A set of hindcast simulations with the new version 3.6 of the Nucleus for European Modelling of the Ocean (NEMO) ocean–ice model in the ORCA1 configuration and forced by the DRAKKAR Forcing Set version 5.2 (DFS5.2) atmospheric data was performed from 1958 to 2012. Simulations differed in their sea-ice component: the old standard version Louvain-la-Neuve Sea Ice Model (LIM2) and its successor LIM3. Main differences between these sea-ice models are the parameterisations of sub-grid-scale sea-ice thickness distribution, ice deformation, thermodynamic processes, and sea-ice salinity. Our main objective was to analyse the response of the ocean–ice system sensitivity to the change in sea-ice physics. Additional sensitivity simulations were carried out for the attribution of observed differences between the two main simulations.
In the Arctic, NEMO-LIM3 compares better with observations by realistically
reproducing the sea-ice extent decline during the last few decades due to its
multi-category sea-ice thickness. In the Antarctic, NEMO-LIM3 more
realistically simulates the seasonal evolution of sea-ice extent than
NEMO-LIM2. In terms of oceanic properties, improvements are not as evident,
although NEMO-LIM3 reproduces a more realistic hydrography in the Labrador
Sea and in the Arctic Ocean, including a reduced cold temperature bias of the
Arctic Intermediate Water at 250
Sea ice is an important part of Earth's climate system because
it effectively regulates the amount of energy being transferred between the
atmosphere and oceans
Additionally, the sea-ice cover and its variability may affect the
large-scale atmospheric circulation, also outside the high latitudes. For
example, some studies suggest that the Arctic sea-ice loss has increased the
frequency of atmospheric blocking events, which then has changed the snowfall
over America and Eurasia
Recently, the version 3.6 of the Nucleus for European Modelling of the Ocean (NEMO)
was released, along with its new sea-ice component, Louvain-la-Neuve Sea Ice Model
(LIM) version 3.6
To support our task, a significant body of literature presenting ocean–ice
model assessments provides us with an important reference when carrying out our
NEMO-LIM assessments. For example, papers of the CORE-II virtual special
issue of the
In the polar context, which is the regional focus of our study, the most
important CORE-II papers include
The paper is divided into six sections. Section 2 describes the two versions of the ocean–ice models, NEMO-LIM2 and NEMO-LIM3, their initial and boundary conditions, model input data and observational reference data. In Sect. 3, we present sea-ice-related results of the reference LIM3 hindcast simulation in comparison with observations and the reference LIM2 hindcast. Section 4 presents results of the NEMO-LIM sensitivity simulations to test the robustness of LIM3 and LIM2 differences for surface freshwater adjustments. In Sect. 4 we also assess how realistic sea-ice-simplified LIM3 single-category ice thickness parameterisation reproduces. In Sect. 5 differences of the ocean characteristics between NEMO-LIM2 and NEMO-LIM3 are discussed. Finally, the most important findings are highlighted in the conclusion section.
All simulations presented here are based on the version 3.6_STABLE
(revision 5918, released on 26 November 2015) of the NEMO-LIM ocean–ice
modelling system
OPA is a finite difference, hydrostatic, primitive equation ocean general
circulation model. Its vertical coordinate system is based on
The simulations are performed on an ORCA-like global tripolar grid with
1
Our ocean–ice configurations only differ in their sea-ice component, all
other experimental conditions being identical. We use the levitating sea-ice
framework, following the convention of
LIM2
LIM3.6
The NEMO model bathymetry is a combination of the 1 arcmin Global Relief
Model
As a standard practice in forced ocean–ice simulations, the mean sea-level
controls were used to prevent the unrealistic drift of the sea surface height
due to freshwater boundary forcing distorted by errors in precipitation,
evaporation and river runoff
NEMO-LIM simulations started from the state of no motion in January 1958,
with initial conditions for ocean temperature and salinity derived from PHC3
NEMO3.6-LIM simulations analysed in this study.
