extent of firn aquifers.
For this analysis, we assume that the extent of firn aquifers is in steady state over the 5 year period to
allow interpolation of aquifer extent between OIB flight lines. To convert the linear distance (km) obtained
by radar profiling to an area estimate (km2), we invoke constraints along the across-flight line direction
and consider four cases (Figure 2). Cases (a) and (b) are “isolated-aquifer detections” in which there is only
one flight line and no crossover points. For aquifer detections over linear distances less than 5 km (case
(a)), we use the along-flight line aquifer length to estimate the aquifer dimension in the across-track direc-
tion, creating a squared area. For linear distances greater than 5 km (case (b)), we limit the aquifer dimen-
sion in the across-flight line direction to 5 km creating a rectangle around the flight line. We use a 5 km
threshold based on examination of the connected-aquifer detection, but the choice remains somewhat
arbitrary. Case (c) is characterized by multiple flight lines with one more crossover points; here we use
the end points of the profiles delineating the aquifer to define a polygon (Figure 2). Case (d) is character-
ized by the two parallel flight tracks closer than 10 km with a detected aquifer. We assume that the area
between the two flight lines also contains water and connect them into a polygon. The two parallel flight
lines cannot be separated by more than 10 km; otherwise, they would be included in case (a) or (b). Finally,
we allow in a given polygon linear gaps less than 10 km, gaps which may potentially be due to aquifer
detection deficiencies during radar data collection, often linked to plane geometry, e.g., airplane roll
greater than 10°.

5.4. The 2-D Lateral Flow Model

We use SEEP2D, a 2-D finite element groundwater flow model within the Groundwater Modeling System
package [Jones, 1999], to help explain lateral changes in the aquifer extent in the upper part of Helheim
Glacier. We combine radar data with laser altimetry to obtain a continuous measurement of the water table
elevation relative to the surface topography along an ice-flow line. We assign a head value to each node
along the water table as a boundary condition. The head value corresponds to the water surface elevation
inferred from the radar (one radar trace every 16 m in average). A uniform hydraulic conductivity range
between 1 × 10
5 and 5 × 10
5 m s
1 is used throughout the aquifer [Fountain and Walder, 1998] to deduce
volumetric flow rate, and we assume a constant aquifer thickness of 25 m [Koenig et al., 2014]. To summarize,
three assumptions are made: (1) unconfined firn aquifer at steady state, (2) constant aquifer thickness, and (3)
a constant hydraulic conductivity value taken from mountain glacier observations. This modeling approach
uses only water table and surface elevations to simulate an aquifer flow pattern at steady state without
accounting for seasonal meltwater input. For future work, adding meteorological forcing, thermodynamics,
crevasses, and third dimension to this model will be essential to better constrain and simulate firn-aquifer
seasonal recharge, local discharge, refreezing, lateral flow, and drainage, but this remains beyond the scope
of this study.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2386

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 2. (a–d) Schematic of the four different aquifer detection cases used to delineate polygons and convert linear aqui-
fer distance derived from radar observations into aquifer area. Flight lines are in blue, firn-aquifer detections in red, and the
deduced area in black. For aquifer detections over 5 km, a set 5 km distance perpendicular to the flight line is used as
threshold to create polygons.

6. Results
6.1. Firn-Aquifer Identification
The last meter-long segments of firn cores ACT11-A (1 May 2011) and ACT11-A2 (30 April 2011) were
retrieved in water-saturated firn at 25.3  0.5 m and at 10.4  0.5 m depth, respectively (Figure 1 for core
location). Within an 11 day time interval, AR and GPR data imaged a bright reflector at similar depths
(Figure 3) [Forster et al., 2014]. The high correlation coefficient (r = 0.995) between the water table depth
observations from the two radars gives us confidence in our ability to map the firn aquifer with only AR
[Forster et al., 2014].
The AR and MCoRDS systems have operated simultaneously onboard the NASA P-3 aircraft since 2010, allow-
ing for a direct comparison of the two systems in aquifer regions (Figure 4, for a profile located between red
arrows in Figure 1). There is good agreement found between the presence of weak, intermittent, or missing
bed echoes in the MCoRDS data and the presence of a water-saturated firn layer as detected by the AR
(Figure 4). From the MCoRDS data only we can infer the presence or absence of water in the firn layer,

Figure 3. (a) The inset represents the surface elevation (black) and surface slope (red) both retrieved from OIB ATM data
[Krabill, 2013]. Direct comparison (correlation coefficient of r = 0.995) between (b) GPR (400 MHz; 2 May 2011) and
(c) AR (750 MHz; 19 April 2011) over an 18 km portion of the ACT-11 traverse (transect is located between the two black
arrows on the map of Figure 1). The three boreholes (black diamonds and vertical lines) confirm the presence of water at
the bright reflector depth. The insets in Figures 3b1 and 3b2 show the details of the GPR data.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2387

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 4. (a and c) Accumulation radar at 750 MHz and (b and d) MCoRDS at 195 MHz profiles over the firn aquifer in the
upper part of Helheim Glacier collected on 17 April 2012 and 5 April 2013. Missing bed reflections in the MCoRDS radar
correlate with the presence of a water-saturated firn layer as illustrated by vertical dashed lines that indicate coincident
transitions in both radar profile. Profiles are located between the two red arrows in Figure 1.

although we cannot estimate the depth to the water table. Other causes for missing bed echoes in the
MCoRDS cannot be excluded, such as possible changes in the englacial and basal condition, but these other
causes seem unlikely, given that the surface return and its multiples do not change in strength substantially
across the aquifer transition. In addition, the relative amplitude differences of ice-bottom scattering are
negligible without the aquifer between two points, as the relative ice-bottom scattering between the same
points is large when the aquifer is present at one point. Finally, internal layers disappear simultaneously with
the loss of the bed reflection, suggesting that the latter cannot be related to changes in basal conditions
alone (Figure 4).

