Sea Ice Sea Ice CMIP6
CMIP6 Multi-Model Mean Context
Comparison with CMIP6 ensemble mean from 7 members.
Contributing models: ACCESS-ESM1-5, CNRM-CM6-1, CNRM-ESM2-1, EC-Earth3, INM-CM5-0, MPI-ESM1-2-LR, MRI-ESM2-0
Synthesis
Related diagnostics
Sea Ice Area March & September Trends
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure evaluates Northern and Southern Hemisphere sea ice area trends (March and September) for 1980–2014, revealing severe biases in the high-resolution models compared to OSI-SAF observations and the CMIP6 multi-model mean.
Key Findings
- IFS-NEMO-ER and IFS-FESOM2-SR exhibit extreme positive biases in the Northern Hemisphere, overestimating sea ice area by >5 million km² in both winter and summer, and failing to capture the observed decline in September minimum.
- ICON-ESM-ER displays an exaggerated seasonal cycle in the Arctic: while winter maxima are closer to observations (though still high), September minima are severely underestimated, dropping to near ice-free conditions (<2 million km²).
- In the Southern Hemisphere, IFS-NEMO-ER shows a massive positive bias year-round (>22 million km² in winter vs ~16 million km² observed); ICON-ESM-ER captures the SH winter maximum magnitude best but underestimates the summer minimum.
- The CMIP6 Multi-Model Mean generally tracks observational magnitudes and trends (particularly the NH September decline) significantly better than the evaluated high-resolution models.
Spatial Patterns
Temporal evolution shows a distinct discontinuity/drift in IFS-FESOM2-SR Southern Hemisphere winter ice around 1988, dropping from ~24 to ~20 million km². In the NH, observations show a clear negative trend in September, whereas the positively biased IFS models show flat trends, indicating a lack of sensitivity to warming likely due to excessive mean state ice thickness.
Model Agreement
Model-observation agreement is poor. The models diverge significantly from observations and each other. IFS variants consistently overestimate ice area (cold bias/ice retention), while ICON tends towards excessive summer melt. CMIP6 MMM provides a much closer fit to observations than any of the individual high-res runs shown.
Physical Interpretation
The extreme positive bias in IFS models suggests a systematic cold bias in polar regions or incorrect sea ice thermodynamics (e.g., albedo tuning) leading to excessive ice volume that survives the summer melt. Conversely, ICON-ESM-ER's 'boom-and-bust' cycle (reasonable winter, collapsed summer) suggests ice that is potentially too thin or overly sensitive to summer radiative forcing or ocean heat fluxes.
Caveats
- The large initialization shock or drift seen in IFS-FESOM2-SR (SH September) complicates trend analysis.
- Sea ice area aggregates concentration; thickness biases (likely high for IFS, low for ICON) are inferred but not directly shown.
Sea Ice Area Seasonal Cycle
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure evaluates the seasonal climatology of Northern and Southern Hemisphere sea ice area for three high-resolution models against OSI-SAF observations and the CMIP6 multi-model mean. The IFS-based models generally exhibit severe positive biases (excessive ice) in both hemispheres, whereas ICON-ESM-ER shows an amplified seasonal cycle with excessive summer melt in the Northern Hemisphere.
Key Findings
- IFS-NEMO-ER and IFS-FESOM2-SR exhibit a massive, year-round positive bias in Northern Hemisphere sea ice area, overestimating the winter maximum by ~6 million km² (+43%) and the summer minimum by ~6 million km² (>100%) compared to observations.
- In the Southern Hemisphere, IFS-NEMO-ER is an extreme outlier, retaining >10 million km² of sea ice in austral summer (vs ~2.5 million km² observed) and reaching a winter maximum of ~23 million km².
- ICON-ESM-ER displays an exaggerated seasonal cycle in the Northern Hemisphere: it grows ~2 million km² too much ice in winter but melts too aggressively in summer, dropping to <2 million km² (significantly below the observed ~5 million km² minimum).
- CMIP6 MMM generally outperforms the high-resolution IFS models in terms of absolute magnitude, tracking observations closely in the NH and slightly underestimating in the SH.
