Global Climatology Biases — Feather — EERIE Model Evaluation

Total Cloud Cover Annual Mean Bias

Variables	clt
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	%
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: 1.40 · Rmse: 4.90
IFS-NEMO-ER	Global Mean Bias: 2.36 · Rmse: 4.85
ICON-ESM-ER	Global Mean Bias: 2.05 · Rmse: 9.41
CMIP6 MMM	Global Mean Bias: -0.08 · Rmse: 5.41

Summary high

Diagnostic maps of annual mean total cloud cover bias relative to ERA5 reveal that the IFS-based models (NEMO and FESOM coupled) significantly outperform ICON-ESM-ER, which exhibits large-scale systematic biases. While IFS models show moderate positive biases over oceans and negative biases over tropical land (RMSE ~4.9%), ICON displays severe overestimation in tropical convergence zones and underestimation over land and high latitudes (RMSE ~9.4%).

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER demonstrate superior performance (RMSE ~4.9%) compared to both ICON-ESM-ER (RMSE ~9.4%) and the CMIP6 Multi-Model Mean (RMSE ~5.4%).
A systematic negative bias over tropical land masses (Amazon, Congo, Maritime Continent) is present in all models, including the CMIP6 MMM, but is most pronounced in ICON.
ICON-ESM-ER exhibits excessive cloudiness (>15% positive bias) along the ITCZ and SPCZ, suggesting overly vigorous tropical convection or moisture retention.
IFS models show positive cloud biases in eastern boundary upwelling regions (e.g., off Peru/Chile, Namibia), contrasting with the typical GCM deficit in stratocumulus decks.

Spatial Patterns

The IFS models display a relatively uniform positive bias over subtropical and tropical oceans, interrupted by negative biases over tropical rainforests. In contrast, ICON-ESM-ER shows a sharp dipole pattern: extreme positive biases in the tropical ocean convergence zones and strong negative biases over continents and the high-latitude Southern Ocean (specifically near Antarctica). The CMIP6 MMM shows a zonal structure with positive biases in the Southern Ocean mid-latitudes and negative biases over tropical land.

Model Agreement

There is strong structural agreement between the two IFS variants, indicating that the atmospheric component dominates cloud physics errors over ocean coupling differences. All models agree on the sign of the error (negative) over tropical land. ICON is an outlier with significantly larger bias magnitudes and different spatial structures in the high latitudes compared to the IFS/CMIP6 group.

Physical Interpretation

The pervasive negative land bias likely points to deficiencies in convective parameterization (e.g., insufficient detrainment or cloud lifetime) over land surfaces. ICON's intense ITCZ cloudiness suggests issues with the deep convection scheme or relative humidity thresholds. The positive bias in IFS stratocumulus regions is notable as models typically struggle to produce enough cloud there; this may indicate over-active low-cloud schemes or biases in the reference ERA5 reanalysis.

Caveats

The observational reference is ERA5 reanalysis, which relies on its own model physics for cloud cover and may contain biases compared to direct satellite observations (e.g., CERES, ISCCP).
Annual mean averaging obscures seasonal shifts in the ITCZ, potentially masking compensating seasonal errors.

Total Cloud Cover DJF Bias

Variables	clt
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	%
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.45 · Rmse: None

Summary high

This figure displays spatial biases in Total Cloud Cover (TCC) for the DJF season relative to ERA5 reanalysis for three high-resolution EERIE models and the CMIP6 multi-model mean.

Key Findings

ICON-ESM-ER exhibits the most severe biases, characterized by a stark contrast: strong underestimation of cloud cover in tropical deep convection zones (> -20%) and strong overestimation over the Southern Ocean and subtropical eastern oceans (> +20%).
IFS-FESOM2-SR and IFS-NEMO-ER show nearly identical bias patterns, confirming that atmospheric physics (IFS) rather than ocean coupling dominates the cloud bias structure.
All three high-resolution models underestimate cloud cover in key deep convection regions (Amazon, Congo, Maritime Continent, and ITCZ) compared to ERA5.
The CMIP6 Multi-Model Mean generally shows smaller magnitude biases but shares the negative bias over the Amazon and positive bias in the Indian Ocean.

Spatial Patterns

The IFS models show a zonal bias structure: negative along the equatorial ITCZ/SPCZ and tropical landmasses, flanked by positive biases in the subtropical trade wind regions and eastern ocean basins. ICON-ESM-ER amplifies this pattern significantly and adds a massive positive cloud bias band over the Southern Ocean (40°S–60°S) and pervasive negative biases over northern mid-latitude landmasses (Europe/Asia).

Model Agreement

There is exceptionally high agreement between the two IFS-based models, indicating robustness to the ocean model used. There is low agreement between IFS and ICON regarding bias magnitude and high-latitude behavior (e.g., Southern Ocean), where ICON is far more biased.

Physical Interpretation

The negative biases in tropical convective zones (Amazon, ITCZ) suggest deficiencies in convective parameterizations, specifically regarding cloud detrainment or lifetime in deep convection (failing to produce sufficient anvil/cirrus clouds). The positive biases in subtropical eastern oceans (IFS and ICON) point to difficulties in correctly simulating stratocumulus-to-cumulus transition regimes, resulting in excessive low cloud cover. ICON's Southern Ocean positive bias is likely related to supercooled liquid water or phase-partitioning parameterizations common in that region.

Caveats

The reference dataset is ERA5 reanalysis, which itself relies on model physics to generate cloud cover; comparisons with direct satellite observations (e.g., CERES-MODIS) might yield different magnitudes.
High-latitude biases (Antarctica) should be interpreted with caution due to greater uncertainties in reanalysis products over polar ice sheets.

Total Cloud Cover JJA Bias

Variables	clt
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	%
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -0.71 · Rmse: None

Summary high

This figure evaluates JJA Total Cloud Cover biases relative to ERA5 for three high-resolution models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) and the CMIP6 Multi-Model Mean. The two IFS-based models exhibit very similar spatial bias structures, contrasting sharply with ICON-ESM-ER, particularly in the Southern Ocean and tropical bands.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER show high structural agreement, characterized by positive cloud biases in the Southern Ocean and eastern subtropical subsidence regions (stratocumulus decks), and negative biases over the Asian Monsoon region and Amazon.
ICON-ESM-ER exhibits a distinct bias pattern with a strong negative cloud cover bias over the Southern Ocean (circumpolar) and North Atlantic, and broad positive biases across the tropical oceans.
The CMIP6 MMM generally shows smaller amplitude biases due to averaging but shares the negative Amazonian cloud bias seen in the IFS models.
A major inter-model divergence exists in the Southern Ocean: IFS models overestimate cloud cover (+10-20%), while ICON underestimates it (<-20%), implying opposite radiative feedback errors in this region.

Spatial Patterns

IFS models display positive biases in eastern boundary current regions (off Peru, Namibia, California) and the Southern Ocean, with negative biases over tropical land masses (Amazon, India, SE Asia). ICON-ESM-ER displays zonal positive biases in the tropical Pacific and Indian Ocean, flanked by strong negative biases in the mid-to-high latitudes (Southern Ocean, North Atlantic, Arctic).

Model Agreement

There is exceptionally high agreement between the two IFS-based simulations, indicating the atmosphere model dominates the cloud bias signal over the ocean model choice. There is low agreement between IFS and ICON, particularly at high latitudes.

Physical Interpretation

The positive bias in IFS stratocumulus regions suggests overly aggressive low-cloud formation or persistence in subsidence zones. The negative bias in ICON over the Southern Ocean likely points to deficits in supercooled liquid water or cloud fraction parameterization in the storm tracks, which would lead to excessive surface solar heating. The negative biases over the Amazon and India in IFS suggest deficiencies in capturing the intensity or extent of deep convection during the summer monsoon/wet season.

Caveats

ERA5 is a reanalysis product; while it assimilates observational radiances, its cloud fraction is model-derived and may have its own biases compared to direct satellite products like CALIPSO/CloudSat.
The analysis is limited to JJA (Northern Hemisphere summer/Southern Hemisphere winter); seasonal shifts in biases (e.g., during DJF) are not assessed here.

Surface Latent Heat Flux Annual Mean Bias

Variables	hfls
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: 3.13 · Rmse: 11.77
IFS-NEMO-ER	Global Mean Bias: 1.20 · Rmse: 10.00
ICON-ESM-ER	Global Mean Bias: 3.90 · Rmse: 16.09

Summary high

This diagnostic evaluates annual mean surface latent heat flux biases in three high-resolution coupled models relative to ERA5, identifying IFS-NEMO-ER as the best performer while ICON-ESM-ER exhibits systematic excessive evaporation.

Key Findings

ICON-ESM-ER shows a strong, widespread positive bias (global mean +3.9 W/m²; RMSE 16.1 W/m²), with particularly intense evaporation excesses over continental landmasses (Amazon, Africa, Eurasia) and mid-latitude oceans.
IFS-NEMO-ER and IFS-FESOM2-SR display very similar bias patterns globally (reflecting their shared atmospheric component), with IFS-NEMO-ER achieving the lowest RMSE (10.0 W/m²) and global bias (+1.2 W/m²).
Western Boundary Current regions (Gulf Stream, Kuroshio) exhibit characteristic dipole bias structures in the IFS models, indicating spatial shifts in current separation and extension paths compared to reanalysis.

Spatial Patterns

The IFS models show zonal banding in the tropical Pacific and negative biases in the North Atlantic subpolar gyre (more pronounced in NEMO). In contrast, ICON-ESM-ER is dominated by positive biases across the Southern Ocean, North Atlantic storm track, and major rainforest basins. All models show positive biases over the Amazon and Congo basins.

Model Agreement

The two IFS-based models (NEMO and FESOM2) show high structural agreement, diverging primarily in the North Atlantic and parts of the Southern Ocean, likely due to ocean model formulation (finite element vs. finite difference). ICON-ESM-ER is an outlier with significantly higher positive biases globally.

Physical Interpretation

The sharp dipoles in the Gulf Stream and Kuroshio in IFS models are typical of high-resolution simulations where the precise path of the current deviates from the assimilated reanalysis, leading to large local heat flux errors. ICON's systematic positive bias suggests overly vigorous turbulent surface exchange, potentially driven by high surface wind speeds or aggressive bulk aerodynamic transfer coefficients. The strong positive land biases across all models suggest challenges in parameterizing evapotranspiration over tropical rainforests.

