ArticlePDF Available

Oceanic Boundary Conditions for Jakobshavn Glacier. Part I: Variability and Renewal of Ilulissat Icefjord Waters, 2001–14



Jakobshavn Glacier, west Greenland, has responded to temperature changes in Ilulissat Icefjord, into which it terminates. This study collected hydrographic observations inside Ilulissat Icefjord and from adjacent Disko Bay between 2001 and 2014. The warmest deep Disko Bay waters were blocked by the entrance sill and did not reach Jakobshavn Glacier. In the fjord basin, the summer mean temperature was 2.8°C from 2009 to 2013, excluding 2010, when it was 1°C cooler. Despite this variability, summer potential densities in the basin were in the narrow range of 27.20 ≤ σ[subscript θ] ≤ 27.31 kg m[superscript −3], and basin water properties matched those of Disko Bay in this layer each summer. This relation has likely held since at least 1980. Basin waters from 2009 and 2011–13 were therefore similar to those in 1998/99, when Jakobshavn Glacier began to retreat, while basin waters in 2010 were as cool as in the 1980s. The 2010 basin temperature anomaly was advected into Disko Bay, not produced by local atmospheric variability. This anomaly also shows that Ilulissat Icefjord basin waters were renewed annually or faster. Time series fragments inside the fjord did not capture the 2010 anomaly but show that the basin temperatures varied little subannually, outside of summer. Fjord velocity profiles from summer 2013 implied a basin renewal time scale of about 1 month. In model simulations of the fjord circulation, subglacial discharge from Jakobshavn Glacier could drive renewal of the fjord basin over a single summer, while baroclinic forcing from outside the fjord could not, because of the sill at the mouth.
Oceanic Boundary Conditions for Jakobshavn Glacier. Part I: Variability and
Renewal of Ilulissat Icefjord Waters, 2001–14*
New York University Abu Dhabi, Abu Dhabi, United Arab Emirates
New York University, New York, New York
Greenland Institute of Natural Resources, Nuuk, Greenland
Technical University of Denmark, Charlottenlund, Denmark
(Manuscript received 11 March 2014, in final form 22 October 2014)
Jakobshavn Glacier, west Greenland, has responded to temperature changes in Ilulissat Icefjord, into which it
terminates. This studycollected hydrographic observationsinside Ilulissat Icefjord and from adjacentDisko Bay
between 2001 and 2014. The warmest deep Disko Baywaters were blockedby the entrance sill and did not reach
Jakobshavn Glacier. In the fjord basin, the summer mean temperature was 2.88C from 2009 to 2013, excluding
2010, when it was 18C cooler. Despite this variability, summerpotential densities in the basin were in the narrow
range of 27.20 #s
#27.31 kg m
, and basin water properties matched those of Disko Bay in this layer each
summer. This relation has likely held since at least 1980. Basin waters from 2009 and 2011–13 were therefore
similar to those in 1998/99, whenJakobshavn Glacier began toretreat, while basin waters in 2010 were as cool as
in the 1980s. The 2010 basin temperature anomaly was advected into Disko Bay, not produced by local at-
mospheric variability.
This anomaly also shows that Ilulissat Icefjord basin waters were renewed annually or faster. Time series
fragments inside the fjord did not capture the 2010 anomaly but show that the basin temperatures varied little
subannually, outside of summer. Fjord velocity profiles from summer 2013 implied a basin renewal time scale
of about 1 month. In model simulations of the fjord circulation, subglacial discharge from Jakobshavn Glacier
could drive renewal of the fjord basin over a single summer, while baroclinic forcing from outside the fjord
could not, because of the sill at the mouth.
1. Introduction
Over time scales of tens to hundreds of millennia, the
Greenland Ice Sheet (GrIS) has undergone massive
changes in response to changing climate. During the
previous interglacial, when boreal summer temperatures
were up to 58C warmer than at present, a smaller GrIS
contributed 2 m to a global-mean sea level that was at
least 4m higher than that of today (Colville et al. 2011;
Dahl-Jensen et al. 2013). On multimillenial time scales,
* Supplemental information related to this paper is available at
the Journals Online website:
Current affiliation: Department of Earth, Atmospheric and
Planetary Sciences, Massachusetts Institute of Technology, Cam-
bridge, Massachusetts.
Corresponding author address: Carl Gladish, MIT Bldg. 54-1423,
77 Massachusetts Ave., Cambridge MA, 02139.
DOI: 10.1175/JPO-D-14-0044.1
Ó2015 American Meteorological Society 3
large future volume changes of the GrIS will probably be
controlled by the feedback between surface elevation and
surface temperature (Levermann et al. 2013). In the
present era of warming climate, however, in which the
GrIS has been contributing to global sea level at a rate of
0.6 mm yr
(Shepherd et al. 2012), ice loss is occurring
largely at Greenland’s marine outlet glaciers (Rignot and
Kanagaratnam 2006;Thomas et al. 2009;Pritchard et al.
2009), which can respond to changing climate, especially
ocean temperature changes, through fast dynamical mech-
anisms (Thomas 2004;Howatetal.2007).
Jakobshavn Glacier (JG), a marine glacier that termi-
nates into the 750–800-m-deep Ilulissat Icefjord (IIf;
698100N, 508300W; see Fig. 1) in west Greenland, has been
losing mass more rapidly than other Greenland outlet
glaciers in recent years (Howatetal.2011;Joughin et al.
2008). It appears that its acceleration and retreat were
initiated by a switch to warmer ocean temperatures along
west Greenland in the late 1990s (Holland et al. 2008;
Motyka et al. 2011;Hansen et al. 2012;Rignot et al. 2012).
Ocean temperature changes in the adjacent Disko Bay
(DB; see Fig. 1) have likely had a controlling influence on
JG for at least the past 100 yr (Lloyd et al. 2011). In gen-
eral, continued observation and further understanding of
glacier–ocean interactions around Greenland and Ant-
arctica is necessary for modeling past and future ice sheet
changes (Joughin et al. 2012a).
The premise that the surrounding ocean modulates the
behavior of JG motivated our collecting a multiyear record
of water properties in DB and IIf. These data include
annual surveys, moorings, and instrumented seals and
Greenland halibut. Similar efforts have been undertaken
at several major Greenland outlet glaciers (Rignot et al.
2010;Mortensen et al. 2011;Straneo et al. 2010;Murray
et al. 2010;Christoffersen et al. 2011;Rignot and Steffen
2008;Johnson et al. 2011). Summer water properties in DB
have been routinely monitored since the mid-twentieth
century (Andersen 1981) and more intensively in recent
decades (Hansen et al. 2012;Ribergaard 2013). IIf, how-
ever, is typically full of icebergs trapped in the deep fjord
interior by the 50–245-m-deep sill that runs across the fjord
mouth (Schumann et al. 2012), and entering the fjord in
a research vessel is typically dangerous or impossible. In
section 2,wedescribeourdataset.
Our main observational result (section 3a) is a record
of summer water temperatures in the IIf basin between
2007 and 2013 and an understanding of the relation to
Disko Bay waters. The IIf basin, which makes up the
major part of the oceanic boundary facing JG, is the
;500-m-thick layer below the depth of the sill’s saddle
point (245 m). Our second main finding (section 3b),
based on disparate fjord time series data, was that the IIf
basin had little subseasonal temperature variability. IIf
therefore differs from Sermilik Fjord, east Greenland,
FIG. 1. The main map shows DB and IIf, west Greenland, with bathymetric contours. Markers
show the locations for CTD, AXCTD, and XCTD stations from 2007 to 2013. Lines show the in-
terpolation path for vertical sections in Fig. 3. Markers are displaced up to 1 min of latitude for
legibility. The villages of Qeqertarsuaq, Aasiaat, and Ilulissat are marked for reference. The light blue
line marks a typical longitude of JG terminus between 2009 and 2011.The black box defines the fjord
mouth region (Table 3). The red box shows the extent of the map in Fig. 2. The pink marker in the
inset shows the location of mooring WG1 on the east side of DS. The WGC is shown schematically.
which is dominated by subseasonal variability (Jackson
et al. 2014).
We found that the mean summer temperature in the
IIf basin from 2009 to 2013 (excluding 2010) was 2.88C,
which is very similar to the temperature of equally dense
DB waters in 1998 and 1999, when JG began to retreat.
Fjord basin temperatures in summer 2010 were conspic-
uously cooler than neighboring years—by about 18C. The
cooling of the IIf basin (along with equally dense DB
waters) from summer 2009 to summer 2010 and the return
to 2009-like conditions in summer 2011 raised several
questions. First, what effect did the cooler water have on
JG, considering that a ;18C warming in the late 1990s
likely triggered the retreat of the glacier? We address this
question briefly in section 4. Second, what dynamics were
responsible for the complete renewal of the fjord basin
from one summer to the next? An initial hypothesis might
be that the IIf basin is renewed only when denser water
appears at the sill depthoutside the fjord (Arneborg et al.
2004). We argue in section 5, however, that the basin is
renewed mainly in summer by an overturning circulation
driven by subglacial discharge and terminus melting.
Subglacial discharge is important in determining melt
rates at marine termini in some instances (Motyka et al.
2003;Jenkins 2011;Xu et al. 2013); here, we investigate its
role in driving the renewal of IIf. If correct, the renewal
mechanisms of IIf and Sermilik Fjord differ. At deep-
silled Sermilik Fjord, subglacial discharge has an impact
on fjord circulation (Straneo et al. 2011), but the fjord is
renewed subseasonally by oscillatory baroclinic currents
driven by external wind forcing (Straneo et al. 2010;
Jackson et al. 2014). In section 6, we present model sim-
ulations of the circulation in an idealized IIf. We used the
MITgcm model (Marshall et al. 1997) with a setup similar
to that of Xu et al. (2012) and Sciascia et al. (2013),but
with an emphasis on how sill geometry affects the efficacy
of these two potential drivers of fjord renewal.
Finally, there is the question of the ultimate origin of
interannual temperature variability in DB/IIf. In a com-
panion paper (Gladish et al. 2015), we address this ques-
tion by examining hydrographic and meteorological data
from the wider region.
2. Observational data
a. Hydrographic profile data
A profile from the north arm of IIf (Fig. 1)wasre-
trieved using an airborne expendable CTD (AXCTD)
probe on 7 August 2007 along with conventional CTD
profiles (7 June 2007) at 12 stations selected for annual
reoccupation (Fig. 1, gray markers) at the fjord mouth
(Holland et al. 2008). On 20 July 2008, profiles at 2 of the
12 standard stations were collected using a Seabird
Electronics (SBE) 19plus V2 CTD. On 3 August 2009, 8
August 2010, 21 July 2011, 28 June 2012, and 19 June
2013, the 12 standard CTD stations were again occupied,
along with additional stations some summers. In 2012 and
2013, our CTD surveys covered much of DB (Fig. 1). One
SBE 19plus V2 CTD was used in 2009 and another was
used for 2010 to 2013. Both instruments were recalibrated
between the 2012 and 2013 seasons and showed levels of
drift that were small compared to the variability we focus
on in this work. After correcting for an assumed uniform
drift rate, the accuracy of these CTD data are estimated
to be close to factory-determined levels of 60.0058Cfor
temperature and 60.01 g kg
for salinity.
Each summer from 2009 to 2013, 9 to 12 expendable
CTD (XCTD) probes (Tsurumi Seiki Co.) were success-
fully deployed from a hovering helicopter into IIf (loca-
tions in Fig. 1). Each probe typically sampled the entire
water column, transmitting a digital signal along a fine
copper wire to a receiver in the helicopter. The accuracy
determined by the manufacturer is 60.028C for tempera-
ture and 60.03 mS cm
for conductivity, which corre-
spondstoasalinityaccuracyof60.05 g kg
survey of the fjord came within 10–15 km of the calving
front of JG most years and within 1–3 km in 2010 and 2013.
A cross comparison of XCTD and CTD profiles at three
standard stations in 2009 showed agreement in temperature
to within 0.18C and in salinity to within 0.04 g kg
25-m depth after vertically shifting the XCTD profiles by
a few meters (XCTD probes only activate after detecting
a minimum conductivity of about 15 mS cm
and depths
are accurate to within 2%, according to the manufacturer).