Spatial Patterns
The seasonal phase is generally captured by all models (NH min in Sept, SH min in Feb), but amplitudes vary wildly. In the SH, IFS-NEMO-ER fails to capture the austral summer melt magnitude, maintaining a 'perennial' ice cover baseline that is far too high. Conversely, ICON-ESM-ER and IFS-FESOM2-SR (in SH) show strong melting phases, with ICON nearly becoming ice-free in NH summer.
Model Agreement
There is poor inter-model agreement. The two IFS variants agree closely in the NH (both high positive bias) but diverge significantly in the SH summer (NEMO retains ice, FESOM melts). ICON-ESM-ER aligns more closely with the CMIP6 MMM in the SH but diverges in the NH with its amplified cycle.
Physical Interpretation
The pervasive positive bias in the IFS models suggests a systematic cold bias in the polar regions or issues with sea ice thermodynamics (e.g., albedo parameters) leading to excessive retention. The difference between IFS-NEMO and IFS-FESOM in the SH suggests the ocean component (NEMO vs FESOM) plays a critical role in Antarctic sea ice persistence, likely due to differences in Southern Ocean mixing or heat uptake. ICON's amplified NH cycle (excessive growth and melt) suggests high sensitivity to radiative forcing or ice-albedo feedbacks, potentially indicating the ice is too thin despite being extensive.
Caveats
- Sea ice area relies on concentration thresholds; biases in marginal ice zone concentrations can inflate area metrics.
- The 'SR' and 'ER' designations imply different resolutions (Standard vs Eddy-Rich), which may confound direct component comparisons.
Sea Ice Area Time Series
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure shows monthly and annual-mean sea ice area time series for the Northern (NH) and Southern Hemispheres (SH), revealing that ICON-ESM-ER closely tracks NH observations but underestimates SH ice, while both IFS models exhibit severe positive biases in sea ice area in both hemispheres.
Key Findings
- ICON-ESM-ER shows excellent agreement with NH observational mean state and variability, closely tracking the observed decline, though it underestimates SH sea ice area by ~2-3 million km².
- IFS-NEMO-ER and IFS-FESOM2-SR exhibit massive positive biases in the NH, maintaining an annual mean area of ~16 million km² compared to the observed ~10-11 million km².
- In the SH, IFS-NEMO-ER shows an extreme positive bias, with an annual mean nearly double the observations (~18 vs ~10 million km²).
- IFS-FESOM2-SR displays a notable initialization drift in the SH, starting with a high positive bias (~13 million km²) that gradually decreases over the simulation period, unlike its stable positive bias in the NH.
Spatial Patterns
While purely spatial patterns are not visible, the hemispheric asymmetry in performance is striking. The NH bias in IFS models is systematic and stable, whereas the SH bias shows significant divergence between the two IFS configurations (stable high bias in NEMO vs. drifting bias in FESOM).
Model Agreement
Inter-model agreement is very poor. The models span a huge range of mean states (e.g., SH spread is ~12 million km² between ICON and IFS-NEMO). Only ICON-ESM-ER falls within the range of the CMIP6 Multi-Model Mean for the NH; the IFS models are significant outliers compared to both observations and the CMIP6 ensemble.
Physical Interpretation
The consistent positive bias in both IFS-based models in the NH suggests a potential driver in the shared atmospheric component (IFS) or sea ice parameterizations (e.g., albedo or freezing temperature settings) leading to excessive ice growth or insufficient melt. The divergence in the SH suggests ocean model formulation (NEMO vs. FESOM) plays a stronger role there, potentially regarding vertical mixing and heat flux from the Southern Ocean. The low SH bias in ICON is a common bias in coupled models (often related to cloud feedbacks or warm ocean biases), which is also reflected in the CMIP6 MMM.
Caveats
- The figure shows area, not volume; it is unclear if the excessive area in IFS models corresponds to thin, extensive ice or thick, multi-year ice.
- The strong drift in IFS-FESOM2-SR SH ice suggests the model had not reached equilibrium from its initial state.