Caveats

Biases are computed relative to ERA5, a reanalysis product which itself relies on model physics for surface fluxes, particularly over ocean areas lacking direct in-situ measurements.
Positive latent heat flux biases can be self-limiting (cooling the SST), so these maps must be interpreted alongside SST bias maps to determine causality.

Surface Latent Heat Flux DJF Bias

Variables	hfls
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014

Summary high

This figure displays the global distribution of Surface Latent Heat Flux (hfls) biases for the DJF season relative to ERA5 reanalysis. The IFS-based models show similar patterns with moderate biases concentrated in boundary currents, whereas ICON-ESM-ER exhibits stronger, distinct biases, particularly over land and tropical oceans.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER show high consistency, sharing positive biases (excessive evaporation) over the Gulf Stream, Kuroshio Extension, and Mediterranean, and weak negative biases in the deep tropics.
ICON-ESM-ER displays significantly larger bias magnitudes globally, with extensive positive biases over tropical oceans and Southern Hemisphere land masses (South America, Africa, Australia during their summer).
A striking contrast exists in the North Atlantic: IFS models show a positive bias along the Gulf Stream extension, while ICON shows a strong negative bias in the subpolar gyre/North Atlantic Current region.

Spatial Patterns

In DJF (Northern Hemisphere winter), Western Boundary Currents are active regions of heat loss; IFS models generally overestimate this flux (red bias), while ICON underestimates it in the specific current extensions (blue) but overestimates in the broader subtropics. Over Southern Hemisphere land (summer), ICON shows very strong positive latent heat flux biases, suggesting excessive evapotranspiration compared to ERA5.

Model Agreement

There is strong agreement between the two IFS-based models (IFS-FESOM2 and IFS-NEMO), indicating that the atmospheric component likely dictates the surface flux errors or that the ocean models share similar SST biases. ICON-ESM-ER diverges significantly from the IFS group in both spatial pattern and magnitude.

Physical Interpretation

Latent heat flux is driven by surface wind speed and the vertical humidity gradient (q_s - q_a). The positive biases in Western Boundary Currents for IFS likely stem from either warm SST biases (increasing q_s) or excessive surface winds/dry air outbreaks. ICON's widespread positive bias in the tropics and over land suggests a fundamental difference in its boundary layer scheme or surface parameterizations leading to more vigorous evaporation/transpiration. The negative bias in the North Atlantic for ICON might relate to a cold SST bias or a misplaced North Atlantic Current.

Caveats

ERA5 is a reanalysis product and relies on its own bulk formulae, so biases are relative to the ERA5 model physics.
Flux direction is upwards positive; red indicates model flux > ERA5 flux (stronger cooling of ocean).

Surface Latent Heat Flux JJA Bias

Variables	hfls
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014

Summary high

This diagnostic compares JJA Surface Latent Heat Flux biases for three high-resolution coupled models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) against ERA5 reanalysis. The two IFS-based models exhibit remarkably similar bias structures marked by tropical deficits and high-latitude excesses, while ICON-ESM-ER shows distinct, widespread positive biases over Northern Hemisphere land and subtropical oceans.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER show strong structural agreement, featuring negative latent heat biases over the Amazon (suggesting a dry/moisture-limited regime) and positive biases over Northern Eurasia and the Kuroshio Extension.
ICON-ESM-ER is characterized by extensive, high-magnitude positive biases (>30 W/m²) over Northern Hemisphere land masses (North America, Europe, Siberia), indicating excessive evapotranspiration in the boreal summer.
A clear model divergence exists over the Amazon Basin: IFS models exhibit a negative bias (too little evaporation), whereas ICON-ESM-ER exhibits a positive bias (too much evaporation).
All models show positive evaporation biases over the Southern Ocean, likely driven by strong westerly winds or warm SST biases, but ICON's positive anomalies are more widespread across the subtropical Atlantic and Pacific.

Spatial Patterns

IFS models show a distinct negative bias in the equatorial Pacific cold tongue and the Amazon. ICON displays a global tendency towards positive biases, particularly strong over NH continents and subtropical ocean gyres, but shares the negative bias over the Maritime Continent seen in IFS.

Model Agreement

The two IFS models (FESOM and NEMO) are nearly identical in their flux bias patterns, confirming that the atmospheric component (OpenIFS) dominates the surface flux characteristics. There is low agreement between the IFS family and ICON-ESM-ER regarding land-surface interaction over South America and the magnitude of NH summer evapotranspiration.

Physical Interpretation

The positive latent heat flux biases over land in ICON (and Northern Eurasia in IFS) suggest excessive soil moisture availability or overly vigorous plant transpiration parameterizations. The negative Amazon bias in IFS is consistent with a 'die-back' or dry bias often seen in this region. Over the oceans, positive biases in the storm tracks and Southern Ocean suggest the models may be maintaining warmer SSTs or stronger surface winds than ERA5, enhancing turbulent heat loss.

Caveats

ERA5 latent heat flux is a derived reanalysis product, not a direct observation, and carries significant uncertainty over tropical land and remote oceans.
Surface flux biases are part of a coupled loop involving precipitation, radiation, and circulation; determining the primary driver (e.g., whether flux error causes or results from SST error) requires further analysis.

Surface Sensible Heat Flux Annual Mean Bias

Variables	hfss
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: 1.74 · Rmse: 6.96
IFS-NEMO-ER	Global Mean Bias: 1.81 · Rmse: 6.60
ICON-ESM-ER	Global Mean Bias: 1.85 · Rmse: 10.43

Summary high

This figure evaluates annual mean surface sensible heat flux (SSHF) biases in three high-resolution coupled models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) relative to ERA5 reanalysis. While all models show a global positive mean bias of ~1.7–1.8 W/m², the spatial distribution of errors varies significantly, with the two IFS-based models behaving similarly and outperforming ICON-ESM-ER in terms of RMSE.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER exhibit highly correlated bias patterns (RMSE ~6.6-7.0 W/m²), confirming that the atmospheric component (IFS) dominates the surface flux characteristics.
All models display prominent negative biases over major tropical rainforests (Amazon, Congo, Maritime Continent), indicating an underestimation of sensible heat flux (likely overestimating latent heat/evapotranspiration) compared to ERA5.
ICON-ESM-ER shows significantly larger errors (RMSE 10.43 W/m²), characterized by intense positive biases over Northern Hemisphere mid-to-high latitude land (Siberia, North America) and strong negative biases over tropical land.
A negative bias 'hole' in the North Atlantic subpolar gyre is present in all models, likely associated with the common cold SST bias in this region which suppresses upward heat flux.
IFS models show a distinct zonal band of positive bias in the Southern Ocean (40°S–60°S), suggesting excessive heat loss to the atmosphere, potentially driven by warm SST biases or excessive wind speeds.

Spatial Patterns

Land biases show a zonal dichotomy, particularly in ICON: strong negative biases in the tropics (rainforests) and positive biases in the extratropics. The IFS models show localized positive biases over orographic features (Rockies, Himalayas, Andes) and western boundary currents. Over the ocean, biases are generally smaller than over land, but organized patterns appear in the Southern Ocean (positive in IFS) and North Atlantic (negative in all).

Model Agreement

There is high agreement between the two IFS-based models (FESOM2 and NEMO ocean variants), indicating that the ocean model choice has a secondary effect on the broad SSHF bias structure compared to the land surface and atmospheric schemes. ICON-ESM-ER diverges significantly, particularly over terrestrial regions.

Physical Interpretation

The land bias patterns strongly suggest differences in surface energy partitioning (Bowen ratio). The negative tropical biases imply that the models partition too much available energy into latent heat (evaporation) rather than sensible heat relative to ERA5. The positive biases over arid/semi-arid and high-latitude land in ICON suggest the opposite (ground drying out or snow/vegetation scheme differences). Over oceans, biases likely follow SST errors; the North Atlantic negative bias is consistent with a 'cold blob' impeding heat transfer to the atmosphere.

Caveats

ERA5 surface fluxes are reanalysis products derived from a forecast model, not direct observations, and carry their own model uncertainties.
The analysis does not separate the contribution of wind speed errors versus temperature gradient (SST - T2m) errors in driving the flux biases.

Surface Sensible Heat Flux DJF Bias

Variables	hfss
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014

Summary high

This diagnostic evaluates DJF climatological biases in Surface Sensible Heat Flux (SHF) for three coupled models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) against ERA5 reanalysis, highlighting discrepancies in major ocean current systems and land surface interactions.

Key Findings

All models exhibit strong dipolar biases (±30-40 W/m²) in the Western Boundary Current regions (Gulf Stream, Kuroshio), indicating misplacement of current separation and heat release zones.
IFS-FESOM2-SR shows a distinct, intense positive bias rimming the Antarctic continent, suggesting excessive heat release likely due to reduced sea ice concentration or thickness in the marginal ice zone.
ICON-ESM-ER displays widespread negative biases over Southern Hemisphere land masses (South America, Southern Africa, Australia) during their summer, indicating potential issues with surface energy partitioning (Bowen ratio).
In the North Atlantic subpolar gyre, ICON-ESM-ER shows a negative bias (reduced heat loss), contrasting with the positive biases seen in both IFS variants.

Spatial Patterns

Biases are largest in regions of strong air-sea interaction: Western Boundary Currents and high-latitude oceans. In the Northern Hemisphere winter, the bias patterns trace the Gulf Stream and Kuroshio extensions. Southern Hemisphere land biases are spatially coherent in ICON but patchy in IFS models.

Model Agreement

The IFS-NEMO and IFS-FESOM models agree well on land patterns and Northern Hemisphere ocean structures, diverging primarily in the Southern Ocean where FESOM's unstructured mesh solution produces unique positive biases. ICON stands apart with stronger negative land biases and a reversed signal in the North Atlantic subpolar gyre.