Our analysis and interpretation also makes use of hy-
drographic data collected in DB by researchers of the
Arctic Station on Disko Island, described in Hansen et al.
(2012). Also, several CTD profiles from DB collected by
the Greenland Institute of Natural Resources (GINR) in
summer of 2007 were used. Finally, we used a large number
of CTD profiles from DB collected annually by the Danish
Meteorological Institute (DMI) and the GINR, which can
be downloaded from the ICES oceanographic database.
For all profile data, downcast segments were manually
selected, avoiding surface and bottom artifacts, and
samples were averaged over 2-m vertical bins before
subsequent analysis. Salinities and other thermody-
namic variables were calculated from conductivity, in
situ temperature, and pressure using the International
Thermodynamic Equation Of Seawater—2010 (TEOS-
10) standard
as recommended by IOC et al. (2010) and as
implemented by available MATLAB routines (McDougall
JANUARY 2015 G L A D I S H E T A L . 5
and Barker 2011). In particular, salinity here refers to
Absolute Salinity S
, the mass fraction of dissolved mate-
rial in a given sample, and is therefore stated with units of
grams per kilogram. In these units, S
differs numerically
from salinities S
on the practical salinity scale of 1978. For
waters of the subpolar gyre and Baffin Bay with S
the relation S
10.16) g kg
a crude but convenient mental translation.
b. Moorings
The names, dates, depths, and locations of all moor-
ings are provided in Table 1. We deployed a mooring
carrying a SBE 37 MicroCAT just inside the fjord
(Fig. 2, red marker) at 400-m depth in August 2009.
Judging by its pressure record, it moved vertically in
a series of essentially discrete jumps, suggesting it was
displaced by icebergs. It could not have crossed the sill at
the fjord mouth before February 2010, although it was
eventually recovered 30 km north of the fjord mouth by
a fisherman in June 2010. We therefore split the data for
this mooring into four separate segments where the
mooring depth was essentially constant (first four series
in Table 1). During segments Fjord 400 m and Fjord
290 m, the mooring must have been east of the sill inside
the fjord, and for segments DB 100 m and DB 58 m the
mooring was at shallower depths and had probably
somehow moved into DB. Mooring DB 296 m, carrying
a Teledyne RD Instruments (RDI) Citadel CTD-NH,
was deployed in DB near the fjord mouth (Fig. 2, orange
marker) in August 2010 and was shortly after caught by
a fisherman. In July 2011, a mooring we deployed near
the fjord mouth was again caught by a fisherman less
than 3 weeks later. A second mooring, DB 350 m, was
deployed in July 2011 with a newly calibrated SBE 37
west of the fjord mouth, out of range of habitual fishing
grounds (Fig. 2, green marker), which we successfully
recovered 1 yr later. These moorings all sampled every
10 min. The most extreme 1% of samples by density
were discarded and interpolated over to remove spuri-
ous spikes.
For the purpose of comparison, we downloaded data
from the Arctic Observing Network archive
a mooring deployed by University of Washington re-
searchers nearly continuously since 2004 at the Greenland
shelf break in eastern Davis Strait (DS). This mooring,
identified as WG1, was located at 150-m depth in the path
of the West Greenland Current (WGC), which is the pri-
mary pathway for warm waters of low-latitude origin to
enter Baffin Bay (Fig. 1, inset). At 150-m depth, WG1 was
shallower than the warm core of the WGC. It did,
however, monitor waters similar in density to IIf basin
waters. Two 1-yr-long segments (October 2009–September
2010 and October 2010–September 2011) were extracted,
with samples every 30 min (Table 1).
c. Virtual moorings from instrumented seals
In both September 2012 and August 2013, three ringed
seals were captured and instrumented with satellite-
telemetered CTD tags provided by the Sea Mammal Re-
search Unit (SMRU) of the University of St. Andrews.
The tags transmitted data from a representative set of
depth levels for each seal dive. The longest-lasting tag
transmitted dive data for 9 months. The horizontal loca-
tions of the seal dives varied over the main fjord and its
north and south arms (Fig. 2). The SMRU tags measure
temperatures with an accuracy of 60.0058C and conduc-
tivities with an accuracy of 60.01 mS cm
, which corre-
sponds to a salinity accuracy of about 0.02 g kg
the seal dive profiles, we synthesized five ‘‘virtual moor-
ing’’ time series made up of data extracted at fixed depths
in IIf, disregarding the horizontal location within the fjord
(Table 1). For instance, Seals 100 m 2013 is made up of data
extracted at 100-m depth from dives during the overwinter
period 2012/13. The temporal spacing was irregular, so the
samples were averaged into 1-day bins and gaps were in-
terpolated over.
TABLE 1. Mooring deployments. Virtual moorings from seal
dives sampled irregularly in time and dive locations were scattered
as shown in Fig. 2. Precise locations are unknown for Fjord 290 m,
DB 100 m, and DB 58 m because of horizontal motion. The term
Dt5duration of deployment (days); T5sample period (median
time for seal dives; units specified).
(8W) Start DtT
Fjord 400 m 69.1859 51.1269 4 Aug 2009 65 10 m
Fjord 290 m 4 Nov 2009 78 10 m
DB 100 m 10 Feb 2010 70 10 m
DB 58 m 15 May 2010 16 10 m
DB 296 m 69.2204 51.1869 7 Aug 2010 14 10 m
DB 350 m 69.1756 51.3762 25 Jul 2011 334 10 m
Seals 100 m 2013 16 Sep 2012 188 5.0 h
Seals 300 m 2013 18 Sep 2012 125 7.0 h
Seals 100 m 2014 2 Aug 2013 268 5.5 h
Seals 300 m 2014 2 Aug 2013 268 6.2 h
Seals 500 m 2014 10 Sep 2013 175 10.5 h
WG1 150 m 2010 67.1054 56.3274 18 Oct 2009 347 30 m
WG1 150 m 2011 67.1057 56.3284 1 Oct 2010 365 30 m
Daily samples were missing for 13% of Seals 100 m 2013, 30%
of Seals 300 m 2013, 5% of Seals 100 m 2014, 26% of Seals 300 m
2014, and 72% of Seals 500 m 2014.
d. Data from instrumented Greenland halibut
Between autumn 2001 and autumn 2003, a total of 181
Greenland halibut (a deep-water flatfish) were caught in
the waters off Ilulissat (Fig. 1) and tagged with temper-
ature and pressure recording devices. There were 12 fish
recovered 50 to 175 days later, either inside IIf or in DB.
These dataare not precisely geolocated, but by examining
the pressure records in relation to the known bathymetry
of the region, we identified multimonth time intervals
when individuals were very likely inside IIf. Further-
more, of the 12 recaptured fish, 3 were recaptured deep
inside the fjord and 3 were just inside the mouth of the
fjord, thus confirming their residence inside the fjord.
These data therefore provide time series of in situ
temperature at 10- to 15-min sampling periods inside the
deep basin of the fjord, generally during winter months.
Further details on tagging, collection, and processing of
these data can be found in Boje et al. (2014).
e. XCP profiles
Four Lockheed Martin expendable current profilers
(XCP) were deployed along the fjord in summer 2013
(locations in Fig. 2). Three of these (V1, V2, and V3)
were deployed on 10 June 2013 within 25 min of each
other. The fourth (V4) was deployed on 20 June 2013.
Velocity measurements were averaged over 5-m vertical
bins to reduce random errors. Using temperature pro-
files recorded by each XCP device and temperatures
from nearby XCTD profiles, vertical offset errors were
removed (relative to the XCTD depths).
f. Meteorological data
We used DMI meteorological data from the airports
in the towns of Ilulissat and Aasiaat from 2007 to 2012.
An automated weather station (AWS) we maintain near
the terminus of JG (Fig. 2) has a nearly complete record
of temperature, pressure, and wind velocity from a site
101 m above sea level since 2007.
g. Bathymetry
Regional bathymetry, which we extracted from In-
ternational Bathymetric Chart of the Arctic Ocean ver-
sion 3 (IBCAO V3) (Jakobsson et al. 2012), plays an
important role in determining which water masses are
able to reach DB and IIf. One important feature is the
Egedesminde Dyb, a 300- to 900-m trough cutting across
the continental shelf from the shelf break into DB (Fig. 1,
main map and inset). The shallowest part of the trough,
southwest of Qeqertarsuaq, is an important barrier for
deep water. We refer to this 300-m-deep barrier as the
Egedesminde Dyb Sill (EDS). In this paper, we focus our
attention to the east of the EDS, while in the companion
paper we consider sources of variability to the west of the
EDS and the controlling influence of the EDS itself on
variability in DB. The second important bathymetric
impediment is the Iceberg Bank sill at the mouth of IIf.
Swath bathymetry from Schumann et al. (2012) maps this
FIG. 2. Background Landsat image (from September 1999) shows iceberg-filled IIf prior to
the breakup of the floating proglacial ice shelf of JG. Calving fronts from 2009 to 2011 are
shown as black lines. Locations of the New York University (NYU) AWS, mooring sites, XCP
velocity stations, repeat CTD stations, and seal dive locations are shown (some appear to be on
land because of intermittent position inaccuracies). The 245-m bathymetric contour from
Schumann et al. (2012) is shown near the mouth of the fjord. The saddle point at 245-m depth is
apparent near the northern side of the fjord mouth.
JANUARY 2015 G L A D I S H E T A L . 7
seafloor ridge and shows that its deepest point (the
saddle point of the sill) is at 245-m depth just southwest
of the town of Ilulissat (Fig. 2). Over the southern half
of the fjord mouth the sill is no deeper than 150 m. In
the following, sill depth refers to the depth of the sill
saddle point.
3. Fjord Variability
a. Interannual variability from synoptic surveys
Sections of potential temperature u, linearly in-
terpolated along the paths shown in Fig. 1 from inner
Egedesminde Dyb to nearly JG, are shown in Fig. 3.In
the fjord basin, the 500-m-deep bathtub-shaped volume
east of Ilulissat, temperatures were quite homogeneous
with less than 1.08C of range. Each summer, the densest
layers in DB that were able to pass over the sill increased
in thickness along the backside of the sill, forming the
nearly homogeneous layer filling the fjord basin. In each
section, the downward slope of isotherms (and iso-
pycnals, not shown) over the backside of the sill is the
signature of a critical flow over the sill and suggests that
water was rapidly pouring over the sill into the fjord
basin (Whitehead 1998). We therefore refer to the sea-
floor slope joining the sill to the floor of the fjord basin as
the spillway.
When profiles from all years are plotted together on
diagram (Fig. 4) several things are apparent.
First, waters below about 300-m depth in the fjord had
properties matching those of waters at intermediate
depths (between 100- and 300-m depth) in DB. The
densest and warmest waters in DB did not cross the sill
(see also Fig. 3). Second, we found that fjord waters in
2009 and 2011–13 were distinctly warmer than in 2007
and 2010. In the warmer years, fjord basin waters were
between 2.58and 3.08C, while in the cooler years they did
not exceed 2.08C. We define two regions of uS
(as shown in Fig. 4 and defined precisely in Table 2),
which we refer to as Warm Fjord Water (WFjW) and
Cool Fjord Water (CFjW). In Fig. 3, waters falling within
the definitions of WFjW and CFjW are enclosed by black
and gray curves, respectively. These curves illustrate the
WFjW or CFjW layers extending continuously across
DB (centered at 200-m depth or shallower) and into IIf.
The exception was 2007, when fjord basin water was be-
tween WFjW and CFjW. Third, we find that, despite
nearly 1.08C variations in temperature, fjord basin waters
always had potential density anomaly s
between 27.20
and 27.31 kg m
. The single AXCTD profile from 2007
had a smaller maximum density partly because of the
shallowness of the fjord’s north arm.
To determine if IIf basin waters have fallen within this
potential density range in the past, we examined CTD
profiles collected since 1980 from the standard Arctic
Station position southeast of Qeqertarsuaq and from
a regularly visited station northwest of Ilulissat. We
extracted s
at 250-m depth and determined the depth
of the s
527.30 kg m
isopycnal from these profiles.
