Atmospheric reanalyses are valuable datasets for driving ocean–sea ice general circulation models and for proposing multidecadal reconstructions of the ocean–sea ice system in polar regions. However, these reanalyses exhibit biases in these regions. It was previously found that the representation of Arctic and Antarctic sea ice in models participating in the Ocean Model Intercomparison Project Phase 2 (OMIP2, using the updated Japanese 55-year atmospheric reanalysis, JRA55-do) was significantly more realistic than in OMIP1 (forced by the atmospheric state from the Coordinated Ocean-ice Reference Experiments version 2, CORE-II). To understand why, we study the sea ice concentration budget and its relations to surface heat and momentum fluxes as well as the connections between the simulated ice drift and the ice concentration, the ice thickness and the wind stress in a subset of three models (CMCC-CM2-SR5, MRI-ESM2-0 and NorESM2-LM). These three models are representative of the ensemble and are the only ones to provide the surface fluxes and the tendencies of ice concentrations attributed to dynamic and thermodynamic processes required for the ice concentration budget analysis. The sea ice simulations of two other models (EC-Earth3 and MIROC6) forced by both CORE-II and JRA55-do reanalysis are also included in the analysis. It is found that negative summer biases in high-ice-concentration regions and positive biases in the Canadian Arctic Archipelago (CAA) and central Weddell Sea (CWS) regions are reduced from OMIP1 to OMIP2 due to surface heat flux changes. Net shortwave radiation fluxes provide key improvements in the Arctic interior, CAA and CWS regions. There is also an influence of improved surface wind stress in OMIP2 giving better winter Antarctic ice concentration and the Arctic ice drift magnitude simulations near the ice edge. The ice velocity direction simulations in the Beaufort Gyre and the Pacific and Atlantic sectors of the Southern Ocean in OMIP2 are also improved owing to surface wind stress changes. This study provides clues on how improved atmospheric reanalysis products influence sea ice simulations. Our findings suggest that attention should be paid to the radiation fluxes and winds in atmospheric reanalyses in polar regions.
Sea ice is an important component of the polar climate system. At high latitudes, the presence of sea ice affects the exchanges of heat, momentum and freshwater fluxes between the atmosphere and the ocean. Sea ice has experienced dramatic changes during recent decades, especially in the Arctic, where the total sea ice extent dramatically decreased over the satellite-observing period (Comiso et al., 2008; Stroeve and Notz, 2018). In the Antarctic, the total sea ice extent increased slightly, but statistically significantly, to a record high in 2014 and decreased dramatically to the lowest value in 2017 over the satellite record (Parkinson, 2019; Fogt et al., 2022). A record low sea ice extent was set in 2022 (Raphael and Handcock, 2022). Sea ice variability can drive changes in the atmospheric energy budget and circulation (Krikken and Hazeleger, 2015; Smith et al., 2017, 2022) as well as surface fluxes into the ocean and ocean circulation (Haumann et al., 2016; Sévellec et al., 2017; Meneghello et al., 2018). A good simulation of sea ice is crucial for improving model predictions and climate change projections. However, limitations still exist in both fully coupled climate models and ocean–sea ice models. For the Arctic, the observed decline in sea ice cover lies within the spread of modeled trends, but the multimodel mean trend is underestimated in the third, fifth and sixth phases of the Coupled Model Intercomparison Project (CMIP, Stroeve et al., 2007; Massonnet et al., 2012; Rosenblum and Eisenman, 2017; Notz and SIMIP Community, 2020). The observed accelerated ice drift speed is not captured in CMIP3 models (Rampal et al., 2011), while the accelerated ice drift speed is produced in winter but not in summer in CMIP5 models (Tandon et al., 2018). Large ice edge and thickness errors in Arctic subregions are identified from the spatial distribution of sea ice in CMIP6 models (Stroeve et al., 2014; Watts et al., 2021). For the Antarctic, the CMIP5 and CMIP6 models fail to capture the slight increase in observed ice extent from 1979 to 2015, and they do not properly simulate the mean state and interannual variability of the ice cover (Mahlstein et al., 2013; Turner et al., 2013; Zunz et al., 2013; Shu et al., 2015, 2020; Roach et al., 2020). Large biases are also noticed in simulations conducted with ocean–sea ice models driven by atmospheric reanalysis data, in particular on the Antarctic sea ice extent variability and the ice thickness and motion in both hemispheres (e.g., Massonnet et al., 2011; Lecomte et al., 2016; Chevallier et al., 2017). By performing sensitivity experiments with these ocean–sea ice models, one can gain some insight into the origins of those biases. The focus of the present study is on quantifying and understanding how the sea ice simulation can be improved by changing atmospheric forcing fields in ocean–sea ice models.