Sea Ice Extent March & September Trends
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
The high-resolution models exhibit severe, opposing biases in sea ice extent compared to observations and the CMIP6 multi-model mean (MMM). IFS variants generally overestimate extent significantly (too much ice), while ICON-ESM-ER underestimates summer and Antarctic ice (too little), despite capturing Northern Hemisphere winter extent well.
Key Findings
- IFS-NEMO-ER and IFS-FESOM2-SR show massive positive biases in the Northern Hemisphere year-round (~5 million km² excess), failing to reproduce the observed decline in September sea ice.
- ICON-ESM-ER matches NH March (winter) observations remarkably well but simulates a near ice-free Arctic in September, indicating an excessive seasonal melt cycle.
- In the Southern Hemisphere, IFS-NEMO-ER drastically overestimates ice extent (especially in summer), whereas ICON-ESM-ER and the CMIP6 MMM systematically underestimate winter maximums.
- None of the high-resolution models accurately capture the observed negative trend in NH September sea ice; IFS models show flat trends (insensitivity), while ICON is effectively at the zero-bound.
Spatial Patterns
Biases are seasonally dependent. IFS-NEMO-ER exhibits a dampened seasonal cycle (retaining excessive ice in summer in both hemispheres). ICON-ESM-ER shows an amplified seasonal cycle in the NH (good winter accretion, total summer loss). The observed rapid decline in Arctic summer ice (NH September) is not reproduced by the IFS models, which remain near 10-12 million km² throughout the period.
Model Agreement
Inter-model agreement is very poor. The models bracket the observations with large errors: IFS models are consistently on the high side (cold bias), while ICON is on the low side (warm bias/excessive melt), except for NH winter. The standard resolution CMIP6 MMM generally tracks observations better than these specific high-resolution runs, particularly in the Arctic.
Physical Interpretation
The positive bias in IFS models suggests a systemic cold bias in the polar regions or excessive ice thickness/albedo, rendering the ice pack insensitive to historical warming trends. The 'sticky' sea ice in IFS-NEMO-ER (SH Summer) may relate to stratification issues or lack of vertical mixing in the Southern Ocean. Conversely, ICON-ESM-ER's rapid summer loss implies thin first-year ice domination or excessive shortwave absorption (ice-albedo feedback) during the melt season.
Caveats
- The magnitude of the biases (often >50% of total extent) complicates the assessment of interannual variability and trends.
- ICON's near-zero summer ice represents a floor effect, preventing trend analysis for that season.
Sea Ice Extent Seasonal Cycle
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure illustrates the climatological seasonal cycle of sea ice extent for the Northern (NH) and Southern Hemispheres (SH), comparing three high-resolution EERIE models against OSI-SAF observations and the CMIP6 multi-model mean.
Key Findings
- In the NH, both IFS-based models (IFS-FESOM2-SR and IFS-NEMO-ER) overestimate sea ice extent significantly year-round, with a positive bias of approximately 4–5 million km².
- ICON-ESM-ER captures the NH winter maximum well but exhibits excessive summer melt, leading to a September minimum of <2 million km² compared to ~6.5 million km² in observations.
- In the SH, IFS-NEMO-ER shows a massive positive bias, particularly in summer where extent remains above 11 million km² (observations ~3.5 million km²), indicating a failure to melt.
- ICON-ESM-ER consistently underestimates SH sea ice extent year-round, while the CMIP6 multi-model mean generally outperforms the high-resolution models, especially in the NH.
Spatial Patterns
The NH shows systematic positive biases for IFS models (shifted baseline) and an amplified seasonal cycle for ICON. The SH reveals divergent model behaviors: IFS-NEMO-ER is consistently too extensive, ICON-ESM-ER is consistently too low, and IFS-FESOM2-SR has an exaggerated amplitude (too high in winter, too low in summer).
Model Agreement
Inter-model agreement is poor, particularly in the SH where the spread between models (e.g., February extent ranging from near-zero to ~11 million km²) exceeds the observational mean state. The CMIP6 ensemble mean provides a much closer fit to observations than these individual high-res realizations.