These are plotted in Figs. 5a and 5b. Except for two
outliers, waters at 250-m depth near Ilulissat were in the
range s
527.20 to 27.33 kg m
in all summers since
1980, and the depth of the s
527.30 kg m
varied between 200 and 350 m. Near Qeqertarsuaq,
water densities at 250 m were slightly higher and the
27.30 kg m
isopycnals were slightly shallower.
We next divided the data intotwo groups on either side
of the data gap in the mid-1990s, since the interval from
1980 to 1991 had cooler temperatures compared to the
interval after 1997. Near Qeqertarsuaq, the s
27.30 kg m
isopycnal mean depth descended from 217
to 230 m from the first time interval to the second, and the
potential density anomaly at 250-m depth slightly de-
creased from 27.332 to 27 .326 kg m
. These changes,
however, are not statistically significant. At the Ilulissat
station, the changes were statistically significant. The
mean depth of the s
527.30 kg m
isopycnal descended
from 249 to 294 m, while the mean s
at 250-m depth
decreased by 0.06 kg m
from 27.304 to 27.248 kg m
Overall, however, it is accurate to say that waters poised
to fill the fjord basin have been in the s
527.2 to
27.3 kg m
interval for at least the past three decades.
We point out that the downward migration of isopycnals
near Ilulissat between the 1980s and 2000s would lead to
a slight cooling of waters entering the fjord if the uS
relationship of DB waters were fixed.
In Fig. 5c, we plot all of these profiles in uS
ordinates, along with the definitions of WFjW and CFjW.
It is apparent that the cool waters of the 1980s essentially
overlap with the CFjWseen in 2010. Therefore, the range
of interannual fjord basin variability from 2009 to 2011
was nearly equal to the full range of variability from the
past three decades.
In Table 3 and Fig. 6, we present the mean, observed,
in situ temperature Tof waters inside the fjord, at the
fjord mouth, and from DB beyond the fjord mouth area
for the years 1980 to 2013 (when available). The earliest
known data, from 1879, are also included for comparison
(Hammer 1883).
These are subdivided into waters with s
typical of the
fjord basin (27.20 #s
#27.31 kg m
) and denser waters.
Waters of typical fjord basin density at the fjord mouth
generally agreed to within 0.208CwithDBtemperatures
(75% of years) and to within 0.168Cwithfjordbasintem-
peratures (all but 2010) in that density class. In 2010, the
perhaps because of the still ongoing exchange between DB
and IIf. In general, it is reasonable to use the temperature
of waters in this density class at the fjord mouth (column
4ofTable 3) as a proxy for the temperature of waters
reaching JG.
In Table 3, we identified 15 years as cool (T#1.78C)
and 7 years as warm (T$2.7) based on the temperature
FIG. 3. Sections of summer potential temperature along paths in Fig. 1. Stations are marked
by triangles. White space between neighboring profiles indicates the stations were either oc-
cupied more than 20 days apart or else were occupied at least 2 days apart and separated by at
least 30 km (i.e., not part of the same synoptic survey). The fjord sill is just west of the red star
marking the longitude of Ilulissat. Inner Egedesminde Dyb is just westof the red circle marking
the longitude of Qeqertarsuaq. A typical position of the JG terminus is indicated by the blue
rectangle. The thin black (gray) contour encloses WFjW (CFjW), as defined in Table 2. The
white dashed line indicates 200-m depth. Inside the fjord, bathymetry is from XCTD profiles.
Outside the fjord, bathymetry is from IBCAO V3.
JANUARY 2015 G L A D I S H E T A L . 9
of waters at the fjord mouth in the 27.20 #s
27.31 kg m
range. This categorization captures the
persistence of cool conditions until the warming that is
apparent in 1997 (Hansen et al. 2012;Holland et al. 2008)
and that becomes particularly marked in 1998 and 1999.
From 2000 to 2007 water temperatures at the fjord mouth
remained warmer than the cool 1980s, but 2001 and 2004–
07 were closer to the cool 1980s than to the extremely
warm 1998/99. Summer temperatures from 2008 to 2013
were warmer than ever, except 2010, which was as cool as
in the 1980s throughout DB and IIf.
The 2010 cool anomaly does not appear to be con-
nected with variability of local vertical heat fluxes to the
atmosphere. In Fig. 7, we show surface air temperature
anomalies from Ilulissat, Aasiaat, and our AWS near
JG. These show that the winter and spring preceding the
summer of 2010 were actually unusually warm over the
DB/IIf region with high pressure (not shown) and gen-
erally low wind speeds (not shown). It might be sup-
posed that warmer surface air temperatures in early
2010 led to reduced sea ice cover that allowed greater
heat loss and hence cooling of DB/IIf. However, in late
2010 to early 2011, surface air temperatures were again
high, and there was low sea ice cover in Disko Bay
, but
this accompanied the rewarming of DB/IIf. We con-
clude that the 2010 cool anomaly was carried into DB/IIf
by ocean currents.
b. Annual and subannual variability
The summer surveys imply nothing about annual or
subannual (periods less than 1 yr) variability occurring
in DB/IIf. Here, we characterize such variability using
data from our various moorings, instrumented seals, and
Greenland halibut tags.
FIG. 4. The uS
curves for all profiles marked in Fig. 1. The plotting sequence was ran-
domized so that profiles of different colors do not obscure one another in a biased way. Circles
(triangles) mark water properties at 100-m (300 m) depth for each profile. Light colors (pink,
light blue, and gray) are profiles from DB, while saturated colors (red, dark blue, and black) are
from inside IIf. Background curves are contours of the surface-referenced potential density
anomaly s
. Black contours enclose uS
regions for WFjW and CFjW (as defined in Table 2).
Dashed green lines are Gade (meltwater mixing) lines.
All mooring and seal dive data are displayed in Fig. 8.
Basic statistics for all time series are provided in Table 4.
We next present the time series chronologically. Find-
ings illuminating the general character of variability and
renewal in the fjord are emphasized.
Over a 65-day interval, temperatures for Fjord 400 m
(located along the spillway; see Fig. 2) varied within
a 0.698C range with a mean of 2.898C, remaining essen-
tially inside the WFjW category traced out by CTD
profiles earlier that summer (2009). Temperature and
potential density were well correlated for this mooring,
suggesting that temperatures varied mainly as a result of
vertical motions of isopycnals outside the fjord, causing
layers of varying density to spill over the sill.
After rising to 29-m depth, still inside the fjord but perched
45 m below the sill depth, Fjord 290 m encountered a cooler
mean (2.608C) and a wider range of temperatures (1.578C
variation over 78 days) in early winter 2009/10. Water
properties varied outside the WFjW category but remained
within the uS
region traced out by CTD profiles in the
previous warm summer. The cool 2010 anomaly definitely
did not arrive inside IIf before 21 January 2010.
In early 2010, the mooring (DB 100 m) then spent 70
days at 100-m depth in eastern DB or on the sill, we believe.
Although cooler, the waters sampled were still not clearly
cooler than equally dense waters in the warm summer 2009
CTD survey. The cool 2010 anomaly probably had still not
reached eastern DB by 21 April 2010.
By late May 2010, when the mooring (DB 58 m) was at
just 58-m depth, the mean potential temperature was
1.088C. This finally fell within the interval (0.758to 1.458C)
occupied by equally dense waters in the cool CTD survey
of August 2010. If the cooling of these near-surface waters
was because of the advection rather than winter vertical
heat flux, we may tentatively conclude that the unusually
cool waters directly observed in the fjord in August 2010
arrived at eastern DB in May of 2010, 2–3 months after
unusually cool waters appeared in Davis Strait in March
(Fig. 8c).
A single physical mooring recorded these four time se-
ries fragments, so they necessarily cover nonoverlapping
portions of the year. If we nevertheless compare the power
spectra of Fjord 400 m, Fjord 290 m, and DB 100 m
(Fig. 8d), it appears that on subseasonal periods (i.e., up to
about 30 days) the variability of these time series are all
similar to that of WG1 150 m in the path of the WGC. This
suggests that subseasonal variability in DB above the sill
depth, perhaps originating in the WGC, is transmitted at
least some distance into the fjord by waters leaking over
the spillway, even outside of summer.
Mooring DB 296 m collected just 14 days of data be-
ginning in August 2010. There was 0.68C of warming over
the two weeks, but this does not signify the beginning of
the returnto warm 2011 conditions since the temperature
change was associated with lifting isopycnals in DB rather
than warming on fixed isopycnals (Fig. 8a).
The mostly temporally disjoint series DB 350m, WG1
150 m 2010,andWG1 150 m 2011 were each about 1 yr in
duration and sampled from overlapping density ranges.
The Disko Bay bottom mooring time series DB 350m,
however, was strikingly less variable than the two WG1
series. The total range of temperature variations for DB
350 m was only 0.738C, including a background warming
trend amounting to 0.378C over the year. This is an order
of magnitude less variation than the 48–58Cannualrange
measured by mooring WG1. In fact, at essentially all
frequencies WG1 150m showed much greater tempera-
ture variability than DB 350 m (Fig. 8d). Some of the
temperature variation at WG1 is associated with the wide
range of potential densities (Table 4), but moorings at
other depths in eastern Davis Strait show that a large
annual cycle of temperature occurs in all WGC isopycnic
layers that can connect to deep DB. Gladish et al. (2015)
argues that the 300-m-deep Egedesminde Dyb Sill blocks
TABLE 2. Water mass definitions.
Water mass Definition Comment
WFjW 27.20 #s
#27.31 Observed in summers of 2008, 2009,
and 2011–13
u#3:25 22:75
27:31 227:20 (su227:20) 12:75
u$2:75 22:25
27:31 227:20 (su227:20) 12:25
CFjW 27.20 #s
#27.31 Observed in summer 2010
27:31 227:20 (su227:20) 11:5
27:31 227:20 (su227:20) 10:5
JANUARY 2015 G L A D I S H E T A L . 11
the warmest WGC waters because those isopycnic layers
migrate below the depthof this outer sill duringthe warm
phase of the annual cycle.
Virtual mooring Seals 300 m 2013 shows that IIf
temperatures at 300 m varied over just a 0.368C interval
from September 2012 until January 2013. Most of this
temperature change was because of the vertical motion
of isopycnals in the fjord, which descended until De-
cember and then ascended (Figs. 8a,c). From 2-day pe-
riods up to 30-day periods, temperature variability at
300 m in the fjord interior was one-tenth the magnitude of
variability in similarly dense waters in the WGC or along
the sill spillway (Fig. 8d). Even though subseasonal var-
iability in DB appears to leak over the sill at least onto the
FIG. 5. (a) Potential density anomaly s
at 250-m depth from CTD profiles at the standard
Arctic Station position near Qeqertarsuaq and near a regular station close to Ilulissat. The
mean values for data up to 1991 and for data after 1998 are plotted separately as solid horizontal
lines. Only summer [June–August (JJA)] data are shown. (b) Depth of the s
527.3 kg m
isopycnal extracted from the same dataset. (c) The uS
curves for all CTD profiles near the
standard Arctic Station position and positions near Ilulissat from 1980 to 2013. WFjW and
regions are shown.
spillway, these incursions appear to have insufficient
volume to affect the fjord basin interior to the east, such
that the fjord basin is quiet on subseasonal time scales
outside of summer.
Virtual mooring Seals 100 m 2013 shows that, at
100-m depth, fjord waters became denser and warmer
from September 2012 until March 2013 as low-salinity
waters in the upper layer were replaced by denser and
warmer waters more characteristic of surface waters
outside the fjord (Figs. 8a,c). At this depth in the fjord,
variability is much closer to WGC or shallow DB
variability on subseasonal time scales (except for the
7- to 15-day band of periods in 2014, and 15- to 30-day
band in 2013, for unknown reasons), indicating that
exchange between the external waters and the above-
sill layer of IIf probably occurs regularly on sub-
seasonal time scales.
During the next overwinter period from autumn 2013
to spring 2014, variability for Seals 100 m 2014 and Seals
300 m 2014 was similar to the previous year, with basin
temperature and density decreasing until December,
followed by rising isopycnals and rewarming. Additional
spring warming was detected in the longer record from
2014 (compared to 2013) at both 100- and 300-m depths.