Atmospheric reanalyses are particularly valuable in polar regions where in situ observations are scarce. However, these reanalyses have their limitations and biases (e.g., Lindsay et al., 2014; Bromwich et al., 2016; Barthélemy et al., 2018; Lin et al., 2018). Previous studies have shown that differences in the atmospheric forcing fields can affect the ocean–sea ice model simulations of the Arctic monthly mean sea ice thickness and total sea ice volume (e.g., Hunke and Holland, 2007; Lindsay et al., 2014; Sterlin et al., 2021), the Arctic and Antarctic sea ice concentration in the marginal ice zones (Chaudhuri et al., 2016) and the Antarctic sea ice extent, motion and thickness (Barthélemy et al., 2018). Wu et al. (2020) also showed the positive impacts of high-frequency (hourly to daily) atmospheric fluctuations on the Antarctic sea ice simulation, which implies that driving an ocean–sea ice model with a reanalysis that is developed at enhanced temporal and spatial resolution can help capture the small-scale atmospheric processes and eventually improve the representation of sea ice.
The CMIP6 Ocean Model Intercomparison Project (OMIP, Griffies et al., 2016)
provides global ocean–sea ice model simulations in two streams of model experiments: OMIP1, forced by the Coordinated Ocean-ice Reference
Experiments, version 2 interannual dataset (CORE-II; Large and Yeager,
2009), and OMIP2, forced by the updated Japanese 55-year atmospheric
reanalysis (JRA55-do; Tsujino et al., 2018). The design of the CMIP6 OMIP
simulations has been coordinated by the World Climate Research Programme
(WCRP) Climate Variability and Predictability (CLIVAR) Working Group on
Ocean Model Development Panel (OMDP), and ongoing research collaboration is
done through the OMDP to further develop OMIP2 (Griffies et al., 2016). The
same configuration is used under two different atmospheric forcing datasets
as mentioned in Tsujino et al. (2020). The JRA55-do atmospheric forcing is
relatively new with major improvements, e.g., increased temporal frequency
(3 h) and horizontal resolution (0.5625
The spatial variability of sea ice concentration and its links with the atmospheric circulation vary with season. The change in the position and strength of the cyclonic or anticyclonic circulation center over the sea ice can affect the sea ice motion and freezing/melting processes (Rigor et al., 2002; Raphael and Hobbs, 2014; Ding et al., 2017). Strong winter wind-driven ice exports in the Eurasian coastal region occur during high North Atlantic Oscillation (NAO) index years, which can have contributed to the reduction in summer Arctic sea ice extent observed during the 1980s and 1990s (Hu et al., 2002). In the Antarctic, the decreases in sea ice concentration generally occur in regions of poleward atmospheric flow, and the increases in sea ice concentration occur in regions of equatorward flow (Renwick et al., 2012). During the seasonal sea ice advance and retreat periods, the spatial ice concentration variability is associated with different atmospheric circulation patterns, and both thermal advection and dynamical forcing are important (Raphael and Hobbs, 2014). The thermodynamic and dynamic processes that contribute to the Antarctic sea ice concentration seasonal evolution are discussed in Barthélemy et al. (2018). These authors conducted three sensitivity experiments with different atmospheric forcing fields using the NEMO-LIM3 ocean–sea ice model. They found that differences in the thermodynamic component of the forcing were mostly responsible for the differences in ice concentration simulated by the model experiments during the melting season, while during the ice-expansion period, both thermodynamic and dynamic components were important. The relationships between spatially averaged observed sea ice drift speed in the central Arctic and ice concentration, ice thickness and wind stress were investigated by Olason and Notz (2014). According to their results, on the seasonal timescales, ice drift speed changes in the central Arctic are primarily attributable to the changes in the ice concentration from June to November and changes in the ice thickness when the ice concentration is high, i.e., from December to March. The relationships between Arctic sea ice drift speed, concentration and thickness are relatively well captured by the NEMO-LIM3 model (Docquier et al., 2017) and the coupled GFDL-ESM2G model (Eyring et al., 2020), with higher drift speed associated with lower concentration and thickness. In the Antarctic, away from the coastline, the mean ice drift is significantly correlated with the wind forcing in the Pacific and Atlantic sectors, with the spatially averaged vector correlation coefficient larger than 0.7 (Kimura, 2004; Holland and Kwok, 2012).