Physical Interpretation
The shared atmospheric component in IFS-NEMO and IFS-FESOM2 leads to similar positive biases in the NH, suggesting an atmospheric cold bias or shared sea ice parameterization issue. However, their divergence in the SH highlights the critical role of the ocean component (NEMO vs. FESOM) in Antarctic sea ice dynamics. ICON-ESM-ER's rapid summer loss in both hemispheres suggests overly aggressive ice-albedo feedbacks or excessive ocean heat flux convergence.
Caveats
- The magnitude of biases suggests these high-resolution configurations are likely untuned compared to the CMIP6 ensemble.
- Extent is a threshold metric; it does not reveal ice thickness or volume, which might show different biases.
Sea Ice Extent Time Series
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
Time series analysis of Northern and Southern Hemisphere sea ice extent reveals substantial mean-state biases in high-resolution EERIE models compared to OSI-SAF observations and the CMIP6 Multi-Model Mean (MMM). While seasonal phases are captured, the absolute magnitudes of ice extent vary drastically between models.
Key Findings
- In the Northern Hemisphere (NH), both IFS-based models (IFS-FESOM2-SR and IFS-NEMO-ER) exhibit a strong positive bias, overestimating annual mean extent by ~4 million km² compared to observations.
- ICON-ESM-ER consistently underestimates sea ice extent in both hemispheres, with NH values ~2 million km² below observations and SH values ~5 million km² too low.
- Southern Hemisphere (SH) results show extreme inter-model spread: IFS-NEMO-ER dramatically overestimates extent (~18 million km² vs ~12 million km² observed), while ICON-ESM-ER is significantly lower than even the negatively biased CMIP6 MMM.
- IFS-FESOM2-SR displays a significant drift or adjustment in the SH during the 1980s, dropping from ~13 to ~11 million km², but ultimately settles closer to observations than the other simulations.
Spatial Patterns
The biases are persistent year-round rather than seasonal. In the NH, the CMIP6 MMM tracks observations far better than the specific high-resolution models shown. In the SH, the 'Antarctic sea ice paradox' (observed stability/slight increase vs. model decline) is overshadowed by massive mean-state offsets.
Model Agreement
Inter-model agreement is poor. The models diverge by up to 10 million km² in the SH (IFS-NEMO vs ICON). Agreement with observations is generally weak, with IFS models tending towards too much ice (especially in NH) and ICON towards too little.
Physical Interpretation
The pervasive positive bias in IFS models suggests a systematic cold bias in polar regions or insufficient ocean heat transport to the surface. Conversely, ICON's negative bias implies a warm bias or excessive vertical mixing/melt. The massive excess in IFS-NEMO-ER SH ice points to specific Southern Ocean coupling issues, likely related to cloud feedbacks or lack of deep convection, which prevents warm deep water from venting to the surface.
Caveats
- The initial sharp drop in IFS-FESOM2-SR (SH) suggests the model may still be experiencing spin-up drift.
- Extent metrics do not reveal ice thickness; a model could have correct extent but incorrect volume (thin ice).
Sea Ice Volume March & September Trends
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | PSC |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure evaluates annual trends in sea ice volume for the Northern and Southern Hemispheres during their respective annual maximum and minimum months (March/September) for three high-resolution models against observations (likely PIOMAS/GIOMAS reanalysis) and the CMIP6 multi-model mean.
Key Findings
- IFS-FESOM2-SR exhibits an extreme positive bias in Northern Hemisphere sea ice volume, estimating ~90,000 km³ in March compared to the observational ~30,000 km³ (a factor of 3 overestimation), indicating excessive multi-year ice accumulation.
- IFS-NEMO-ER also shows a strong positive bias in the NH (~2x observations) and is the only model to substantially overestimate Southern Hemisphere ice volume, particularly in the SH summer (March) where it retains ~11,000 km³ vs ~3,000 km³ observed.
- ICON-ESM-ER generally exhibits a negative 'thin ice' bias; it closely tracks NH winter observations but loses nearly all ice in NH summer (September), suggesting an overly sensitive thermodynamic response or insufficient summer persistence.
- The CMIP6 Multi-Model Mean (MMM) performs significantly better than the high-resolution IFS variants in the Northern Hemisphere, tracking observational magnitude and trends much more closely.