Physical Interpretation

The dipolar ocean biases suggest latitudinal shifts in sharp SST fronts (Western Boundary Currents). The positive Antarctic bias in FESOM implies insufficient insulation by sea ice, allowing the relatively warm ocean to heat the cold atmosphere. The negative land biases in ICON (SH summer) suggest the model favors latent heat flux or reflects less solar absorption compared to ERA5.

Caveats

Surface fluxes in ERA5 are model-derived estimates constrained by assimilation, not direct observations.
Biases in dynamic regions like the Gulf Stream are of similar magnitude to the mean signal.

Surface Sensible Heat Flux JJA Bias

Variables	hfss
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014

Summary high

This diagnostic compares JJA Surface Sensible Heat Flux (SSHF) biases of three high-resolution models against ERA5 reanalysis. The IFS-based models exhibit similar, moderate bias patterns, whereas ICON-ESM-ER displays widespread, strong positive biases over Northern Hemisphere land, suggesting distinct differences in land-surface energy partitioning.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER show high consistency, with moderate positive biases over Northern Eurasia and the Amazon, and negative biases in monsoon regions (Sahel, India).
ICON-ESM-ER exhibits significantly larger positive biases (>30 W/m²) across most Northern Hemisphere land masses (North America, Eurasia) and the Amazon, indicating a systematic overestimation of sensible heat release (likely linked to a high Bowen ratio/dry bias).
All models display a prominent zonal dipole bias in the Southern Ocean (SH winter), with a band of strong positive bias (excess heat release) adjacent to a negative bias band, characteristic of sea ice edge location errors.

Spatial Patterns

In JJA (NH summer), land biases dominate the Northern Hemisphere. ICON shows a continental-scale positive bias over North America and Eurasia. The IFS models have a more regional signal, with positive biases in high latitudes and the Amazon, but negative biases in the African Sahel and India. In the Southern Hemisphere (winter), the dominant feature is the high-latitude oceanic dipole structure surrounding Antarctica.

Model Agreement

There is strong agreement between IFS-FESOM2-SR and IFS-NEMO-ER, confirming that atmospheric and land surface physics (IFS/HTESSEL) dominate the surface flux biases rather than the ocean model choice (FESOM vs NEMO). ICON-ESM-ER diverges significantly, particularly over land, with much higher bias magnitudes.

Physical Interpretation

The widespread positive SSHF bias in ICON-ESM-ER over NH summer land suggests the model partitions too much available energy into sensible rather than latent heat, implying a limitation in soil moisture availability or evapotranspiration (a 'hot/dry' bias). In the Southern Ocean (winter), the band of positive bias likely corresponds to regions where models have less sea ice than ERA5, exposing relatively warm waters to the cold overlying atmosphere and driving excessive upward heat flux.

Caveats

Reference data is ERA5 reanalysis, which itself relies on model parameterizations for surface fluxes.
Biases in sensible heat flux are strongly coupled to latent heat flux errors (not shown); analyzing the Bowen ratio would confirm the dry bias hypothesis.

Total Precipitation Rate Annual Mean Bias

Variables	pr
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	kg/m2/s
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: 0.00 · Rmse: 0.00
IFS-NEMO-ER	Global Mean Bias: -0.00 · Rmse: 0.00
ICON-ESM-ER	Global Mean Bias: 0.00 · Rmse: 0.00
CMIP6 MMM	Global Mean Bias: 0.00 · Rmse: 0.00

Summary high

IFS-NEMO-ER demonstrates the highest fidelity to observations with the lowest RMSE, while ICON-ESM-ER exhibits extensive positive precipitation biases in the tropics and over topography. The substitution of the ocean component (FESOM2 vs. NEMO) in the IFS model significantly alters tropical precipitation patterns, exacerbating dry biases over the Amazon and Maritime Continent.

Key Findings

IFS-NEMO-ER outperforms other models with the lowest RMSE (8.63e-6 kg/m²/s) and minimal global mean bias, showing a reduced double-ITCZ signature compared to CMIP6 MMM.
ICON-ESM-ER displays severe wet biases (RMSE 1.84e-5 kg/m²/s) across the entire Tropical Pacific, Indian Ocean, and major mountain ranges (Andes, Himalayas), suggesting over-active convection.
IFS-FESOM2-SR shows a marked degradation compared to the NEMO configuration, characterized by a strong Indian Ocean dipole bias (wet West, dry East) and a more intense dry bias over the Amazon.

Spatial Patterns

All models show traces of the 'Double ITCZ' bias (wet southern tropical Pacific). ICON-ESM-ER is distinctively wet globally in the tropics. IFS-FESOM2-SR exhibits a strong zonal asymmetry in the Indian Ocean (wet western IO, dry Maritime Continent) and a dry stripe along the equatorial Pacific cold tongue. IFS-NEMO-ER has the weakest bias amplitude but retains some wet bias in the southern tropical Atlantic and Pacific.

Model Agreement

IFS-NEMO-ER agrees best with ERA5. There is significant divergence between models: ICON is generally too wet, while IFS-FESOM2 has specific regional dry/wet dipoles. The CMIP6 MMM captures the structural biases (Amazon drying, double ITCZ) seen in the high-res models but underestimates the magnitude of errors seen in ICON.

Physical Interpretation

The contrast between IFS-NEMO and IFS-FESOM2 highlights the role of SST biases driven by ocean model formulation; FESOM2 likely produces SST errors (e.g., cold bias in the Maritime Continent, warm in West IO) that shift convective centers. ICON's strong wet bias over steep topography (Andes, Himalayas) suggests issues with orographic precipitation parameterization or grid-scale vertical motion at eddy-rich resolutions. The ubiquitous Amazon dry bias usually relates to errors in moisture transport from the Atlantic or land-surface feedbacks.

Caveats

ERA5 precipitation is a reanalysis product and relies on its own model physics, which may influence bias estimates over data-sparse oceans compared to satellite-only products like GPCP.
Units are in kg/m²/s; a bias of 3e-5 is roughly 2.6 mm/day, which is a substantial error for climatological means.

Total Precipitation Rate DJF Bias

Variables	pr
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	kg/m2/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.00 · Rmse: None

Summary high

This figure evaluates DJF precipitation rate biases in high-resolution coupled models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) and the CMIP6 Multi-Model Mean against ERA5 reanalysis. While IFS variants exhibit nearly identical bias patterns dominated by atmospheric physics, ICON displays a distinctly different error structure, particularly in the tropical Pacific and Maritime Continent.

Key Findings

IFS models (FESOM2 and NEMO) show a strong dry bias over the Amazon and Maritime Continent, with a wet bias extending zonally along the equatorial Pacific.
ICON-ESM-ER exhibits a 'double ITCZ-like' bias structure in the Pacific, with a dry equator flanked by wet biases to the north and south, and is wet over the Maritime Continent (opposite to IFS).
All models, including the CMIP6 MMM, struggle to capture the intensity of the Amazon wet season, showing significant dry biases in this region.
The similarity between IFS-FESOM2 and IFS-NEMO confirms that precipitation biases are primarily driven by the atmospheric component (IFS) rather than the ocean model formulation.

Spatial Patterns

The IFS models feature a dipole bias in the Indian Ocean (wet west/dry east) and a zonal wet strip along the Pacific equator. In contrast, ICON shows a dry equatorial Pacific strip with wet off-equator bands and strong wet biases in the Northern Hemisphere storm tracks (North Atlantic and Pacific). Both IFS models and CMIP6 show varying degrees of spurious wet bands in the Southern Ocean/SPCZ region.

Model Agreement

There is extremely high agreement between the two IFS configurations, indicating robustness to ocean resolution/grid choice. There is significant disagreement between IFS and ICON regarding the sign of biases in the Maritime Continent and Equatorial Pacific. However, all models agree on the dry bias over South America.

Physical Interpretation

The persistent Amazon dry bias suggests common deficiencies in land-atmosphere coupling or deep convection triggering over tropical land during the wet season. The contrasting Pacific patterns (IFS wet equator vs. ICON dry equator) point to fundamental differences in convective parameterizations and their interaction with SST gradients (e.g., cold tongue biases). The wet biases in ICON's storm tracks may indicate excessive moisture transport or overly active synoptic systems in the mid-latitudes.

Caveats

ERA5 precipitation is a model-derived product (reanalysis) and shares the same atmospheric model lineage as the IFS simulations, potentially influencing the apparent bias magnitude.
The analysis is limited to DJF (Austral summer/Boreal winter), so seasonal shifts in the ITCZ bias are not captured.

Total Precipitation Rate JJA Bias

Variables	pr
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	kg/m2/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.00 · Rmse: None

Summary high

This figure displays JJA climatological precipitation biases for three high-resolution models and the CMIP6 multi-model mean against ERA5 reanalysis. The analysis highlights significant challenges in representing the Asian Summer Monsoon and the Intertropical Convergence Zone (ITCZ), with distinct error characteristics between the IFS and ICON model families.

Key Findings

All models exhibit a problematic 'dry land / wet ocean' dipole over the South Asian Monsoon region (India/Bay of Bengal vs. Indian Ocean), indicating a failure to propagate monsoon rainfall fully onto the subcontinent.
IFS-NEMO-ER demonstrates the lowest bias magnitudes (palest colors) globally, suggesting that the eddy-rich ocean resolution (compared to the SR configuration) or specific coupling configuration significantly improves precipitation fidelity.
ICON-ESM-ER shows strong positive (wet) biases in the tropical Atlantic and Pacific ITCZ, contrasting with the southward shift/double-ITCZ dipole pattern seen in the IFS models and CMIP6 MMM.
The CMIP6 MMM exhibits classic double-ITCZ biases (wet southern branch, dry northern branch in the Pacific) which are mirrored in IFS-FESOM2-SR but reduced in IFS-NEMO-ER.

Spatial Patterns

The most prominent features are zonal bias dipoles in the tropical Pacific and Atlantic. IFS models and CMIP6 show a negative (dry) bias along the observed ITCZ latitude (~10°N) and a positive (wet) bias just south of it or south of the equator, indicative of a southward ITCZ shift or double-ITCZ bias. ICON, conversely, shows intense positive biases centered on the ITCZ regions. In the Indian Ocean, a meridional dipole exists with dry biases over India/Indochina and wet biases over the equatorial Indian Ocean.

Model Agreement

There is strong inter-model agreement on the sign of the error over the South Asian Monsoon (dry India, wet ocean). However, the models diverge in the tropical convergence zones: IFS models align with the systematic biases seen in CMIP6 (dipole patterns), whereas ICON diverges with excessive precipitation intensity in the main convective bands.

Physical Interpretation

The Indian Monsoon bias suggests dynamical errors in cross-equatorial moisture transport or land-sea thermal contrast, causing precipitation to 'hang up' over the ocean rather than penetrate inland. The Pacific/Atlantic dipoles in IFS/CMIP6 are signatures of the 'Double ITCZ' problem, often linked to warm SST biases in the southeastern tropical oceans. ICON's widespread wet bias suggests overly aggressive convective parameterization. The reduced bias in IFS-NEMO-ER compared to IFS-FESOM2-SR implies that higher ocean resolution (resolving eddies and fronts) may mitigate these coupled systematic errors.

Caveats

ERA5 is a reanalysis product and may have its own uncertainties in precipitation over the open ocean compared to direct satellite retrieval (e.g., GPCP/IMERG).
Biases are climatological means; they do not reveal whether the errors stem from frequency or intensity of precipitation events.

Mean Sea Level Pressure Annual Mean Bias

Variables	psl
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	Pa
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: -5.87 · Rmse: 101.64
IFS-NEMO-ER	Global Mean Bias: 11.27 · Rmse: 100.14
ICON-ESM-ER	Global Mean Bias: -70.87 · Rmse: 498.55
CMIP6 MMM	Global Mean Bias: -0.29 · Rmse: 86.99

Summary high

The figure evaluates annual mean Mean Sea Level Pressure (MSLP) biases, revealing that IFS-based models reproduce global circulation patterns with high fidelity (RMSE ~1 hPa), comparable to the CMIP6 multi-model mean, whereas ICON-ESM-ER exhibits severe zonal circulation biases.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER show excellent agreement with ERA5, with low RMSE values (~100-101 Pa) and biases generally below 2-3 hPa, closely matching the performance of the CMIP6 Multi-Model Mean.
ICON-ESM-ER displays major circulation errors (RMSE ~498 Pa), characterized by a strong negative pressure bias over Antarctica (<-15 hPa) and a positive bias band in the Southern Hemisphere mid-latitudes.
In the Northern Hemisphere, ICON-ESM-ER shows strong positive biases over the North Pacific and North Atlantic, indicating a significant weakening of the Aleutian and Icelandic Lows.
The choice of ocean model (FESOM2 vs. NEMO) has negligible impact on the mean atmospheric pressure field in the IFS coupled system, as both configurations show nearly identical bias patterns.

Spatial Patterns

IFS models exhibit weak, regionally scattered biases, with a slight positive bias band in the Southern Ocean (~50°S). In contrast, ICON-ESM-ER shows distinct zonally symmetric bias structures: a deep pressure deficit over the Antarctic continent bordered by a high-pressure ring in the mid-latitudes, and broad high-pressure anomalies over the northern subpolar oceans.

Model Agreement

There is strong agreement between the two IFS configurations and the CMIP6 MMM. ICON-ESM-ER is a significant outlier with biases nearly 5 times larger than the other models.

Physical Interpretation

ICON-ESM-ER's pattern in the Southern Hemisphere suggests an excessive meridional pressure gradient and likely overly strong zonal westerlies (positive SAM-like bias). In the Northern Hemisphere, the positive biases over the subpolar gyres indicate that ICON fails to sufficiently deepen the semi-permanent low-pressure systems, which would weaken the associated storm tracks. The IFS models successfully capture the intensity and position of these major circulation features.

Caveats

MSLP reduction to sea level over high topography (e.g., Antarctica, Himalayas) can introduce artifacts, though the oceanic biases in ICON are clearly dynamical.
Annual means may obscure seasonal biases in semi-permanent pressure systems (e.g., wintertime lows).

Mean Sea Level Pressure DJF Bias

Variables	psl
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	Pa
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 10.25 · Rmse: None

Summary high

This diagnostic evaluates Mean Sea Level Pressure (MSLP) biases in DJF for three high-resolution models and the CMIP6 multi-model mean against ERA5. The IFS-based models exhibit remarkably low biases compared to ICON-ESM-ER, which shows severe large-scale circulation errors exceeding 15 hPa in major semi-permanent pressure centers.

Key Findings

ICON-ESM-ER displays large-magnitude biases (>1500 Pa) indicating major stationary wave errors: a significantly too-weak Aleutian Low (positive bias) and a too-deep Icelandic Low (negative bias).
IFS-NEMO-ER shows the best performance with minimal biases globally, mostly constrained within ±200 Pa.
IFS-FESOM2-SR shows slightly larger biases than IFS-NEMO-ER, specifically a deepening of the Amundsen Sea Low (negative bias) and higher pressure over Antarctica.
ICON-ESM-ER exhibits a strong circum-Antarctic negative bias band, indicating an overly deep Southern Ocean circumpolar trough.
CMIP6 MMM shows distinct positive biases over high topography (Himalayas, Andes, Greenland) likely related to pressure reduction artifacts, which are less pronounced in the high-resolution IFS/ICON simulations.

Spatial Patterns

In the Northern Hemisphere (DJF), ICON-ESM-ER shows a strong zonal wave-number 2 bias pattern with high pressure anomalies over the North Pacific and low pressure anomalies over the North Atlantic and Arctic. In the Southern Hemisphere, ICON shows a zonally symmetric negative bias around 60°S. The IFS models are much more spatially consistent with observations, though IFS-FESOM2-SR shows a localized negative bias in the Pacific sector of the Southern Ocean.

Model Agreement

There is a strong divergence between the IFS models (high skill) and ICON-ESM-ER (low skill) for this metric. The IFS-NEMO-ER and IFS-FESOM2-SR are much closer to ERA5 than the CMIP6 MMM is in terms of topographic artifacts, but ICON-ESM-ER is an outlier with much larger broad-scale biases than even the CMIP6 mean.

Physical Interpretation

The ICON-ESM-ER patterns suggest systematic errors in the general circulation. The positive North Pacific bias implies a weakened storm track or blocking dominance in the Aleutian sector, while the negative North Atlantic bias suggests a too-zonal, intensified jet stream (positive NAO-like state). The deep Southern Ocean bias in ICON indicates overly vigorous westerlies or excessive cyclogenesis in the circumpolar trough. The difference between IFS-FESOM2 and IFS-NEMO (which share the same atmosphere) points to the influence of ocean coupling (FESOM vs NEMO) on surface pressure patterns, particularly in the Southern Ocean.

Caveats

Biases over high orography (e.g., Himalayas in CMIP6) strongly depend on the method used to reduce surface pressure to sea level and may not represent dynamical errors.
The extreme magnitude of ICON biases suggests this model version may require significant tuning of orographic drag or cloud-radiative processes.

Mean Sea Level Pressure JJA Bias

Variables	psl
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	Pa
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -12.17 · Rmse: None

Summary high

This figure evaluates June-July-August (JJA) Mean Sea Level Pressure (MSLP) biases relative to ERA5 for three high-resolution coupled models and the CMIP6 Multi-Model Mean. The comparison reveals stark differences in Southern Hemisphere circulation simulation between the ICON and IFS model families.

Key Findings

ICON-ESM-ER exhibits severe zonal biases in the Southern Hemisphere, characterized by a massive negative bias over Antarctica (<-15 hPa) and strong positive biases in the subtropics (>10 hPa).
IFS-FESOM2-SR and IFS-NEMO-ER display very similar bias patterns (reflecting their shared atmospheric component), with a moderate positive bias in the Southern Ocean (~5-10 hPa) and negative anomalies near New Zealand.
The CMIP6 MMM shows a weak positive bias in the Southern Ocean, similar in sign to the IFS models but with smaller magnitude, indicating the IFS models are closer to the ensemble consensus than ICON.
In the Northern Hemisphere, ICON generally overestimates pressure over the North Pacific and Atlantic, while IFS models show smaller, regionally localized biases (e.g., negative bias over Europe/Azores).

Spatial Patterns

The dominant spatial feature is the zonal banding in the Southern Hemisphere. ICON shows a distinct 'high-subtropics, low-pole' bias dipole, intensifying the meridional gradient. Conversely, IFS models show a positive bias in the high-latitude Southern Ocean (filling the circumpolar trough) and weaker biases in the subtropics.

Model Agreement

There is strong agreement between the two IFS-based models (FESOM and NEMO), suggesting atmospheric physics dominates these MSLP biases. There is strong disagreement between IFS and ICON, particularly in the representation of the Southern Hemisphere polar trough.

Physical Interpretation

The bias patterns relate directly to the strength of the Southern Hemisphere Westerlies. ICON's enhanced meridional pressure gradient (positive subtropical bias, negative polar bias) implies excessively strong westerly winds and a potential positive SAM-like mean state bias. The IFS models' positive polar bias suggests a 'too weak' circumpolar trough, likely associated with weaker westerlies or an equatorward shift of the storm track.

Caveats

The ICON-ESM-ER biases are exceptionally large (>15 hPa), suggesting potential issues with mass conservation or tuning in this specific simulation.
Differences in orography handling over high terrain (e.g., Himalayas, Antarctica) may contribute to localized MSLP interpolation artifacts.

Surface Downwelling Longwave Annual Mean Bias

Variables	rlds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: 0.02 · Rmse: 6.70
IFS-NEMO-ER	Global Mean Bias: -6.25 · Rmse: 8.20
ICON-ESM-ER	Global Mean Bias: 1.38 · Rmse: 13.02
CMIP6 MMM	Global Mean Bias: 1.00 · Rmse: 6.35

Summary high

The models exhibit distinct biases in surface downwelling longwave radiation: IFS-NEMO-ER shows a systematic global negative bias, IFS-FESOM2-SR is well-balanced globally but with a specific Southern Ocean excess, and ICON-ESM-ER displays a strong latitudinal dipole with large regional errors.

Key Findings

IFS-NEMO-ER underestimates downwelling longwave radiation globally (mean bias -6.25 W/m²), suggesting a widespread deficit in atmospheric emissivity (water vapor or cloud cover).
ICON-ESM-ER has the highest spatial error (RMSE ~13.0 W/m²), characterized by strong positive biases in high latitudes and strong negative biases in tropical/subtropical oceans.
IFS-FESOM2-SR achieves the lowest global mean bias (~0.02 W/m²) but exhibits a prominent positive bias band over the Southern Ocean that is absent in the NEMO configuration.

Spatial Patterns

ICON-ESM-ER shows a clear zonal contrast: excessive downward radiation over land and high-latitude oceans (red), and insufficient radiation over tropical oceans (blue), particularly in marine stratocumulus regions (e.g., off Peru, Namibia, California). The IFS models share similar small-scale features (e.g., along the ITCZ) but differ in background offset; IFS-FESOM2-SR shows a strong positive anomaly in the Southern Ocean, whereas IFS-NEMO-ER is uniformly cooler/negative.

Model Agreement

Inter-model agreement is low regarding spatial structure. The two IFS variants correlate in pattern (except the Southern Ocean) but offset significantly in magnitude. ICON-ESM-ER presents a fundamentally different error structure compared to the IFS models and the CMIP6 MMM.

Physical Interpretation

Downwelling longwave radiation is driven by lower-tropospheric temperature, water vapor, and cloud base emission. ICON's negative tropical biases likely reflect deficits in low cloud cover (marine stratocumulus) or humidity, while its high-latitude positive biases suggest excessive cloudiness or warm lower-atmosphere biases. The contrast between IFS-FESOM2 (positive Southern Ocean bias) and IFS-NEMO (neutral/negative there) highlights the impact of the ocean model coupling on Southern Ocean cloud/surface flux feedbacks. IFS-NEMO's global negative offset implies a systematic dry or clear-sky bias compared to ERA5.

Caveats

Biases are relative to ERA5 reanalysis, which itself relies on model physics for radiative transfer.
The strong Southern Ocean bias in IFS-FESOM2-SR versus IFS-NEMO-ER suggests sensitivity to sea ice or SST coupling in that region.

Surface Downwelling Longwave DJF Bias

Variables	rlds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 1.51 · Rmse: None

Summary high

During boreal winter (DJF), the models exhibit distinct regional biases in surface downwelling longwave radiation (RLDS) relative to ERA5, with ICON-ESM-ER showing the largest deviations. Common features include positive biases over the Southern Ocean and negative biases over North Africa, while polar biases diverge significantly between the IFS configurations.

Key Findings

ICON-ESM-ER displays severe biases: a strong negative deficit (<-40 W/m²) over North Africa and the tropical Atlantic, contrasting with a strong positive excess (>40 W/m²) over the Southern Ocean and Australia.
IFS-FESOM2-SR is characterized by a prominent positive bias over the Antarctic continent (>40 W/m²) and widespread positive biases over the Southern Ocean.
IFS-NEMO-ER generally shows the lowest bias magnitudes globally but exhibits a distinct negative bias (< -20 W/m²) over the Arctic Ocean and North Atlantic.
All models, including the CMIP6 Multi-Model Mean, show positive biases over the Southern Ocean, indicating a systematic systematic overestimation of atmospheric longwave emission in the Southern Hemisphere summer.

Spatial Patterns

The Southern Ocean features a zonally oriented band of positive bias across all models. Tropical arid regions (North Africa, Middle East) consistently show negative biases, most extreme in ICON. High-latitude patterns are model-dependent: IFS-FESOM2 has a strong positive bias over Antarctica, whereas IFS-NEMO has a strong negative bias over the Arctic. ICON shows a hemispheric asymmetry with strong positive anomalies in the SH mid-latitudes and negative anomalies in the tropics.

Model Agreement

IFS-NEMO-ER shows the best agreement with ERA5 in terms of global spatial patterns and magnitude, barring the Arctic. ICON-ESM-ER diverges most significantly, particularly in the tropics and Southern Ocean. Inter-model agreement is highest regarding the sign of the Southern Ocean bias (positive) and North African bias (negative).

Physical Interpretation

Positive biases over the Southern Ocean and Antarctica likely result from excessive cloud fraction or cloud optical depth (potentially phase errors with too much supercooled liquid water) trapping outgoing radiation and re-emitting it downwards. The strong negative biases over North Africa in ICON suggest an overly dry atmosphere or insufficient cloud cover, reducing the greenhouse effect of water vapor. Differences between IFS-FESOM2 and IFS-NEMO (which share atmospheric physics) in polar regions highlight the strong influence of the underlying ocean/sea-ice models (FESOM2 vs. NEMO) on lower tropospheric temperature and cloud formation.

Caveats

ERA5 is a reanalysis product; while robust for radiation, it is not a direct observation.
Biases in downwelling longwave radiation often compensate for shortwave biases (e.g., cloud cover errors affect both), so net radiation errors may differ.

Surface Downwelling Longwave JJA Bias

Variables	rlds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.54 · Rmse: None

Summary high

This figure evaluates biases in JJA Surface Downwelling Longwave Radiation (rlds) relative to ERA5 for three high-resolution models and the CMIP6 multi-model mean. The models exhibit widely diverging bias patterns, ranging from systematic underestimation in IFS-NEMO-ER to strong regional dipoles in ICON-ESM-ER.

Key Findings

IFS-NEMO-ER exhibits a pervasive negative bias (blue) globally, indicating a systematic underestimation of downwelling longwave radiation, particularly over Northern Hemisphere land and the Southern Ocean.
ICON-ESM-ER displays high-amplitude regional contrasts: strong negative biases in tropical convective zones (ITCZ, Amazon, Congo) and strong positive biases over subtropical deserts (Sahara, Australia) and the Southern Ocean.
IFS-FESOM2-SR shows a mixed bias structure that differs significantly from IFS-NEMO-ER despite sharing the same atmospheric component, notably featuring positive biases over the Arctic and parts of the Southern Ocean.
The CMIP6 MMM generally shows positive biases in the Southern Ocean (40°S–60°S) and negative biases in the tropics, a pattern that is spatially exaggerated in ICON-ESM-ER.

Spatial Patterns

Biases are strongly regionally dependent in ICON (convective zones vs. deserts) but more spatially uniform (negative) in IFS-NEMO-ER. The Southern Ocean consistently shows positive biases in ICON, FESOM, and CMIP6, likely linked to cloud cover issues, whereas IFS-NEMO-ER shows negative biases near Antarctica.

Model Agreement

Low inter-model agreement. The two IFS-based models diverge significantly (NEMO is generally 'colder'/less emissive than FESOM), suggesting strong sensitivity to the underlying ocean/sea-ice coupling or state. ICON-ESM-ER has the largest local bias magnitudes.

Physical Interpretation

Downwelling longwave radiation is primarily driven by lower tropospheric temperature, water vapor, and cloud base emissivity. The widespread negative bias in IFS-NEMO-ER suggests a generally colder or drier lower atmosphere, or reduced cloud cover, potentially linked to cold SST biases. The sharp contrasts in ICON-ESM-ER (negative in ITCZ, positive in deserts) point to specific deficiencies in tropical convection parameterization (underestimating cloud radiative effect) and potentially excessive heating or emissivity over arid regions. The Southern Ocean positive bias in most models is consistent with known difficulties in simulating cloud phase and cover in this region.

Caveats

ERA5 is a reanalysis product and may have its own biases in data-sparse regions like the Southern Ocean.
The analysis is limited to JJA (Northern Hemisphere summer); biases may shift seasonally.

Surface Downwelling Shortwave Annual Mean Bias

Variables	rsds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: -1.27 · Rmse: 9.31
IFS-NEMO-ER	Global Mean Bias: -1.00 · Rmse: 8.28
ICON-ESM-ER	Global Mean Bias: 1.03 · Rmse: 14.27
CMIP6 MMM	Global Mean Bias: 4.06 · Rmse: 9.40

Summary high

This figure evaluates biases in annual mean surface downwelling shortwave radiation (rsds) for three high-resolution EERIE models and the CMIP6 multi-model mean against ERA5 reanalysis. The IFS-based models generally exhibit smaller root-mean-square errors (RMSE) and negative oceanic biases, whereas ICON-ESM-ER displays larger biases with distinctive positive polarity over the Southern Ocean and stratocumulus regions.

Key Findings

IFS-based models and ICON show opposite biases in the Southern Ocean: IFS models are negatively biased (too little surface SW, implying excessive cloudiness), while ICON and CMIP6 MMM are positively biased (too much surface SW, implying insufficient cloudiness).
All models consistently overestimate surface downwelling shortwave radiation over major tropical landmasses (Amazon, Central Africa) and Eurasia, indicating a systematic underestimation of continental cloud cover or optical thickness.
ICON-ESM-ER exhibits the highest spatial RMSE (14.27 W/m²) due to strong positive biases in stratocumulus regions (Eastern Pacific, Atlantic) and the Southern Ocean, contrasting with the lower RMSE of IFS-NEMO-ER (8.28 W/m²).

Spatial Patterns

The Intertropical Convergence Zone (ITCZ) appears as a band of negative bias (excessive cloud shielding) in all simulations. Stratocumulus decks off the coasts of Peru, Namibia, and California show strong positive biases in ICON (too much sunlight reaching surface), a classic 'too few low clouds' bias, whereas IFS models handle these regions with smaller or mixed biases. High-latitude biases diverge significantly: the Southern Ocean is a region of excess insolation in ICON/CMIP6 but deficit insolation in IFS.

Model Agreement

IFS-FESOM2-SR and IFS-NEMO-ER show very similar spatial patterns (dominated by the shared atmospheric physics), differing mainly in magnitude. There is inter-model agreement on the positive land bias (Amazon/Africa) and negative ITCZ bias. The models disagree strongly on the sign of the bias in the Southern Ocean and North Pacific.

Physical Interpretation

The positive biases in ICON and CMIP6 over the Southern Ocean and eastern subtropical oceans likely stem from a deficiency in low-level cloud fraction or liquid water path (common in many ESMs). The negative bias in IFS models suggests their cloud parameterizations might be too aggressive in generating optical depth in these regions. The widespread positive land bias suggests all models struggle to maintain sufficient cloud cover over tropical forests.

Caveats

The observational reference is ERA5, which is produced using a version of the IFS model. This likely confers an advantage to the IFS-based simulations (IFS-NEMO/IFS-FESOM) regarding RMSE and spatial pattern similarity.
Annual mean averaging may obscure compensating seasonal errors, particularly in monsoonal regions.

Surface Downwelling Shortwave DJF Bias

Variables	rsds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 2.60 · Rmse: None

Summary high

This figure evaluates biases in DJF surface downwelling shortwave radiation relative to ERA5 for three high-resolution EERIE models and the CMIP6 multi-model mean. The analysis reveals systematic errors in cloud radiative effects, most notably excessive insolation over the Amazon and eastern subtropical oceans across all models.

Key Findings

All models exhibit a strong positive bias (>40 W/m²) over the Amazon basin during the local wet season (DJF), indicating a severe underestimation of cloud cover and consequent excessive solar heating.
IFS-FESOM2-SR and IFS-NEMO-ER show a prominent zonal band of negative bias (<-30 W/m²) in the Southern Ocean (~50°S–60°S), implying these models simulate more optically thick cloud cover than ERA5, contrasting with the positive 'too much sun' bias often found in standard resolution models.
Classic positive biases persist in eastern boundary upwelling regions (off Peru/Chile, Namibia/Angola) across all simulations, reflecting the longstanding challenge of resolving marine stratocumulus decks.
ICON-ESM-ER displays significantly higher amplitude biases than the IFS models, characterized by intense negative biases along the ITCZ/SPCZ (excessive cloudiness) and storm tracks, and strong positive biases in the eastern tropical Pacific.

Spatial Patterns

The IFS variants display nearly identical spatial structures, dominated by the Southern Ocean negative bias and continental positive biases. ICON-ESM-ER shows a more granular and high-contrast pattern, particularly in the tropical Pacific where it exhibits a strong dipole (positive in cold tongue, negative in convergence zones) suggestive of ITCZ structural issues. The CMIP6 MMM broadly captures the continental and subtropical deficits but lacks the sharp Southern Ocean negative band seen in the IFS runs.

Model Agreement

Agreement is exceptionally high between IFS-FESOM2-SR and IFS-NEMO-ER, confirming that these surface radiation biases are dictated by the shared atmospheric physics (OpenIFS) rather than the ocean model formulation. ICON-ESM-ER diverges in bias magnitude and regional morphology, particularly in the storm tracks and tropics.

Physical Interpretation

Positive biases (red) indicate excessive downwelling shortwave radiation, driven by insufficient cloud fraction or optical depth; this is critical over the Amazon (convection parameterization failure) and subtropical eastern oceans (boundary layer cloud deficit). Negative biases (blue), such as those in the IFS Southern Ocean or ICON storm tracks, imply excessive cloud shielding compared to ERA5.

Caveats

Biases are calculated relative to ERA5 reanalysis, which relies on model physics for radiative transfer and has its own uncertainties in surface fluxes.
The color scale saturates at ±40 W/m², potentially masking the true peak magnitude of the massive Amazonian bias in ICON-ESM-ER.

Surface Downwelling Shortwave JJA Bias

Variables	rsds
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	W/m2
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 5.78 · Rmse: None

Summary high

This figure evaluates JJA Surface Downwelling Shortwave Radiation (rsds) biases relative to ERA5. The IFS-based models exhibit moderate biases driven by atmospheric physics (excess insolation over rainforests and boreal land), while ICON-ESM-ER displays high-amplitude zonal and land-sea contrasts, significantly overestimating insolation over NH continents and the Southern Ocean.

Key Findings

IFS-FESOM2-SR and IFS-NEMO-ER share nearly identical bias patterns, confirming that the IFS atmospheric physics dominates the radiative error profile regardless of the ocean coupling.
All models, especially ICON-ESM-ER, show positive biases (excess shortwave reaching surface) over Northern Hemisphere land masses (Eurasia, North America) and tropical rainforests (Amazon, Congo), suggesting underestimated cloud cover or optical depth.
ICON-ESM-ER exhibits a strong dipole bias in the Northern Hemisphere: severe excess insolation over land (+30-50 W/m²) versus strong deficits over the North Pacific and North Atlantic oceans (-30-50 W/m²), indicating issues with land-sea cloud regime partitioning.
The CMIP6 MMM and ICON-ESM-ER show the classic 'too much shortwave' bias over the Southern Ocean, likely due to phase partitioning errors (lack of supercooled liquid clouds), whereas IFS models show reduced biases in this region.

Spatial Patterns

Northern Hemisphere summer (JJA) is characterized by positive biases over continents (Eurasia/NA) and negative biases over the ITCZ and desert regions (Sahara/Arabia) across most models. ICON presents a sharper land-sea contrast with deep blue negative biases over NH ocean basins and dark red positive biases over land. The Southern Ocean shows a zonal band of positive bias in ICON and CMIP6, less pronounced in IFS.

Model Agreement

The two IFS variations agree closely. All models agree on positive biases over the Amazon and Eurasia. Disagreement is strongest in bias magnitude (ICON is much larger) and over the NH oceans (ICON is strongly negative, IFS is mixed/weakly negative).

Physical Interpretation

Positive biases over land imply a 'summer hot bias' mechanism where insufficient cloud cover allows excess solar heating, potentially driving positive feedbacks in soil moisture and temperature. The negative biases over the Sahara and Arabian peninsula (seen in all models) suggest these models may have higher aerosol (dust) optical depth than ERA5, attenuating solar radiation. The ICON bias structure suggests a systematic error in cloud formation, potentially producing too few convective clouds over land and optically thick stratiform clouds over cold oceans.

Caveats

Reference data is ERA5 reanalysis, which itself relies on model physics and assimilation; biases in aerosol representation in ERA5 vs. ESMs can influence the desert comparisons.
The strong negative bias in ICON over NH oceans suggests a specific parameterization issue (e.g., cloud optical properties) that requires specific tuning.

2m Temperature Annual Mean Bias

Variables	tas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	K
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: -0.03 · Rmse: 1.49
IFS-NEMO-ER	Global Mean Bias: -1.34 · Rmse: 1.80
ICON-ESM-ER	Global Mean Bias: -0.38 · Rmse: 2.30
CMIP6 MMM	Global Mean Bias: 0.04 · Rmse: 1.09

Summary high

This figure evaluates annual mean 2m temperature biases relative to ERA5 for three high-resolution models and the CMIP6 multi-model mean (MMM). While the CMIP6 MMM shows the lowest RMSE (1.09 K), the high-resolution models exhibit distinct bias characteristics ranging from systematic global cooling (IFS-NEMO-ER) to high-magnitude regional dipoles (ICON-ESM-ER).

Key Findings

IFS-NEMO-ER suffers from a severe, systematic cold bias globally (mean -1.34 K), which is most extreme in the Arctic (> -6 K).
ICON-ESM-ER displays the largest regional contrasts (highest RMSE of 2.30 K), notably a strong warm bias in the North Atlantic subpolar gyre (> 6 K) and high-latitude continents, versus cold biases in the tropics.
IFS-FESOM2-SR has the smallest global mean bias (-0.03 K) among the single models but features a prominent warm bias surrounding Antarctica (> 4 K).
CMIP6 MMM exhibits classic warm biases in eastern boundary upwelling regions (e.g., off Peru/Chile, Namibia), which appear less pronounced in the IFS-based high-resolution simulations.

Spatial Patterns

The Arctic is a region of major disagreement: IFS-NEMO-ER is excessively cold, while ICON-ESM-ER shows substantial warming over northern high-latitude land and the Labrador Sea. The Southern Ocean consistently shows warm biases in IFS-FESOM2-SR and ICON-ESM-ER, whereas IFS-NEMO-ER remains cold there. The North Atlantic 'warming hole' (cold bias) seen in CMIP6 and IFS-FESOM2-SR is replaced by a strong warm bias in ICON-ESM-ER, suggesting fundamentally different circulation or mixing dynamics.

Model Agreement

There is low inter-model agreement among the high-resolution runs regarding the sign of biases in high latitudes. The CMIP6 MMM provides the most spatially spatially consistent performance (lowest RMSE), benefitting from error cancellation inherent to ensemble averaging. Among the specific models, IFS-FESOM2-SR is closest to observations in the tropics and mid-latitudes but diverges near the poles.

Physical Interpretation

The extreme cold bias in IFS-NEMO-ER's Arctic suggests issues with sea ice parameterization (too extensive/insulating) or cloud radiative forcing. ICON-ESM-ER's localized warm bias in the North Atlantic subpolar gyre is physically distinct and likely indicates an overshoot of the Gulf Stream or excessive deep convection, preventing the formation of the observed 'cold blob'. The Southern Ocean warm biases in IFS-FESOM2 and ICON are typical of coupled models, often attributed to insufficient reflection by supercooled liquid clouds or excessive ocean mixing bringing heat to the surface.

Caveats

The CMIP6 MMM benefits statistically from being an average of many models, smoothing out internal variability that is present in the single-realization high-resolution runs.
The analysis relies on annual means, potentially masking seasonal biases (e.g., wintertime inversions vs summertime melting).

2m Temperature DJF Bias

Variables	tas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	K
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.03 · Rmse: None

Summary high

This figure displays spatial maps of climatological biases in December-January-February (DJF) 2m temperature relative to ERA5 for three high-resolution models (IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER) and the CMIP6 multi-model mean.

Key Findings

IFS-FESOM2-SR and ICON-ESM-ER exhibit severe warm biases (>10 K) over the Antarctic continent and Southern Ocean during austral summer (DJF), whereas IFS-NEMO-ER shows a slight cool bias in this region, indicating strong sensitivity to the ocean/sea-ice component.
Over Northern Hemisphere land masses (boreal winter), models diverge significantly: ICON-ESM-ER shows a strong warm bias (>5-8 K) over North America, while IFS-NEMO-ER displays a widespread cold bias over both North America and Eurasia.
A persistent cold bias is observed in the North Atlantic subpolar gyre region across all models, including the CMIP6 MMM, likely linked to common Sea Surface Temperature (SST) biases in the Gulf Stream extension.
Tropical biases are generally smaller, though ICON-ESM-ER shows a distinct cold bias (~3-5 K) over Northern Africa (Sahara/Sahel) and the Arabian Peninsula.

Spatial Patterns

Biases are most pronounced in polar regions and over continental land masses. The Southern Hemisphere shows a strong contrast between the intense warming in IFS-FESOM2/ICON-ESM and the neutral/cool state of IFS-NEMO. In the Northern Hemisphere, biases follow continental boundaries, with specific regional features like the 'warming hole' cold bias in the North Atlantic.

Model Agreement

Inter-model agreement is low in high latitudes, particularly regarding the sign of the bias over Antarctica and North America. However, there is broad qualitative agreement on the cold bias over the North Atlantic and Greenland (except ICON's mixed Greenland signal). IFS-NEMO-ER is generally the coldest model, while ICON-ESM-ER appears the warmest over land.

Physical Interpretation

The striking difference in Antarctic bias between IFS-FESOM2 and IFS-NEMO (which share the same atmospheric component) points to the sea-ice/ocean model (FESOM vs. NEMO) as the primary driver, likely involving ice albedo or melt processes during austral summer. The warm winter bias in ICON over North America suggests potential issues with snow cover parameterization (albedo feedback) or atmospheric circulation preventing cold air outbreaks. The North Atlantic cold bias is a classical coupled model feature often attributed to weak AMOC or errors in the path of the North Atlantic Current.

Caveats

ERA5 reanalysis has higher uncertainty in polar regions, though the >10 K model biases exceed typical observational error margins.
This analysis is restricted to DJF (austral summer/boreal winter); biases often exhibit strong seasonality and may reverse in JJA.

2m Temperature JJA Bias

Variables	tas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	K
Period	1980–2014
CMIP6 MMM	Global Mean Bias: 0.06 · Rmse: None

Summary high

This figure presents JJA 2-meter temperature biases relative to ERA5 for three high-resolution coupled models and the CMIP6 multi-model mean, highlighting severe and divergent bias patterns across the different modeling systems.

Key Findings

ICON-ESM-ER exhibits extreme warm biases (>4 K) over high-latitude Northern Hemisphere land (Canada, Siberia) and Antarctica, contrasting with strong cold biases over tropical oceans.
IFS-NEMO-ER shows a pervasive global cold bias, most pronounced over the Southern Ocean and Northern Hemisphere continents, deviating significantly from the CMIP6 warm Southern Ocean bias.
IFS-FESOM2-SR displays a distinctive, intense warm bias anomaly in the Weddell Sea sector of the Southern Ocean, likely indicating a recurring polynya or sea-ice deficit, while remaining colder elsewhere.
CMIP6 MMM biases are generally lower in magnitude and spatially distinct (e.g., warm US Great Plains, warm Southern Ocean) compared to the specific, higher-amplitude biases of the high-resolution simulations.

Spatial Patterns

Spatial contrasts are stark: ICON shows a land-warm/ocean-cold pattern; IFS-NEMO is dominated by global cooling; IFS-FESOM is spatially heterogeneous with specific high-latitude anomalies. All models struggle with topography-related biases over the Himalayas/Tibetan Plateau.

Model Agreement

Inter-model agreement is notably poor. The models do not converge on a common bias pattern, differing in sign over major regions (e.g., NH land is strongly warm in ICON but cold in IFS-NEMO).

Physical Interpretation

ICON's NH summer warm bias suggests issues with land-surface coupling (e.g., soil moisture deficits limiting evaporative cooling) or insufficient cloud radiative reflection. The Weddell Sea warm bias in IFS-FESOM is a signature of excessive heat flux from the ocean to the atmosphere during austral winter (JJA), caused by missing sea ice (open polynya). IFS-NEMO's widespread cold bias suggests a systematic drift in global energy balance or ocean surface cooling efficiency.

Caveats

The extreme magnitude of ICON's warm bias suggests a potential configuration or tuning issue rather than structural resolution limits.
JJA represents austral winter, meaning Antarctic biases are driven by different physics (sea ice insulation) than Arctic biases (insolation/albedo).

10m U Wind Annual Mean Bias

Variables	uas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: -0.01 · Rmse: 0.54
IFS-NEMO-ER	Global Mean Bias: -0.11 · Rmse: 0.41
ICON-ESM-ER	Global Mean Bias: 0.26 · Rmse: 1.67
CMIP6 MMM	Global Mean Bias: 0.01 · Rmse: 0.62

Summary high

This diagnostic evaluates annual mean 10m zonal wind biases, revealing that IFS-NEMO-ER significantly outperforms other models with minimal bias, while ICON-ESM-ER shows a systematic overestimation of wind strength and IFS-FESOM2-SR exhibits latitudinal shift biases similar to the CMIP6 mean.

Key Findings

IFS-NEMO-ER shows exceptional agreement with ERA5 (RMSE ~0.41 m/s), largely eliminating the systematic errors seen in other models.
ICON-ESM-ER exhibits a severe systematic bias (RMSE ~1.67 m/s) where surface winds are excessively strong globally: trade winds are too easterly (blue tropical bias) and mid-latitude westerlies are too westerly (red extratropical bias).
IFS-FESOM2-SR and CMIP6 MMM both display an equatorward shift of the Southern Hemisphere westerly jet, characterized by a dipole bias pattern (too westerly around 45°S, too easterly/weak around 60°S).
Tropical trade winds are simulated as too weak in IFS-FESOM2-SR and CMIP6 MMM (positive/westerly bias), but too strong in ICON-ESM-ER (negative/easterly bias).

Spatial Patterns

The dominant spatial feature is the zonal banding in the bias maps. IFS-FESOM2-SR and CMIP6 MMM show a distinct positive bias band in the Southern Ocean (~40-50°S) and negative bias poleward of that, indicating an equatorward displacement of the storm track. ICON-ESM-ER's bias map effectively mirrors the climatological wind pattern (red in westerly regions, blue in easterly regions), indicating a global amplification of wind speeds rather than a spatial shift.

Model Agreement

IFS-NEMO-ER is the outlier in terms of high skill, showing very little structure in its bias map. IFS-FESOM2-SR aligns more closely with the CMIP6 multi-model mean in terms of bias structure (SH jet shift, weak trades), likely reflecting standard resolution limitations. ICON-ESM-ER diverges from the group with its unique high-wind-speed regime.

Physical Interpretation

The equatorward shift of the SH westerlies in IFS-FESOM2-SR and CMIP6 is a classic bias in lower-resolution coupled models, often linked to insufficient resolution of ocean thermal fronts or atmospheric eddy-mean flow interactions. The superior performance of IFS-NEMO-ER suggests that its eddy-rich resolution or specific coupling physics successfully anchors the jet poleward. The ICON-ESM-ER bias (winds too strong everywhere) suggests a fundamental issue with surface drag parameterization or boundary layer coupling, resulting in insufficient momentum dissipation.

Caveats

The magnitude of the ICON-ESM-ER bias is unusually large for a modern ESM, raising the possibility of a configuration issue or a discrepancy in the diagnostic height (e.g., reporting winds at a level higher than 10m).
Comparison of 'SR' (Standard Resolution) and 'ER' (Eddy Rich) variants implies resolution dependency, but the specific resolution of the ocean component in IFS-FESOM2-SR relative to IFS-NEMO-ER is a key factor in the SH jet position.

10m U Wind DJF Bias

Variables	uas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -0.05 · Rmse: None

Summary high

This diagnostic evaluates DJF 10m zonal wind biases relative to ERA5. The IFS models demonstrate a poleward shift of the Southern Hemisphere westerlies, contrasting with the classic equatorward bias seen in the CMIP6 multi-model mean, while ICON-ESM-ER exhibits excessively strong zonal circulation globally.

Key Findings

ICON-ESM-ER shows severe systematic biases with significantly overpowered circulation features: mid-latitude westerlies are too strong (positive bias) and tropical trade winds are too strong (negative bias adds to negative climatology).
IFS models (both FESOM2 and NEMO) exhibit a dipole bias in the Southern Ocean (negative around 45°S, positive around 60°S), indicating a poleward shift of the westerly jet compared to ERA5.
The CMIP6 MMM displays the opposite Southern Ocean bias pattern (positive at 45°S, negative at 60°S), reflecting the well-known systematic equatorward jet bias in standard-resolution models.
IFS-NEMO-ER performs best overall, showing the lowest bias magnitudes, particularly in the North Atlantic and Tropics, though it shares the poleward shift tendency with the FESOM2 configuration.
Tropical Trade winds are generally too weak (westerly bias) in the IFS models but too strong in ICON-ESM-ER.

Spatial Patterns

The dominant spatial feature is the zonal banding of biases. In the Southern Hemisphere, this manifests as dipoles indicating meridional shifts of the westerlies. In ICON, the pattern mimics the climatology itself (amplifying it). The IFS-FESOM2-SR shows a noticeable westerly bias (red) in the tropical Pacific and Atlantic, indicating weakened trade winds.

Model Agreement

There is significant inter-model divergence. IFS and CMIP6 disagree on the direction of the Southern Hemisphere jet shift (poleward vs. equatorward). ICON disagrees with all other models regarding the intensity of the general circulation, showing much higher wind speeds globally.

Physical Interpretation

The contrast in Southern Ocean jet latitude likely relates to horizontal resolution; higher resolution (IFS-ER/SR) typically reduces the equatorward bias found in lower-resolution models (CMIP6), though here IFS appears to overshoot slightly poleward. ICON's globally excessive wind speeds suggest potential issues with surface drag parameterization, boundary layer momentum mixing, or coupling fluxes, as the atmosphere appears overly energetic. The weakened trades in IFS may point to issues in the Walker circulation intensity or tropical coupling.

Caveats

The extreme magnitude of ICON biases makes fine-scale comparison with other models difficult on the same color scale.
Biases are for DJF only; seasonal shifts in other months might display different error characteristics.

10m U Wind JJA Bias

Variables	uas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -0.01 · Rmse: None

Summary high

This diagnostic evaluates JJA 10m zonal wind biases relative to ERA5, revealing systematic errors across all models in the strength of tropical trade winds and the latitudinal position of the Southern Hemisphere westerlies.

Key Findings

ICON-ESM-ER exhibits the most severe biases, characterized by drastically weakened tropical trade winds (strong positive bias >3 m/s in easterly regions) and a substantial equatorward shift of the Southern Hemisphere westerly jet.
IFS-FESOM2-SR and IFS-NEMO-ER display nearly identical spatial bias patterns, indicating the errors are primarily driven by their shared OpenIFS atmospheric component rather than the ocean model; their bias magnitudes are generally lower than ICON's.
A distinct dipole bias pattern exists in the Southern Ocean across all models (including CMIP6 MMM), with positive biases around 45°S and negative biases poleward (60°S), signaling a systematic equatorward displacement of the westerly wind belt compared to observations.

Spatial Patterns

The dominant spatial features are zonally coherent bands: extensive positive biases (red) throughout the tropical Pacific and Atlantic indicating weak easterlies, and a dipole structure in the Southern Hemisphere mid-latitudes (red equatorward/blue poleward). Comparison with the observation panel confirms these represent a weakening of the trades and a northward shift of the 'Roaring Forties'.

Model Agreement

There is high structural agreement between models (and the CMIP6 MMM) regarding the location and sign of biases, suggesting common physical deficiencies. However, they disagree on amplitude, with ICON-ESM-ER showing significantly stronger errors than the IFS-based configurations.

Physical Interpretation

The positive bias in the tropical easterlies implies a weak surface component of the Walker and Hadley circulations. The meridional dipole in the Southern Ocean reflects a classic 'equatorward jet bias,' common in coupled models, where the eddy-driven jet is positioned too far north, potentially due to errors in SST gradients or cloud-radiative feedbacks.

Caveats

The CMIP6 MMM panel shows stippling/texture which likely indicates regions of model agreement or grid interpolation artifacts.
Biases are evaluated against ERA5, which is a reanalysis product but highly reliable for surface winds.

10m V Wind Annual Mean Bias

Variables	vas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
IFS-FESOM2-SR	Global Mean Bias: -0.01 · Rmse: 0.46
IFS-NEMO-ER	Global Mean Bias: 0.03 · Rmse: 0.34
ICON-ESM-ER	Global Mean Bias: -0.06 · Rmse: 0.85
CMIP6 MMM	Global Mean Bias: -0.05 · Rmse: 0.54

Summary high

This diagnostic evaluates annual mean 10m meridional (V) wind biases relative to ERA5. IFS-NEMO-ER demonstrates superior performance with the lowest RMSE (0.34 m/s), while ICON-ESM-ER and CMIP6 MMM exhibit significant systematic biases in the tropical trade wind belts indicative of weakened convergence.

Key Findings

ICON-ESM-ER displays the largest errors (RMSE 0.85 m/s), characterized by a strong dipole bias across the tropics (positive anomaly north of the equator, negative south) that opposes the climatological flow.
IFS-NEMO-ER shows excellent agreement with observations, lacking the systematic tropical biases seen in other models and achieving a global RMSE significantly lower than the CMIP6 ensemble mean.
IFS-FESOM2-SR exhibits similar tropical bias patterns to ICON and CMIP6 (weakened trade winds) but with reduced magnitude (RMSE 0.46 m/s) compared to the CMIP6 baseline.

Spatial Patterns

The dominant error pattern in ICON-ESM-ER and CMIP6 MMM is a zonal dipole in the tropical Pacific and Atlantic: red (positive) biases in the Northern Hemisphere trades and blue (negative) biases in the Southern Hemisphere trades. IFS-NEMO-ER avoids this zonal structure, showing only weaker, regionally scattered biases.

Model Agreement

There is significant divergence in model performance. IFS-NEMO-ER agrees well with ERA5, whereas ICON-ESM-ER diverges substantially in the tropics. IFS-FESOM2-SR offers an improvement over the CMIP6 MMM baseline but shares its structural biases.

Physical Interpretation

The tropical bias pattern in ICON-ESM-ER (positive bias in northerly NH trades, negative bias in southerly SH trades) represents a weakening of the meridional wind component. This indicates insufficient surface convergence into the Intertropical Convergence Zone (ITCZ), a systematic error often linked to weak Hadley circulation intensity or double-ITCZ biases in coupled models.

Caveats

Meridional wind biases near the ITCZ are highly sensitive to small latitudinal shifts in the convergence zone, which can manifest as large dipole errors.
The 'texture' in the CMIP6 MMM plot may indicate grid interference or stippling not explicitly defined in the legend.

10m V Wind DJF Bias

Variables	vas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -0.07 · Rmse: None

Summary high

This figure evaluates DJF 10m meridional (V) wind biases relative to ERA5. A striking contrast exists between models: ICON-ESM-ER exhibits a systematic, global overestimation of wind speeds, while IFS variants show weaker circulation features, particularly in the trade winds.

Key Findings

ICON-ESM-ER displays a large-scale bias pattern that mirrors the climatological flow, indicating a systematic overestimation of meridional wind magnitude (both northerlies and southerlies are too strong) by >2 m/s globally.
IFS-FESOM2-SR and IFS-NEMO-ER exhibit similar bias structures, primarily characterized by a positive (southerly) bias in the tropical North Pacific, which indicates an underestimation of the strength of the climatological northerly NE trade winds.
IFS-NEMO-ER shows slightly reduced bias magnitudes compared to IFS-FESOM2-SR, particularly in the Southern Ocean and North Atlantic.
CMIP6 MMM biases are spatially smoother but show distinct tropical banding (zonal asymmetry) typical of ITCZ positioning errors, comparable in magnitude to the IFS biases.

Spatial Patterns

The ICON bias map essentially replicates the observation map's sign structure (blue bias in blue regions, red bias in red regions), implying an amplification of the mean flow. The IFS models show a zonal band of positive bias around 5-15°N in the Pacific and Atlantic, opposing the climatological northerly flow. Eastern boundary currents (e.g., Peru/Chile) show strong positive (southerly) biases in ICON, strengthening the coastal jet.

Model Agreement

High agreement between the two IFS configurations (FESOM2 vs NEMO) regarding the spatial distribution of errors. Strong divergence between IFS (winds too weak) and ICON (winds too strong).

Physical Interpretation

The systematic overestimation in ICON-ESM-ER suggests potential issues with surface drag parameterization or momentum coupling, leading to excessive surface wind speeds. Conversely, the IFS positive bias in the winter hemisphere (NH) trades suggests a slightly sluggish Hadley circulation inflow or insufficient momentum mixing in the boundary layer. The strong meridional biases along eastern boundaries in ICON would drive excessive Ekman upwelling.

Caveats

The ICON-ESM-ER bias is so strongly correlated with the mean field that it warrants verification of surface drag parameters or potential unit/scaling issues in the output.

10m V Wind JJA Bias

Variables	vas
Models	IFS-FESOM2-SR, IFS-NEMO-ER, ICON-ESM-ER, CMIP6 MMM
Obs Dataset	ERA5
Units	m/s
Period	1980–2014
CMIP6 MMM	Global Mean Bias: -0.02 · Rmse: None

Summary high

This diagnostic evaluates JJA 10m meridional wind (V) biases relative to ERA5. The comparison highlights systematic errors in the Asian Summer Monsoon circulation and significant differences in trade wind representation between the IFS and ICON model families.

Key Findings

All models, including the CMIP6 MMM, systematically overestimate the strength of the Somali Jet and South Asian Summer Monsoon flow, indicated by prominent positive (southerly) biases in the Arabian Sea and western Indian Ocean.
ICON-ESM-ER exhibits the largest biases, appearing to significantly overestimate the strength of the zonally averaged Hadley circulation, with excessive northerly flow (blue bias) in the NH tropics and excessive southerly flow (red bias) in the SH tropics reinforcing the trade winds.
IFS-NEMO-ER demonstrates the best performance with the lowest amplitude biases globally.
IFS-FESOM2-SR shares spatial bias structures with IFS-NEMO-ER (reflecting their common atmospheric component) but generally exhibits higher bias magnitudes, particularly in the Southern Ocean and Indian Ocean.

Spatial Patterns

The dominant spatial feature is the intensification of the monsoon circulation (positive bias in the western Indian Ocean). In the tropical Pacific and Atlantic, ICON-ESM-ER displays a dipole bias pattern (negative north of equator, positive south) that aligns with and amplifies the climatological trade wind convergence zone. The IFS models show more regionally localized biases, such as a positive bias band in the North Atlantic and Southern Hemisphere mid-latitudes.

Model Agreement

There is strong inter-model agreement on the sign of the bias in the Indian Ocean (too strong southerly monsoon flow). The IFS variants agree closely on spatial patterns, differing primarily in magnitude. ICON-ESM-ER is an outlier, showing much stronger, basin-scale biases that suggest overly energetic meridional overturning in the tropics.

Physical Interpretation

The positive bias in the Somali Jet region suggests that the models simulate an overly energetic cross-equatorial flow or an overly deep monsoon trough in JJA. The stark difference between ICON and IFS suggests that atmospheric core formulation plays a larger role in these wind biases than ocean resolution, though the degradation from IFS-NEMO to IFS-FESOM implies that air-sea coupling or SST biases in the FESOM configuration may be amplifying atmospheric errors. The 'dipole' bias in ICON suggests a 'Double ITCZ' or overly strong ITCZ convergence mechanism.

Caveats

The ICON-ESM-ER biases are saturated at the colorbar limits (+/- 2 m/s) over large areas, indicating errors significantly larger than the other models.
CMIP6 MMM stippling creates visual artifacts that may obscure finer bias details.

Evaluation Global Climatology Biases CMIP6

CMIP6 Multi-Model Mean Context

Synthesis

Related diagnostics

Total Cloud Cover Annual Mean Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Total Cloud Cover DJF Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Total Cloud Cover JJA Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Latent Heat Flux Annual Mean Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Latent Heat Flux DJF Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Latent Heat Flux JJA Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Sensible Heat Flux Annual Mean Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Sensible Heat Flux DJF Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Surface Sensible Heat Flux JJA Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Total Precipitation Rate Annual Mean Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation

Caveats

Total Precipitation Rate DJF Bias

Summary high

Key Findings

Spatial Patterns

Model Agreement

Physical Interpretation