This paper complements a companion publication (Lin et al., 2021) that documents a new Sea Ice Evaluation Tool (SITool v1.0) and applies this tool to assess the sea ice simulations in CMIP6 OMIP models. In that study, the improved Arctic and Antarctic ice concentration and drift simulations in CMIP6 OMIP2 compared to OMIP1 were highlighted from performance metrics and diagnostic spatial maps. In the present study, the thermodynamic and dynamic processes that contribute to the improved ice concentration simulation in OMIP2 compared to OMIP1 are assessed. The related surface sensible and latent heat fluxes, net shortwave and longwave radiation fluxes and surface wind stress on sea ice are investigated to trace the origin of simulated sea ice differences back to the forcing datasets. Meanwhile, the sensitivity of ice drift simulation to the changes in ice concentration, ice thickness and surface wind stress is examined to help understand the factors responsible for improving the ice drift simulation. This paper is organized as follows. The CMIP6 OMIP models, observational references and atmospheric reanalysis data are described in Sect. 2. The sea ice concentration simulations and the effects of the thermodynamic and dynamic components of the atmospheric forcing are presented in Sect. 3.1. The ice drift simulation and the connections to ice concentration, ice thickness and wind stress are discussed in Sect. 3.2. Finally, in Sect. 4, conclusions and a discussion are provided. Appendix A presents some extra sea ice diagnostics.
Five CMIP6 OMIP models have been forced by both CORE-II (OMIP1) and JRA55-do (OMIP2) reanalysis data so far, and they are marked as
The details of the five CMIP6 OMIP sea ice models evaluated in the study. Some information can be found at the following link:
Two sets of observational references for the sea ice concentration, thickness and ice drift are used for comparison. The two sea ice concentration products are derived from the passive microwave data by using the NASA Team algorithm (NSIDC-0051, Cavalieri et al., 1996) and the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Ocean and Sea Ice Satellite Application Facility algorithm (OSI-450, Lavergne et al., 2019), respectively. The first observed ice thickness data are derived from the measurements of the ESA's Environmental Satellite (Envisat) radar altimeter (Guerreiro et al., 2017), and the second one is derived from measurements of NASA's Ice, Cloud, and land Elevation Satellite (ICESat) Geoscience Laser Altimeter System (GLAS) (Yi and Zwally, 2009; Kurtz and Markus, 2012). The first observed ice drift product is processed by the NSIDC and enhanced by the Integrated Climate Data Center (ICDC-NSIDCv4.1, Tschudi et al., 2019), and the second ice drift data (KIMURA) are processed by Kimura et al. (2013). More information on the observational references can be found in Sect. 2.2 of Lin et al. (2021). The evaluation period is chosen according to available historical model outputs and observations and is consistent with the analysis in Lin et al. (2021). The ice concentrations, concentration tendencies and their relations to surface heat fluxes and wind stress are evaluated from 1980 to 2007, while the ice drift and its links to the ice concentration, ice thickness and wind stress are assessed from 2003 to 2007.
The monthly mean surface air temperature, specific humidity, downward shortwave and longwave radiation fluxes and wind speed in the CORE-II and JRA55-do reanalysis datasets from 1980 to 2007 are used to evaluate the differences between two forcing datasets.