Differences between LIM2 and LIM3 initial sea-ice and ice dynamics parameter
values originate from the fact that they are recommended values according to
the default NEMO3.6 configuration. Instead of setting the LIM2 initial values
and ice dynamics identical to LIM3, for example, we took the point of view
that we compare two different sea-ice models, each with its own specific and
optimised tuning, and with no specific focus on ice dynamics. This is the
point of view that was adopted by
Even though the sea-ice initialisation differs slightly in terms of
hemispheric snow thickness, it does not impact our results, which focus on the
last decade 45 years after the initialisation. The lower LIM2 ice strength
In addition to the two reference simulations, we carried out sensitivity
simulations to determine how significant and systematic differences between
LIM3 and LIM2 are. In these sensitivity simulations, processes related to
ocean–ice interactions were regulated and adjusted. In this way, we were
able to isolate the impacts of individual processes and quantify their
significance. First, we switched NEMO-LIM3 into its single-category mode
that employs a virtual ice thickness distribution parameterisation, which
make the model simpler and computationally cheaper than with multiple
categories. Then, instead of using a prognostic salinity profile, we set the
LIM3 sea-ice salinity to a constant value of 4
The second set of sensitivity experiments were performed to examine the
impact of ocean surface boundary conditions on ocean–ice properties, and
therefore to see how robust our LIM3–LIM2 comparison results are. For this,
we carried out NEMO-LIM2 and NEMO-LIM3 simulations where the mean sea-level
controls were switched off by setting
When quantitatively assessing the modelled sea-ice and upper ocean realism,
we included comparisons with satellite-based and reanalysis products of
sea-ice concentration, thickness and velocity. Since 1979, space-borne
passive microwave sensors have produced a nearly continuous and consistent
record of ice concentration, which provides a good basis for model validation.
For sea-ice concentration we used the joint National Oceanic and Atmospheric
Administration and National Snow and Ice Data Center (NOAA/NSIDC) Climate
Data Record of Passive Microwave Sea Ice Concentration, version 2
For hydrographic comparisons, we decided to use two observational data sets.
First, we selected aforementioned Polar Hydrographic Climatology version 3
(PHC3), which is a global climatology with a combination of National
Oceanographic Data Center (NODC) 1998 world climatology, the Environmental
Working Group (EWG) Arctic Ocean Atlas and selected Canadian data provided
by the Bedford Institute of Oceanography
Modelled mixed layer depths (MLD) were compared with the observational
climatologies by
In this section, we analyse how well LIM reproduces large-scale climatological
sea-ice properties (ice areal coverage, volume and drift). In order to evaluate
the new sea-ice model, we compare LIM3 results to satellite observations,
reanalysis data as well as to the equivalent LIM2 simulation. All mean
fields are computed over the last decade of integration, from 2003 to 2012. As
the LIM3 sea-ice properties have already been analysed by
In the NH in September, the geographical distribution of LIM3 sea-ice
concentration presents high values in the Canadian Arctic Archipelago with a
realistic latitudinal decrease toward the Eurasian Arctic (Fig. 1a). LIM3
tends to generally underestimate the ice concentration, by
Geographical distribution of Arctic sea-ice concentration averaged
for September
Simulated (coloured lines) and observed
Mean seasonal cycles of the modelled sea-ice extents are shown in Fig. 2a and c together with the NSIDC observations, all averaged over the years 2003–2012. In the NH, the LIM3 sea-ice extent closely follows the observed data and represents a clear improvement compared to LIM2, particularly in summer (Fig. 2a). The respective LIM2 values are too high. LIM2 does not manage to melt enough ice and systematically overestimates the NH sea-ice extent. On the contrary, the LIM3 multi-category sea-ice thickness distribution allows for larger rates of melting due to its thin ice categories compared to the single-category LIM2, and enhances the seasonal cycle of sea-ice extent bringing it closer to observations.
Associated with the better mean seasonal cycle, the inter-annual time series
of sea-ice extent is improved in LIM3 compared to LIM2 (Fig. 2b). Both the
maximum and minimum sea-ice extent are well reproduced by LIM3, as shown by
the time series in Fig. 2b that closely follow the NSIDC data in 1979–2012.
Moreover, LIM3 realistically captures most of the summer minimum extents,
including the 2007 record minimum. In contrast, LIM2 systematically
overestimates yearly minimum, maximum and mean sea-ice extents during the
whole period of integration. For example the 2007 minimum is overestimated by
50 %. The two LIM models show comparable negative sea-ice extent trends
in March, which are less negative than satellite observed trends. In
September, the LIM3 trend is close to the observed one, while the LIM2
negative trend is too small. As concluded by
In the SH, the LIM3 sea-ice edge is generally well located in the austral summer and the geographical distribution is correctly represented (Fig. 1g, h). LIM3 sea ice is mostly confined to the western Weddell Sea, the southern Bellingshausen and Amundsen seas and the south-east Ross Sea. Some differences with satellite observations are present. Notably, LIM3 underestimates the narrow fringe of sea ice around the eastern Antarctic coast and its sea ice also disappears excessively in the western Weddell Sea, where the model has a lower sea-ice concentration than observed, also indicating that its sea ice is too thin regionally. The LIM2 sea-ice concentration is everywhere larger than the LIM3 one and the observed one across most of the Southern Ocean, with the largest differences in the Ross Sea and the eastern Weddell Sea (Fig. 1i, k).