6.2. Firn-Aquifer Extent

We use data from five consecutive OIB seasons (2010–2014) to map the presence of firn aquifers over the
Greenland ice sheet. To estimate the total aquifer area, 110 polygons were created with areas ranging from
1 km2 to 9600 km2 (Figure 5); polygons less than 1 km2 were discarded. Roughly half of the firn-aquifer area is
found in southeastern Greenland with a total of 10,690 km2 (Figure 5c). The southern tip of the ice sheet
represents the second largest aquifer region (35%) with a total area of 7230 km2 (Figure 5d). In eastern
Greenland, smaller aquifers are detected for a total area of 2200 km2 (10%), mainly south of Kangerlussuaq
Glacier, with a few isolated aquifers on the Geikie Plateau (Figure 5). For western Greenland, the
Sukkertoppen ice cap has high accumulation rates and a small aquifer (150 km2) persists for at least the
4 years of our study (2010–2014). In a lower and narrower elevation band (1070  150 m), a few isolated
aquifers (5%) are found in northwestern Greenland (total area of 1230 km2). No aquifers were identified in
the northeast portion of the ice sheet (Figure 5).
With a mean elevation at 1615  300 m and a mean accumulation rate of 1.0  0.4 m w.e. yr
1, most of the
firn aquifers are found below the 2000 m contour line, except for the southern tip of the ice sheet where

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2388

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 5. The ice sheet wide map shows the firn aquifers mapped from AR data for 2010–2014 (colored by detection year).
OIB flight lines are in light gray. (a–d) The four different insets present the depth to the water table for the 5 year combined
data set. The two black arrows (Figure 5c) show the AR profile location of Figure 10. The polygons contoured in blue
are used to estimate aquifer extents for each inset. Figures 3b and 3c have a 40° angle from geographic north. All inset
maps are at the same spatial scale. Coastline, ice mask, and 1 km DEM are from Rignot and Mouginot [2012], Howat et al.
[2014], and Helm et al. [2014], respectively.

aquifers extend up to the 2200 m contour (Figures 5 and 6). A total linear distance of 3930 km of firn aquifers
were imaged in the five spring airborne surveys (Table 1), yielding a total estimated area of 21,900 km2. Due
to the varying nature of OIB flight lines, sampling bias always exists and surface elevations are not equally
sampled across flight campaigns (Figure 6a). The depth to the water table is normal distributed over the
5 year survey period with a mean of 22  7 m, implying a minor bias for this time period (Figure 6b). The mean
accumulation in aquifer regions ranges between 0.4 and 1.6 m w.e. yr
1, based on a Calibrated Polar MM5
simulation [Burgess et al., 2010] (Figure 6c). The total radar-derived firn-aquifer area is about 3.5 times less
than the Regional Atmospheric Climate Model (RACMO2) simulation (70,000 km2) of the water-saturated firn
layer for April 2011 [Forster et al., 2014]. To explain this discrepancy, we suggest that the radar-derived aquifer
extent is a lower bound estimate, as it is biased by flight line locations and the need for sufficient water to be
detected in the radar data. In contrast, the RACMO2 liquid-water-content estimate is more likely to represent
an upper bound estimate for aquifer total area. RACMO2 simulates water overwintering in firn for locations
where no aquifers are detected in the AR, but this may be due to the fact that the RACMO2 aquifer-extent
estimate includes relatively small water amounts (<200 kg m
2) integrated over the entire firn column thick-
ness [Forster et al., 2014]. This could likely be below the detection limits of the AR since no water table might
be present but only small amount of water held by capillary forces.

6.3. Temporal Evolution

Radar measurements repeated over the same locations allow the interannual monitoring of firn-aquifer var-
iations (Figure 1). Two transects are selected, one located at our Ridgeline field site with ground-based radar
profiles and the other on a Helheim Glacier ice-flow line, where airborne radar repeat profiles exist in close

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2389

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 6. Firn-aquifer (a) surface-elevation, (b) depth-to-water, and (c) mean-accumulation rate distributions for 5 years of
airborne radar data: 2010 (orange), 2011 (red), 2012 (blue), 2013 (green), and 2014 (magenta). Mean and 1 standard
deviation are given for each year. The mean accumulation rate is simulated by Calibrated Polar MM5 between 1958 and
2007 [Burgess et al., 2010].

proximity to our Ridgeline Site (~20 km). Since the radar tracks are not exactly superposed between years (up
to 220 m of horizontal differences for AR), we linearly interpolate the radar data to a given profile to avoid
image distortions.
At our Ridgeline Site, a GPR profile (~2.5 km) was first collected in May 2011 and then resurveyed in April 2013
and April 2014 between two firn-core locations (ACT11-A and ACT11-A2; Figures 1 and 7). In 2012, the same
transect was overflown by OIB allowing us to fill the temporal gap with AR data. We record temporal changes
of the water table elevation as the difference between the snow surface elevation and the depth to the water
table between two consecutive years (Figure 7b). Along this transect the average snow surface lowered by
0.4, 1.4, and 1.0 m for 2011–2012, 2012–2013, and 2013–2014, respectively. The water table elevation does
not follow the same trend. After a 0.6 m average lowering between 2011 and 2012, the water table rose by
1.1 m in 2013, illustrating filling from the more extensive melt year of 2012. Later, in 2014, the mean water
table is 2.5 m deeper than in 2013. To put this lowering in context, we look at possible differences in the
amount of seasonal snow but find that seasonal snow depth was about 280 cm on 6 April 2013 and
300 cm on 10 April 2014, rather similar. In addition, we compare the vertical stratigraphy from four firn cores
extracted above the water table to observe changes in ice lenses to firn ratio over time (Table 2 and Figure 1

Figure 7. Temporal evolution of the snow surface and the water table depth from GPR data for 2011, 2013, and 2014 and
from AR data for 2012*, along profile between two firn-core sites (ACT11-A to ACT11-A2; Figure 1). (a) The snow-surface
elevations, water-surface elevations (solid lines), and the water table depths (dashed lines) for the 4 years. (b) The water
table elevation changes between two consecutive years. The uncertainty bounds (Figure 7a) are estimated at both ends of
the profile.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2390