The 1980–2007 September and February mean spatial distributions of the Arctic and Antarctic sea ice concentrations from the NorESM2-LM simulations and observational reference OSI-450 compared to the observational reference NSIDC-0051 are shown in Fig. 1, and figures from CMCC-SR5-CM2, MRI-ESM2-0, EC-Earth3 and MIROC6 are displayed in Figs. A1 and A2. The model biases are much larger than observational differences (Fig. 1, second column). Olason and Notz (2014) suggested that, for concentrations above 80 %, variations in sea ice state variables (concentration and thickness) greatly affect the ice drift speed. To study the drivers of the ice concentration and drift speed changes, we divided the regions into two parts for each month, with ice concentration larger (interior) and smaller (exterior) than 80 % in the NSIDC-0051 observational reference. The black lines in Fig. 1 exhibit September and February contours of 80 % concentration in the NSIDC-0051 data. Spatial averages of the 1980–2007 September and February mean sea ice concentration biases are given in Tables 2 and 3 for the Arctic and Antarctic, respectively. The spatial averages over the interior and exterior regions are calculated with data closer than 75 km to the coast removed to reduce the spatial noise.
Spatial averages of the 1980–2007 September mean Arctic sea ice concentration (SIC) biases (vs. NSIDC-0051, Figs. 1, A1 and A2), ice concentration tendencies through thermodynamic and dynamic processes (Figs. 2, 3, A3 and A4) and surface heat fluxes and surface stress on sea ice (Figs. 4 and A5) over March–August. The results derived from five model groups under the OMIP1 (/C) and OMIP2 (/J) runs are listed. The spatial averages over the interior region in the Arctic and the Canadian Arctic Archipelago (CAA) in summer are given. The improvements in ice concentration simulations and the contributions from the thermodynamic process and the surface heat flux to sea ice are marked in bold for the interior region and the CAA region.
Spatial averages of the 1980–2007 February and September
mean Antarctic SIC biases (vs. NSIDC-0051, Figs. 1, A1 and A2), the ice concentration tendencies through thermodynamic and dynamic processes (Figs. 2, 3, A3 and A4) and surface heat fluxes and surface stress on sea ice (Figs. 4 and A5) over October–January and March–August. The results derived
from five model groups under the OMIP1 (/C) and OMIP2 (/J) runs are listed. The spatial averages over the interior region in the Antarctic (52 to
60
1980–2007 September and February mean Arctic
By applying the mean ice concentration difference metric developed in Lin et al. (2021), one finds that the Arctic mean ice concentration biases in OMIP1
simulations from 1980 to 2007 are reduced in OMIP2. The improvements are
primarily in the boreal summer in the interior region and the Canadian
Arctic Archipelago (CAA) region as shown in Figs. 1, A1 and A2. In
September, the ice concentration is primarily underestimated in the interior
region and overestimated in the CAA region in OMIP1 simulations, and those biases are reduced in OMIP2 runs. The spatial mean ice concentration biases
in the interior region are reduced from
The Antarctic mean ice concentration biases in OMIP1 simulations from 1980
to 2007 are also diminished. The improvements are mainly over the coastal
regions of the western Weddell Sea and the Amundsen Sea in the austral
summer and over exterior regions from 70 to 180
To understand the differences in the simulated sea ice concentration noted in Figs. 1, A1 and A2, we analyze the thermodynamic and dynamic processes contributing to the concentration tendencies during the melt and growth seasons under different atmospheric forcings (Figs. 2, 3, A3 and A4). The idea is close to the sea ice concentration budget proposed in Holland and Kwok (2012) and applied in Uotila et al. (2014), Lecomte et al. (2016) and Barthélemy et al. (2018). The contributing thermodynamic processes to the concentration tendencies are freezing or melting, whereas the relevant dynamic processes are ice advection, divergence/convergence and mechanical redistribution (rafting/ridging). The tendencies of ice concentration due to dynamic and thermodynamic processes are available as standard SIMIP diagnostics in the three models (Notz et al., 2016). Spatial averages of the Arctic and Antarctic ice concentration tendencies due to thermodynamic and dynamic processes are listed in Tables 2 and 3, respectively.