Both LIM models have a seasonal cycle of sea-ice extent with too large
amplitudes (Fig. 2c). Periods of sea-ice growth are shorter, and sea-ice
growth/melt rates are faster than observed. In LIM3, the monthly minimum
sea-ice extent in February is less than the observed, while the maximum
sea-ice extent in September is overestimated with a seasonal amplitude of
The time series of annual-mean sea-ice extent of LIM3 is rather well reproduced and closely follows observations (Fig. 2d), but the sea-ice summer retreat is systematically too strong and summer extent too low. The LIM3 winter sea ice is on the average thicker than the LIM2 sea ice, while in summer their average thicknesses are close to each other (not shown here). On the other hand, the average LIM3 sea-ice concentration is systematically about 1–10 % smaller than the LIM2 one, even in the central ice pack. As a result, the LIM3 sea-ice extent is smaller, particularly in summer.
The processes explaining the low LIM3 summer sea-ice extent compared to LIM2
are related to (1) the steeper decline of LIM3 mean sea-ice thickness and
(2) to its systematically lower sea-ice concentration. Arguably the most
important process is the positive ice-albedo feedback, which is governed by
the fast melting of thin ice enabling an effective penetration of solar
energy into the upper ocean. Negative sea-ice-related feedbacks are the ice
thickness–ice strength relationship and the ice thickness–ice growth rate
relationship, which is important during the growth period. These processes
affect the ice evolution in both models. However, models with sub-grid-scale
ice thickness distribution, as in LIM3, have a less resistant ice pack to
convergence resulting in thicker ice than a single-category model, as in LIM2,
under similar conditions
In the SH in September, both LIM models present statistically significantly
increasing sea-ice extent anomaly trends, estimated with the linear
least-squares fit at the 5 % significance level, consistent with
observations. However, these modelled September trends are larger than the
observed trend (not shown here). The increase of the Antarctic sea-ice extent
has been explained by a range of mechanisms. Many studies attribute the
increase of sea-ice extent to changes in the atmospheric dynamics, mainly by
the increasing trend of the Southern Annular Mode, which in turn has
strengthened westerly winds around the Antarctic continent and deepened the
Amundsen Sea Low. Stronger westerlies effectively spread sea ice to north and
a deeper Amundsen Sea Low increases the sea-ice production in the Ross Sea
In the NH, the monthly mean LIM3 sea-ice volume, which is the domain integral
of the sea-ice thickness multiplied by sea-ice area per grid cell, varies
from the minimum of
In the SH, the LIM models' monthly mean sea-ice volume reaches its maximum in
October and then decreases to 1600
As with the sea-ice extent in the SH, the annual-mean LIM3 sea-ice volume has
a statistically significant positive trend of
There are important differences between P/GIOMAS reanalyses and NEMO-LIM models, explaining the systematic deviation of their sea-ice volume from each other. First, P/GIOMAS uses the National Centers for Environmental Prediction (NCEP)-based atmospheric forcing compared to the DFS one used in the NEMO-LIM simulations. Second, P/GIOMAS assimilates sea-ice concentration and sea surface temperature (SST) data, while the NEMO-LIM simulations do not. Finally, P/GIOMAS ocean and sea-ice models and the computational grid are different from NEMO-LIM ORCA1 configurations along with numerous physical parameterisations implemented in the models.