Sporadic seal dives down to 500-m depth (Seals 500 m
2014) show that temperatures deeper in the basin were
also slowly warming over a 0.638C range from December
2013 to March 2014, with s
increasing from 27.06 to
27.18 kg m
over the same time interval. These changes
occurred on a slower-than-seasonal time scale, so the
fjord basin can still be described as quiet on subseasonal
to seasonal time scales (Fig. 8d). We propose that the lack
of subseasonal variability in the fjord basin was because
of the weak exchange between DB and the fjord basin
TABLE 3. Mean in situ temperature for waters in three nonoverlapping geographical areas (fjord interior, fjord mouth, and DB) and two
different density categories in summer (JJA). Temperatures from fjord mouth waters in the typical basin density range, the fourth column,
make up the basin proxy record. Temperatures at the fjord mouth cooler than or equal to 1.78C are shown in italics and those warmer than
or equal to 2.78C are shown in bold.
Year IIf typical s
IIf s
.27.31 Fjord mouth typical s
Fjord mouth s
.27.31 DB typical s
DB s
1879 — 0.6
1980 — 0.40 0.38 0.74 1.21
1981 — 0.76 0.75 0.09
1982 — 1.46 1.75 1.54 1.78
1983 — 1.34 1.63 1.36 1.72
1984 — 1.40 1.77 1.33 1.76
1985 — 0.94 1.29 0.93 1.60
1986 — 1.23 1.67 1.49 1.94
1987 — 1.32 1.56 0.87 1.70
1988 — 0.96 1.28 0.95 1.52
1989 — 1.50 1.90 1.60 2.17
1990 — 1.45 1.69 1.27 1.88
1991–93 —
1994 — 1.08 1.32 1.13 1.52
1995–96 —
1997 — 2.23 2.29 3.11
1998 — 2.77 3.02 2.70 3.28
1999 — 2.79 3.26 2.74 3.44
2000 — 2.37 2.71 2.18 3.13
2001 2.01
2.07 2.19 2.58
2002 2.39
2.60 2.89 2.66 3.11
2003 2.18
2.47 3.11 2.41 2.97
2004 — 1.77 2.60 1.67 2.47
2005 — 2.11 3.09 1.85 2.69
2006 — 1.61 2.49 1.68 2.59
2007 1.83 1.97 2.81 2.18 3.14
2008 — 2.94 3.32 3.12 3.62
2009 2.85 2.95 2.77 3.28 2.70 3.41
2010 1.67 2.01 1.44 1.90 1.05 2.35
2011 2.68 2.84 3.20 3.07 3.56
2012 2.92 3.12 3.08 3.27 2.89 3.66
2013 2.64 2.77 3.10 2.73 3.35
Mean value between 150–250-m depth.
Mean value below 250-m depth.
Mean temperature below 300-m depth, from winter beginning that calendar year.
JANUARY 2015 G L A D I S H E T A L . 13
outside of the summer glacial melt season. IIf basin
temperatures still tracked DB temperatures interann-
ually because the basin was renewed each summer.
The instrumented Greenland halibut datasets from
2001 to 2004 do not contain salinity and were therefore
not incorporated into the above analysis. In Fig. 9,we
show time series of daily averaged temperatures from
all measurements below 300-m depth inside IIf. The
fish were always initially captured in autumn and re-
captured several months later after having swum in IIf
most of the winter. The mean temperatures from each
overwinter time interval were entered in Table 3 next
to DB measurements from the previous summer, since
the winter fjord temperatures correlate with fjord
mouth temperatures from the series of preceding
summers (the winter temperatures being somewhat
cooler), but not with the series of following summers.
This suggests that the interannual signal is carried into
the fjord when the fjord is renewed in summer and is
still not overwhelmed by higher-frequency variability
several months later. For each overwinter period, the
halibut temperature time series ranged over a ,0.48C
interval (ignoring 6 outlier days), which is consistent
with there being weak subseasonal to seasonal vari-
ability in the fjord basin outside of summer.
4. Relation of fjord variability to JG behavior
We found that summer-to-summer temperature varia-
tions in the IIf basin can be nearly as large (about 1.08C) as
the entire range of temperature variability (1.58C) of DB in
the past 100 yr, as determined from foraminiferal proxies
(Lloyd et al. 2011). Fully investigating the effect of this
variability on JG is beyond the scope of this work. Here,
we give only qualitative remarks on the relation between
IIf basin temperature variability and the behavior of JG.
As justified above, we refer to the temperature of fjord
mouth waters in the 27.20 #s
#27.31 kg m
as our
basin proxy in years lacking direct measurements in the
fjord. On 25 August 1997 the basin proxy was 2.238C, the
warmest going back to at least 1980 (Table 3). Our basin
proxy record is in precise agreement with Motyka et al.
(2011), who estimated that a 1.18C warming of the deep
fjord water in 1997, relative to the 1980s average, increased
melt rates under the then present ice shelf of JG by 25%
and initiated the retreat. Hansen et al. (2012) report even
warmer water (above 3.08C) at 200-m depth in DB near
Qeqertarsuaq in summer 1997, but their Fig. 8 shows that
the incursion may have been brief, with waters closer to
28C probably filling DB and IIf by the end of the summer.
In summer 1998, the basin proxy reached 2.778C. It was in
spring of 1998 that the glacier speed underwent a step
change increase in velocity along with a significant retreat
of the ice shelf (Luckman and Murray 2005).
After 1999, however, glacial variability does not corre-
late clearly with basin temperature variability. For instance,
the near-terminus glacier speed in summer of 2000 was
25% greater than in 1998 (Luckman and Murray 2005), yet
the basin proxy was 0.48C cooler that summer compared to
1998 (Table 3). On the other hand, a slight readvance and
deceleration of JG in 2001 (Motyka et al. 2011;Joughin
et al. 2008) coincided with further cooling to 2.078C. Gla-
cier speeds accelerated from 2000 until 2007 (Joughin et al.
FIG. 6. Temperatures from Table 3. Fjord mouth temperatures in
the typical s
range make up our basin proxy record.
FIG. 7. Monthly-mean surface air temperature anomalies (rela-
tive to 2007–12 monthly means) from DMI stations in (a) Ilulissat,
(b) Aasiaat, and the (c) NYU automated weather station next to
JG (Fig. 2). Anomalies greater (less) than one standard deviation
are colored red (blue).
2008), although the proxy record implies that fjord basin
temperatures were likely never as warm in those years as
they were in 1998 and 1999. The generally high fjord basin
temperatures from 2008 to 2013 coincide with a period in
which JG continued to retreat and accelerate (Joughin
et al. 2012b). We suggest that after 1999 the glacier became
less sensitive to basin temperature changes, perhaps be-
cause of the loss of its ice shelf or because it was still re-
sponding to the sudden perturbation of 1997–99.
The only apparent effect of the cooler waters that arrived
in the fjord basin by the summer of 2010 was an apparently
more abrupt shutdown in autumn 2010 of the usual en-
hanced summer glacier velocities and a delayed return of
fast flow the following summer (Joughin et al. 2014). As our
overwinter halibut data show for prior years, the fjord basin
must have remained cool through early 2011. This may
have led to reduced undercutting of the terminus, a re-
duced tendency for calving to occur, and therefore slower
glacier flow (Joughin et al. 2008). If the seasonal modula-
tion of calving is controlled by the strength of the proglacial
mélange (Amundson et al. 2010), then it seems unlikely
that the cool basin temperatures were involved, unless re-
duced melting of the largest icebergs in the mélange (those
that penetrate into the basin) could make the mélange
stronger than usual. The possible controlling role of fjord
surface waters on the glacier, via inuences on the ice
mélange, is beyond the scope of this work. The lack of clear
correlation between basin waters and glacier behavior since
1999 does suggest looking to surface waters or elsewhere
for the main controls on JG behavior in recent years.
5. Fjord renewal
Implicit in our discussion of fjord variability and its pos-
sible influence on JG is the premise that the temperature
FIG. 8. (a ) Daily averaged uS
properties for mooringsdeployed
in the fjord and in DB as well as virtual moorings assembled from
properties extractedat 100- and 300-mdepth from instrumentedseal
dives. Potential density and WFjW/CFjW definitions are shown
using black contours. (b) Dates randomly selected from each of
these time series are plotted on a timeline. The date of the 2010
summer survey is marked by a blue star. (c) The 5-day-averaged
temperature time series for moorings and seal dives. No individual
time series were longer than a year, so data are plotted without
ambiguity using day-of-year only. (d) Power spectra for the time
series longer than 2 months. Pure sinusoids of identical frequency
and amplitude, but different duration or sampling frequency would
appear here as nearly identical spikes.
TABLE 4. Mooring time series statistics: mean, standard de-
viation, and total range for potential temperature u(8C) and po-
tential density anomaly s
(kg m
). Subtidal refers to the fraction
of the variance of uin frequencies f,0.9 day
(u)DuSubtidal su
Fjord 400 m 2.89 0.15 0.69 0.77 27.28 0.03 0.14
Fjord 290 m 2.60 0.29 1.57 0.65 27.18 0.07 0.41
DB 100 m 1.64 0.16 1.73 0.62 27.03 0.04 0.26
DB 58 m 1.07 0.22 1.63 0.18 27.07 0.05 0.31
DB 296 m 2.03 0.32 1.24 27.40 0.02 0.10
DB 350 m 3.40 0.14 0.73 0.93 27.32 0.03 0.11
Seals 100 m 2013 1.06 0.48 4.22 26.61 0.32 1.48
Seals 300 m 2013 1.94 0.08 0.36 27.09 0.02 0.14
Seals 100 m 2014 0.90 0.59 3.73 26.72 0.18 0.82
Seals 300 m 2014 1.92 0.27 1.31 27.07 0.06 0.29
Seals 500 m 2014 2.14 0.13 0.63 27.09 0.03 0.13
WG1 150 m 2010 2.67 1.23 5.12 0.93 27.24 0.12 1.20
WG1 150 m 2011 4.20 1.14 4.51 0.89 27.15 0.15 1.82
JANUARY 2015 G L A D I S H E T A L . 15
profile T
(z) of the fjord basin interior is imposed as an
ambient boundary condition on the outer edge of a turbu-
lent plume rising next to the glacier terminus. The dynamics
of the plume then determine the melt rate (Jenkins 2011;
Xu et al. 2013). In this view, it is understood that the plume
temperature T(z) varies, depending on the melt rate and
entrainment rate. Yet, it is tempting to assume that T
determined by external sources only, which is to say that
glacier melting does not cool the ambient basin. Indeed, if
there were no detrainment of plume water and no re-
circulation of the plume outflow back down into the basin,
then T
would be unaffected by glacier melting. Alterna-
tively, T
might be unaffected by mixing with detrained or
outflowing plume waters if the fjord basin were an effec-
tively infinite heat reservoir.
The real basin is not effectively infinite, however, as
a simple calculation shows. Suppose the terminus of JG
melts at 4 m day
, an upper bound for marine glaciers
just north of JG (Rignot et al. 2010). Then 1.4 km
ice melts because of contact with the lateral face of the
50 km 38km 3500 m basin layer during a 90-day sum-
mer. The latent heat required to melt 1.4km
of ice is
4.5 310
J. The heat capacity of the 200 km
basin layer
is 8.2 310
. If a dam were installed across the fjord
mouth at the beginning of summer, by the end of summer
the fjord basin would be 0.58C cooler, which is compa-
rable to the observed interannual variability. By contrast,
the 27.20 #s
#27.31 kg m
layer in DB has a volume
of approximately 100 km 3100 km 3100 m 51000 km
and extracting 4.5 310
J from this layer would cool it by
only 0.18C. The DB layer feeding the IIf basin is effec-
tively an infinite heat reservoir for JG, while the IIf basin
itself is not.
If the renewal of the IIf basin were rapid (i.e., on av-
erage, basin water parcels leave before cycling more than
once through the plume), then the ambient basin tem-
perature T
would be close to the fjord mouth proxy
temperature. Melting at the terminus would then draw
from the inexhaustible heat reservoir of DB, limited only
by the heat flux across the plume. If the renewal were slow
(i.e., on average, water parcels cycle repeatedly through
the plume before being exchanged with DB), T
become significantly cooler than DB. In fact, the fjord
waters are cooler than DB waters at equal depths. Above
about 100-m depth, we attribute this to iceberg melting,
which modifies the upper fjord water faster than it can be
renewed by warmer DB waters (Fig. 4). If basin waters
were cooled by interaction with the glacier terminus or
deep-drafted icebergs, their uS
properties would lie
along a meltwater mixing line, which they nearly do, be-
low about 300-m depth (Fig. 4). However, DB waters with
$27.1 kg m
also lie along the meltwater mixing line.
