In this section, first the ice thickness reproduction ability within the CAA
of the NEMO-LIM2 configurations used in this study is examined via
comparisons with the ECCC thickness. Second, the detailed thermodynamic
and dynamic ice thickness changes, both the spatial distribution and temporal
evolution at selected sites (Cambridge Bay and Resolute), based on the
simulation outputs will be presented. Ice volume balance,
focusing on the thermodynamics contribution and lateral transport, in the
northern CAA, Parry Channel, and Baffin Bay will also be included at the end.
Ice thickness comparison
Simulated ice thickness at each selected ECCC thickness site
(Fig. and Table , unit: metres) from
January 2003 to December 2016 (orange: ANHA4-CORE simulation; green:
GLORYS2v3 product; blue: ANHA4-CGRF simulation; red: ANHA12-CGRF simulation)
against weekly ECCC observations (black dots). Note the GLORYS2v3 product is
a monthly mean field while the rest of the simulations use 5-day averages.
Different y axis scales are used.
Similar to Fig. but for ice thickness seasonal cycle
(starting from 17 September to the next 12 September, averaged over 2003 to
2016; ANHA4-CORE ends by 2009). Note observations are not averaged over time
because the sampling time is different from year to year).
Figure shows the ice thickness comparison with observations.
In general, both ANHA4-CGRF (blue lines) and ANHA12-CGRF (red lines)
simulations produce similar seasonal and interannual variations in ice
thickness, which compare reasonably well at some sites (i.e., Cambridge Bay,
Coral Harbour, Hall Beach, Resolute, and Iqaluit) but not at the others (Eureka,
Alert, and Alert LT1). The sites where the model produced much thicker ice are
likely where significant concentrations of old ice exists .
Although the observations are missing in the sea ice melting season, an
asymmetric seasonal cycle (a shorter faster melting period follows a
relatively longer slow growth period) is evidenced by the available data
and reproduced by the simulations. This is clearly shown in the ice thickness
seasonal cycle plot (Fig. ). Taking account of the
model resolution, the interpolated simulated ice thickness reflects
the variability some distance off the coast rather than the exact observation
locations. The geographic location differences, which are also related to
model resolution, could also lead to discrepancies in the comparisons here.
Thus, if the model can capture the seasonal cycle (e.g., multiple data points
in both ice growth and melting seasons), the model is likely capable of
simulating the process.
At Iqaluit, the model did a good job in most years during the initial ice
growth period but failed to catch the thick sea ice in the next April and May
(Fig. ), particularly in 2014, 2015, and 2016
(Fig. h). This could be a local atmospheric forcing field bias
or a model resolution issue, i.e., the measurements captured very localized
extremes beyond the ability of model to resolve. Similar behaviour happens at
Coral Harbour and Hall Beach (Fig. c and d). Further
investigation is needed.
At Eureka, Alert, and Alert LT1 sites (Figs. and
e, f, and g), there are clear differences between the
simulated ice thickness and the observations (∼ 2 m at Alert/Alert LT1
and ∼ 1 m at Eureka). Note that neither ANHA4 nor ANHA12 has the capability
to resolve the difference between Alert and Alert LT1; thus, the same
simulated values are shown on the figure for both sites. The differences
between simulations and observations could be an initial value problem,
particularly at Eureka (Fig. g). However, given high
concentrations of old ice are at these sites, observations represent the
immobile level first-year ice only. Thus, the model and the observations may
not be representing the same type of ice. At Alert and Alert LT1, both ANHA4-CGRF
(blue line) and ANHA12-CGRF (red line) show similar interannual trends to
that in GLORYS2v3 (which extends back to 1993, green line), meaning it is
likely a pure initial value problem rather than the model equilibrium issue
mentioned in . In addition, the seasonal cycle is
not clear in the GLORYS2v3 product. The issue is also present in some years,
i.e., 2005–2007, in the ANHA4-CGRF and ANHA12-CGRF simulations. ANHA4-CORE
(orange line) is generally improved compared with the observations in both
the amplitude and seasonal cycle. However, this improvement was achieved by
accident and is related to a snow depth issue in this simulation. The
snowfall data from CORE-II have a monthly resolution, which is possibly too
coarse temporally (Hakase Hayashida, personal communication, 2017). This leads the snow depth to drop
to close to zero quickly during the first year of the simulation but not
in the CGRF simulations with hourly snowfall. Thus, it does not indicate that
CORE-II forcing is performing better than other atmospheric forcing datasets
in this region. The equilibrium issue, i.e., ice thickness keeping
increasing, might happen at Eureka in our simulations with either CGRF or
CORE-II forcing. The upward trend over 2005 to 2007 is also present in the
observations but is missing in the GLORYS2v3 although GLORYS2v3 has a small
thickness (which is likely due to data assimilation in GLORYS2v3 or an
atmospheric forcing issue in 2005). Its trend does not reflect the real
change or variability.