In general, closer similarities between the LIM2 and LIM3 sea-ice volume
distributions in the SH compared to the NH emphasise the importance of the
ocean model dictating the evolution of sea ice, while the level of
sophistication of sea-ice model has a smaller importance. This is, at least
partly, due to the divergent large-scale sea-ice motion where sea-ice
deformation remains small
The simulated March and September mean (2003–2012) sea-ice velocities are
shown in Fig. 3, together with the OSI SAF sea-ice drift product
In the NH, the LIM3 mean drift pattern in March consists of an offshore
motion over Siberian shelves (4–6
The two LIM models perform somewhat differently in terms of sea-ice speed,
LIM2 sea ice being generally faster, in particular in the Beaufort Gyre
(Fig. 3e). The 10-year-mean Arctic sea-ice velocity in March is 4.6
(4.8)
In the SH, the LIM models feature similar and realistic looking distribution
of the September ice drift (Fig. 3). They show realistic patterns of the
Weddell and Ross gyres, the westward coastal and eastward offshore
circumpolar currents. The observed OSI SAF drift generally compares well with
the modelled ones in terms of their large-scale velocity field patters
although the modelled speeds appear faster than observed, particularly along
the ice edge. That suggests that LIM models simulate the Antarctic sea-ice
drift reasonably well albeit somewhat too fast, which seems to be a consistent
ocean–ice model bias
As in the Arctic, the two LIM models have similar sea-ice velocity magnitude
within the central ice pack, but larger differences appear close to the ice
edge, where the LIM3 ice drift is
One important new feature in LIM3 is the prognostic sea-ice salinity compared
to the constant 4 ppm sea-ice salinity in LIM2
It is reasonable to assume that to some extent the more realistic LIM3 sea
ice might be due to the advanced salinity-dependent halo-thermodynamics and a
more realistic seasonal cycle of sea-ice salinity, and associated upper ocean
freshwater fluxes. In winter, newly formed LIM3 sea ice preserves a higher
salinity than in LIM2 (Fig. 4). In contrast, in summer, the remaining LIM3
sea ice has a 2–4 ppm lower salinity than LIM2 in the Arctic (not shown
here). However, during the Antarctic summer, the LIM3 sea-ice salinity stays
relatively high, except at the ice edge (not shown here). This is due to the
fact that even in summer air temperature remains at freezing over the coastal
Antarctic seas. As in
Based on rather descriptive analysis of differences between the LIM models, presented in the previous section, we have gained a relatively comprehensive understanding of how their global sea-ice distributions compare. In this section, we address what makes LIM3 sea ice different from LIM2 sea ice. Model grid and atmospheric forcing are identical, sea-ice differences can only arise from differences in sea-ice model physics parameterisations and these differences can be further amplified by ocean–ice feedback processes. To find out which parameterisations are of importance in producing LIM model differences, we performed and analysed some additional simulations.
The sea-ice salinity difference (in ppm) between LIM3 and LIM2
LIM3 differs from LIM2 in two important aspects; LIM3 has a multi-category sub-grid-scale sea-ice thickness distribution and multi-layer halo-thermodynamics scheme with prognostic non-constant sea-ice salinity profile. We tested the effect of these parameterisations by carrying out a LIM3 simulation with a single-category sea-ice thickness distribution having a virtual ice thickness distribution and a constant 4 ppm sea-ice salinity. Importantly, by setting the LIM3 sea-ice salinity constant, along with its two vertical ice layers and one snow layer, its thermodynamics scheme becomes similar to the LIM2 one. However, the initialisation procedure of LIM3 is different from the one used in LIM2, as explained in Sect. 2.3. We denote the LIM3 single-category simulation as LIM3SC.
In terms of NH sea-ice concentration and extent, LIM3SC is located between LIM3 and LIM2 (Figs. 1d, f and 2b). In the SH, LIM3SC annual-mean sea-ice extent follows closely to the one of LIM2 (Fig. 2b, d). However, the monthly sea-ice extent climatology of LIM3SC is distinctly closer to LIM3 and does not have the distorted shape of LIM2 monthly sea-ice extent climatology (Fig. 2a, c). Furthermore, the summer minimum and winter maximum of LIM3SC sea-ice extents clearly differ from LIM2 ones. This result suggests that the use of the single-category and constant salinity parameterisations brings LIM3 sea ice closer to LIM2 output, as expected, but apparent differences remain.
The LIM3SC sea-ice volume relative to two other LIM simulations is more different in the SH than in the NH (not shown here). In the Southern Hemisphere, the LIM3SC sea-ice volume immediately diverts from LIM2 and LIM3, although its annual-mean sea-ice extent remains rather close to LIM2 with a seasonal variability closer to the LIM3 one. It is possible that strong ocean–ice feedback processes in LIM3SC affect the melting and freezing rates during its first simulation year, and associated fluxes of salt and freshwater. This in turn modifies the upper ocean stratification and oceanic heat, which result in further differences in LIM3SC sea-ice volume that adjusts above the LIM2 level. The 20 cm thicker LIM3SC initial snow might have contributed to the differences in sea-ice thickness between LIM2 and LIM3SC by reducing the spring melt at the end of the first simulation year resulting in a relatively high sea-ice volume minimum in summer that persists through the simulation. After this, the high LIM3SC sea-ice volume seems to be in a balance with the upper ocean adjusted during the first years of the simulation.