1980–2007 March–August
Same as Fig. 2 but for the 1980–2007 October–January
Compared to OMIP1 runs, changes in thermodynamic processes in the Arctic
Ocean and the CAA region contribute to the ice concentration changes in
OMIP2 runs during March to August (Figs. 2 and A3). The differences between
OMIP1 and OMIP2 simulations in the contributions from dynamic processes are small. As shown in Table 2, the spatial mean ice concentration tendencies
due to thermodynamic processes (10
The major Arctic winter ice concentration biases are located in the exterior regions in OMIP1 simulations, with a minor reduction in OMIP2 runs (Figs. 1h to j and A1g to l). The winter ice concentration simulation in exterior regions is complicated because both dynamic and thermodynamic processes are important. The contributions from thermodynamic and dynamic processes are anticorrelated in these regions, with the dynamic processes increasing ice concentration through the expansion of sea ice and the thermodynamic processes contributing to ice melt. During October to January, the increased Arctic ice concentration is dominated by dynamic processes in exterior regions in OMIP1 simulations (Figs. 2 and A3). Compared to OMIP1 runs, these dynamic processes in OMIP2 runs contribute to the decreased ice concentration in exterior regions in the east of Greenland, while thermodynamic processes contribute to the increased ice concentration. This contributes to minor winter ice concentration differences between OMIP1 and OMIP2 simulations.
During October to January, the thermodynamic processes contribute to the
decreased ice concentration in the Southern Ocean, except in some coastal
regions, and the dynamic processes contribute to the decreased ice
concentration in the inner region in the three OMIP1 runs (Figs. 3 and A4).
Compared to the OMIP1 runs, the thermodynamic processes dominate the increased ice concentration in the coastal regions of the western Weddell Sea and
Amundsen Sea in the three OMIP2 simulations and the decreased ice concentration in the CWS in CMCC-SR5-CM2/J and MRI-ESM2-0/J. As shown in
Table 3, the spatial mean ice concentration tendency due to thermodynamic
processes (10
During March to August, the thermodynamic processes contribute to the
increased Antarctic ice concentration, except for some exterior regions, and
the dynamic processes contribute to the increased ice concentration
primarily in the exterior region (Figs. 3 and A4). Compared to OMIP1 runs,
the dynamic processes dominate the decreased ice concentration in exterior
regions in the three OMIP2 simulations. As shown in Table 3, the spatial
mean ice concentration tendencies related to dynamic processes (10
In general, by changing the atmospheric forcing from CORE-II to JRA55-do,
the improvements in the simulation of summer Arctic and Antarctic sea ice
concentrations within the pack are driven by differences in the thermodynamic tendency terms, while the improvements in Antarctic winter concentration
simulation in the exterior region from 70 to 180
To trace the origin of the differences in thermodynamic and dynamic tendency
terms noted in the previous section, the surface heat and momentum fluxes
available from the standard OMIP1 and OMIP2 model outputs are compared. The
sign convention for flux in this study is that a downward flux towards the
surface is positive. The net surface heat flux is downward (positive) in the
Arctic during March to August and in the Antarctic during October to January
in OMIP1 runs (Figs. 4 and A5). Compared to OMIP1 simulations, the net
surface heat fluxes in OMIP2 simulations are smaller in the central Arctic
Ocean and over the coastal regions of the western Weddell Sea and larger in the CAA and CWS regions. As shown in Tables 2 and 3, the spatial mean net
surface heat flux from OMIP1 to OMIP2 simulations decreased in the Arctic interior region (from 27.4 to 14 W m
1980–2007 March–August and October–January mean Arctic
To study which part dominates the surface heat flux changes from OMIP1 to
OMIP2 simulations, the surface sensible and latent heat fluxes and the net
shortwave and longwave radiation fluxes are computed (Figs. 5 and A6).