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

a
Table 2. Depth to Water and Ice Fraction of the Different Firn Cores Collected During Fieldwork (Figure 1)
Firn Cores Elevation (m) Depth to Water (m) Ice Fraction (%)

ACT11-A 1589 25 14
ACT11-A2 1559 10 7
FA13-A 1563 12.2 9
FA13-B 1563 12.2 8
a
The two-digit numbers in the firn-core names give the year of collection. Note that FA13-A and B are only located 2 m
apart.

for core locations). From the snow surface to the water table, ice lenses represent less than 10% of the dry firn
column, and no obvious increase is observed over the short 2 year period, even though it included the stron-
ger 2012 melt year (Table 2). We hypothesize that the water table lowering observed in 2014 is due to a com-
bination of a decrease in surface meltwater production while lateral water flow and drainage remained more
or less constant.
The second study transect is located in the upper part of Helheim Glacier between 2010 and 2014 (Figures 4
and 8). The first measurement in spring 2010 was made shortly before the above-average 2010 summer melt
[Fettweis et al., 2013] (Figure 1c) that produced an average rise of 1.8  0.9 m in the water table elevation by
spring 2011. The water table elevation lowered by 1.0  0.6 m between 2011 and 2012, and a 3 km upstream
extension of the aquifer is observed in 2012. At the lower elevation end, the water table undulates, following
the surface topography, before encountering crevasses, where the bright return from the top of the water
table disappears (~19 km in Figure 4a). In 2013, consistent with MCoRDS observations, the water table is sub-
stantially lowered or missing between 13 and 18 km (Figures 4c and 4d). In contrast, on the upper end of the
profile (5 to 12 km), the water table elevation increased by 3.3  0.4 m between 2012 and 2013 and finally
lowered by 0.8  0.8 m between 2013 and 2014. All the water table elevation changes between 2010 and
2014 are compiled in Figure 8. In addition, from the ATM elevation data collected over the 12 km long trans-
ect simultaneously with the AR, we observe an average 1.0  0.4 m surface lowering between the springs of
2011 and 2012, and a 1.5  0.5 m surface lowering between springs 2010–2011 and 2012–2013, in agreement
with the two above-average summer melt years of 2010 and 2012. For those two years, large meltwater pro-
duction increased meltwater percolation and therefore increased recharge of the firn aquifer.

Figure 8. Temporal evolution of the snow surface and the water table depth from AR data for 2010–2014 for the upper
Helheim Glacier (between red arrows in Figure 1). (a) The snow-surface elevations, water-surface elevations (solid lines),
and water table depths (dashed lines) for the 5 years. (b) The water table elevation changes between two consecutive years.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2391

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 9. Lower end of the firn aquifer in relation to crevasses for the upper part of Helheim Glacier (starts at 11 km in
Figure 8). (a) The WorldView01 image from ©DigitalGlobe with a 0.5 m resolution. The colors indicate the depths to the
water table in meters. The flight lines in red, blue, and green represent the OIB data from 2011, 2012, and 2013, respectively.
The dashed line indicates the last upstream crevasse visible in close proximity to the firn-aquifer edge.

6.4. Lateral Boundaries

The lateral limits of the Ridgeline Site aquifer are studied for the two transects using the GPR (upper elevation
limit) and AR (lower elevation limit) data. The AR can cover crevassed terrain, whereas the GPR provides
greater penetration depth and finer detail at the upper end of the firn aquifer. We find that the upper eleva-
tion transition (~1700 m along the ACT traverse) between dry and wet firn is gradual and diffuse. In Figure 3
b1, two brighter reflectors are found in the GPR profile at depths between 30 and 40 m but with some short
discontinuities. This suggests that the upglacier firn-aquifer expansion is sensitive to surface meltwater input
and surface topography. Also, we note that these two brighter reflectors observed in the GPR data are not
well defined in the AR data (Figure 3); we postulate that the AR might be less sensitive to small amounts
of water (<200 kg m
2), particularly at greater depths (>30 m). The low-elevation end of the aquifer is inves-
tigated using AR data, coupled with ~0.5 m resolution visible satellite imagery (WorldView-1, DigitalGlobe©).
The upper crevasses are imaged by the AR, as vertical stripes locally disrupting the internal layers (Figure 9).
Combining the AR profile with the satellite images, we note that the lower elevation edge of the water table is
a sharp boundary and corresponds to the initial encounter of the crevassed area, implying that water drains
into them. The exact depths of the crevasses are unknown; however, due to the loss of the aquifer reflector, it
can be assumed that they at least reached the water table. Our current measurements do not allow us to
investigate the fate of the water once it reaches the crevasses, but it is likely that some water enters the eng-
lacial network, possibly reaching the bedrock. In Figure 9, the upper crevasses (4–5 km) are located on a steep
section, and a bright reflector is observed lower down, where the slope decreases and there no visible cre-
vasses nearby (5–6 km). The relationship between firn aquifers, crevasses, and the impact of water flow are
further addressed in section 7.

6.5. Unconfined Firn Aquifer

Radar imagery shows stacked sequences of internal reflection horizons in the dry firn above the water table,
layers that represent density contrasts from the accumulation and melt histories. Throughout the study area
we find places where these internal reflection horizons intersect the water table at an angle (Figure 3b2).
After showing that the water table fluctuates in the radar data from year to year, the fact that the water table
intersects preexisting internal layering implies that lateral flow of water is influenced by the surface slope and
not bounded by possible laterally extensive refrozen ice layers. This observation, coupled with the interann-
ual repeat radar profiles, shows the free vertical bound of the water table and gives an unconfined behavior

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2392

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 10. (a) AR profile (2011) corrected for topography using the ATM elevation data [Krabill, 2013]. Profile location is
shown in Figure 5c and is oriented perpendicular to elevation contours. (b) The inset corresponds to the water table
depth (red), the surface slope (black), and water table slope (green). The vertical dashed lines illustrate that local minimum
water table depths are observed when the surface slope intersects 0.