In addition to sea-ice thermodynamics, the sea-ice salinity scheme modifies the ocean–ice salt and freshwater exchange, and upper ocean heat fluxes, which influence the evolution of sea ice. Compared to LIM2, the LIM3 multi-category sea-ice is saltier in winter due to its prognostic sea-ice salinity (Fig. 4). This implies a smaller ocean-to-ice salt flux during freezing and a more stably stratified ocean surface layer, particularly in the Southern Ocean and in the Barents Sea where LIM3–LIM2 winter salinity differences seem particularly large (Fig. 4). If the LIM3 prognostic salinity was of primary importance, we would expect a higher sea-ice volume in the LIM3 prognostic sea-ice salinity simulation than in the LIM3SC constant sea-ice salinity simulation due to smaller salt rejection rates and associated ocean convection in the Southern Ocean. As this is not the case, the importance of the sea-ice salinity scheme, in modifying the sea-ice evolution by affecting upper ocean heat fluxes, appears to be a secondary one compared to the effects of sea-ice salinity scheme on sea-ice thermodynamics and especially to the effects of sub-grid-scale ice thickness parameterisation.
Following a common practise when carrying out forced ocean–ice simulations, we applied a freshwater budget adjustment and SSS restoring in our simulations. The freshwater budget, evaporation minus precipitation minus river runoff, was adjusted from the previous year's annual-mean budget to zero at the beginning of each simulation year. Additionally, we added a SSS-dependent flux correction term on freshwater fluxes. This flux correction term practically damps the model top-level salinity towards the PHC3 top level salinity everywhere, also under sea ice, in LIM2, LIM3 and LIM3SC simulations. These treatments prevent an unrealistic drift of the sea surface height due to errors in the prescribed freshwater budget components.
In addition to the common practise, we completed two otherwise identical
integrations, one for LIM2 and one for LIM3, where we turned off the two
freshwater adjustment mechanisms to see what kind of effect they have on our
results. As expected, the ocean salinity drift became remarkable in the
non-adjusted simulations, being strongest in the top layer, increasing its
salinity by 0.4
Perhaps interestingly and in contrast to the North Atlantic, the Southern Ocean mixed layers were deeper without freshwater adjustments (LIM3FW and LIM2FW). Importantly, for the scope of this study, the effects of freshwater adjustments on sea-ice evolution were minuscule. LIM models produced almost identical sea ice independent of whether the freshwater adjustments were turned on or off. Mutual oceanic differences between the LIM3 and LIM2 simulations and between the LIM3FW and LIM2FW simulations did not change drastically, as we will show. However, as 54-year simulation is rather short from the ocean circulation perspective, it is possible that in longer simulations the differences between the simulations in terms of oceanic circulation increase to the point that they start to modify the sea-ice characteristics remarkably.
We now move on to explore differences in ocean properties between the two LIM versions. Figure 5 shows LIM3 Arctic SSS and SST averaged over the last decade of the simulation, 2003–2012, and LIM3 and LIM2 departures from PHC3, WOA13 and LIM2. The LIM models' SSS is closer to that of PHC3 than WOA13 due to their surface salinity restoring towards PHC3. LIM models' differences from PHC3 and WOA13 are much larger than their mutual differences, highlighting the fact that the LIM version has a secondary impact on the Northern Hemispheric ocean properties.
LIM3 surface salinity distribution realistically reflects the fact that the Arctic Ocean has a low salinity surface layer in contrast to the much saltier surface layer of the North Atlantic (Fig. 5a). Compared to PHC3 and WOA13, both NEMO-LIM versions are too fresh in the North Atlantic, Labrador Sea and Nordic seas, although the LIM3 Labrador Sea surface is saltier than that of LIM2, in particular close to the Greenland coast (Fig. 5b–f). In some parts of the Eurasian Basin, LIM3 is saltier than PHC3, which is partly associated with its negative sea-ice concentration bias and the lack of fresh melt water (Fig. 5b). Compared to WOA13, LIM3 SSS is much higher due to the SSS restoring toward PHC3, which indicates disagreements in terms of Arctic SSS due to the lack of observations and that the PHC3 SSS is higher than WOA13 SSS. As PHC3 was carefully constructed for the Arctic, it is plausible that its SSS climatology is closer to the truth. LIM ocean salinity biases mainly arise from the NEMO ocean model configuration, and the applied boundary conditions, such as the atmospheric forcing, river runoff and freshwater adjustments.