Spatial Patterns
There is a distinct hemispheric asymmetry in model performance. IFS-FESOM2-SR is 'High NH / Low SH', while IFS-NEMO-ER is 'High NH / High SH'. ICON-ESM-ER is 'Low NH / Low SH'. The seasonal cycle amplitude in IFS-NEMO-ER is notably damped in the Antarctic, leading to the large summer bias.
Model Agreement
Inter-model spread is extremely large, exceeding 60,000 km³ in the Arctic (greater than the total observed volume), indicating high uncertainty in sea ice physics parameterizations at these resolutions. The models diverge significantly from observations and each other.
Physical Interpretation
The massive volume overestimation in the IFS models (particularly FESOM) suggests a fundamental imbalance in the thermodynamic growth vs. melt rates or insufficient dynamic ice export, leading to unrealistic retention and thickening of multi-year ice. Conversely, ICON's near-zero NH summer volume implies strong ice-albedo feedbacks or excessive summer melting. The biases suggest that higher resolution alone does not solve sea ice volume issues without appropriate tuning of thermodynamic parameters.
Caveats
- The 'Obs' dataset for volume is typically a reanalysis product (e.g., PIOMAS/GIOMAS) rather than direct satellite observation, carrying its own uncertainties.
- The metadata lists 'PSC (ice_conc)' but the plot clearly shows Volume (km³), confirming the use of volume reanalysis data for the black line.
Sea Ice Volume Seasonal Cycle
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | PSC |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This diagnostic evaluates the seasonal cycle of sea ice volume in the Northern and Southern Hemispheres for three high-resolution models against observations and the CMIP6 multi-model mean. The models exhibit drastic disparities in mean state volume, with errors reaching factors of 3–4 relative to observations.
Key Findings
- IFS-FESOM2-SR exhibits an extreme positive bias in Northern Hemisphere (NH) sea ice volume, peaking at ~90×10³ km³ compared to the observational maximum of ~28×10³ km³, suggesting excessive ice thickness accumulation.
- IFS-NEMO-ER shows a systematic high bias in both hemispheres; it overestimates NH volume by ~2x and Southern Hemisphere (SH) volume by ~30–50%, notably failing to melt back sufficiently during SH summer.
- ICON-ESM-ER consistently underestimates sea ice volume, with a weak seasonal cycle in the SH (peaking at <10×10³ km³ vs ~20×10³ km³ obs) and a near-total loss of volume in the NH summer (September minimum near zero).
- The CMIP6 Multi-Model Mean tracks NH observations (likely PIOMAS) remarkably well but underestimates the SH winter maximum compared to the reference dataset.
Spatial Patterns
The seasonal phase (timing of maximum/minimum) is generally consistent across models and observations (March max for NH, September max for SH). However, the amplitude and mean state vary wildly. In the SH, IFS-NEMO-ER's minimum (summer) volume is ~10×10³ km³, four times the observed ~2.5×10³ km³, indicating a failure to melt seasonal ice.
Model Agreement
Inter-model agreement is very poor. In the NH, model estimates for annual mean volume range from ~10×10³ km³ (ICON) to ~80×10³ km³ (IFS-FESOM2-SR). None of the high-resolution models match the observational baseline as well as the CMIP6 MMM in the NH.
Physical Interpretation
The massive volume in IFS-FESOM2-SR (NH) implies unchecked growth of multi-year ice thickness, potentially due to albedo biases, insufficient ocean heat flux, or initialization issues. Conversely, ICON-ESM-ER appears dynamically or thermodynamically too warm (or has too low albedo), preventing ice thickening. IFS-NEMO-ER's high SH summer minimum suggests difficulty melting ice or excessive production, leading to a permanent thick ice pack where observations show mostly seasonal ice.
Caveats
- Observation metadata lists 'PSC (ice_conc)' but plots volume; this implies a derived product (likely PIOMAS for NH and GIOMAS or similar reanalysis for SH) which carries its own uncertainties, particularly in the SH.
- The extreme values in IFS-FESOM2-SR are physically unrealistic for the historical period, exceeding geometric bounds for typical ice areas.