It is also possible, therefore, that the IIf basin contained
essentially unmodified DB water at the time of the sur-
veys and that the depressed isopycnals/isotherms simply
reflected the sill being a control point for DB waters
pouring into the basin. We will argue from a single sparse
velocity survey in the fjord that the latter picture is cor-
rect, implying that rapid renewal of the IIf basin, in
summer, exposes JG to the effectively infinite heat res-
ervoir of DB (and justifying our proposal of the proxy
time series).
a. Fjord circulation from XCP data
Velocity profiles V1, V2, and V3 (Fig. 2) revealed
a baroclinic flow with waters below approximately 300 m
flowing toward the ice at 0.05–0.15 m s
and water
above this flowing seaward (Fig. 10). Profile V4 near the
glacier terminus had a somewhat different character,
with weak currents down to 600-m depth and a flow of
speed 0.15 m s
toward the terminus of the main trunk
of the glacier (i.e., toward the south east) in the deepest
100 m of the profile.
To interpret these velocity profiles we must consider
their degree of contamination by tidal signals. A quanti-
tative analysis in appendix A shows that the barotropic
tide was much smaller than the XCP signal at station V1
and was at the level of noise for the other stations to the
east. We also show that first baroclinic mode internal tidal
velocities in the fjord do not dominate the velocity pro-
files. We conclude that the XCP profiles containthe signal
of a quasi-steady overturning circulation in the fjord.
In Table 5, we record zonal volume and salt fluxes
determined from each XCP station. Volume fluxes are
vertical integrals of zonal velocity u, and salt fluxes are
vertical integrals of S
ru. Salinity S
was taken from the
nearest XCTD profile, using extrapolation if necessary
to cover the depth range for the XCP station. We found
FIG. 9. Daily averaged in situ temperatures below 300-m depth
inside IIf from instrumented Greenland halibut.
FIG. 10. (a) Profiles of zonal and meridional velocity in summer 2013 at station V1 (Fig. 2).
Error bars are estimates for individual samples as determined by the instrument. The fjord sill
depth is shown as a dashed line. (b) Temperature profile (red line) from the XCTD station
nearest to V1 and temperatures from the XCP probe at station V1 (thin blue line, shown with
a 0.58C offset for clarity). (c)–(h) As in (a) and (b), but for stations V2,V3, and V4.
JANUARY 2015 G L A D I S H E T A L . 17
an unbalanced volume flux in all the XCP profiles (es-
pecially V2, naturally, which omitted the upper water
column). If, for instance, the velocity profile at station
V3 was representative of the 8-km width of the fjord, the
net volume flux into the fjord would be 14 km
This would double the volume of water in the fjord in
less than a month. For comparison, a continuously rising
spring tide would double the fjord volume in 6 months.
For station V3, the missing outflow could be in the up-
permost 11 m that was not sampled. However, the re-
quired speed would be 2 60.1 m s
, which seems
exceptionally large. At each station there was also a net
salt flux into the fjord. We calculated the salinity of the
missing outflow needed to eliminate the net salt flux at
each station (Table 5), but the error range was too large
to confirm that the missing outflow was from the surface.
The first inner Rossby radius in the fjord varied from 9 to
12 km depending on the location and year, which is just
slightly larger than the fjord width. It is therefore pos-
sible that across-fjord variations account for some of the
missing volume outflux (Cushman-Roisin et al. 1994).
Given the lack of closure in the volume fluxes, we
could not estimate fluxes of meltwater or runoff into the
fjord. However, for stations V2 and V3 the fluxes and
zonal velocity toward the glacier were nearly equal be-
low the sill depth. Therefore, we suppose the deep zonal
circulation was essentially uniform along the fjord, at
least in the central basin. Given that a vertical planar
slice along the fjord has area 5 310
and averaging
the fluxes at V2 and V3, a volume equivalent to the
whole fjord flows past the central fjord basin every 24 6
0.5 days. If water parcels did not recirculate then this
would be the residence time of typical fjord waters.
From Fig. 10a it appears that perhaps one-third of the
westward-flowing water recirculated below the sill
depth. In this case, water parcels spent an average of 36
days in the fjord. We conclude that the fjord basin was
on course to be renewed in about 1 month at the time of
the XCP survey in June 2013.
b. Discussion of renewal in silled fjords
For the remainder of this paper, we are concerned
with how the fjord circulation glimpsed by the XCP
survey could be driven. The challenge is to explain how
all the water in the IIf basin, essentially a huge bathtub
with no drain, could be exchanged over the sill in less
than a year (or, indeed, in as little as 1 month). This
requires explanation since basin waters in midlatitude
silled fjords are typically renewed episodically only
when dense waters appear at the fjord sill to uplift wa-
ters in the basin that, meanwhile, have gradually become
less dense because of vertical buoyancy fluxes (Inall and
Gillibrand 2010;Stigebrandt 2012). When denser water
becomes available at the sill, basin renewal in some lo-
cations can occur quickly. For instance, at Gullmar
Fjord, renewal by inflowing dense water occurs on a time
scale of days to weeks (Arneborg et al. 2004). However,
the interval between renewal episodes (for instance, in
certain Scottish lochs) could be much longer than 1 yr
(Gade and Edwards 1980).
Mortensen et al. (2011) studied Godthåbsfjord, south
of JG along the coast of Greenland, and described four
modes of circulation that can transport heat to the head
of the fjord and also bring about fjord–shelf exchange.
First is the estuarine mode of circulation present in all
fjords in which freshwater runoff enters the fjord at the
surface. As this freshwater flows out of the fjord, en-
trainment along the underside of this layer increases
the volume export and hence induces a compensating
inflow just beneath. In IIf, there is terrestrial runoff, but
any estuarine circulation must occur above the very
sharp pycnocline at 100-m depth and is therefore not
involved in renewal of the basin. Second is the flow
driven by subglacial discharge. At Godthåbsfjord,
subglacial discharge is released into the fjord at a rela-
tively shallow depth and an inner sill restricts the depths
at which basin waters ow toward the glacier to re-
plenish the water entrained by the subglacially driven
plume. At IIf, the fjord basin is essentially at and this
mode, as we will argue, is the main circulation mode.
Third, at Godthåbsfjord in winter there are episodic
dense water inow events similar to those observed at
1 to 3 months. During these events, isopycnals in the
fjord rise as existing waters are lifted by the intrusion of
denser waters. At IIf, our DB bottom mooring recorded
TABLE 5. Volume and salt fluxes from XCP and XCTD profiles:z
5depth (m) of first measurement; F
5glacierward volume fluxes
per unit width (m
); F
5glacierward salt fluxes per unit width (10
kg m
); F(in) 5integrated over all depths with positive zonal
velocity; F(out) 5integrated over depths with negative zonal velocity; F(net) 5integrated over all depths; F(deep) 5integrated over all
depths greater than 250 m; and S
(out) 5salinity required in missing outflow to balance salt flux (g kg
Station z
(in) F
(out) F
(net) F
(deep) F
(net) S
V1 5 43.4 60.6 228.3 60.5 15.2 60.8 38.7 60.6 5.5 60.3 35.3 62.4
V2 172 24.6 60.3 24.3 60.3 —
V3 11 25.4 60.8 25.6 61.0 19.8 61.1 22.8 60.3 7.0 60.4 34.4 62.8
V4 13 28.5 61.5 23.3 60.4 25.4 61.5 17.8 60.5 8.9 60.8 34.0 63.6
no episodes of signicantly enhanced density (Fig. 4),
and the isopycnals in the fjord (down to 500 m) appear
to steadily fall until January. Rather than being epi-
sodic, the dense inflow into the IIf basin (associated
with rising isopycnals) from January until spring ap-
pears to be slow and continuous. The fourth mode is
termed the intermediary circulation, which is essen-
tially any flow driven by horizontal density gradients
at intermediate depths at the fjord mouth. Such gra-
dients may be because of the tilting of existing iso-
pycnals (Klinck et al. 1981;Arneborg 2004)or
because of the formation of new density classes at
intermediate depths. At Godthåbsfjord, tidally in-
duced mixing creates a warm water mass in summer
that is transported into the fjord because of horizontal
density gradients.
The intermediary circulation mode rapidly flushes
some midlatitude fjords (Klinck et al. 1981;Arneborg
2004;Stigebrandt 2012), and it has been shown that
alongshore wind variability drives an intermediary
circulation that dominates the velocity field of Sermilik
Fjord. This circulation causes the waters of that glacial
fjord to exchange with external waters on a subseasonal
time scale and hence creates large subseasonal vari-
ability in the fjord (Straneo et al. 2010;Jackson et al.
2014) We point out, however, that Sermilik Fjord does
not have a high sill like IIf. Stigebrandt (2012) states
that intermediary circulation does not normally in-
fluence basin water.
In a series of two-layer numerical simulations,
Klinck et al. (1981) found that intermediary circulation
created large deep-layer velocities regardless of the sill
height above the seafloor, suggesting that an in-
termediary circulation could be responsible for re-
newing the IIf basin. It is physically intuitive, however,
that in a sufficiently stratified two-layer system with
a sufficiently high sill, deep-layer currents will not be
generated on one side of the sill by interface oscilla-
tions on the other side of the sill. For instance, the sill
could separate the deep layer into disconnected vol-
umes. In appendix B, we show that the indifference of
the circulation to tall sills in the model of Klinck et al.
(1981) is possibly spurious.
In the seal dive data, there was little subseasonal
variability in IIf basin temperatures. A Sermilik-like
intermediary circulation is therefore probably absent in
the IIf basin (but is probably important above the sill).
Also, in our model simulations described below, pyc-
nocline oscillations outside of the mouth of an idealized
IIf did not drive renewal of the basin waters on an an-
nual or subannual time scale. The annual or subannual
renewal of the IIf basin must be driven by another
The glacier terminus at the head of IIf is a buoyancy
source because of the melting at the ice face and the dis-
charge of subglacial meltwater. Subglacial discharge can
significantly enhance melting at glacier termini (Motyka
et al. 2003;Xu et al. 2013;Straneo et al. 2011;Jenkins
2011). Here, we consider the possibility that subglacial
discharge also drives the renewal of IIf basin waters.
Unlike during an observed renewal episode at
Gullmar Fjord (Arneborg et al. 2004), we never ob-
served a steplike jump in the density profile for near-
bottom waters inside the IIf basin. If exchange was
actually occurring in 2013 during the XCP/XCTD sur-
veys, then the inflowing waters were not lifting the old
density as in the Gullmar Fjord renewal episode.
Rather, the inflowing waters were at most 0.1 kg m
denser than overlying basin waters with no sharp in-
terface (Fig. 4). Furthermore, the basin water of 2011
less dense than in 2010, which suggests
that the waters of 2010 had first to be modified before
they were flushed. We propose that, at the time of the
XCP/XCTD surveys, the plume next to the glacier was
entraining massive amounts of basin waters and eject-
ing the resulting mixture up and out of the fjord.
To estimate the amount of glacial meltwater required
to renew the fjord basin in this way, we make four as-
sumptions: 1) All fjord water below the outflow depth z
must be modified so that its density equals the density
at the outflow depth. 2) This modification is entirely
due to mixing with freshwater, not vertical fluxes. 3) The
outflow density is not very different from the density of
basin waters, so that the ratio of freshwater to seawater
is small in the outgoing water. 4) We assume a linear
thermodynamic equation of state for seawater. Then,
integrating over all water parcels below the outflow
depth, the volume Vof freshwater required to flush the
fjord is
dV , (1)
where r
is the density of the incoming freshwater.
Analysis of water properties from the 2009 survey
indicates that waters at 150-m depth were especially
JANUARY 2015 G L A D I S H E T A L . 19
influenced by mixing with freshwater (Straneo et al.