At Cambridge Bay, simulations (red and blue lines in Fig. b)
with CGRF forcing show very good agreement with the observations except
during the winters of 2013 and 2014. Considering the horizontal resolution of
our simulations is not capable of resolving the inner bay at Cambridge Bay,
the match in ice thickness between the simulation and observations indicates
the variation of ice thickness within the inner Bay might be small. Both
ANHA4-CORE and GLORYS2v3 simulations underestimated the maximum values in
winters by ∼ 0.5 m. This indicates CGRF forcing might provide more
realistic surface inputs in this region.
At Resolute, it is more complicated (Fig. a). Prior to the
significant sea ice melting in 2007, none of the simulations show ice-free
conditions in this region in summer. GLORYS2v3 shows relatively thinner ice
in summer months but it is still 0.5 to 1.5 m thick. It could be the
initial value problem. However, high-frequently variations even in winter in
the ANHA simulations suggest that the ice growth process is not dominated by
a smoothly changing physical process (e.g., air temperature). Thus, it is
likely due to another physical process such as advection from surrounding
areas. This will be discussed more in the following section. Post-2007, the
seasonal cycle in the sea ice field is more distinct, although ice-free
summer conditions do not happen every year. After 2010, simulations produce
winter sea ice thickness much closer to the observations.
Thermodynamic and dynamic ice thickness change
In the real world, both the thermodynamic and dynamic ice thickness processes
are coupled together (occurring at the same time). However, with the
assistance of numerical model, the two processes can be decoupled (shown in
Eq. ) to better understand the relative importance of each
process.
ΔHtotal=ΔHthermal+ΔHdynamic,
where ΔHtotal is the total ice thickness change over a
specific time interval, ΔHthermal is the ice thickness
change due to vertical heat fluxes (through the atmosphere–ice–ocean
interfaces), and ΔHdynamic is the ice thickness change due to
dynamic processes. In practice, a simple approach is utilized to compute the
two terms on the right side. ΔHthermal is calculated based
on the model thermal ice production. ΔHdynamic is taken as
the residual from the ΔHtotal.
Spatial distribution
Upper panel shows the thickness (unit: metres) averaged over
2003–2016 at the beginning of December (a) and at end of
April (b). Lower panel shows the thermodynamics component
(c) and dynamic component (d) ice thickness contribution (unit: metres)
between December (a) and the following April (b) averaged
over 2003–2015. ANHA12-CGRF simulation is used here.
Here we focus on ice growth process between December and April of the
following year. Figure a and b show the simulated ice
thickness in ANHA12 at the beginning of December and at the end of April,
respectively. Geographically, at the end of April, very thick sea ice is
located in the northern CAA (∼ 4 m by the end of April) with regional
maximum (> 4.5 m) at the openings to the Arctic Ocean. This is consistent
with the ICESat and Cryosat-2 estimations
e.g.,.
Less thick sea ice covers western, and central Parry Channel (just to the
west of the site Resolute) and M'Clintock Channel with a thickness of 2.5
to 3 m. These values are similar to previous observations from airborne
electromagnetic surveys and satellite
. Relatively thin ice (<2 m) is
mainly in the southern CAA, eastern Parry Channel, coast of Baffin Island,
and within Hudson Bay. Invasion of the Arctic Ocean ice pack through the
northern CAA openings and the advection from there into central Parry Channel
are clearly shown in the figures, consistent with previous studies
e.g.,.