LIM3 has a fresher surface than LIM2 in many areas on the Siberian shelf, Barents Sea and Greenland Sea, associated with the smaller ice–ocean salt flux, thicker ice in winter and larger melt rates during spring (Fig. 5d). By contrast, in LIM2, fresher ice and reduced spring melt result in an increased ice–ocean salt flux and therefore higher SSS in those regions. However, along the eastern Greenland coast, thicker LIM2 sea ice is associated with higher melt rates and result in a fresher surface also in the Labrador Sea, to where the ice and freshwater drifts. These differences in surface salinity, associated with sea-ice differences, have potential implications for the strength of AMOC. Hence, although mutual hydrographic differences between the freshwater adjusted LIM simulations are small compared to their observational biases, they may potentially have an impact on the convective processes in the North Atlantic.
The experiments without freshwater adjustments, LIM2FW and LIM3FW, display larger SSS differences with the expanded or changed regions of statistically significant difference (not shown here). For example, the region north of Greenland in the Arctic Ocean, which is significantly saltier in LIM3 than in LIM2 (Fig. 5d), is not significantly saltier in LIM3FW than in LIM2FW. However, LIM3FW–LIM2FW SSS differences remain clearly smaller than the corresponding SSS differences with the observational climatologies. In general the geographical patterns of SSS differences remain rather similar, mainly the magnitudes of SSS differences change. In any case, this indicates that the freshwater flux corrections reduce the salinity differences originating from two different sea-ice models.
As with SSS, SST differences between the LIM models in the Arctic are small
compared to their differences from PHC3 and WOA13 (Fig. 5h–l). The LIM
models have a distinct cold bias in the North Atlantic and in the Greenland
Sea. These cold biases are related to the common atmospheric ERA-Interim-based
forcing, which is known to have a cold anomaly over the Fram
Strait–Svalbard region
Related to the smaller LIM3 sea-ice extent, LIM3 SST is warmer than LIM2 SST across most of the Arctic Ocean, along the eastern Greenland coast, in Baffin Bay and in the Labrador Sea. In contrast, SST in the Norwegian Sea and Barents Sea is lower in LIM3 than LIM2, associated with a lower salinity. In these regions, saltier LIM2 surface waters release less heat to the atmosphere before reaching the critical density and sinking down, which explains the warmer LIM2 SST. Additionally, LIM3 and LIM2 mixed layer depths show remarkable differences across the Nordic seas, as we will soon show. These SSS, SST and MLD differences signify the varying locations of effective upper ocean convection in the LIM simulations.
In the SH, LIM3/LIM2 and PHC3/WOA13 SSS differences are smaller than in the NH (Fig. 6b, c, e, f). In the regions covered by sea ice, LIM models' ocean surfaces are fresher than PHC3, except in some areas along the eastern Antarctic coast. These coastal differences are smaller between the LIM models and WOA13 than between the LIM models and PHC3 (Fig. 6b, c, e, f), which could be related to a larger number of better quality coastal Antarctic observations included in WOA13 and the fact that the simulation analysis period temporally better matches with WOA13 than PHC3. LIM3 SSS differences with LIM2 have smaller magnitudes than the LIM models' differences with the observational climatologies (Fig. 6d). Now, LIM3 Antarctic sea ice is less extensive, but thicker than LIM2 sea ice, on the average. Hence, off the Antarctic coast where the ice melts, more freshwater is released per area in LIM3 than in LIM2 resulting in a lower LIM3 SSS. Close to the Antarctic coast the LIM3 ocean surface is saltier than the LIM2 ocean surface. This is likely to be due to the greater winter ice formation rates in LIM3, i.e. associated larger salt flux from ice to ocean.
Processes related to the LIM3/LIM2 and PHC3/WOA13 SSS discrepancies in the Southern Ocean are likely to be associated with the other freshwater sources rather than the sea-ice-related freshwater exchange. Again, this is because the LIM and PHC3/WOA13 differences are of larger magnitudes than the differences between two LIM models. Most important external freshwater sources in the Southern Ocean are precipitation and melt water fluxes from the Antarctic continental ice sheet, and both of these sources are known to have large uncertainties. Given these observational freshwater and SSS uncertainties, we can expect significant differences between NEMO-LIM and PHC3/WOA13 ocean surface characteristics. However, as the NEMO-LIM simulations applied the SSS restoring, LIM and PHC3 sea surface salinities did not evolve very far apart compared to the simulations where the freshwater was not adjusted (not shown here). As for the Arctic, the LIM experiments without freshwater adjustments display larger mutual differences although their geographical patterns do not essentially change (not shown here).