2012), so we assume z
52150 m. Using a fjord width of
8 km, mean depth of 750 m, and length of 47 km (from the
foot of the spillway to the terminus), the 2009 survey
implies V54.3 km
and the 2010 survey implies V5
4.9 km
. The dependence on z
is strong since the outflow
density changes rapidly with depth. With z
52200 m,
only about 1 km
of freshwater is required. Outflow at
5210 m would require more than 30 km
Echelmeyer and Harrison (1990) estimated that,
over the drainage basin for JG, 650–1350 m
surface meltwater was produced and entered the gla-
cier upstream of the grounding line from late June to
mid-August (observations from July/August 1984–88).
Assuming a 90-day melt season, this implies that 5–
10 km
of freshwater potentially crossed the grounding
line at JG each summer. More recently, Mernild et al.
(2010) used a model constrained by weather station
observations and estimated the total amount of glacial
runoff from the JG drainage basin to be 1.81–5.21 km
for 2000–07 (mean of 3.4 61.1 km
), distributed over
roughly a 3-month melt season each year. Given the
complete lack of observations of subglacial discharge
along the grounding line at JG, we will take 2–10 km
a range of possible values.
We next report estimates of the freshwater flux from
terminus melting. Jenkins (2011) used a plume model to
obtain a summer melt rate of 3.2 m day
near the base of
JG, assuming a planar subglacial discharge of 0.17 m
[equivalent to the 1350 m
estimate from Echelmeyer
and Harrison (1990) distributed uniformly over a 8-km-
long grounding line]. This estimate assumed a thermal
driving of 4.378C based on rather cool temperatures in
the north arm of IIf from 2007. From 2009 to 2013 (ex-
cluding 2010), the thermal driving at JG below 700-m
depth was actually 5.28–5.68C. Using observed water
column properties from 2009, we determined that the
mean melt rate from Jenkins’ plume model is 3.9 m day
over the deepest 630 m, using the same subglacial dis-
charge rate. Rignot et al. (2010) estimated mean melt
rates ranging from 0.7 60.02 m day
at Eqip Sermia
to 3.9 60.8 m day
at Sermeq Kujatdleq and Sermeq
Avangnardleq in August 2008. At these nearby west
Greenland glaciers they were better able to capture the
heat, salt, and volume fluxes toward the glacier than we
were at IIf because of the heavier ice conditions at IIf. The
thermal driving next to Sermeq Kujatdleq/Avangnardleq
was 3.88C. From these results, a reasonable estimate of the
summer melt rate at JG terminus is 4 m day
. Assuming
this rate holds over the deepest 600 m and over a width of
8 km for 90 days, an additional 2 km
of liquid freshwater
enters IIf during the summer.
The total amount of liquid freshwater entering IIf at
JG terminus in the summer is therefore 4–12 km
, which
is 1 to 3 times the amount required to flush the fjord
basin by the freshening mechanism described above
(V54–5 km
). This is consistent with the roughly
monthly renewal time scale derived from XCP data.
6. Model simulations of fjord circulation
We have provided evidence that the IIf basin is
renewed at least once each summer by a circulation
driven mainly by subglacial freshwater discharged into
the fjord basin. We also suggested that intermediary
circulation is probably weak in the IIf basin and not
responsible for renewal. To test these hypotheses, we
carried out two-dimensional, idealized simulations of
the fjord using the MITgcm ocean model (Marshall et al.
a. MITgcm fjord model setup
Our model setup was similar to those in Xu et al.
(2012) and Sciascia et al. (2013). New elements in our
model setup were our introduction of idealized bathymetry
to represent the sill at IIf, our introduction of external
baroclinic forcing, longer-duration simulations, and our
emphasis on the question of renewal.
The model domain was designed to capture the gross
bathymetric features of DB/IIf, including a 45-km-wide
fjord basin, a glacier terminus making up its right ver-
tical boundary, a sill at the mouth, and an open bound-
ary connecting to an idealized DB on the left. In the
most realistic model run, there was a 350-m-deep DB,
250-m-deep sill, and an 800-m-deep fjord basin (Fig. 11).
In the terminology of Fig. 11, this corresponds to a DB
topographic height H
5450 m and a sill topo-
graphic height H
5550 m. We also varied H
in some cases to investigate a hypothetical deep-
silled geometry.
IIf is narrower than the first Rossby radius so we
assumed that cross-fjord gradients are not important
for the along-fjord circulation. We therefore config-
ured the model to run in a vertical plane and neglected
the Coriolis acceleration. Nonhydrostatic momentum
equations were used along with a linearized free surface.
The fully nonlinear equation of state for seawater was
used. No attempt was made to include the mechanical or
thermodynamical influence of icebergs.
The model used eighty 10-m vertical levels. The need
to resolve vertical gradients at all depths (including over
the sill) and our omission of surface forcing led to our
choice of uniform vertical resolution. Horizontal grid
size varied gradually from 500 m outside the fjord to
100 m inside the fjord to 10 m in the final 500 m next to
the model glacier (Fig. 11). Our finest grid spacing is
identical to the near-glacier grid size of Sciascia et al.
(2013). The nominal width of our grid in the third (un-
modelled) dimension was 1 m.
The horizontal eddy viscosity and diffusivity were set to
0.25 m
, in accordance with the discussion in Sciascia
et al. (2013) in the context of Sermilik Fjord. Vertical
mixing made use of the KPP scheme (Large et al. 1994)
with conventional parameters and background vertical
diffusivities and viscosity of 1 310
The time step for all runs with subglacial discharge
was 2 s, while other cases used a 5-s time step. This
choice was constrained by the high vertical velocities of
the plume (exceeding 1 m s
) in cases with subglacial
discharge. The model fluid was initially at rest and all
water columns were initialized using a summer 2009
CTD profile from near the mouth of IIf. Outside the
sill, the CTD profile was extended in depth as neces-
sary by repeating the deepest observation. Inside the
basin the initial temperature and salinity were set
equal to the values at the sill depth. All cases were run
for 100 days.
We applied no surface wind stress or surface fluxes of
heat or freshwater. Sea ice was not included. Along the
left boundary, which represents the connection to far-field
Disko Bay, Orlanski open boundary conditions (Orlanski
1976) were applied in the momentum equations, which
allowed fluid to leave the domain. Temperature and sa-
linity were prescribed on the left boundary using the same
summer 2009 CTD profile used to initialize the model.
Along solid boundaries, friction was parameterized using
a quadratic drag law with nondimensional drag coefficient
52.5 310
At the beginning of each run two passive tracers were
initialized. Both tracers were initially zero left (outside)
of the fjord mouth. Inside the fjord, the basin tracer was
initially equal to 1 below the sill depth and 0 above, while
the above-sill tracer was initially equal to 1 above the sill
depth and 0 below. The tracers were allowed in principle
to leave the domain because of the Orlanski open
boundary, while tracer values on the boundary were set to
zero so that any fluid entering the domain was tracer free.
The purpose of the tracers was to determine the fraction
of fjord water exchanged in response to different forcing
and bathymetry.
The right boundary of the domain was a vertical wall
representing the glacier terminus. In some model runs,
ice–ocean thermodynamical interactions (i.e., melting)
were activated along this wall using the MITgcm module
described in Xu et al. (2012). In certain runs, subglacial
discharge was added by injecting freshwater into the
deepest and rightmost fluid cell in the domain (Fig. 11).
Unlike Xu et al. (2012) or Sciascia et al. (2013), we did not
account for the shape or distribution of subglacial chan-
nels but rather assumed that discharge is uniform along
the grounding line. Subglacial discharge added to the
model was assumed to enter the domain with no kinetic
energy and with temperature and salinity of zero. The
true amount of summer subglacial discharge into IIf is
highly uncertain. Given the 2–10 km
range of estimates
described above, we took 5 km
as a central estimate and
10 km
as an upper estimate. Assuming summer sub-
glacial discharge is distributed uniformly along the
grounding line of JG (of nominal width L58km) over
a 90-day period, we therefore used a planar subglacial
discharge rate F50.08 m
or F50.16 m
in our
two-dimensional model. In the model, the area flux Fwas
truly added to the deepest and rightmost 10 m 310 m
model cell, such that the net flux out of this square was F.
FIG. 11. MITgcm model domain. Pink shading indicates the zone
of 500-m horizontal grid size, yellow shading indicates 100-m grid
size, and the black stripe is the zone of 10-m grid size. Light gray
lines show every tenth grid cell boundary in both directions. The
quantities H
and H
from Table 6 are defined here.
JANUARY 2015 G L A D I S H E T A L . 21
An area flux Fwas therefore also removed evenly along
the open (left) boundary.
To induce an intermediary circulation, we forced the
pycnocline to oscillate at the left boundary in some
model runs. In these cases, we altered the boundary
values of temperature and salinity to vertically shift the
interface between the cool and low-salinity Polar Wa-
ter (PW) layer and the warmer and more saline At-
lantic Water (AW) layer (Fig. 12). Over a period of 10
days, the interface was shifted downward by 200 m and
brought back to its original position. Forced pycnocline
oscillations were meant to induce an intermediary cir-
culation as observed in Sermilik Fjord. Our 10-day
period is at the upper end of periods observed there
(Straneo et al. 2010;Jackson et al. 2014). The 200-m
amplitude is extreme, but, because of interaction with
the Orlanski boundary condition, was necessary to in-
duce the pycnocline to oscillate a more reasonable
50 m a few kilometers east of the open boundary. The
observations necessary to determine whether similar
baroclinic forcing exists outside IIf have not been
made. Here, we investigated what role such forcing
could play in renewing IIf, if it exists.
b. Model runs
Table 6 describes our set of model runs. Control ex-
periments (labeled C) were run without subglacial dis-
charge or pycnocline oscillations, but with melting
activated. Cases with subglacial discharge used either the
equivalent of 5 km
(S5) or 10 km
(S10) total summer
discharge. Cases with forced pycnocline oscillations are
denoted P. The shape of the bathymetry was varied as
well. The designation H indicates a high sill, similar to the
real IIf sill,while L indicates a low sill, submerged deep in
the lower layer like at Sermilik Fjord. A final m indicates
that a case had glacier melting activated. Otherwise, the
glacier wall was treated as a regular solid boundary,
though possibly with subglacial discharge. By running
without melting we could determine if the additional
buoyancy supplied by melting was important compared
to that from subglacial discharge.
c. Model results
Snapshots from five model runs are shown in Fig. 13.
Model outputs are from day 50 or from the first episode of
peak surface layer inflow following day 50 in the forced
pycnocline runs. Contours of the instantaneous stream-
function care shown for each case, where c50byde-
nition along the fjord basin bottom and glacier terminus
wall. Throughout the domain, the horizontal velocity uand
vertical velocity ware given by (u,w)5(2›c/z,›c/x),
so the difference in cyields the area flux (m
) between
any two streamlines.
In case C_Hm (Fig. 13a), melting was permitted, but
there was no subglacial discharge or external baroclinic
forcing to initiate motion near the glacier wall. Mathe-
matically, no buoyant plume will develop in an initial
state of rest that is horizontally homogeneous, since the
melt parameterization does not represent the molecular
heat transfer that would initiate convection in this situ-
ation (Xu et al. 2012). In case C_Hm, however, an up-
welling meltwater plume did develop next to the glacier
within the first 6 model hours in response to a tiny
minimum value imposed on the rate of heat transfer into
the ice.
The meltwater plume eventually reached
a maximum vertical speed of 7 cm s
and a maximum
vertical area flux of 2.5 m
but did not rise above the
sill depth. Most of the plume outflow recirculated back
into the fjord basin.
In case S5_Hm (Fig. 13b), which had a realistic sill,
active glacier melting, and realistic subglacial discharge,
there was a vigorous overturning circulation in the fjord
that caused rapid exchange with DB. As in the obser-
vations, the warmest DB waters did not enter the IIf
basin and isopycnals sloped downward slightly over the
FIG. 12. Pycnocline oscillations designed to induce an in-
termediary circulation. Shown here are the prescribed temperature
profiles during the descent of the pycnocline for cases P_L/P_Lm.
Points above z
5250 m and below z
52300 m were fixed. The
point at z
52150 m (red star) was moved down to z
2d52350 m
(blue star) and then back up to z
over the 10-day period, varying
sinusoidally in time. Segments joining z
to z
and z
were vertically
stretched or compressed to maintain continuity.