Sea Ice Volume Time Series
| Variables | siconc, sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | PSC |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
The figure presents monthly and annual-mean sea ice volume time series (1980–2014) for the Northern and Southern Hemispheres, revealing extreme biases in the mean state of high-resolution models compared to reanalysis-based observations (likely PIOMAS/GIOMAS) and the CMIP6 multi-model mean.
Key Findings
- In the Northern Hemisphere (NH), IFS-FESOM2-SR and IFS-NEMO-ER exhibit massive positive biases, with annual mean volumes of ~80×10³ km³ and ~50×10³ km³ respectively, compared to observations of ~20–25×10³ km³.
- ICON-ESM-ER consistently underestimates sea ice volume in both hemispheres, with NH volume around 10×10³ km³ (roughly 50% of observed) and SH volume near 3×10³ km³.
- In the Southern Hemisphere (SH), model spread is similarly large but rank orders differ: IFS-NEMO-ER overestimates volume (~19×10³ km³ vs ~12×10³ km³ Obs), while IFS-FESOM2-SR underestimates it (~5×10³ km³).
- Despite mean state biases, all models capture the declining trend in NH sea ice volume, though the absolute rate of loss varies.
Spatial Patterns
While spatial maps are not shown, the magnitude of the volume biases (3-4x observations for IFS-FESOM NH) implies severe and widespread excessive ice thickness across the Arctic basin, rather than just extent errors.
Model Agreement
Inter-model agreement is very low regarding the mean state. The spread covers an order of magnitude in the NH (from ~10k to ~80k km³). The CMIP6 multi-model mean generally tracks observations much better than these specific high-resolution runs, suggesting the high-resolution configurations require significant tuning.
Physical Interpretation
The extreme volume in IFS models (especially NH) suggests thermodynamic imbalances, likely excessive winter growth or insufficient summer melt/export, leading to multi-year ice accumulation far exceeding reality. ICON's low volume points to warm biases in the polar oceans or atmosphere. The divergence between IFS-NEMO (high SH volume) and IFS-FESOM (low SH volume) highlights the strong influence of the ocean model component (NEMO vs. FESOM) on sea ice preservation.
Caveats
- Sea ice volume observations are derived from reanalysis (e.g., PIOMAS/GIOMAS) rather than direct satellite measurement, carrying higher uncertainty than concentration/extent data, particularly in the SH.
- The metadata mentions 'siconc' but plots volume, confirming reliance on a derived thickness product.
Arctic Sea Ice Concentration
| Variables | siconc |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure compares Arctic sea ice concentration climatologies (1980–2014) for March and September across three models against OSI-SAF observations, revealing distinct biases: IFS-FESOM2-SR closely matches observations, while IFS-NEMO-ER and ICON-ESM-ER show opposing positive and negative sea ice biases, respectively.
Key Findings
- IFS-NEMO-ER exhibits a substantial positive sea ice bias (excessive extent), particularly in March where ice incorrectly surrounds Iceland and extends excessively south in the Labrador Sea.
- ICON-ESM-ER displays a strong negative bias (insufficient ice), characterized by a largely ice-free Barents Sea in March and a depleted, low-concentration central pack in September.
- IFS-FESOM2-SR is the best-performing model, capturing the spatial distribution of the winter marginal ice zone and the summer minimum extent with high fidelity relative to observations.
Spatial Patterns
In March, the North Atlantic ice edge varies drastically: IFS-NEMO-ER extends past Newfoundland and Iceland, while ICON-ESM-ER retreats north of Svalbard. In September, IFS-NEMO-ER retains ice in shelf seas (Chukchi/East Siberian) that are open in observations, whereas ICON-ESM-ER shows extensive low-concentration areas (<60%) within the central pack.
Model Agreement
Inter-model agreement is low, particularly in the Atlantic sector (Barents, Greenland, and Labrador Seas). IFS-FESOM2-SR is the only model that aligns well with the observational baseline in both seasons, sitting between the cold bias of IFS-NEMO-ER and the warm bias of ICON-ESM-ER.