to the aquifer. This is consistent with the unconfined firn aquifers observed on other glaciers [Fountain and
Walder, 1998; Christianson et al., 2015]. The thickness of the unconfined aquifer is unknown; at the GPR
and AR frequencies the bottom of the aquifer, where water-saturated firn transitions to glacial ice, is
not detected.
The unconfined aquifer assumption allows continuous hydraulic head estimates to be made directly from the
depth of the water table along the radar profile. Figure 10 illustrates the water table imaged by the AR over a
25 km along-flow transect, corrected for topography. Small-scale surface undulations (<5 km) with, for exam-
ple, 50 m of elevation changes are associated with water table fluctuations up to 20 m (Figure 10). Over the
full length of the profile (~25 km), the water table is deeper in the upper part, with a gradual decrease in the
depth to the water table; consistent with those made on mountain glacier aquifers, although the latter have
shorter aquifer lateral extents [Fountain, 1989]. Along our profile, the average surface slope is 0.64° and the
averaged hydraulic gradient is 0.01 (water table slope of 0.61°), 10 times lower than at Storglaciären (0.1
for a water table slope of 7.1°) [Schneider, 1999]. Minima in the depth to the water table coincide with areas
where the surface slope is flat; this is found extensively in our airborne data as well as along other
radar transects.

6.6. The 2-D Simulated Flow

SEEP2D simulates the flow of water within the firn aquifer while using the variables, assumptions, and bound-
ary conditions described in section 5.4. Figure 11 shows the simulated local flow cells [Tóth, 1963]; their size
depends on surface undulations, and the spacing between flow lines decreases with increasing water velo-
city. To interpret the groundwater simulation, we assume that Darcy’s law is valid and that there is no flow
into or out of the plane of the cross section (i.e., 2-D flow). Following Darcy’s law, the volumetric flow rate
Q (m3 s
1) of an incompressible fluid is calculated under stationary conditions within a porous medium of
the cross-sectional profile area A (m2), and length ΔL (m) under a head difference ΔH (m), which represents
the hydraulic gradient [Schneider, 1999]. These three variables (A, L, and H) are taken from the AR profile from
spring 2012 over an ~12 km long section (Figure 11).
ΔH
Q¼K A (3)
ΔL

We obtain an averaged total flow rate (Q) ranging between 3300 and 16,500 m3 yr
1 per meter width of the
profile, based on a hydraulic conductivity (K) range between 1 × 10
5 and 5 × 10
5 m s
1, taken from

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2393

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Figure 11. (a) The steady state flow lines (solid) simulated using 2-D flow model using water table and an assumed aquifer
thickness (25 m) as boundary conditions (dashed). The vertical arrows represent the local discharge locations that occur
at steep/flat transitions. (b) The AR profile used to retrieve the depth to water (similar to profile of Figure 4a, located
between red arrows in Figure 1).

mountain glacier studies [Fountain and Walder, 1998]. The flow rate needs to be taken with caution, as it is
directly proportional to the assumed value of K, which has not been measured directly at our field site.
Fluctuations in head translate to variations in drainage rates; as such, the drainage rate is not constant
throughout the aquifer. Spatial undulations of the water table can imply that local flow cells are created in
the aquifer with a similar scale to undulations of the surface topography (typically less than 10 km). The local
flow cells are strongly related to the local topography (Figure 11). Local surface depressions (vertical arrows in
Figure 11) may act as local discharge locations of the firn aquifer, as is discussed in section 7.2.

7. Discussion
7.1. Firn-Aquifer Variability
We study the temporal and spatial variabilities of the firn aquifer on AR profiles in the upper part of Helheim
and at our field location (Ridgeline site), where repeated data are available. At the field location, the firn-
aquifer lateral extent appears stable between the springs of 2011 and 2012, but both snow-surface elevation
and water table elevation decreased (Figure 7). The relatively constant depth to the water table between
these 2 years implies that water discharge was able to accommodate additional meltwater input, if any.
The vertical rise (+1.1 m) of the water table observed in spring 2013 is likely due to a combination of an
increase in surface melt in summer 2012, compared to 2011, and a reduction in snowfall prior to data collec-
tion. Data from nearby AWSs confirm that accumulated snowfall was 40% less in 2013, compared to the
2008–2014 mean, and that the annual mass loss in 2011 was ~5% below average, whereas 2012 was ~20%
above average. However, the rise of the water table observed in 2013 may not simply represent an increase
in water volume storage. For the Storglaciären firn aquifer, Schneider [1999] observed that the rise of the
water table was similar to the rise of firn/ice transition (~1.5 m yr
1), meaning that the aquifer thickness
was relatively constant. In our study, the lack of radar signal penetration through the aquifer prevents us from
evaluating this conclusion. The following winter (2013–2014) corresponded to another low snowfall year
(50% less than the 2008–2014 average), from the AWS data, but the aquifer water level dropped (
2.5 m).
This lowering can be explained by lateral water flow in the aquifer, associated with lower input of new melt-
water, depleting the water from these elevations. In case of shallow water tables, usually located less
than10 m from the surface, it is not excluded that refreezing from the winter cold wave might contribute
to lowering the water table. From the AWS data, 2013 was a relatively low melt year, with ablation rates
20% below average, implying that less meltwater was produced.
The OIB repeat radar surveys in the upper part of Helheim Glacier present a strong case for partial drainage of
the firn aquifer after spring 2012. Between 2012 and 2013, the aquifer disappeared at low elevations, while
the water surface rose at higher elevations (Figures 4 and 8). The absence of radar reflectors can be due to
(1) water drainage, (2) refreezing, or (3) reduced penetration of the AR signal. We favor the drainage