Without the imposed minimum, an identical plume still de-
veloped after 20 model hours, initiated simply by nonzero veloci-
ties associated with roundoff errors from the discretization.
sill spillway. The fact that the surface layer in the model
was thinner and warmer than in reality must be due, at
least partly, to the absence of icebergs in the model. The
most extreme streamfunction value near the glacier was
c5222.5 m
(see magnified view of S5_Hm in Fig. 14),
implying that the initial subglacial discharge flux of
0.08 m
had been amplified by a factor of 280 in the
upwelling plumelike boundary layer. The maximum ver-
tical velocity in the plume was 1.04 m s
at 240-m depth.
Below 170-m depth and above the sill, 15.6 m
into the fjord over the sill, with a compensating outflow
above this. This inflow renewed most of the fjord (except
for a shallow stagnant layer near the surface), which had an
area of 4.3 310
in the model. At this rate of flux, all
the basin water would have left the fjord after 32 days (28
days if we remove the upper 100 m), which is in agreement
with the renewal time-scale estimate from the XCP pro-
files. In the interior of the basin, cattained values of 224
to 248 m
on contours that were closed inside the
basin. These streamlines did not pass through the plume
but were associated rather with a slower whole basin
overturning current. This overturning was a product of the
model having not yet reached equilibrium with the exter-
nal water column. During this adjustment period, basin
waters had an artificially long residence time (because they
were slightly too dense) and turned over and over in the
basin until they had cooled and freshened enough to leave
the basin.
In case P_H (Fig. 13c), with no glacier melting or sub-
glacial discharge, pycnocline forcing at the left boundary
induced a transient circulation that was most energetic
above the sill depth. A flux of 9.9 m
was flowing out
over the sill at the time of this snapshot, with an equal flux
into the fjord surface layer above. Given that the area of
the model fjord above the sill depth was 1.5 310
flux would renew the upper fjord layer in 17.5 days if it
were sustained. Since the flow was actually oscillatory,
much of the fjord surface water that left during one-half of
the cycle flowed back into the fjord during the next half,
making the renewal inefficient, in the sense of Arneborg
(2004). Gradually, renewal of the upper layer did occur,
however, while the basin waters exhibited seiche-like
motions but remain trapped in the basin (see tracer ani-
mations in the supplemental information).
In case P_Hm (Fig. 13d), melting was activated and
the result was roughly the sum of cases C_Hm and P_H.
The meltwater plume had little effect on the circulation
above sill depth (a c50 contour cut off the basin from
the overlying waters in this snapshot), but caused over-
turning in the basin. Case P_Hm is perhaps similar to IIf
during the winter. Given the basin volume of 2.75 3
, it would take 145 days for all the water in the
basin to circulate through the meltwater plume. From
autumn until the return of subglacial discharge the fol-
lowing spring, the basin waters could all be modified
(not necessarily flushed) by this mechanism.
Cases P_H and P_Hm indicate that intermediary cir-
culation does not renew the IIf basin because of the
shallow sill. To show that this is not a vacuous result, it is
important to show that our model can create a reason-
able intermediary circulation in the case of a deep sill. In
case P_L (Fig. 13e), the sill was still 100 m higher than
DB, but DB was artificially sunk to 800-m depth so the
sill was submerged in unstratified deep water. In this
case, pycnocline oscillations alone (no melting) were
able to induce flow in the warm, deep water (including
the thin below-sill basin layer). The instantaneous flux
below the thermocline and over the sill was 24 m
this snapshot. Over the 10 pycnocline cycles, the mean
total influx below 300 m and over the sill was 2.5 3
during each event, which is 24% of the volume
flushed out of Sermilik Fjord during each of the 16
largest observed pulses.
After 10 cycles (100 days) of
pycnocline oscillations, the intermediary circulation in
case P_L had therefore moved an area equivalent to the
fjord’s deep, warm layer into the fjord over the sill. As in
P_H, renewal of the deep, warm layer in P_L was less
efficient than this calculation suggests (see Fig. 17 below
and supplemental information).
For case P_L, we also show Hovmöller diagrams of
along-fjord velocities in Fig. 15, which can be compared
to similar figures in Jackson et al. (2014). They reported
RMS velocities of 10–22 cm s
in the middle of Sermilik
Fjord and 5 cm s
in the upper fjord, which they attri-
bute to externally forced intermediary circulation. The
TABLE 6. MITgcm model run definitions.
Case Melting
oscillations H
(m) H
C_Hm U0 — 450 550
C_Lm U0 — 0 100
S5_Hm U5 — 450 550
S5_H — 5 450 550
S5_Lm U5 — 0 100
S10_Hm U10 — 450 550
P_Hm U0U450 550
P_H — 0 U450 550
P_Lm U0U0 100
P_L — 0 U0 100
Scaled by the 8-km width of the fjord, the outgoing volume in
our model is 2.0 310
, compared to 8.5 310
from Jackson
et al. (2014).
JANUARY 2015 G L A D I S H E T A L . 23
RMS speeds in our model at 100-m depth were 7.4 cm s
at the fjord mouth, 6.2 cm s
at mid fjord, and 4.0 cm s
near the head of the fjord. At 300-m depth, the model
RMS speeds at the same sites were 4.1, 3.7, and 2.8 cms
respectively. Our simulated intermediary circulation for
a hypothetical deep-silled IIf is similar to that observed in
Sermilik Fjord in terms of vertical structure and temporal
variability and in the weakening currents toward the head
of the fjord. The model’s RMS speeds are smaller by
a factor of 3. The amplitude of the pycnocline oscillations
we imposed in the model was arbitrary, so agreement in
speeds is not to be expected, but since those oscillations
were large, the lower speeds in the model probably reflect
the weaker PW/AW density contrast in IIf (in summer)
compared to at Sermilik Fjord (in winter).
To relate our model runs to the observed XCP ve-
locities, profiles of horizontal velocities for shallow-
silled cases S5_Hm and P_H are shown in Fig. 16.In
S5_Hm, subglacial discharge and glacier melting led to
a fast glacierward current along the fjord basin and
a compensating outflow above, as in the XCP velocity
profiles (Fig. 10). The deep influxes in observed profiles
V2 are V3 are both .20 m
(Table 5), which is about
half of the flux along the bottom of the model fjord
FIG. 13. Snapshots of temperature and circulation in five key model runs. Colors show po-
tential temperature (8C). Black lines are contours of the streamfunction c(m
). Case
definitions are shown in Table 6. (a) Case C_Hm at t550 days. (b) Case S5_Hm averaged over
t550–52 days. The flow was mostly steady but was averaged over a 2-day period to smooth
some high-frequency variability near x5250 km. (c) Case P_H at t558 days. (d) Case P_Hm
at t558 days. (e) Case P_L at t556 days.
basin. Some of this flux is an artifact of the initial ad-
justment of the fjord basin interior, however, and the
flux of 20 m
along the lower 350 m of the fjord basin
(Fig. 14) near the glacier agrees with the observed deep
basin current flux. The surface-intensified character of
currents in P_H is not a good match for the observed
velocity profiles.
We do not present the other five cases in equivalent
detail, but we incorporate all runs in our analysis of the
passive tracer experiments. (Animations of the passive
tracer evolution for all cases are provided in the sup-
plemental information.) We plot the time-varying frac-
tion of each tracer remaining inside the fjord for all cases
in Fig. 17. To obtain these curves, the tracer concen-
trations were integrated over the region to the right of
the fjord sill (over all depths) at each time and divided by
the integral from the initial time.
In the control case C_Hm, 96% of the basin tracer
remained in the fjord by the end of the 100-day simu-
lation while the above-sill tracer had dropped to 68% of
its original total after 100 days. Melting alone therefore
did not renew either the basin or above-sill waters of the
model IIf on a subseasonal to seasonal time scale. In case
S5_Hm, renewal was rapid above and below the sill
depth. The half-life for the basin tracer, meaning the
time required for half of the tracer to leave the fjord, was
27.5 days. This includes several days of spinup, and the
slope of the tracer curve is not constant. Nevertheless,
this compares reasonably well with the 32-day renewal
time scale estimated from the model snapshot at day 50
and the 24–36-day renewal time scale estimated from
the XCP velocities that did not take into account mixing
or transient aspects of the flow. Above the sill, renewal
was slightly faster, with a half-life of 20.5 days for the
above-sill tracer.
Switching off melting made renewal somewhat slower
(case S5_H compared to S5_Hm), with the basin tracer
half-life lengthening to 31.5 days (23 days for above-sill
tracer). Remarkably, when the subglacial discharge rate
was doubled (case S10_Hm compared to S5_Hm), the
half-life of the basin tracer lengthened to 40.5 days (23.5
days for above-sill waters). Lacking a complete expla-
nation for this, we point out that in S10_Hm the plume
outflow was initially at the surface, while outflow in
S5_Hm was below the surface. In S10_Hm, the above-
sill tracer level initially plummeted but then slowed after
10 days when the plume outflow switched to a subsurface
mode and the surface waters from outside the fjord
reclaimed the fjord surface. Throughout the 100-day
run, there was a complex confrontation at the surface
between plume outflow and the external surface waters,
and this seems to have impacted the rate of exchange
above and below the sill (not shown in detail, but par-
tially clear from tracer animations).
Deepening the sill made melting alone (case C_Lm)
quite effective at renewing both basin (half-life of 41
days) and above-sill waters (half-life of 56 days). In-
troducing subglacial discharge to the deep-silled sce-
nario (case S5_Lm) shrank both the tracer half-lives to
14 days.
When there was a high sill, intermediary circulation
was just slightly more effective than melting alone at
renewing basin waters, with or without melting activated
(cases P_H and P_Hm compared to C_Hm). With a deep
sill, intermediary circulation alone (P_L) did not evacu-
ate the basin tracer effectively, but with melting activated
(P_Lm) did slightly accelerate basin renewal compared to
the melting-only control case (C_Lm). Intermediary cir-
culation alone was generally ineffective at renewing basin
waters compared to the renewal that would already occur
because of melting at the glacier terminus.
Intermediary circulation was more effective at com-
pletely renewing waters above the sill and, especially,
waters above the pycnocline. With a high sill, in-
termediary circulation alone (P_H) removed half of the
above-sill tracer from the fjord after the full 100-day run.
The limitations of the model probably make the above-
sill renewal time scale in case P_H longer than it should
be since much of the tracer-laden waters leaving the
fjord during one phase of the cycle reentered during the
next half of the cycle. Waters just inside the fjord mouth
FIG. 14. Zoomed-in view of S5_Hm at t550 days next to JG
terminus (pale blue region on the right). Background colors are
fjord temperatures (nonlinear colorscale). Black contours are of
streamfunction c(m
). White markers are the centers of model
grid cells (every other one vertically).
JANUARY 2015 G L A D I S H E T A L . 25
were completely renewed during the 100-day simulation,
but renewal near the head of the fjord was less complete
(see the supplemental information). In a three-dimensional
model (and in reality), transverse currents outside the
fjord might sweep away the tracer-laden water on their
outward excursion and allow new water to enter the fjord
on the next inflow phase, thereby rectifying the tracer
outflow and accelerating the renewal rate. Activating
melting in the high-silled case (P_Hm) caused little
change to above-sill exchange compared to the no melt-
ing case (P_H) since the meltwater plume did not rise
above the sill depth. With a low sill and no melting (P_L),
80% of the above-sill tracer remained after 100 days.
The total exchange over the sill exceeded the volume of
the fjord (as shown above), but the oscillatory nature of
the current made the exchange inefficient.
d. Validity of model plume
The meltwater plume next to the JG terminus is the
most challenging aspect of the fjord circulation to model
correctly and also the most important part for the prob-
lem of renewal. If there were no mixing between sub-
glacial discharge and ambient fjord basin water then the
freshwater would simply rise to the surface and exit the
fjord at the surface without driving any renewal. In our
model runs with subglacial discharge, the initially fresh-
water was highly diluted by mixing with ambient basin
water (with a ratio of 280 to 1 in case S5_Hm), so the
plume outflow was composed almost entirely of basin
water and renewal was very effective.