Physical Interpretation
The excessive winter ice in IFS-NEMO-ER, especially in the Labrador Sea, suggests a 'cold hole' bias likely driven by suppressed deep convection or fresh surface biases. Conversely, the ICON-ESM-ER retreat implies excessive heat input, potentially from strong Atlantic Water inflow into the Barents Sea or strong ice-albedo feedback amplifying summer melt.
Caveats
- Analysis is based on concentration only; sea ice thickness/volume would provide further insight into the mass budget.
- Regional biases in the marginal seas may be sensitive to specific ocean resolution and bathymetry details not fully detailed here.
Antarctic Sea Ice Concentration
| Variables | siconc |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | OSI_SAF |
| Units | 0-1 |
| Period | 1980–2014 |
Summary high
This figure compares Antarctic sea ice concentration climatologies (March and September) from three high-resolution coupled models against OSI-SAF observations (1980–2014). The models exhibit divergent behaviors, ranging from severe overestimation of summer ice to general underestimation throughout the year.
Key Findings
- IFS-NEMO-ER displays a critical bias in the seasonal cycle: it fails to melt sea ice during austral summer (March), retaining a near-winter-like extent that encircles the entire continent.
- ICON-ESM-ER underestimates sea ice concentration in both seasons, with a retracted ice edge in September and a nearly ice-free Southern Ocean in March compared to observations.
- IFS-FESOM2-SR shows the best agreement with observations, capturing the Weddell Sea remnant ice in March and a reasonable September extent, though it is slightly less extensive than observations in the Indian Ocean sector.
- All models tend to show sharper gradients at the ice edge (Marginal Ice Zone) compared to the more diffuse transition seen in the satellite observations.
Spatial Patterns
In September (winter maximum), IFS-NEMO-ER extends ice furthest north, while ICON-ESM-ER shows significant open water anomalies in the eastern Weddell and Indian Ocean sectors. In March (summer minimum), the contrast is stark: Observations show ice restricted primarily to the Weddell and Ross Seas; IFS-NEMO-ER shows a massive, circumpolar ice shield similar to its winter state; IFS-FESOM2-SR closely matches the observational pattern; and ICON-ESM-ER loses almost all ice, including in the Weddell Sea.
Model Agreement
There is very poor inter-model agreement in the Southern Hemisphere. The models occupy three distinct regimes: severe positive bias (IFS-NEMO-ER), negative bias (ICON-ESM-ER), and realistic representation (IFS-FESOM2-SR).
Physical Interpretation
The persistence of winter-like ice in summer for IFS-NEMO-ER suggests a fundamental issue with the Southern Ocean heat budget, possibly due to excessive surface stratification preventing deep ocean heat from reaching the surface (shutdown of convection) or strong cold biases in atmospheric forcing. Conversely, ICON-ESM-ER likely suffers from a warm SST bias or excessive vertical mixing that melts ice too aggressively. The sharp ice edges in models may reflect high-resolution dynamical fronts that are smoothed in lower-resolution observational products.
Caveats
- The analysis relies on visual concentration maps; calculating integrated sea ice area or extent metrics would quantify the magnitude of the biases.
- The exact cause of IFS-NEMO-ER's non-melting state requires inspection of ocean temperature profiles and surface fluxes.
Arctic Sea Ice Thickness (m)
| Variables | sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | PSC |
| Units | m |
| Period | 1980–2014 |
Summary high
Diagnostic maps of Arctic sea ice thickness reveal striking opposing biases: IFS-based models simulate excessive multi-year ice thickness year-round, whereas ICON-ESM-ER produces an unrealistically thin ice pack that nearly vanishes in summer.
Key Findings
- IFS-FESOM2-SR and IFS-NEMO-ER exhibit a strong positive thickness bias, with central Arctic ice exceeding 4 m in March and retaining >3.5 m thickness over large areas in September, far exceeding the observed ~2–3 m maxima.
- ICON-ESM-ER displays a severe negative bias, failing to form the thick multi-year ice wedge north of Greenland in winter and showing a nearly ice-free state (<0.5 m) in September.
- The spatial distribution of the thickest ice (north of Greenland/CAA) is qualitatively captured by IFS models but quantitatively exaggerated, while ICON lacks this gradient almost entirely.