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2394

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

hypothesis by ruling out the other two possibilities. Refreezing seems unlikely, after the above-average sum-
mer melt of 2012; it would be difficult to refreeze substantial amounts of water in the firn, which would have
been temperate over a few years, particularly since, based on the AWS temperature data, winter tempera-
tures were not abnormally cold. Reduced penetration of AR radar signal also seems unlikely. We use a con-
tinuous bright radar reflector in the AR data to infer the presence of a water table; its absence might be
due to thick ice layers preventing the radar signal from reaching greater depths (>10 m). But even if the
AR signal (750 MHz) was attenuated, the lower-frequency MCoRDS system (195 MHz) would have been able
to indicate water. However, the MCoRDS profile shows clear bed-echo returns over the lower part of the pro-
file (Figure 4). Drainage of a 5 km stretch of the aquifer implies that hydraulic gradients and hydraulic conduc-
tivity must be high enough to mobilize a large volume of water and remove it, presumably into crevasses.
This lower part of the aquifer appears to be decoupled from the upper part, at least temporarily, until it
can be recharged either from seasonal surface meltwater production or from upstream water flowing later-
ally downglacier. Based on nearby AWS data, summer 2013 melt was below the 2008–2014 average, and the
low-elevation end of the aquifer stayed dry in spring 2014.

7.2. Lateral Water Flow

Hydraulic gradient is the main driver of the water flow over the relatively flat surface of the Greenland ice
sheet. For mountain glacier aquifers, hydraulic gradients can be up to 10 times greater than the ones for
the ice sheet but over smaller lateral extents; the steeper gradients route meltwater more efficiently, thus lim-
iting water from overwintering. The hydraulic gradient is related to the firn-aquifer thickness and surface
slope [Jansson et al., 2003]. For our study transect, the hydraulic gradient is 0.01 on average, an order of mag-
nitude lower than Storglaciären [Schneider, 1999], implying the presence of thicker firn aquifers within the ice
sheet. However, as the hydraulic gradients vary spatially following ice sheet surface undulations, the aquifer
thickness is expected to be variable.
In the percolation zone, the presence of ice lenses complicates water percolation. Horizontal ice lenses dis-
rupt the vertical flow of water, forcing it to flow horizontally [Conway and Benedict, 1994]. On the Devon
Ice Cap and in West Greenland, thick and continuous ice lenses force meltwater to pool above them and flow
laterally downstream either to run off or refreeze at the end of the melt season [Gascon et al., 2013; Machguth
et al., 2016]. At our field location, dur-
ing the spring 2014, the GPR profile
suggests slow downstream migration
of water, following hydraulic gradi-
ents (Figure 12). In this figure, the
water surface happens to diverge
from the water table while the sur-
face slope reaches 0°. A thick ice
layer, which creates an impermeable
boundary, could have been encoun-
tered by the water. The water pools
on this horizon and expands laterally,
slowly flowing downstream at the
firn ice interface, saturating the
above firn. This feature is thought to
be a perched water-saturated firn
layer on top of the firn aquifer. The
sharp termination of the reflector on
the flatter portion of the radar profile
implies that lateral water flow was
active, progressing downslope (white
arrow in Figure 12). In concordance
Figure 12. GPR profile from April 2014, corrected for surface elevation, over
a 1.5 km segment (Figure 1). A perched water table is observed ~1–2 m
with low hydraulic gradients, we infer
above the firn-aquifer surface, interpreted as water ponding on an ice layer a slow lateral motion of water during
and traveling downglacier under low hydraulic gradients. the winter, but we did not monitor its

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2395

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

changes in real time. This observation is consistent with observations from Humphrey et al. [2012], showing
that deep pore water slowly migrates downglacier under low hydraulic gradients.
Based on our hydraulic simulations, flow cells are revealed and the transition “steep to flat” marks the succes-
sion of local water discharge areas within the larger aquifer (~12 km), indicating local flow patterns (2–3 km;
Figure 11). This pattern was expected due to the undulating water table inferred from the radar data. These
local discharge features are located in surface depressions and seem to be transient, increasing the water
level temporally due to additional water coming from a steeper uphill. This process seems to happen
throughout the fall and winter (after the melt season shutdown) as the lateral flow evidence observed in
Figure 12 for early spring. However, our 2-D simulation does not include 3-D effects in water flow occurring
if flight lines do not intersect surface undulations at the lowest point of a given depression. Such a scenario
would make water flow in a direction that is not parallel to an ice-flow line, therefore not parallel to our 2-D
flow model, moving spatially the simulated discharge locations. If the hydraulic gradients were steep enough,
we would have observed water tables intersecting the surface in a few instances elsewhere over the ice sheet
in the AR data. Additionally, Christianson et al. [2015] reported water near or at the surface in a topographic
depression for the Holtedahlfonna firn aquifer in winter, consistent with low backscatter radar signals from
satellites. As a result, the near-surface water becomes sensitive to the cold winter temperature and partial
to complete refreezing would occur based on the amount of water. In west Greenland, water discharge
has also been observed, and Humphrey et al. [2012] documented the presence of pingo-type features (usually
found in permafrost) forming at similar locations to our local water discharge simulations (Figure 11), where
the slope transitions from steep to flat.