Compared to OMIP1 simulations, the net shortwave radiation flux and latent
heat flux in OMIP2 are smaller in the central Arctic Ocean and the coastal
region of the western Weddell Sea, the net shortwave radiation flux is
larger in the CAA and CWS regions and the sensible heat flux is larger in
the CAA region. As shown in Table 4, the decreased net shortwave radiation
flux in OMIP2 simulations (
Spatial averages of the 1980–2007 mean Arctic
(March–August) and Antarctic (October–January) net surface heat fluxes (Figs. 4 and A5), sensible and latent heat fluxes, net shortwave and longwave radiation fluxes (Figs. 5 and A6), downward and upward shortwave fluxes
(Fig. A7) and downward and upward longwave fluxes over the interior region in the Arctic and Antarctic (52 to 60
1980–2007 March–August mean Arctic
The changes in the shortwave radiation flux are crucial for the summer ice
concentration changes in the OMIP2 simulations in the Arctic interior region
and the CAA and CWS regions. The downward and upward shortwave radiation
fluxes in NorESM2-LM and CMCC-SR5-CM2 (Fig. A7) and spatial averages (Table 4) are displayed. The decreased downward shortwave radiation flux in
OMIP2 simulations (
Compared to other regions, the surface stress on ice along the eastern coasts of Greenland, Svalbard and Baffin Island, near the Bering Strait from 60 to
70
The improvement in the winter Arctic ice concentration in the exterior region is not as clear. Compared to OMIP1 simulations, the surface wind stress in OMIP2 simulations is smaller along the eastern coasts of Greenland, Svalbard and Baffin Island (Figs. 4g and h, A5i to l). This is consistent with the decrease in ice concentration in the exterior region in OMIP2 simulations due to the dynamic processes away from the eastern coast (Figs. 2l, A3u and A3x). However, the thermodynamic processes in OMIP2 simulations contribute to the increase in ice concentration, which is close to the decrease due to the dynamic processes in these regions (Figs. 2k, A3t and A3w). The different contributions of the thermodynamic processes to the winter ice concentration tendency in the exterior region between OMIP1 and OMIP2 simulations are primarily related to the dynamic processes, while the surface heat flux difference on the sea ice is small.
To identify how the differences in the atmospheric forcings are transferred
to the model results, the 1980–2007 mean surface air temperature, specific
humidity, downward shortwave and longwave radiation fluxes during melting
months and wind speed during freezing months in CORE-II and JRA55-do are shown in Fig. 6. The selection to show these variables during
melting/freezing months is because in general the ice concentration
simulations are improved from OMIP1 to OMIP2 in summer due to surface heat
flux changes and in winter due to wind stress changes. Compared to CORE-II,
the downward shortwave radiation flux and specific humidity in the central
Arctic Ocean (Fig. 6g and h) and specific humidity in the coastal region of the western Weddell Sea (Fig. 6q) in JRA5-do are smaller, the downward
shortwave radiation flux in the CAA and CWS regions (Fig. 6h and r) and the
air temperature in the CAA region are larger (Fig. 6f) and the surface wind speed on Antarctic sea ice in the inner part of the exterior region from
70 to 180
1980–2007 March–August mean Arctic and October–January mean Antarctic surface air temperature (first column) and specific humidity (second column), downward shortwave (third column) and longwave radiation (fourth column) fluxes, and October–January mean Arctic and March–August mean Antarctic surface wind speeds (fifth column). The first and third rows correspond to CORE-II, and the second and fourth rows are differences between JRA55-do and CORE-II.
The Arctic and Antarctic ice drift magnitude and direction simulations are
improved from OMIP1 to OMIP2 (Lin et al., 2021). To understand the factors
responsible for this feature, the sensitivity of the ice drift magnitude
simulation to the changes in ice concentration, ice thickness and surface
wind stress is investigated. The mean kinetic energy (MKE) is calculated to
measure the ice drift magnitude,
2003–2007 monthly mean and spatially averaged Arctic ice kinetic energy
(MKE)
Same as Fig. 7 but for the Antarctic. The Envisat ice thickness data are provided from May to October.