During the winter, sea ice grows everywhere in the CAA regions due to the
thermodynamic cooling (Fig. c). But the total increase
over the winter is not evenly distributed in space, and ice growth is not largest
in the north. Large thermodynamic ice growth is seen in the eastern CAA
(eastern Parry Channel, Nares Strait, Baffin Island coast, and western Hudson
Bay), Amundsen Bay, and many coastal regions (e.g., western coast of Banks
Island, northern coast of western Parry Channel). Regions covered by thick
sea ice (i.e., northern CAA, west-central Parry Channel, and M'Clintock
Channel) show less thermodynamic ice growth over the winter. This is
particularly true in the northern CAA, likely due to the existence of already
thick ice reducing the heat exchange between the ocean and atmosphere.
The dynamic contribution to sea ice thickness is mainly negative (reduces
local ice thickness) within the CAA (Fig. d). Large
positive values (0.4 to 0.7 m) are shown along the Arctic Ocean coast
off the CAA and within the Beaufort Sea. This is consistent with known sea
ice convergence or strong advection of thick ice from upstream regions
. Within the northern CAA,
west-central Parry Channel, and M'Clintock Channel, there is ∼ 0.25 m
thick ice loss locally due to the dynamics. Note the positive values
occurring in the south of M'Clintock Channel, suggesting a net convergence
there which contributes to the local ice thickening in winter. In the eastern
CAA (e.g., eastern Parry Channel, Nares Strait, and northwest corner of Baffin
Bay), there are large negative dynamic thickness contributions, implying
strong ice advection.
Although the North Water (NOW) Polynya
e.g., region is
still ice covered by the end of April (Fig. b), the
spatial distribution of negative dynamic ice thickness (which helps to remove
local ice) captures the shape of NOW Polynya well. Weaker advection of sea
ice at Smith Sound and to its south, which is likely to be caused by ice
jamming, is also simulated by the model.
Seasonal cycle at Cambridge Bay and Resolute
Two sites, Cambridge Bay and Resolute, were selected to further study the
seasonal cycle of the thermodynamic and dynamic ice thickness changes.
Seasonal cycle (averaged over 2003 to 2016) of ice thickness
(a, unit: metres), dynamic (b), and thermodynamic
(c) ice thickness changes (unit: metres per 5-day) at Cambridge Bay
from the ANHA12-CGRF simulation. Note each x grid line indicates the
beginning of each month.
Figure shows the seasonal cycle of ice thickness,
5-day ΔHdynamic and ΔHthermal averaged
between 2003 and 2016 at Cambridge Bay. Sea ice reaches its maximal thickness
(∼ 2 m) in late May with ice-free conditions for about 2 months
(August and September). As the sea ice starts to form (October and November),
both thermodynamics (e.g., due to cold temperature) and dynamics (e.g., local
advection) play a role in the production of the ice thickness although with
opposite contributions (Fig. b and c). Starting from
December through to the end of the next May, it is almost a pure
thermodynamic process that controls the ice thickness change. Note that the
thermodynamic ice production is not constant in time; it is about 3 times
larger in the first period (∼ 0.03 m per day) than in the later
period (∼ 0.01 m per day). The steady thermodynamic growth in the
second period contributes to about half of the total ice thickness. During
the ice melting period (June and July), the thermodynamics is the major
player as well (Fig. c).
Same as Fig. but at Resolute.
Same as Fig. but only considering 2012 for
ANHA12-CGRF.
At Resolute, on average, there is no ice-free period (∼ 2.5 m at the
end of May and ∼ 1 m in August and September)
(Fig. a), although there is large interannual variability
(Fig. a). For example, in 2012 there is an ice-free period in
the middle of September (Fig. a). The freeze-up
date is about half a month earlier at Resolute than that at Cambridge Bay.