7.3. Impacts of Firn Aquifers

The impact of firn aquifers on broader ice sheet mass balance and hydrologic systems must be considered
since stored water within the aquifer influences firn densification, the mass of the firn column, and influences
meltwater runoff routing and timing. Currently, firn models implemented in regional climate models have a
water-retention scheme that only allows water to be held in the firn by capillary forces [e.g., Kuipers Munneke
et al., 2014]. This model assumption results in underestimation of water storage in firn-aquifer regions, as it
neglects the storage term at the firn-ice transition, where water tends to pool. Additionally, drainage from
the firn aquifer, such as we observe between 2012 and 2013, can impact ice dynamics by delivering melt-
water to the ice sheet bed or by increasing the ice temperature, contributing to cryo-hydrologic warming
[Phillips et al., 2013].
Firn aquifers are important sites for storage of the increased melt observed in Greenland. In the percolation
zone of West Greenland, Harper et al. [2012] reported available pore space in the firn being filled with melt-
water that refreezes, leading to an increased number of ice lenses, until capacity is reached. Recent observa-
tions, in a similar region [Machguth et al., 2016], suggested that formation of near-surface thick ice layers
renders pore space at depth inaccessible during intense melt years (i.e., 2012), promoting early runoff. For
the aquifer at our Ridgeline site, however, even if we observe surface lowering, we did not observe a substan-
tial interannual increase in ice fraction in the snow/firn column above the water table, not even after the
warm summer of 2012. If the additional meltwater is not contributing to an increase in the ice fraction of
the firn above the water table by refreezing, then this implies that there is slow runoff of the water at depth
under low hydraulic gradient year-round. However, our ice-fraction observations are spatially limited (four
firn cores only), and in places where water is close to the snow surface (less than 10 m), we suspect that freez-
ing would occur from top-down as the cold wave penetrates at depth during winter. The fate of the water
drained from the firn aquifer is not clear from the radar data set only. In crevassed areas, the absence of bright
reflectors in radar profiles suggests that the water could have drained into the crevasses either refreezing or
finding englacial pathways toward the glacier bed. On mountain glaciers, the firn aquifer is intimately linked
to the englacial network via crevasses and moulins [Fountain and Walder, 1998], allowing meltwater to access
the bed after being delayed in the aquifer. In Greenland, englacial pathways from the surface to the bed have
been inferred [e.g., Catania and Neumann, 2010]; however, there are no observations yet of meltwater exiting
the aquifer to reach the bed. The consequences of delaying or suppressing meltwater flow to the bed can be
related to seasonal changes in glacier velocity in Southeast Greenland [Moon et al., 2014]. For the studied
aquifer location, the ice thickness is >800 m and the formation of moulins, necessary for bringing water to
greater depths, depends on the water amount and rate. Indeed, to bring aquifer water to the bed, enough

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2396

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

water would need to flow at a high rate to keep up with the crevasse penetration rate [e.g., van der Veen,
2007]. However, an initial modeling effort using similar flow rates to the ones calculated in section 6.6 shows
that the aquifer can release sufficient water into downstream crevasses for fractures to propagate to the bed
[McNerney, 2016].

8. Conclusions
Widespread firn aquifers have been detected directly or inferred over high accumulation regions of the
Greenland ice sheet by a variety of radars systems: a 195 MHz radar depth sounder, a 750 MHz accumulation
radar (AR), and a 400 MHz ground-penetrating radar. A total linear distance of 3930 km of firn aquifers,
inferred to represent a total area of 21,900 km2, has been observed and mapped using the AR between spring
2010 and spring 2014 in the lower part of the percolation zone (wet-snow zone). The firn aquifers are mainly
observed in the southeast and south sectors of the ice sheet, representing an area of 17,920 km2. Firn aquifers
are found at an average elevation of 1615  300 m (1σ), an average accumulation rate of 1.0  0.4 m w.e.
yr
1 (1σ), and the average depth to the water table (top of the aquifer) is 22  7 m (1σ). Driven by low
hydraulic gradients (0.01), water in the aquifer flows downglacier following ice sheet surface undulations.
Within the firn aquifer, local flow cells are simulated and their lateral limits are related to the surface topogra-
phy with water discharge located at the steep-to-flat transitions. The firn-aquifer observed upstream of
Helheim Glacier responds to the recent above-average melt years and is expanding toward the ice sheet
interior. We also observed drainage of its lower elevation portion between spring 2012 and spring 2013.
The movement of meltwater from the aquifer into nearby crevasses would warm the ice in the vicinity by
refreezing and releasing latent heat. If the amount of water draining out of the aquifer into a crevasse is suffi-