As Fig. 5, but for the Southern Hemisphere.
The surfaces of the LIM simulations are colder than PHC3 and WOA13 around the eastern Antarctic and in the Ross Sea (Fig. 6h, i). As these differences are associated with fresher surface and lower than observed ice concentration, it is likely that the more stable LIM surface stratification decreases the upward oceanic heat flux and increases the surface heat loss to the atmosphere due to larger open-water areas. Consistent with this explanation, the somewhat higher LIM2 sea-ice concentration and SSS seem to result in a higher SST than the one of LIM3 around the eastern Antarctic (Fig. 6j).
In the Arctic Ocean, approximately at 250 m depth, lies the relatively warm
AIW layer that originates from the Atlantic Ocean (Figs. 7 and 8). AIW is
below the halocline and therefore saltier than waters above it (Fig. 8b). The
NEMO-LIM models simulate too fresh and cold waters at 250
As Fig. 5, but for the Arctic Intermediate Water (AIW) at
250
LIM3 AIW remains warmer than LIM2 (Fig. 8a), which indicates a somewhat
larger Atlantic warm water inflow into the Arctic Ocean (Fig. 7j). In the
Nordic seas and Barents Sea, LIM3 and LIM2 are warmer than PHC3 and WOA13 at
250
Vertical profiles of
Outside the polar regions, small temperature and salinity differences emerge
as the LIM simulations proceed. For example, LIM3 has a saltier Atlantic
Ocean than LIM2 at the layer from the surface to 1000 m depth (not shown
here). LIM salinity differences vary in time with maximum values up to
0.05
Oceanic convection, vertical heat transport and deep water formation are
intimately related to the MLD. We keep in mind that the observational MLD
uncertainties are particularly large in ice covered oceans, because of sparse
observations, and therefore limit our comparisons to the North Atlantic and
the Southern Ocean. In Fig. 9, the mean winter MLD is presented for LIM3
along with its difference from the observed climatologies of
In the North Atlantic, the observational climatologies agree generally rather
well, although their spatial coverage vary;
In the Norwegian and Barents seas, the LIM3 and LIM2 MLDs are larger than the observational MLD estimates indicating stronger oceanic convection. This is at least partly due to the cold, non-responsive prescribed winter atmosphere acting as an infinitive heat sink to the relatively warm ocean surface layer and associated with relatively cold LIM3 and LIM2 SSTs (Sect. 5.2). Elsewhere in the North Atlantic, in particular in the Labrador Sea and the Greenland Sea, the LIM3 and LIM2 MLDs are smaller than the observational estimates. This indicates weak convection and is associated with relatively fresh LIM3 and LIM2 ocean surfaces, as discussed in Sect. 5.1. As we will show in the next section, the NEMO-LIM AMOC is weak compared to observational estimates. It is likely that the weak deep water formation in NEMO-LIM, as manifested by low MLDs in the Labrador Sea and the Greenland Sea, is the main reason for its weak AMOC.
In the Southern Ocean, as in the North Atlantic, the observational mixed
layer depth climatologies show similar large-scale features when compared to
LIM3 and LIM2 MLDs (Fig. 9h, i, k, l). The LIM3 and LIM2 mixed layers are
usually deeper than the observed mixed layers outside the regions covered by
ice, particularly in the Pacific sector (Fig. 9h, i, k, l). There is one
larger region, north of 60
Although the LIM3 MLD generally appears to be relatively close to the LIM2 one in the Arctic, it is statistically significantly deeper in some Arctic Ocean locations, such as in the Canadian Basin, where the LIM3 ocean surface is somewhat warmer and fresher (Figs. 9d, 5d and j). In the Southern Ocean, LIM2–LIM3 MLD differences are quite small, with the statistically significantly regions visible in the marginal ice zone off the coast of eastern Antarctica and in along the Antarctic coast (Fig. 9j). As these coastal regions are the Antarctic Bottom Water formation regions, MLD differences indicate differences in the locations and rates of the deep water formation between the two LIM simulations. This variability in the deep water formation changes the deep water properties, which is manifested as slowly emerging differences in abyssal temperature and salinity, decades after the beginning of simulations, as discussed earlier in Sect. 5.6.