The ice production is a little larger at the beginning (October to December),
∼ 0.02 m per day, than later (Fig. b),
∼ 0.01 m per day, but the difference is not as noticeable as at
Cambridge Bay (Fig. b). The relatively faster
thermal growth lasts longer at Resolute than that at Cambridge Bay, likely
due to local advection. These features are also applicable to a specific
year, e.g., 2012 (Fig. ). The non-thermodynamic
contribution is more significant than at Resolute
(Fig. c) but basically plays a negative role, i.e.,
slowing ice thickness increase during the winter season. Similarly to
Cambridge Bay, the thermodynamics is the dominant factor melting the sea ice,
with a melting peak in July. During the melting season, more ice can be
advected to Resolute and melts later locally
(Fig. c) than that at Cambridge
(Fig. c).
Ice volume budget
Northern CAA
Sea ice volume balance in the northern CAA (location see
Fig. ). (a) Maximum total ice volume (black bars, unit:
km3) in each seasonal cycle (17 September to next 12 September).
(b) The net ice volume change (black bars) between 2 consecutive
years, thermodynamic ice volume change (red bars) and lateral ice volume
transport (blue bars) in cubic kilometres.
Figure a shows the maximum total ice volume
(referred as “ice volume” hereafter if not mentioned specifically) in the
northern CAA (solid black polygon shown in Fig. ), which is
covered by thick ice most of the year. An increase of 14 % (from 695 to
789 km3) in the ice volume is shown in the first 3 years. This is
similar to the equilibrium problem we see at Eureka (Fig. g).
During this period, the thermodynamic growth is the main contributor
(203 km3) while the net lateral ice volume transport (-138 km3 per
year) is out of this region (Fig. b), particular
through Byam Martin Channel (Fig. ). The sign
convention is defined as positive means transport into the northern CAA
regions. The ice volume stabilizes at high values for 4 years until 2008.
After that, a shrinkage of about one-third (792 to 535 km3) in ice volume is
simulated over 2008–2013. This reduction is due to large net lateral
transport (Fig. b), e.g., in 2008
(-125 km3), 2009 (-207 km3), and 2012 (-78 km3). The large lateral
transport is not always due to large outflow to the south, e.g.,
-102 km3 in 2008 and -96 km3 in 2010 through Byam Martin Channel
and -48 km3 in 2010 through Penny Strait, but also could be caused by
less import (e.g., 8 km3 in 2012) or even export (-134 km3 in 2009)
through the northern gates (Fig. ). It also shows
that large import of ice through the northern gates is usually accompanied by
large export to the south, mainly via Byam Martin Channel but also through
Penny Strait in some years, e.g., 2010, 2013, and 2014. Both the thermodynamic
and lateral transport (contribution through each major gate as well)
experience significant interannual variations.
Lateral sea ice volume transport (unit: km3, black bars: north
gates; red bars: Byam Martin Channel; blue bars: Penny Strait; light gray
bars: the rest of the lateral gates) in the northern CAA (location see
Fig. ). (a) Over the same period defined in
Fig. .
Parry Channel
Parry Channel (dashed black polygon shown in Fig. ) is the main
water channel that connects the Arctic Ocean and Baffin Bay through the CAA
(Fig. ). It starts from M'Clure Strait on the west, running to
east by the mouth of Lancaster Sound before entering Baffin Bay. Over the
whole simulation, a decrease of 15.2 km3 per year (r2=0.66,
p=0.0004) in the maximum ice volume is present
(Fig. a). Even ignoring the initial increase over the
first 3 years (from 554 km3 in 2003 to 665 km3 in 2005), the
downward trend is similar (14.6 km3 per year with r2=0.58,
p=0.0065). However, this decline is not steady but with interannual
variability. The minima are found in 2012 (407 km3) and 2013
(398 km3), which are more than 20 % lower than the average,
524 km3. Similar to the northern CAA, thermodynamic growth is the main
contributor to the ice volume increase from year to year while net lateral
transport functions to deplete the sea ice (Fig. b).
Large inflows from M'Clure Strait are simulated in the first 2 years
(Fig. ), but the direction of sea ice flow can switch from
year to year (e.g., -132 km3 in 2011 and 120 km3 in 2013). The
outflow events in 2007 and 2011 are consistent with the ice area flux study
in . As significant interannual variability
in the amount of this ice volume flux is also present, the overall
contribution of ice volume into Parry Channel from M'Clure Strait is small,
which also supports the finding in .