In the Arctic interior region, the ice-motion MKE in NorESM2-LM/C (Fig. 7a,
solid orange) is larger than those in KIMURA (solid blue) and ICDC-NSIDCv4.1 data (solid purple), and this positive bias is slightly reduced in
NorESM2-LM/J (solid green) from January to April and September. The largest
improvement in the interior ice-motion MKE occurs in September (Fig. 7b,
solid black). It is mostly caused by the increased ice concentration and
thickness in NorESM2-LM/J (Fig. 7c to f), while the changes in the surface
wind stress are very small (Fig. 7g to h). The September ice concentration in NorESM2-LM/J is close to NSIDC-0051 and OSI-450 data (Fig. 7c). The observational Arctic ice thickness data in September are not
available for comparison (Fig. 7e). The ice thickness observations during
2003–2007 are restricted to a few months per year in both the Envisat and ICESat datasets. The Envisat ice thickness data are provided from November to April for the Arctic with coverage up to 81.5
In the Arctic exterior region, the ice-motion MKE in the five OMIP1
simulations (Figs. 7a, A8a to d, dashed orange) is much larger than those in KIMURA (dashed blue) and ICDC-NSIDCv4.1 data (dashed purple), and the positive biases in OMIP1 simulations are largely reduced in OMIP2
simulations (dashed green) from November to April. The decreased Arctic
ice-motion MKE in OMIP2 simulations in the exterior region from November to
April is mainly induced by the decreased surface wind stress (Figs. 7g and
h, A8m and n), while the changes in ice concentration and thickness are very small (Figs. 7c to f and A8e to l). There is no consistent improvement
in the representation of sea ice concentration and thickness during November to April from OMIP1 to OMIP2 simulations. We average modeled ice thickness
limited up to 81.5
In the Antarctic interior region, the ice-motion MKE in NorESM2-LM/C (Fig. 8a, solid orange) is larger than those in KIMURA (solid blue) and ICDC-NSIDCv4.1 data (solid purple), and this positive bias is reduced in April in NorESM2-LM/J (solid green, smaller than
The significant vector correlation coefficients during 2003–2007 at a
level of 99 % between modeled ice drift (NorESM2-LM/C) and two
observational data (KIMURA and ICDC-NSIDCv4.1), respectively, and between
NorESM2-LM/C-modeled ice drift and surface wind stress in the Arctic
In the Antarctic exterior region, the ice-motion MKE in NorESM2-LM/C (Fig. 8a, dashed orange) is larger than those in KIMURA (dashed blue) and ICDC-NSIDCv4.1 data (dashed purple), and this positive bias is reduced in
NorESM2-LM/J from July to September (dashed green, smaller than
We finally aim at determining to what extent the change in atmospheric forcing may lead to an improvement in the simulated ice drift direction (independently of the improvements in sea ice drift magnitude noted in the previous section). To that end, the vector correlations between simulated and observed ice drift fields (KIMURA and ICDC-NSIDCv4.1 data) are diagnosed, as done in Lin et al. (2021). In general, the vector correlation coefficients between the modeled ice drift and observational data during 2003–2007 are larger in NorESM2-LM/J than those in NorESM2-LM/C in the Arctic (Fig. 9d and e) and Antarctic (Fig. 9j and k).
The links with the surface wind stress are assessed. The vector correlation coefficients between modeled ice drift and surface wind stress are much larger in NorESM2-LM/J than those in NorESM2-LM/C in the Beaufort Gyre area (Fig. 9f) and the Pacific and Atlantic sectors of the Southern Ocean (Fig. 9l). Those regions correspond to large improvements in the ice vector direction simulation in NorESM2-LM/J (Fig. 9d, e, j and k). This suggests that the improved ice vector direction simulation is related to the changed surface wind stress in NorESM2-LM/J. These improvements can also be found in CMCC-CM2-SR5/J, MRI-ESM2-0/J, EC-Earth3/J and MIROC6/J in both hemispheres (Fig. A10), but the improvements in MRI-ESM2-0/J, EC-Earth3/J and MIROC6/J are smaller than those in NorESM2-LM/J and CMCC-CM2-SR5/J in the Arctic.
OMIP provides useful datasets for reconstructing sea ice evolution over the past decades. Lin et al. (2021) have shown that the accuracy of the reconstruction depends on the atmospheric forcing used. This paper attempts to explain why this is so by conducting surface momentum and heat flux analyses. The two atmospheric reanalysis products are different in both dynamical and thermodynamical components for the Arctic and Antarctic, such as the air temperature and winds, which contribute to heat flux and momentum flux differences in the ocean–sea ice models. We studied the dynamic and thermodynamic processes contributing to the ice concentration tendencies and their links with surface heat and momentum fluxes as well as the connections between the simulated ice drift and the ice concentration, ice thickness and wind stress.