Acknowledgments cient for hydrofracturing, the water could reach the bed, influencing ice dynamics. To further our understand-
We would like to thank Scientific Editor ing of firn-aquifer drainage, we recommend modeling work joint with water discharge measurements to
Bryn Hubbard for handling our manu-
script and for his comments, as well as characterize these englacial pathways and subglacial connections. Regarding firn-aquifer formation and evo-
an Associate Editor, three anonymous lution, we also recommend additional field observations to better constrain both the hydraulic variables
reviewers and Jack Kohler for many (hydraulic conductivity, water age, and residence time) and the water volume mobilized in the aquifer.
insights and suggestions that substan-
tially improved this manuscript. Thanks Although many firn-aquifer effects on the ice sheet still need to be further characterized, we have shown that
to NASA Operation IceBridge and firn aquifers are extensive features covering a substantial portion of the ice sheet percolation zone. They are
CReSIS at the University of Kansas for long-lived, dynamic systems responding to surface melt variations that will likely increase and play a more
the high-quality airborne data that
made this study possible. CReSIS data significant role in future warmer climate scenarios.
were generated with support from NSF
(ANT-0424589) and NASA
(NNX13AD53A). C. Miège acknowledges
References
NASA ESS Fellowship program Ahlstrøm, A. P., and the PROMICE project team (2008), A new programme for monitoring the mass loss of the Greenland ice sheet, Review of
(NNX11AL64H). NSF with grants Survey activities 2007, Geol. Surv. Den. Greenl. Bull., 15, 61––64.
OPP-0909499 and EAGER-1311655 Benson, C. S. (1962), Stratigraphic studies in the snow and firn of the Greenland ice sheet, CRREL Res. Rep. 70.
supported fieldwork in 2011 and 2013. Bradford, J. H., J. T. Harper, and J. Brown (2009), Complex dielectric permittivity measurements from ground-penetrating radar data to
L. Brucker and L. Koenig were supported estimate snow liquid water content in the pendular regime, Water Resour. Res., 45, W08403, doi:10.1029/2008WR007341.
by NASA’s Cryospheric Science Burgess, E. W., R. R. Forster, J. E. Box, E. Mosley-Thompson, D. H. Bromwich, R. C. Bales, and L. C. Smith (2010), A spatially calibrated model of
Program. R. Forster acknowledges NASA annual accumulation rate on the Greenland ice sheet (1958–2007), J. Geophys. Res., 115, F02004, doi:10.1029/2009JF001293.
grant NNX13AK23G. J. Box and R. Fausto Catania, G. A., and T. A. Neumann (2010), Persistent englacial drainage features in the Greenland Ice Sheet, Geophys. Res. Lett., 37(2), L02501,
are supported by the Det Frie doi:10.1029/2009GL041108.
Forskningsråd of Denmark grant Christianson, K., J. Kohler, R. B. Alley, C. Nuth, and W. J. J. van Pelt (2015), Dynamic perennial firn aquifer on an Arctic glacier, Geophys. Res.
DFF–4002-00234. Thanks to the 2011 Lett., 42, 1418–1426, doi:10.1002/2014GL062806.
and 2013 field team members: Colbeck, S. C. (1975), A theory for water flow through a layered snowpack, Water Resour. Res., 11, 261–266, doi:10.1029/WR011i002p00261.
B. Ballard, T. Gacke, and J. Kyne. Thanks Coléou, C., and B. Lesaffre (1998), Irreducible water saturation in snow: Experimental results in a cold laboratory, Ann. Glaciol., 26, 64–68.
to the following institutions for the Conway, H., and R. Benedict (1994), Infiltration of water into snow, Water Resour. Res., 30, 641–649, doi:10.1029/93WR03247.
fieldwork support: CH2MHILL, IDDO, Enderlin, E. M., I. M. Howat, S. Jeong, M.-J. Noh, J. H. van Angelen, and M. R. van den Broeke (2014), An improved mass budget for the
109th Air National Guard, Air Greenland, Greenland ice sheet, Geophys. Res. Lett., 41, 866–872, doi:10.1002/2013GL059010.
and UNAVCO. Thanks to G. Roth at the Fettweis, X., B. Franco, M. Tedesco, J. H. van Angelen, J. T. M. Lenaerts, M. R. van den Broeke, and H. Gallée (2013), Estimating the Greenland
Polar Geospatial Center for preparing ice sheet surface mass balance contribution to future sea level rise using the regional atmospheric climate model MAR, Cryosphere, 7(2),
the DigitalGlobe© images. Thanks to 469–489, doi:10.5194/tc-7-469-2013.
J. Collins at CReSIS for the careful Forster, R. R., et al. (2014), Extensive liquid meltwater storage in firn within the Greenland ice sheet, Nat. Geosci., 7(2), 95–98, doi:10.1038/
grammatical edits on this manuscript. ngeo2043.
Thanks to Xavier Fettweis at the Fountain, A. G. (1989), The storage of water in, and hydraulic characteristics of, the firn of South Cascade Glacier, Washington State, U.S.A.,
University of Liège for the availability of Ann. Glaciol., 13, 69–75.
his MAR output data. The firn-aquifer Fountain, A. G. (1996), Effect of snow and firn hydrology on the physical and chemical characteristics of glacial runoff, Hydrol. Processes, 10(4),
map and depth to the water table are 509–521.
available for download at Doi: 10.18739/ Fountain, A. G., and J. S. Walder (1998), Water flow through temperate glaciers, Rev. Geophys., 36, 299–328, doi:10.1029/97RG03579.
A2985M or by contacting the lead author Gascon, G., M. Sharp, D. Burgess, P. Bezeau, and A. B. G. Bush (2013), Changes in accumulation-area firn stratigraphy and meltwater flow during a
([email protected]). period of climate warming: Devon Ice Cap, Nunavut, Canada, J. Geophys. Res. Earth Surf., 118, 2380–2391, doi:10.1002/2013JF002838.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2397