To assess the validity of the upwelling plume in our
model, we compared it to the plume model in Jenkins
(2011). To calculate a Jenkins-style plume, we used
ambient water properties from the same summer 2009
CTD profile used in the MITgcm model and the same
planar subglacial discharge rate as in case S5_Hm
(0.08 m
). All turbulence parameters were those in
Jenkins (2011).
The numerical solution of Jenkin’s model equations
yielded a plume that rose 594 m up the terminus wall of
JG before detaching. The thickness grew nearly linearly
to a maximum of 24.5 m, and the upward velocity was
0.72–0.85 m s
with a mean of 0.83 m s
. The upward
flux grew nearly linearly and reached a maximum of
17.8 m
at 206-m depth. The plume temperature
varied from 2.468to 2.808C with a mean of 2.758C after
rapidly warming in the deepest 20 m.
If we define the width of the plume in our model as the
width in which the upwelling velocity is at least 25% of
the maximum speed next to the wall, then the width of
our plume incase S5_Hm grew gradually from 10 m at the
base of the terminus to30 m at 200-m depth. Below 200-m
depth, our model’s upward velocity averaged over this
defined width varied from 0.41 to 0.72m s
of 0.62 m s
. At 210-m depth, the vertical flux was
22.5 m
. The temperature varied from 2.698to 2.758C
FIG. 15. Along-fjord velocities in case P_L. (a) At 60 km from the glacier (over the sill). (b) At
35 km from the glacier. (c) At 10 km from the glacier.
with an average of 2.728C. All of these are in reasonable
agreement with Jenkins’ model.
If the nondimensional entrainment coefficient E
Jenkins’ model [Eq. (6)] is increased by 65% from 3.6 3
to 5.9 310
, then Jenkins’ model attains the same
flux as our model at 265-m depth and attains a width of
35 m. The mean upwelling speed decreases to 0.71 m s
in better agreement with our model results. Therefore,
our model reproduced a plume that entrains ambient
water somewhat more readily than Jenkins’ model using
best-estimate parameters and hence was slower, thicker,
and carried greater flux. From the XCP observations,
the true glacierward flux in the deep layer at station V4
was 17 m
, which is in reasonable agreement with
both models.
7. Summary and conclusions
Our summer observations, theoretical arguments, and
numerical simulation results can be synthesized into the
schematic picture of Ilulissat Icefjord (IIf) in summer in
Fig. 18. Summer profiles show that the warmest sub-
surface waters in Disko Bay (DB) are blocked by the
245-m-deep sill and do not enter IIf. The intermediate
depth DB waters that do enter the 750–800-m-deep fjord
basin are generally in the narrow potential density range
27.20 #s
#27.31 kg m
. The nearly homogeneous
basin waters make up most of the ambient thermal
boundary condition of Jakobshavn Glacier (JG) at the
head of the fjord. Near the surface there is an extremely
fresh layer in the upper 100 m of IIf in the summer that
does not extend far out into DB (a fresh surface cur-
rent that flows north past the town is known to Ilulissat
Warm Fjord Water (WFjW) observed in the basin in
summer 2009 and 2011–13 was similar to the (inferred)
basin waters of 1997/98 that triggered the retreat of JG
(Holland et al. 2008;Thomas 2004;Motyka et al. 2011).
In the August 2010 fjord survey, however, WFjW was
completely absent and instead the fjord was filled with
water more than a degree cooler. Anomalously cool
water appeared in eastern Davis Strait in March 2010,
and the cool anomaly seems to have arrived at the
mouth of the fjord in May 2010.
Cool Fjord Water (CFjW) in 2010 was nearly the same
density as WFjW from other summers (27.20 #s
27.31 kg m
). This density class is found between the
temperature minimum of cool and fresh Polar Water and
the warmer layer below fed by the West Greenland Cur-
rent (Fig. 4). This interface occupies various depths in DB
but is often above or straddling the 200-m line (Fig. 3).
Therefore, as Myers and Ribergaard (2013) suggested, the
variability of waters shallower than 200 m (i.e., Polar Wa-
ters) in DB are a major, sometimes principle, source of
variability for the thermal boundary condition of JG, de-
spite the great depth of IIf (see also Gladish et al. 2015).
The interannual shift from WFjW (2009) to CFjW
(2010) to WFjW (2011) throughout DB and IIf proved
that the fjord basin was renewed at least once per year
by DB waters. We therefore investigated the possible
mechanisms and time scales of renewal. Velocity pro-
files from summer 2013 showed basin waters flowing
glacierward at 5–15 cm s
, with compensating currents
flowing out of the fjord above. Paper and pencil calcu-
lations, along with MITgcm simulations, suggest the
interpretation that basin waters were flowing toward the
glacier where they were entrained into an upwelling
plume driven by fresh subglacial discharge and melting
along the terminus wall, as seen at some other marine
glaciers (Motyka et al. 2003;Rignot et al. 2010;Straneo
et al. 2011). The implied basin renewal time scale is
approximately 1 month.
Temperature data from instrumented seal and Green-
land halibut dives inside IIf show that subseasonal tem-
perature variability was small in the basin compared to
similarly dense waters at the shelf break to the west or just
inside the fjord mouth along the spillway. The basin waters
were dominated by interannual variability, and renewal of
the fjord basin as rapid as that seen in Sermilik Fjord
(Straneo et al. 2010;Jackson et al. 2014) did not likely
FIG. 16. Vertical profiles of model zonal velocity. Case S5_Hm
reached a nearly steady state, so velocity profiles are from a fixed
time (day 50) from various locations (0.5, 10, 20, 30, and 40 km west
of the glacier terminus). Case P_H was unsteady and the profiles
shown are snapshots taken every 2 days over a 10-day period be-
ginning on day 50 at a position 30 km west of the terminus.
JANUARY 2015 G L A D I S H E T A L . 27
occur outside of summer. Our model results show that
external baroclinic forcing, which drives subseasonal var-
iability and renewal at Sermilik Fjord, is prevented from
driving renewal of the IIf basin by the shallow sill.
Variability associated with the seasonal cycle in the IIf
basin was also small, as was variability in eastern DB at
350-m depth. Here, small means small compared to the
38C annual range of temperatures observed at 200-m
depth in DB (Hansen et al. 2012) or to the even larger
annual range of temperatures in West Greenland Cur-
rent waters on fixed isopycnals (Curry et al. 2014;
Gladish et al. 2015). From autumn to midwinter, fjord
isopycnals gradually descended (the basin became
slightly cooler and fresher), and at the same time the
fresh summer surface layer disappeared. The gradual
basin cooling may have been produced by the shutdown
of exchange with DB and the cycling of basin waters
through a winter terminus plume (driven only by melt-
ing) that was not buoyant enough to escape the fjord
basin. From January until late spring the fjord isopycnals
gradually rose, possibly in response to isopycnals outside
the fjord rising in the course of the regional seasonal
cycle (Gladish et al. 2015).
Below the sill depth, any large temperature changes in
IIf probably occur after the onset of superficial melting
over the drainage basin of JG. Once this water crosses the
grounding line, the extra buoyancy provided to the ter-
minus plume allows it to rise above the sill depth and
makes it energetic enough to entrain the massive amounts
of basin water required to renew the basin. As new water
replaces the old, potentially new water properties come
into contact with the glacier.
If the residence time of basin waters in summer is 30
days and the thermal forcing for inflowing waters is 58C,
the heat available for melting that enters the fjord each
month is 40 310
J. The heat removed by melting at
JG terminus in that time, assuming a rate of 4 mday
1.5 310
J. Only 4% of the heat entering the fjord is
consumed by melting. Besides the fact that the sill blocks
the very warmest DB waters from entering the IIf basin,
the fjord circulation therefore does not significantly re-
strict the rate at which heat is brought near the glacier in
summer. It is the turbulent dynamics of the boundary
plume that throttles the rate of heat transfer toward the
ice, not the total heat content of DB/IIf.
The CFjW in the IIf basin in 2010 was nearly as cold as
the (inferred) basin waters of the 1980s, before the
massive retreat of JG. The effect of temporarily reversing
the sudden warming of 1997/98 appears to have been
small, however, with glacier velocities from summer 2010
to spring 2011 (when CFjW was likely in the basin)
bearing little sign of changed oceanic influence (Joughin
et al. 2014). A key question for the future of JG is the
following: Is calving at the vertical terminus controlled
by melting (O’Leary and Christoffersen 2013)orby
purely internal glacier dynamics [certain mechanisms in
Benn et al. (2007)]? After all, if calving were somehow
to stop, the glacier would advance to fill the entire fjord
in just a few years and would surely thicken and slow
(Joughin et al. 2008). The role ocean temperatures play
in the delicate near balance between calving velocity
and ice velocity must be understood before it will be
possible to predict the future response of JG to ocean
temperature variability.
Acknowledgments. NYU observations at Ilulissat
were supported by NSF Office of Polar Program Grants
ARC-0806393 and ARC-1304137, NASA Polar Pro-
grams Grant NNX08AN52G, and the NYU Abu Dhabi
Center for Global Sea Level Change Grant G1204.
Several CTD profiles collected in 2007 were contributed
by Søren Rysgaard. CTD data at the fjord mouth were
collected by Ian Howat in summer 2008. CTD data for
2011 in DB were collected by the GINR. CTD data in
FIG. 17. Time evolution of the fraction of passive tracers remaining in the fjord in all model runs from Table 6.
DB collected by DMI (led by Mads Ribergaard in recent
years) on behalf of the GINR were downloaded from the
ICES Oceanographic database. Marc Hansen and Torkel
Nielsen (Tech University of Denmark) made the Arctic
Station CTD dataset available to us for this work. XCP
velocity data in the fjord during 2013 were collected by
Denise Holland. Mitch Bushuk assisted with the NYU
CTD survey in summer 2012. Ole Eigaard processed the
instrumented halibut data. MITgcm computations were
carried out using NYU High Performance Computing
clusters in New York and Abu Dhabi. One-dimensional
plume solutions were computed using code written by
Adrian Jenkins. Comments from two anonymous reviewers
led to major improvements.
Tides in IIf
The gravest barotropic fjord seiche mode has a pe-
riod of 4L/ffiffiffiffiffiffiffi
p545 min, with a fjord depth H5800 m
and length L560 km. MacAyeal et al. (2012) show that
this period may be lengthened to a few hours when
deep-drafted icebergs are present throughout the fjord.
Nevertheless, this seiche mode is too high frequency to
resonate with the semidiurnal barotropic tide. The sea
surface throughout the fjord therefore rises and falls
uniformly and barotropic tidal currents in the fjord at
a given location are no larger than y5ARp/(CT), where
Ais the surface area of the fjord landward of the given
location, R52.75 m is the maximum tidal range (from
the pressure record for Fjord 400 m), Cis the vertical
cross-sectional area of the fjord at the given location, and
T512 h is the semidiurnal period. For stations V2, V3,
and V4, which are inland of the areally large north and
south arms of the fjord, we take A5DW and C5WH,
where Dis the distance to the ice front, Wis the width of
the fjord, and His the depth. Then y5DRp/(HT), with
tidal velocities decreasing to zero toward the ice front.
At V2, we estimate y50.0075 m s
and at V3 and V4, y
is even smaller. At the three landward stations, baro-
tropic tidal velocities are insignificant compared to the
signals in the XCP profiles. At V1, making the crude
approximation that the north and south arms together
double the area of the fjord (compared to the main east–
west channel alone), we estimate y50.03 m s
. This is
not negligible, but it is still smaller than the 0.15 m s
current flowing toward the glacier in the bottom 100 m
or the 0.07 m s
current flowing out of the fjord over the
sill (Fig. 10a).
Internal tides generated at the fjord mouth also do
not dominate the XCP velocity profiles. To estimate
the size of internal tidal velocities in the fjord basin, we
conceptually model the fjord as a two-layer system
(Fig. A1).
In the fjord interior (x.0), the governing equations,
linearized about a state of rest, are
x, (A1)
x, (A2)
x, and (A3)
x, (A4)
where g05g(r
is the reduced gravity. We let u
denote the barotropic tide, which applies to both layers,
and we suppose that u
and u
constitute a baroclinic
mode solution. Baroclinic mode waves have phase speed
FIG. 18. Schematic diagram of IIf based on our findings.
JANUARY 2015 G L A D I S H E T A L . 29