In general, the sea ice concentration and ice drift magnitude and direction
simulations are improved from OMIP1 to OMIP2, and improvements in the Arctic
are larger than those in the Antarctic. The net surface heat fluxes are decreased in the interior region, with ice concentration above 80 %, and
increased in the CAA and CWS regions during March to August (Arctic) and
October to January (Antarctic) in OMIP2 compared to OMIP1 simulations. This
can explain the improved OMIP2 ice concentration simulations in the summer,
pointing to the important role played by the thermodynamic processes during
the ice-melting season. The changed net shortwave radiation fluxes from OMIP1 to OMIP2 simulations are crucial for improving the OMIP2 summer ice concentration simulations in the Arctic interior, CAA and CWS regions.
The decreased surface wind stress in the inner part of the exterior region
during March to August in OMIP2 compared to OMIP1 contributes to the
improved (decreased) Antarctic September OMIP2 ice concentration simulation
in the exterior region from 70 to 180
From our analysis, the differences in the atmospheric forcing are transferred to the modeled surface fluxes and contribute to the improved ice concentration simulation. However, tuning is also a key aspect in climate models that can explain differences in performance. It is possible that some groups could have tuned for OMIP2 and then used the same setup for OMIP1, so part of the improvement with OMIP2 could be due to this experimental setup choice. While this paper reiterates the importance of the atmospheric forcing for the representation of the sea ice state, it is expected, based on the conclusions, that errors in the atmospheric forcing will also affect the ocean through modified heat, freshwater and momentum fluxes between the ice and the ocean. These errors can thus eventually affect the representation of ocean temperature, salinity and currents.
In this Appendix, extra sea ice diagnostics from CMCC-CM2-SR5, MRI-ESM2-0, EC-Earth3 and MIROC6 are given to help support the conclusions derived from NorESM2-LM. The ice concentration simulations from four models are provided in Figs. A1 and A2. The effects of the thermodynamic and dynamic components of the atmospheric forcing in CMCC-CM2-SR5 and MRI-ESM2-0 are presented in Figs. A3 to A6. The surface heat flux on sea ice is not provided for MRI-ESM2-0, and the corresponding figures are not included in Figs. A5 and A6. The downward and upward shortwave radiation fluxes in NorESM2-LM and CMCC-CM2-SR5 are added in Fig. A7. The ice drift simulation and the relationship with ice concentration, ice thickness and wind stress in CMCC-CM2-SR5, MRI-ESM2-0, EC-Earth3 and MIROC6 are provided in Figs. A8 to A10.
1980–2007 September and February mean Arctic
1980–2007 September and February mean Arctic
1980–2007 March–August
Same as Fig. A3 but for the 1980–2007 October–January
1980–2007 March–August and October–January mean Arctic
1980–2007 March–August mean Arctic
1980–2007 March–August mean Arctic
2003–2007 monthly mean and spatially averaged Arctic ice kinetic energy
(MKE)
Same as Fig. A8 but for the Antarctic.
Differences in significant vector correlation coefficients during 2003–2007 at a level of 99 % between model/J and model/C in the Arctic
CMIP6 OMIP data are freely available from the Earth
System Grid Federation. The download links to the observational references
used in this paper are listed in Table 3 of Lin et al. (2021). The Coordinated Ocean-ice Reference Experiments, version 2 interannual data (CORE-II) are available at
XL and FM developed the concept of the paper. XL performed the analysis and led the writing of the paper. All the authors contributed to the discussion of the study and the editing of the manuscript.
The contact author has declared that none of the authors has any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We are grateful to Noriaki Kimura and Sara Fleury for providing and introducing the ice drift and ice thickness datasets, respectively. We thank the sea ice observational groups and the climate modeling groups for producing and making available their output. Xia Lin is a F.R.S.-FNRS scientific collaborator. François Massonnet is a F.R.S.-FNRS research fellow.
This research has been supported by the Copernicus Marine Environment Monitoring Service (CMEMS) SI3 project. The CMEMS is implemented by Mercator Ocean International in the framework of a delegation agreement with the European Union. Xia Lin also received support from the National Natural Science Foundation of China (grant nos. 41941007, 41906190 and 41876220) and the Innovation Group Project of the Southern Marine Science and Engineering Guangdong Laboratory (Zhuhai) (grant no. 311021008).
This paper was edited by Jari Haapala and reviewed by two anonymous referees.