 Journal of Geophysical Research: Earth Surface 10.1002/2016JF003869

Harper, J., N. Humphrey, W. T. Pfeffer, J. Brown, and X. Fettweis (2012), Greenland ice-sheet contribution to sea-level rise buffered by melt-
water storage in firn, Nature, 491(7423), 240–243, doi:10.1038/nature11566.
Helm, V., A. Humbert, and H. Miller (2014), Elevation and elevation change of Greenland and Antarctica derived from CryoSat-2, Cryosphere,
8(4), 1539–1559, doi:10.5194/tc-8-1539-2014.
Howat, I. M., A. Negrete, and B. E. Smith (2014), The Greenland Ice Mapping Project (GIMP) land classification and surface elevation data sets,
Cryosphere, 8(4), 1509–1518, doi:10.5194/tc-8-1509-2014.
Humphrey, N. F., J. T. Harper, and W. T. Pfeffer (2012), Thermal tracking of meltwater retention in Greenland’s accumulation area, J. Geophys.
Res., 117, F01010, doi:10.1029/2011JF002083.
Jansson, P., R. Hock, and T. Schneider (2003), The concept of glacier storage: A review, J. Hydrol., 282(1–4), 116–129, doi:10.1016/S0022-1694
(03)00258-0.
Jones, N. L. (1999), SEEP2D Primer, Environ. Model. Res. Lab., Brigham Young Univ., Provo, Utah.
Koenig, L. S., C. Miège, R. R. Forster, and L. Brucker (2014), Initial in situ measurements of perennial meltwater storage in the Greenland firn
aquifer, Geophys. Res. Lett., 41, 81–85, doi:10.1002/2013GL058083.
Kovacs, A., A. J. Gow, and R. M. Morey (1995), The in-situ dielectric constant of polar firn revisited, Cold Reg. Sci. Technol., 23(3), 245–256,
doi:10.1016/0165-232X(94)00016-Q.
Krabill, W. B. (2013), IceBridge ATM L2 Icessn Elevation, Slope, and Roughness, version 2 [2010–2014], NASA DAAC at the Natl. Snow and Ice
Data Cent., Boulder, Colo.
Krabill, W. B., W. Abdalati, E. B. Frederick, S. S. Manizade, C. F. Martin, J. G. Sonntag, R. N. Swift, R. H. Thomas, and J. G. Yungel (2002), Aircraft
laser altimetry measurement of elevation changes of the Greenland ice sheet: Technique and accuracy assessment, J. Geodyn., 34(3–4),
357–376, doi:10.1016/S0264-3707(02)00040-6.
Kuipers Munneke, P., S. R. M. Ligtenberg, M. R. van den Broeke, J. H. van Angelen, and R. R. Forster (2014), Explaining the presence of per-
ennial liquid water bodies in the firn of the Greenland ice sheet, Geophys. Res. Lett., 41, 476–483, doi:10.1002/2013GL058389.
Leuschen, C., C. Lewis, P. Gogineni, F. Rodriguez-Morales, J. Paden, and J. Li (2014a), IceBridge accumulation radar L1B geolocated radar echo
strength profiles, version 2 [2010–2014], NASA DAAC at the Natl. Snow and Ice Data Cent., Boulder, Colo.
Leuschen, C., P. Gogineni, R. Hale, J. Paden, F. Rodriguez-Morales, B. Panzer, and D. Gomez (2014b), IceBridge MCoRDS L1B geolocated radar
echo strength profiles, version 2 [2010–2014], NASA DAAC at the Natl. Snow and Ice Data Cent., Boulder, Colo.
Machguth, H., M. MacFerrin, D. van As, J. E. Box, C. Charalampidis, W. Colgan, R. S. Fausto, H. A. J. Meijer, E. Mosley-Thompson, and
R. S. W. van de Wal (2016), Greenland meltwater storage in firn limited by near-surface ice formation, Nat. Clim. Change, 6, 390–393,
doi:10.1038/nclimate2899.
McGrath, D., W. Colgan, N. Bayou, A. Muto, and K. Steffen (2013), Recent warming at Summit, Greenland: Global context and implications,
Geophys. Res. Lett., 40, 2091–2096, doi:10.1002/grl.50456.
McNerney, L. (2016), Constraining the Greenland firn aquifer’s ability to hydrofracture a crevasse to the bed of the ice sheet, pp. 1–54 .
Miège, C., R. R. Forster, J. E. Box, E. W. Burgess, J. R. McConnell, D. R. Pasteris, and V. B. Spikes (2013), Southeast Greenland high accumulation
rates derived from firn cores and ground-penetrating radar, Ann. Glaciol., 54(63), 322–332, doi:10.3189/2013AoG63A358.
Moon, T., and I. Joughin (2008), Changes in ice front position on Greenland’s outlet glaciers from 1992 to 2007, J. Geophys. Res., 113, F02022,
doi:10.1029/2007JF000927.
Moon, T., I. Joughin, B. Smith, M. R. van den Broeke, W. J. van de Berg, B. Noël, and M. Usher (2014), Distinct patterns of seasonal Greenland
glacier velocity, Geophys. Res. Lett., 41, 7209–7216, doi:10.1002/2014GL061836.
Oerter, H., and H. Moser (1982), Water storage and drainage within the firn of a temperate glacier (Vernagtferner, Oetztal Alps, Austria), in
Hydrological Aspects of Alpine and High-Mountain Areas, Proceedings of the Exeter Symposium, vol. 138, pp. 71–81, IAHS Publ. no. 138.
Pfeffer, W. T., and N. F. Humphrey (1996), Determination of timing and location of water movement and ice-layer formation by temperature
measurements in sub-freezing snow, J. Glaciol., 42(141), 292–304.
Phillips, T., H. Rajaram, W. Colgan, K. Steffen, and W. Abdalati (2013), Evaluation of cryo-hydrologic warming as an explanation for increased
ice velocities in the wet snow zone, Sermeq Avannarleq, West Greenland, J. Geophys. Res. Earth Surf., 118, 1241–1256, doi:10.1002/
jgrf.20079.
Raymond, C. F., and K. Tusima (1979), Grain coarsening of water-saturated snow, J. Glaciol., 22, 83–105.
Rignot, E., and J. Mouginot (2012), Ice flow in Greenland for the International Polar Year 2008–2009, Geophys. Res. Lett., 39, L11501,
doi:10.1029/2012GL051634.
Rignot, E., I. Velicogna, M. R. van den Broeke, A. Monaghan, and J. T. M. Lenaerts (2011), Acceleration of the contribution of the Greenland and
Antarctic ice sheets to sea level rise, Geophys. Res. Lett., 38, L05503, doi:10.1029/2011GL046583.
Rodriguez-Morales, F., et al. (2013), Advanced multifrequency radar instrumentation for polar research, IEEE Trans. Geosci. Remote Sens., 52,
2824–2842, doi:10.1109/TGRS.2013.2266415.
Schneider, T. (1999), Water movement in the firn of Storglaciaren, Sweden, J. Glaciol., 45(150), 286–294.
Schneider, T., and P. Jansson (2004), Internal accumulation in firn and its significance for the mass balance of Storglaciären, Sweden,
J. Glaciol., 50(168), 25–34, doi:10.3189/172756504781830277.
Smith, B. M., and S. Evans (1972), Radio echo sounding: Absorption and scattering by water inclusion and ice lenses, J. Glaciol., 11, 133–146.
Spikes, V. B., G. S. Hamilton, S. A. Arcone, S. Kaspari, and P. A. Mayewski (2004), Variability in accumulation rates from GPR profiling on the
West Antarctic plateau, Ann. Glaciol., 39(1), 238–244, doi:10.3189/172756404781814393.
Tedesco, M., X. Fettweis, T. Mote, J. Wahr, P. Alexander, J. E. Box, and B. Wouters (2013), Evidence and analysis of 2012 Greenland records from
spaceborne observations, a regional climate model and reanalysis data, Cryosphere, 7(2), 615–630, doi:10.5194/tc-7-615-2013.
Tóth, J. (1963), A theoretical analysis of groundwater flow in small drainage basins, J. Geophys. Res., 68, 4795–4812, doi:10.1029/
JZ068i016p04795.
van der Veen, C. J. (2007), Fracture propagation as means of rapidly transferring surface meltwater to the base of glaciers, Geophys. Res. Lett.,
34, L01501, doi:10.1029/2006GL028385.

Erratum
The Acknowledgments section of this article has been amended to include an updated link to the data. This
may now be considered the authoritative version of record.

MIÈGE ET AL. GREENLAND FIRN AQUIFERS 2398