Major sea ice volume exchanges (Fig. ) occur at Byam
Martin Channel (inflow from the north), M'Clintock Channel (outflow to the
south), and Lancaster Sound mouth (outflow to Baffin Bay). On average, annual
ice volume fluxes through the first two routes nearly cancel each other
(92 km3 vs. 94 km3), which indicates a relatively volume conservation
due to high concentrations. The averaged sea ice transport at the east end
(via Lancaster Sound mouth) is an export into Baffin Bay (92 km3 per
year), which is closed to an early estimation (102 km3 per year) from
.
Similar to Fig. but within Parry
Channel.
Similar to Fig. but for Parry Channel
region (black bars: M'Clure Strait; red bars: Byam Martin Channel; blue bars:
M'Clintock Channel; light gray bars: Lancaster Sound Mouth; yellow bars: the
rest of the lateral gates).
Baffin Bay
Baffin Bay (dotted black polygon shown in Fig. ), bounded by
Smith Sound in the north, Jones Sound and Lancaster Sound in the west, and
Davis Strait in the south (∼ 672×103 km2), is covered by
seasonal sea ice with an averaged maximum ice volume of 895 km3. No
obvious decline is found over the simulation period
(Fig. a). Although both the local thermodynamic ice growth
and lateral ice volume flux show remarkable interannual variability
(Fig. b), the balance between the contributions results in
a relatively stable ice volume within the Bay.
Figure shows the lateral ice volume flux is dominated
by the inflow from the northern (Smith Sound) and outflow from the south
(Davis Strait). On the west side (via Lancaster Sound and Jones Sound), the
direction of ice flux is mainly into Baffin Bay, but the total amount is
much smaller than ice volume flux via either Smith Sound or Davis Strait.
This is consistent with other studies
e.g.,. The
averaged export of ice volume flux through Davis Strait is 702 km3 per
year with a standard deviation of 147 km3 per year. This number is larger
than estimates in :
500 and 424 km3, respectively. But they are not very different, taking
account of the uncertainties in observations, large interannual variability,
and difference in integration period (Baffin Bay ice volume maximals are used
to determine the integration period in this study). It is more comparable to
the 530–800 km3 per year estimated by . In
addition, the low outflow event in 2004 and high outflow event in 2008 agree
with . The inflow of ice flux from Smith
Sound is 377 km3 per year, which is much larger than the long-term mean
(9 km3 per year) in a coarse simulation done by , but
closer to their estimate through southern Smith Sound section, i.e.,
170 km3. It indicates sea ice in this region is more dynamic in our
simulation (Fig. ). This dynamic feature is also
evidenced in ice motion vector fields derived from enhanced resolution
Advanced Microwave Scanning Radiometer (AMSR-E) imagery in
. Relatively large ice fluxes (e.g.,
110 km3 per year in 1977–1978 and 136 km3 in 1974–1975) through
Smith Sound were also estimated based on satellite images and a mean ice
thickness of 2.5 m by . Another way to
estimate the ice flux through Smith Sound is based on the ice flux through
the north end of Nares Strait (i.e., Robeson Channel). Note ice flux through
Smith Sound usually is larger than the sea ice influx through Nares Strait
. estimated the annual
mean ice volume flux to 141 km3 per year over 2003–2008. The large
outflow (254 km3) event in 2007 through Nares Strait reported by
is also seen in our simulation
(Fig. ). Both and
attributed the
much-lower-than-observation ice flux through Nares Strait to wind forcing,
which does not have enough resolution to resolve the along-strait winds. With
a high-resolution wind forcing, were able to
reproduce much more reasonable ice flux through this narrow channel.
Similar to Fig. but within Baffin Bay.
Similar to Fig. but for Baffin Bay (black
bars: Lancaster Sound Mouth; red bars: Jones Sound Mouth; blue bars: Smith
Sound; light gray bars: Davis Strait).