Spatial Patterns
Observations show a clear gradient from thick ice (>3 m) north of the Canadian Arctic Archipelago/Greenland to thinner ice on the Siberian shelf. IFS models amplify this pattern, extending the >3.5 m thick ice zone across the entire central basin. Conversely, ICON shows a diffuse, uniform field of thin ice (~1.5–2 m) in March and lacks any significant thick ice accumulation zones.
Model Agreement
There is a clear dichotomy: the two models sharing the IFS atmosphere (IFS-NEMO and IFS-FESOM) agree on a 'too thick' bias despite different ocean grids, while ICON is a distinct outlier with a 'too thin' bias.
Physical Interpretation
The shared positive bias in IFS-NEMO and IFS-FESOM suggests the driver may lie in the shared atmospheric forcing (e.g., wind stress causing excessive dynamic piling or cloud radiative biases favoring thermodynamic growth) or common sea ice parameterizations. ICON's inability to sustain multi-year ice points to excessive summer melt (albedo feedback) or insufficient winter growth/retention.
Caveats
- Thickness observations (labeled PSC, likely PIOMAS reanalysis) have higher uncertainty than concentration data.
- Averaging over 1980–2014 smooths out the strong observed thinning trend, potentially masking recent regime shifts.
Antarctic Sea Ice Thickness (m)
| Variables | sithick |
|---|---|
| Models | IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER |
| Obs Dataset | PSC |
| Units | m |
| Period | 1980–2014 |
Summary high
This diagnostic compares Antarctic sea ice thickness climatologies for September (annual maximum) and March (annual minimum) across three high-resolution coupled models against observations. There is a stark divergence in model performance: IFS-NEMO-ER significantly overestimates thickness and extent year-round, while IFS-FESOM2-SR and ICON-ESM-ER severely underestimate them, resulting in a nearly ice-free Southern Ocean in March.
Key Findings
- IFS-NEMO-ER exhibits a strong positive bias, simulating excessive ice thickness (>2.5 m) and extent in September, and retaining a large, overly thick ice pack in March that exceeds observations.
- ICON-ESM-ER and IFS-FESOM2-SR both suffer from severe negative biases, particularly in summer (March), where they fail to reproduce the observed perennial ice in the Weddell Sea.
- In September, ICON-ESM-ER shows a pattern of very thin pack ice (<0.5 m) over most of the domain, with thicker ice artificially confined to narrow coastal bands, unlike the broader distribution in observations.
- The observational reference shows moderate thickness (generally 1-2 m) with key perennial retention in the Weddell Sea, a feature completely missed by ICON and FESOM2 but exaggerated by NEMO.
Spatial Patterns
In September, IFS-NEMO-ER shows a massive extension of thick ice into the Southern Ocean, particularly in the Weddell sector. Conversely, ICON-ESM-ER and IFS-FESOM2-SR show retracted ice edges. In March, the discrepancy is extreme: the 'Obs' panel shows localized thick ice in the western Weddell Sea; IFS-NEMO-ER fills the entire Weddell Sea with thick ice, while the other two models show almost zero sea ice cover.
Model Agreement
Inter-model agreement is very poor. The models bracket the observations with large errors in opposite directions (NEMO too thick/extensive, others too thin/retreating). No model captures the seasonal cycle amplitude correctly.
Physical Interpretation
The 'missing Antarctic summer sea ice' in ICON and FESOM2 is a common bias in coupled models, often linked to warm biases in the Southern Ocean surface waters, insufficient cloud shading (SW radiation bias), or vertical mixing issues bringing warm deep water to the surface. Conversely, the cold/thick bias in IFS-NEMO-ER suggests the model is either too cold in the Southern Ocean or has excessive dynamic piling of ice. The difference between IFS-NEMO and IFS-FESOM (which share an atmosphere) points strongly to the ocean model component (NEMO vs FESOM) and its grid/mixing formulation as the primary driver of these differences.
Caveats
- Antarctic sea ice thickness observations (labeled PSC) generally have higher uncertainty than Arctic data due to snow loading effects on satellite altimetry.
- The 'Obs' panel may be a reanalysis product (like GIOMAS) rather than direct observation, which implies its own model biases.