An important characteristic of a global ocean model is the strength and
extent of its AMOC. The observational-based estimates of its average strength
vary between 16.9
In addition to AMOC temporal evolution, our mean AMOC transport patterns in
depth–latitude space well resemble the NEMO ORCA1 ones of
Deviations between the LIM2 and LIM3 simulations in terms of their AMOC are
minor. There are, however subtle, statistically non-significant, differences,
as seen from Fig. 10a, where the LIM2 annual maximum AMOC within the
50–53
As expected, the AMOC differences between LIM3 and LIM2 become more apparent
when comparing simulations without freshwater adjustments, LIM3FW and LIM2FW.
Both LIM3FW and LIM2FW have a statistically significantly lower AMOC at the
5 % level than the ones with freshwater adjustments, LIM3 and LIM2
(Fig. 10a). As the LIM3 AMOC is on the average smaller than the LIM2 AMOC,
also the average LIM3FW AMOC is smaller (by 0.7
In addition to AMOC, we calculated time series of volume, heat and salinity
transports through a number of oceanic transects: the Australia–Antarctica
transect, the Bering Strait, the Denmark Strait, the Drake Passage, the
Florida Strait, the Gibraltar Strait and the Greenland–Norway transect at
60
A set of hindcast simulations (1958–2012) was performed with
the newest NEMO3.6 model using the global ORCA1 grid forced by the DFS5.2
atmospheric data. The primary objective was to diagnose the sensitivity of
the NEMO-LIM ocean–ice system to the representation of physics in the
sea-ice model. Results of such analysis have not been published for the
newest NEMO in the nominal 1
To assess the performance of two LIM versions, we compared their
climatological sea-ice distributions mutually and with observational
estimates. In terms of global sea ice, LIM3 compares clearly better with
available observations, while LIM2 deviates more, producing too much ice in
the Arctic, for example. The better representation of the ice-albedo feedback
makes LIM3 more capable in simulating the September minimum of extent than
LIM2, including the 2007 extremely low Arctic value. These sea-ice findings
are consisted with the ones of
We mostly restricted our analysis to the last decade of the 54-year simulations. By analysing a 10-year period means that the effect of multi-decadal variability is not taken into account. However, earlier and longer analysis periods would have been more impacted by the model spin-up from its initial state in 1958. Looking at the multi-decadal sea-ice extent and ocean transport time series (Figs. 2 and 10), the LIM2 and LIM3 simulations stay systematically apart. Accordingly, it is sensible to assume that the respective LIM3–LIM2 differences are not very sensitive to the multi-decadal variability, at least during the last few decades of the simulations. Furthermore, it is reasonable to assume that the upper ocean LIM3–LIM2 differences behave like the sea-ice and the oceanic transport ones.
It is worth noting that no specific NEMO-LIM tuning was done for our
experiments. It is likely that after some adjustments, such as controlled
changes in the sea-ice albedo or ice strength, the NEMO-LIM3 and NEMO-LIM2
sea-ice performance will to some extent increase
Our model evaluation focussed on the upper ocean properties and to some extent oceanic transports across major transects of the World Ocean, such as the Drake Passage, along with its meridional overturning circulation. This has not been systematically done before for NEMO3.6. In general, ocean hydrographic differences, such as temperature and salinity, between the two LIM versions are confined to the upper ocean and near the sea-ice zone. In terms of large-scale ocean circulation, differences between the two LIM versions remained small, but kept increasing over the decades, also in the extra-polar regions.
As a further sensitivity experiment, we repeated the NEMO-LIM3 hindcast simulation after setting its sea-ice distribution to the single-category mode. At large and as expected, this single-category configuration resulted in a shift of LIM3 sea-ice distribution towards the LIM2 one, but encouragingly the LIM3 single-category sea ice remained clearly more realistic than the LIM2 one. This result indicates that one option for modellers, who are considering in upgrading from LIM2 to LIM3, is to start using the single-category LIM3 as an intermediate step. Based on these findings, we conclude that NEMO3.6 is ready as a stand-alone ocean–ice model and as a component of coupled atmosphere–ocean models.
The NEMO version 3.6 version incorporates LIM2 and LIM3.6 sea-ice models, and
can be downloaded from the NEMO website (
We acknowledge the creators of low-resolution sea-ice drift product of the
EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF,