Impact of the ocean mesoscale on the atmospheric mean state

Deliverable D7.1 Impact of the ocean mesoscale on the atmospheric mean state Deliverable ID 7.1 Work Package Reference WP7 Issue v1.0 Due Date of Deliverable 30/09/2025 Submission Date 26/09/2025 Dissemination Level Public Lead Partner UOXF Contributors Kristian Strommen, Chris Roberts, Matthias Aengenheyster, Bianca Mezzina, Ipshita Dey and the rest of the WP7 members. Grant Agreement No 101081383 Call ID HORIZON-CL5-2022-D1-02-02 Page 2 of 76 Prepared by Reviewed by Approved by Kristian Strommen, Chris Roberts, Matthias Aengenheyster, Bianca Mezzina, Ipshita Dey, and other WP7 members SSC ECT Issue Date Description Author 1.0 26/09/2025 First issue of the deliverable Kristian Strommen (UOXF) Christopher Roberts (ECMWF) Matthias Aengenheyster (ECMWF) Ipshita Dey (UOR) Bianca Mezzina (ECMWF) Page 3 of 76 Index Index 3 Executive Summary 4 1. Introduction 5 1.1. Background and motivation 5 1.2. The goals of this report 5 2. Data and Methods 6 2.1. Model data 6 2.1.1. EERIE coupled models 7 2.1.2. Sensitivity experiments 8 2.1.3. CMIP multimodel data 9 2.2. Observational data 9 2.3. Methods 10 3. Assessment of the mean state and biases 10 3.1. Sea surface temperatures and sea ice 11 3.2. Surface temperature and precipitation 16 3.2.1. Global biases 17 3.2.2. Gulf Stream Precipitation 25 3.2.3. The double ITCZ bias 27 3.3. Vertical profiles of temperature and zonal winds 28 3.4. Energy and water budgets 31 3.5. The stratospheric polar vortex 35 3.6. Tropical and extratropical cyclones 36 3.7.1 Tropical cyclones 39 3.7.2 Extratropical cyclones 42 4. The role of eddies in shaping the mean state 45 4.1 Background 46 4.2. Local atmospheric response to ocean eddies 47 4.3. Impact of extratropical ocean eddies on the atmospheric mean state 51 4.3.1. Assessment of basic impact 51 4.3.3. Gulf Stream precipitation 54 4.3.3. Remote changes and dynamical response to eddies 56 4.4. Summary 60 5. Discussion and Conclusions 61 References 63 Supplementary Material 72 Page 4 of 76 Executive Summary We examine the atmospheric mean state in the four EERIE eddy-rich coupled models (IFS-FESOM, IFS-NEMO, ICON and HadGEM), focusing on the historical period 1979-2015. The biases in key variables such as two-meter temperature, precipitation and zonal winds are, with the exception of ICON, comparable to the biases of past CMIP6 models at lower resolution. IFS-FESOM and HadGEM perform notably well, consistently outperforming the CMIP6 median. IFS-NEMO has a small but consistent cold bias, which is sufficient to make it slightly colder than the coldest CMIP6 model, though the bias at any given gridpoint is small. ICON has particularly large biases in the atmospheric mean state, likely because it does not use convection parameterisations, which results in large biases in the tropics that affect the entire global circulation. However, all models, ICON included, simulate the global top-of-atmosphere energy budget very well, placing them at the forefront of CMIP models in this respect. No clear improvement to the double ITCZ bias is found in the EERIE simulations. The direct effect of mesoscale eddies on the atmospheric mean state is estimated using the smoothed-SST experiments and compared against the signal in the coupled models. This analysis suggests that the direct effect of mesoscale eddies on the atmospheric mean state is small, though some consistent effects are identified. Firstly, the presence of eddies results in enhanced latent heat fluxes (~4 to 10 W/m2) and precipitation (~0.3 to 0.4 mm/day) over eddy-rich regions such as the Gulf Stream and Kuroshio. The enhanced precipitation appears to trigger robust shifts in the North Atlantic and North Pacific jets, though the signals are confined to particular seasons and hard to detect when assessing extratropical cyclone tracks. Secondly, a reduction in total cloud cover of roughly 1-2% is found over the southern ocean. These changes are supported by eddy composite analysis, which show that eddies have a clear local impact on the atmosphere. In all these cases we emphasise that the magnitude of the changes are modest compared to interannual and inter-model variability. In particular, they are much smaller than the spread between the coupled EERIE models, and sometimes opposite to the direction of change between the coupled EERIE models and CMIP models at lower resolution. This implies that the dominant impact of ocean eddies on the atmosphere likely arises indirectly, via the way they change the large-scale SST biases in models by modulating ocean heat and salinity transports, rather than via the direct forcing attributable to transient mesoscale eddies. Page 5 of 76 1. Introduction 1.1. Background and motivation The EERIE eddy-rich coupled “frontier” simulations are now complete. These comprise the first ever set of HighResMIP-style climate simulations of the period 1950-2015 at ~10km atmospheric horizontal resolution and an ocean resolution high enough to resolve mesoscale eddies, and thus offer a unique opportunity to assess the effect of resolving mesoscale eddies on the climate of the atmosphere. While it might be anticipated that some of the more exciting impacts are on atmospheric variability, for example through the generation of particularly intense heatfluxes across eddy-generated fronts, it is essential to begin by assessing the atmospheric mean state in the eddy-rich simulations. This is because, firstly, in many cases the variability itself is fundamentally dependent on the mean state, and thus the impact of eddies on the atmosphere cannot be clearly understood or attributed without taking into account the model mean state. Secondly, assessing the role of mesoscale eddies on the atmosphere requires a comparison against models run at lower, non-eddying resolutions. Because of the unprecedentedly high atmospheric and oceanic resolution of the EERIE simulations, there is no a priori guarantee that the atmospheric mean state biases of the eddy-rich simulations are comparable to those of lower resolution models (e.g., CMIP6). Indeed, due to the importance of model tuning in setting the mean state, and the difficulty of tuning high resolution models, the atmospheric mean state biases could in theory be substantially larger than those of lower resolution models. Thus a thorough assessment of the atmospheric mean state biases in the EERIE simulations is an important first step when trying to understand the effect of mesoscale ocean eddies on the atmosphere. 1.2. The goals of this report There are two main goals of this report: 1. To document the EERIE model atmospheric mean state biases and compare them to typical CMIP-model biases. Is there a step change in the simulated climates in EERIE models or are they roughly comparable to CMIP6? 2. Discuss which atmospheric mean state changes seen in EERIE models compared to non-eddying CMIP models may be related to the presence of meso-scale ocean eddies. Page 6 of 76 For the first goal, we will show and discuss the EERIE biases, and systematically compare these against CMIP6 biases. For the second goal, we rely on two methods. Firstly, we look at how the bias of interest varies with resolution across a wide range of coupled models spanning from CMIP5 to HighResMIP and EERIE, in order to pick out biases that are sensitive to resolution, and also identify where the performance of EERIE models appears to be notably closer to observations or not. Secondly, we make use of the sensitivity experiments carried out in WP9, described in Deliverable Reports 9.3 and 9.4, which explicitly isolate the direct effect of eddies in targeted AMIP-style simulations. The strength of the first method is it allows for a large ensemble size spanning a wide range of resolutions, but its weakness is that many of the relationships one finds there will be due to changes in atmospheric resolution as opposed to better resolved mesoscale eddies. The strength of the second method is that it explicitly detects the impact of eddies, but its weakness is that the ensemble size is small and it only uses two distinct climate models. These two methods thereby complement each other. We emphasise that the sensitivity experiments are designed to quantify the direct effect of the eddies given the same underlying ocean mean state. Thus they do not quantify the indirect effect the eddies may have on modulating the ocean mean state in the models, and as such will underestimate the total effect of resolving mesoscale eddies. This is discussed further at the end. For the sake of brevity, the focus of this report is in general on the winter season December-January-February (DJF) and the summer season June-July-August (JJA), though in some cases all seasons are considered. 2. Data and Methods 2.1. Model data We use the EERIE coupled model Frontier Simulations for the historical period 1950-2015, with a focus on the satellite period 1979-2015 for validation against observations. We also make use of the AMIP-style sensitivity experiments. We now briefly describe both sets of experiments. Further details are given in Deliverable D4.1. Page 7 of 76 2.1.1. EERIE coupled models The four EERIE models are ICON (MPI-M), IFS-FESOM (AWI), HadGEM-GC5E (MOHC; we refer to this as just “HadGEM” for short) and IFS-NEMO (BSC). ICON (“ICOsahedral Nonhydrostatic”) is the new framework for the MPI-M Earth system model, described by Hohenegger et al. (2023) and Jungclaus et al. (2022). The ocean component solves the primitive equations (hydrostatic) using conservative schemes on a grid with triangular elements and C-type staggering (Korn, 2017). The sea ice model is based on the finite element sea ice model FESIM (Danilov et al., 2015). A global configuration of ICON-O with a nominal resolution of 10km has been described in (Korn et al., 2022). The atmospheric model uses a nonhydrostatic dynamical core, and is described in Cruger et al. (2018) and Giorgetta et al. (2018). Notably, the configuration of ICON involves turning off all convection parameterisations (the so-called “sapphire” configuration of the model; Hohenegger et al., 2023). IFS-FESOM uses version 2.5 of FESOM (Finite Element Sea Ice Ocean Model) as its ocean component, configured with an NG5 unstructured triangular grid and 70 depth levels. This configuration provides a minimum horizontal resolution of 5 km in eddy-rich midand high-latitude regions and a resolution of approximately 13 km over the tropics (Rackow et al. 2025). The sea ice model is FESIM (Danilov et al. 2015). The atmospheric component is the Integrated Forecast System (IFS) cycle 48r1, developed by the European Center for Medium-range Weather Forecasts (ECMWF), a spectral model run on the triangular-cubic octahedral grid Tco1279 (~10km) with 137 vertical levels. The IFS model configuration closely follows Rackow et al. (2025), except for the treatment of deep convection. While Rackow et al. (2025) used a modified scheme with reduced cloud-base mass flux (see Rackow et al., 2025), we applied the operational IFS treatment. HadGEM3-CG5 is the latest version of the HadGEM family developed by the UK Met Office Hadley Center (Xavier et al., 2024, in prep.) The ocean-ice component is based on the NEMO modelling framework version 4.04. The configuration is GOSI9 (Guiavarc’h et al., 2024). It includes the sea ice model SI3 (Sea Ice modelling Integrated Initiative, Blockley et al., 2024). We focus here on the HH (high atmospheric resolution, high oceanic resolution) version. The IFS-NEMO model uses version 4.0.7 of the ocean model component NEMO, which includes the sea ice model SI3, both running on a tripolar global orthogonal curvilinear mesh with 75 vertical levels and a horizontal resolution of 1/12° (ORCA12), coupled to the same version of the IFS used by IFS-FESOM. It thus differs from IFS-FESOM only in its oceanic component. Coupled simulations with ICON, IFS-NEMO, and IFS-FESOM were carried out following the HighResMIP protocol (Haarsma et al. 2016). In line with this protocol, a 50-year coupled spin-up was performed using 1950 CMIP6 radiative forcing. From the final state of the spin-up, two Page 8 of 76 simulations were initiated in parallel: a control run and a historical run. The historical run applied CMIP6 historical forcing from 1950 to 2014 and was integrated for 65 years. The coupled simulations with HadGEM3-GC5 used the CMIP6 protocol (Eyring et al. 2016), with a 200-year spinup with 1850 CMIP6 forcing, followed by a 200-year piControl run and the historical run integrated from 1850-2014. The analysis in this report primarily draws on data from the historical simulation. 2.1.2. Sensitivity experiments We help isolate the impact of ocean eddies on the large-scale atmospheric circulation using idealised high-resolution global atmosphere simulations conducted with higher and lower resolution configurations of the IFS and HadGEM. The main novelty of these simulations compared to past work is the comparison of results from two state-of-the-art global atmospheric models at unprecedented horizontal resolutions, including multi-decadal IFS simulations with an average horizontal grid spacing of ~9km, which matches the resolution of ECMWF operational medium-range weather forecasts. These coordinated simulations are designed to isolate the response of the large-scale atmospheric circulation (in the absence of ocean feedbacks) to the direct thermodynamic forcing attributable to extratropical SST anomalies associated with transient mesoscale features. In this deliverable, we explore the impact of extratropical ocean eddies using the following EERIE atmosphere-only experiments: ● Reference: EERIE atmosphere-only reference simulations with ocean and sea-ice boundary conditions specified using high-resolution satellite-based estimates of daily-mean sea surface temperatures (SST) and sea-ice concentration. ● SmoothAnom: As Reference, but daily-mean SST boundary conditions are modified to suppress transient mesoscale SST anomalies in the extratropics. Climatological SSTs are not changed. Full details of the IFS and HadGEM3 model configurations and experimental design are provided in EERIE deliverables 9.3 and 9.4. Page 9 of 76 2.1.3. CMIP multimodel data In order to place EERIE model biases in context, and to explore the effect of increased model resolution, we leverage existing CMIP coupled model data. We use data drawn from a total of 135 historical model simulations, consisting of multiple ensemble members drawn from a range of CMIP6 models (Eyring et al. 2016), CMIP5 models (Taylor et al. 2012), and High-Resolution Model Intercomparison Project (HighResMIP) models (Haarsma et al. 2016). The CMIP5 and CMIP6 historical experiments consist of coupled uninitialized climate runs forced with historical greenhouse gas and aerosol forcings over the twentieth century, after a spinup from a free-running preindustrial control run. The HighResMIP models were initialized in 1950, following a short 50-yr spinup, and they span the 65 years between 1950 and 2015 and use the same historical forcings as CMIP6. The data are in all cases restricted to the December–February seasons between 1979–01 and 2014–12. Thus, CMIP5 data cover 1979–2005 and CMIP6 and HighResMIP cover 1979–2014. The HighResMIP models are sometimes referred to as the “Process-Based Climate Simulation: Advances in High-Resolution Modeling and European Climate Risk Assessment (PRIMAVERA)” models, in reference to the European Union Project under which they were produced. Tables detailing the exact models and ensemble members used are given in Tables 1, 2 and 3 of the Supplementary Material. In figures of spatial biases, we often show the “CMIP6 median bias”. This involves determining the median across all the CMIP6 models at each individual gridpoint separately. The drawback of this method is that some of the links between different biases will be obfuscated. However, many large scale biases are relatively systematic across models, so this obfuscation is perhaps not too severe. 2.2. Observational data Our observational product for precipitation is GPM-IMERG (Huffman et al. 2019), a data set which combines satellite measurements to create a gridded product at 0.1 degree resolution spanning 2000 to present. For top-of-the-atmosphere (TOA) radiative quantities, we use CERES-EBAF (Loeb et al. 2018), also spanning 2000 to present. For other quantities we use the reanalysis product ERA5 (Hersbach et al. 2020). For comparison against ERA5 variables the data is always restricted to 1979-2015, while for comparison against GPM-IMERG and CERES the data is restricted to 2000-2015. When assessing tropical and extratropical cyclones, the Japanese Reanalysis for Three Quarters of a Century (JRA-3Q; Kosaka et al. 2024) is also used, along with Page 16 of 76 To get a better sense of how the EERIE model biases in SST and SIC compare to typical CMIP6 biases, we show in Figure 6 histograms of the absolute bias of these variables, pooled across all gridpoints: for CMIP6 we have combined all models to make a single histogram. For SSTs, HadGEM and IFS-FESOM perform relatively well, and ICON performs poorly. Despite having a consistent cold bias, IFS-NEMO’s biases at any single gridpoint are entirely comparable to typical CMIP6 biases: its overall cold bias is thus a result of the consistent negative sign of the bias as opposed to the bias at any particular location being particularly negative. For SIC, ICON and HadGEM perform well, but IFS-FESOM and IFS-NEMO both have biases exceeding the full CMIP6 range. Figure 6: Histograms of absolute SST and SIC biases. In (a): the absolute bias of various models against ERA5 for DJF SSTs across all gridpoints. In (b): the same but for JJA SSTs. In (c) and (d): the equivalent plots but for SIC. The filled grey histogram is obtained by combining the absolute biases across all the CMIP6 models used. The line contours are always IFS-FESOM (purple), IFS-NEMO (yellow), ICON (blue) and HadGEM (green). Note the log scale on the y-axes. 3.2. Surface temperature and precipitation We first look at global biases, and then briefly examine the Gulf Stream region and the `double ITCZ’ bias in closer detail. Page 17 of 76 3.2.1. Global biases Figures 7 and 8 show the EERIE model biases in DJF two-meter temperature (T2M) and precipitation, along with the corresponding CMIP6 median bias. Biases in T2M mirror biases in SSTs over the oceans. A variety of biases are also present over land, but comparison against the CMIP6 median suggests the magnitudes of these biases are not unusual, with two exceptions. Firstly, the ICON model has substantial biases over, for example, North America, Africa, Australia and the tropical Pacific that are atypical in the CMIP6 ensemble. Secondly, the IFS-NEMO T2M biases are entirely analogous to its SST biases: while the biases at any given location are not particularly unusual, the consistency of the cold bias is noteworthy. Figure 7: In (a): DJF averaged T2M in ERA5. In (b): the CMIP6 median DJF T2M bias against ERA5. In (c) through (f): DJF T2M biases for the four EERIE simulations (IFS-FESOM, IFS-NEMO, ICON and HadGEM). Readers are advised that ERA5 is known to be biased warm over the poles: see main text for more. Page 18 of 76 With regards to the large cold biases seen over the Arctic in several models, it is important to note that ERA5 is known to be biased warm in the polar regions by up to a few degrees (Tian et al. 2024; Bromwich et al. 2024). Thus cold biases over these regions in models are likely to be exaggerated when measured against ERA5. Finally, we note the warm bias over Antarctica in IFS-FESOM. Investigations confirmed that this is caused by a bug in the snow depth initial conditions of the model, which resulted in unrealistically low snow depth values and consequently unrealistically warm surface temperatures. Figure 8: In (a): DJF averaged precipitation in GPM-IMERG. In (b): the CMIP6 median DJF precipitation bias against GPM-IMERG. In (c) through (f): DJF precipitation biases for the four EERIE simulations (IFS-FESOM, IFS-NEMO, ICON and HadGEM). Units are mm/day. Page 19 of 76 For precipitation (Figure 8), the CMIP6 bias is dominated by the perennial ‘split ITCZ’ bias in the tropics, and biases related to the western boundary currents, especially the Gulf Stream and Kuroshio. For these latter biases, the changes to the boundary currents in the EERIE models leads to consistent changes in precipitation by altering the location of warm SSTs, and potentially also due to the resolution of mesoscale eddies. In the tropics, none of the EERIE models notably alleviate the split ITCZ bias in the Pacific and Atlantic. As noted in Section 2, ICON does not use a convection parameterisation, but it is apparent that 10km atmospheric resolution is insufficient to accurately resolve convection, resulting in precipitation biases that are particularly large. The corresponding plots for JJA are shown in Figures 9 and 10 for completeness. The features are broadly similar to those in DJF but with a seasonal shift. Note that the ICON biases are not as large in JJA as in DJF. Figure 9: T2M biases in JJA: as in Figure 7 but for JJA. Page 20 of 76 Figure 10: Precipitation biases in JJA: as in Figure 8 but for JJA. To further assess the biases in T2M, we show in Figure 11 timeseries of globally and annually averaged T2M for the EERIE models, ERA5 and CMIP6 models. IFS-FESOM is remarkably close to the observed timeseries, while HadGEM is somewhat too warm. Interestingly, despite substantial regional T2M biases, ICON’s globally averaged T2M is only marginally biased cold. This suggests that the overall energy budget is likely realistic, and that it is rather the distribution of this energy throughout the system which is biased, something we examine further in the section on energy budgets. IFS-NEMO’s noteworthy cold bias is made fully apparent in Figure 11, with the global average falling outside the range of CMIP6 models. Page 21 of 76 Figure 11: Time series of annually averaged globally averaged T2M (1950-2015). The grey lines are the various CMIP6 models. The other lines are IFS-FESOM (purple), IFS-NEMO (yellow), ICON (blue) and HadGEM (green). In fact, it falls outside the range of the wider ensemble of CMIP6, HighResMIP and CMIP5. This can be seen in Figure 12a, which shows global mean T2M averaged over the period 1979-2015 for a number of models in this ensemble, sorted according to horizontal atmospheric resolution. Figure 12b shows the analogues plot but for the T2M trend over the period 1950-2015. The trends are visibly in the typical CMIP range, even for the cold-biased IFS-NEMO. This figure also shows that increased resolution has little to no systematic impact on global mean T2M, and a small impact on trends. We expect both these reflect model tuning, on the one hand towards realistic surface temperatures and on the other towards trends that are not too large. To further assess the precipitation biases we show in Figure 13 zonally averaged precipitation for EERIE models, observations (GPM-IMERG) and CMIP6. The CMIP6 models show a wide spread of biases in DJF. The consistent overestimation of precipitation in the southern tropical belt is the signature of the `double ITCZ’ bias (Zhang et al. 2015, Tian and Dong 2020), and the EERIE models evidently have a mixed impact on this bias. In JJA the biases are generally smaller for all models. In general, the EERIE models fall within the spread of typical CMIP6 models. One interesting exception is HadGEM, which produces far too much precipitation in the ITCZ in winter: this bias is clearly visible both in Figure 13a and 7f. The other exception is, as before, the ICON model. Page 22 of 76 Figure 12: In (a): scatter plot of globally averaged T2M in coupled climate models versus their atmospheric horizontal resolution (km). The black dots are models from the CMIP multimodel ensemble (CMIP5, CMIP6 and HighResMIP). The coloured symbols are IFS-FESOM (purple square), IFS-NEMO (yellow triangle pointing down), ICON (blue diamond), HadGEM (green triangle pointing left) and ERA5 (red star). To the right is a histogram of the global mean T2M values for the CMIP multimodal data, with the coloured lines corresponding to the EERIE models and ERA5. The value of M is the mean across the CMIP multimodal ensemble. The period used is 1979-2015. In (b): the equivalent plot but for the trend in global mean T2M computed over the period 1950-2015. To get a better sense of how the EERIE model biases in T2M and precipitation compare to typical CMIP6 biases, we show in Figure 14 histograms of the absolute bias of these variables, pooled across each gridpoint: for CMIP6 models we have combined all models into a single histogram. These suggest that IFS-FESOM and IFS-NEMO perform very well for precipitation in both winter and summer, and IFS-FESOM and HadGEM perform well for T2M in both seasons. Interestingly, IFS-NEMO also performs quite well for T2M, suggesting that while it has an overall cold bias, the bias at any given gridpoint is relatively small, as was the case for its SST biases. HadGEM’s biases Page 23 of 76 in summer precipitation are also clear here. In fact, the maximum error attained by HadGEM here is more than 100mm/day, achieved at a single coastal gridpoint near Burma (not shown), vastly exceeding the errors of any of the CMIP6 models we considered here. The origin of these rare but large biases is not clear. Figure 13: Zonally averaged precipitation averaged over the DJF season in (a) and the JJA season in (b). Shown are IMERG (red), IFS-FESOM (purple), IFS-NEMO (orange), ICON (blue) and HadGEM (green). Also shown is the CMIP6 mean (black) and the 2 sigma spread of individual CMIP6 models around this mean. Another way to assess the magnitude of the EERIE model biases is to compare to biases in AMIP-simulations, where the SSTs are prescribed. Such simulations are not expected to be equivalent to having a bias-free dynamic ocean model, because the lack of coupling in AMIP-simulations impacts the simulated variability (Barsugli and Battisti 1998). However, they give useful insight into the approximate magnitude of bias you might still have even with such a bias-free dynamic ocean. Page 24 of 76 Figure 14: Histograms of absolute precipitation and T2M biases. In (a): the absolute bias of various models against IMERG for DJF precipitation across all gridpoints. In (b): the same but for JJA precipitation. In (c) and (d): the equivalent plots but for T2M biases against ERA5. The filled grey histogram is that obtained by combining the absolute biases across all the CMIP6 models used. The line contours are always IFS-FESOM (purple), IFS-NEMO (yellow), ICON (blue) and HadGEM (green). Note the log scale in (a) and (b). Figure 15 shows the DJF precipitation biases in four such AMIP simulations (the high and low-resolution IFS and HadGEM models). Comparison with Figure 7 suggests that the biases are actually comparable in magnitude in many areas. In particular, comparing the HadGEM AMIP simulations with the HadGEM coupled simulation, considerable overlap between the biases is apparent, suggesting a purely atmospheric origin for much of the precipitation bias. On the other hand, comparing the IFS AMIP simulations with the coupled IFS-FESOM and IFS-NEMO simulations in Figure 7 suggests the coupled biases are somewhat different and often larger in magnitude. Thus while a large amount of the precipitation bias is due to the atmospheric model, inadequacies in the ocean or atmosphere-ocean coupling can easily change the bias, including making it worse. Page 25 of 76 Figure 15: DJF biases in precipitation against IMERG for the four AMIP Reference simulations: IFS-AMIP-LR in (a), IFS-AMIP-HR in (b), HadGEM-AMIP-LR in (c) and HadGEM-AMIP-HR in (d). 3.2.2. Gulf Stream Precipitation Precipitation in the Gulf Stream region in DJF has been linked to high-impact atmospheric events, such as European Blocking, and precipitation in this region has also been shown to be modulated by the SST fronts generated by eddies. We therefore briefly examine precipitation in this region more closely. Figure 16 shows the DJF precipitation biases in the North Atlantic sector, with the black box highlighting a Gulf Stream region, which we define as 35-45N, 70-50W following past conventions. Figure 17(a) shows the mean precipitation in this region across a range of CMIP models at various atmospheric resolutions, while (b) shows the 99th percentile of daily precipitation, computed across all grid points in the domain, again as a function of atmospheric resolution. Page 32 of 76 Figure 21: In (a): scatter plot of globally averaged net TOA energy imbalance in coupled climate models versus their atmospheric horizontal resolution (km). The black dots are models from the CMIP multimodel ensemble (CMIP5, CMIP6 and HighResMIP). The coloured symbols are IFS-FESOM (purple square), IFS-NEMO (yellow triangle pointing down), ICON (blue diamond), HadGEM (green triangle pointing left) and CERES-EBAF (red star). To the right is a histogram of the TOA imbalance values for the CMIP multimodal data, with the coloured lines corresponding to the EERIE models and CERES-EBAF. The value of M is the mean across the CMIP multimodal ensemble. The period used is 2000-2015. In (b): the equivalent plot but for the trend in global mean T2M over the period 1950-2015. Turning to the water budget, Figure 23 shows both globally averaged precipitation and global precipitation minus evaporation (P minus E) over the historical period 1979-2015. There has been an overall tendency for precipitation to increase with resolution, plausibly due to higher resolution models simulating larger precipitation extremes. Page 33 of 76 Figure 22: In (a): scatter plot of globally averaged TOA shortwave radiation in coupled climate models versus their atmospheric horizontal resolution (km). The black dots are models from the CMIP multimodel ensemble (CMIP5, CMIP6 and HighResMIP). The coloured symbols are IFS-FESOM (purple square), IFS-NEMO (yellow triangle pointing down), ICON (blue diamond), HadGEM (green triangle pointing left) and CERES-EBAF (red star). To the right is a histogram of the net shortwave values for the CMIP multimodal data, with the coloured lines corresponding to the EERIE models and CERES-EBAF. The value of M is the mean across the CMIP multimodal ensemble. The period used is 2000-2015. In (b): the equivalent plot but for the new longwave radiation. However, the large spread suggests that the details of a model’s convection parameterisation plays a big role in the total amount of precipitation simulated; orography also likely plays a role. In general all models overestimate global precipitation relative to IMERG, especially over the tropics (Figures 7 and 10), though it must be kept in mind that there are large uncertainties in the IMERG observational estimate. Page 34 of 76 Figure 23: As in figure 22 but for (a) globally averaged precipitation and (b) globally averaged precipitation minus evaporation. In (a) we have marked both IMERG (red star)as our best estimate as well as ERA5 (grey star) for additional context. In (b) we have shown ERA5 (red star) for context (the value in the real world is 0). Interestingly, the higher resolution models are broadly aligned with the estimate from ERA5, suggesting that perhaps the more recently developed, higher resolution models are becoming more realistic at capturing the large scale water budget (which reanalysis products capture by virtue of their construction), but that big biases remain in the drizzle problem (Chen et al. 2021) resulting in excessive precipitation. This would be consistent with the results of Zhou et al. (2022). HadGEM and ICON are notable here in that they substantially overestimate total precipitation, occupying the outer range of past CMIP models. In terms of water conservation, all EERIE models conserve water near-perfectly. However, this is due to the implementation of mass-fixers in these models, which artificially forces perfect conservation. It can be seen that a large number of CMIP Page 35 of 76 models also have such near-zero P-E, consistent with the widespread use of such mass-fixers, or the construction of models with energy and mass conservation as key targets. Models that do not use such fixers, or have not been built from the onset to be conservative, display a wide range of large positive and negative values, indicative of the general difficulty of conserving mass and energy. Indeed, constraining the models with observations would still not guarantee conservation, as demonstrated by the P-E bias of ERA5 shown in the figure. 3.5. The stratospheric polar vortex Polar vortices are a prominent feature of the winter stratospheric climatology at high latitudes, with two counterparts in the Northern Hemisphere (NH) and Southern Hemisphere (SH). Their representation in climate models is known to be sensitive to both vertical and horizontal resolution, and many state-of-the-art models struggle to reproduce an adequate strength and/or lifecycle of the vortices, especially in the Southern Hemisphere (e.g. Rao et al. 2022, Feng et al 2024). It is therefore constructive to evaluate how the EERIE models capture the climatological seasonal cycle of these two vortices. To describe the mean state of the NH and SH vortex we examine the monthly climatology of the zonal-mean zonal wind at 10 hPa at 60°N and 60°S, respectively. We compare the results for the EERIE models with ERA5 and with 6 HighResMIP models (CMCC-CM2-VHR4, CNRM-CM6-1-HR, HadGEM3-GC31-HM, MPI-ESM1-2-XR, EC-Earth3P-HR, ECMWF-IFS-HR) with horizontal resolution between 25 and 100 km. Figure 24(a) displays the results for the NH. The EERIE models show an overall good performance and capture well the amplitude of the winter peak, in line with results from HighResMIP. A notable exception is again ICON, which simulates a too strong vortex, likely a result of the broader atmospheric biases arising from the lack of convection parameterisations. The EERIE models also simulates properly the timing of the vortex formation (August-September), peak (December-January) and breakdown (around April), except for IFS-NEMO that tends to anticipate them by 1-2 months. The corresponding results for the SH are presented in Fig. 24(b). Here, the vortex is expected to start forming around March, peak in August and break down in November, with an overall much stronger amplitude than its NH counterpart. The EERIE models deliver mixed performances. Page 36 of 76 IFS-FESOM is remarkably close to ERA5, while IFS-NEMO reproduces the correct amplitude but again tends to shift the full cycle by about 1 month. HadGEM exhibits a too strong and persistent vortex, although still within the range of the HighResMIP models. In contrast, ICON performs again poorly compared to both the EERIE and HightResMIP models: not only is the vortex consistently too strong, but it also seems to never fully dissolve as no real reversal of the zonal wind occurs. Figure 24: Climatological seasonal cycle of the stratospheric polar vortex in the Northern Hemisphere (left) and Southern Hemisphere (right). Note the different scale on the vertical axis. 3.6. Tropical and extratropical cyclones Extreme weather events such as tropical and extratropical cyclones (TCs and ETCs) have direct socio-economic impacts including floods, wind damage, and storm surges. These systems transport heat, momentum, and moisture across large distances and produce much of the seasonal precipitation in both tropics and midlatitudes. Storm characteristics such as size, maximum intensity, and frequency therefore need to be realistically simulated in GCMs to ensure their accuracy and reliability in future climate projections. The key physical processes affecting storm development often occur at scales smaller than those resolved by the GCMs, which has previously restricted their use for global storm climatology assessments (Lockwood et al. 2025, Bourdin et al. 2024, Baker et al. 2024, Roberts et al. 2020, Priestley et al. 2022, Jiaxiang et al. 2020). Recent studies have highlighted the need for global storm-resolving models, with resolutions in the range 28–2.8 km in the atmosphere and 28–5 km in the ocean to improve the representation of tropical and extratropical cyclones. The impact of Page 37 of 76 horizontal resolution on tropical and extratropical cyclones have been investigated through the CMIP6 High Resolution Model Intercomparison Project (HighResMIP; Haarsma et al. 2016). Significant improvements in simulated tropical and extratropical cyclone mean state such as frequency, intensity, spatial distribution and wind structure with increasing resolutions from 100 km to 25 km have been reported (Baker et al. 2024, Roberts et al. 2020, Priestley et al. 2022, Jiaxiang et al. 2020). The EERIE models employ a much higher resolution (~10 km) and can explicitly resolve mesoscale ocean eddies and convective processes. In addition to assessing the improvements across models and resolutions, the sensitivity experiments offer the opportunity to isolate the impacts of ocean mesoscale eddies on the mean storm climatology. The EERIE models used in the evaluation of storm climatology are summarized in Table 3.7.1. Tropical and extratropical cyclones are diagnosed from the re-gridded model outputs using a feature-tracking algorithm TempestExtremes (Ullrich and Zarzycki 2017; Zarzycki and Ullrich 2017). Candidate TCs (or nodes) are identified every 6 hours by using a closed-contour criterion, i.e. a sea‐level pressure increase of ≥2 hPa over 5.5° from pressure minima. The presence of an upper‐level warm core characteristic of TC is identified by a geopotential thickness, Z250–Z500, decrease of ≥6 m (58.8 m2s2) over 6.5° from a reference Z250–Z500 maximum within 1° of a given pressure minimum. The candidate TCs within 6° are stitched in time into cyclone tracks with a minimum duration of 60 hours and a minimum track length of 8°, which eliminates stationary features and spurious shallow lows. Candidate ETCs are also identified similarly, except the closed-contour criterion is that the sea‐level pressure increases by ≥2 hPa over 6° from pressure minima, and the candidates with a warm-core are rejected. The candidate ETCs within 6° are stitched in time into cyclone tracks with a minimum duration of 60 hours and a minimum track length of 12°. We also diagnose TC/ETC tracks from two reanalysis datasets: ERA5 and the JRA-3Q. In addition, the tracks derived from the model outputs from the HighResMIP/PRIMAVERA project are also presented for comparison. We have only included TC tracks from the coupled (hist-1950) HighResMIP/PRIMAVERA models that were run at a higher resolution of 25-50 km in the comparison (Roberts et al. 2020), which are also summarized in Table 3.7.1. Note that the corresponding ETC tracks are not currently available from HighResMIP/PRIMAVERA. Page 38 of 76 Institution Model Model type Horizontal resolution Period MOHC HadGEM3-GC5 Atmosphere-only (UM) MR (N216) atm: ~60 km HR (N640) atm: ~20 km 1980-2023 ECMWF ECMWF-IFS Atmosphere-only (IFS) LR (tco399) atm: ~28 km HR (tco1279) atm: ~9 km 1980-2023 MOHC HadGEM3-ORCA Coupled (NEMO v4.0.4) MM (N216) ocean: 0.25o atm: ~60 km HM (N640) ocean: 1/12o atm: ~20 km 1850-2014 AWI IFS-FESOM Coupled (FESOM 2.5) HR (tco1279) ocean: 13-4.5 km atm: ~9 km 1950-2014 MPI-M ICON Coupled (ICON-O v2.6.6) HR (R2B8) ocean: ~5km atm: ~10 km 1950-2014 MOHC/ UoR/ NERC HadGEM3-GC3.1 Coupled HM (N512) atm: ~25 km 1950-2014 KNMI / SMHI / BSC EC-Earth3P Coupled HR (T1511) atm: ~36 km 1950-2014 CERFACS CNRM-CM6.1 Coupled HR (T1359) atm: ~50 km 1950-2014 MPI MPI-ESM1.2 Coupled XR (T255) atm: ~34 km 1950-2014 ECMWF ECMWF-IFS Coupled HR (tco399) atm: ~25 km 1950-2014 Table 3.7.1. Description of EERIE and HighResMIP/PRIMAVERA models used for TC/ETC tracking. The blue text highlights the HighResMIP data. The model simulated TCs are compared with the IBTrACS data set. Following Roberts et al. (2020), a tropical storm (excluding all subtropical storms or SS) with 1-min maximum sustained winds of 34 kt (17.5 ms-1) or higher is identified as an observed tropical cyclone. The ETC tracks derived from ERA-5 reanalysis is used to compare the model simulated ETCs, due to the lack of available observed ETC tracks. Page 39 of 76 3.7.1 Tropical cyclones Key findings: ● ICON significantly overestimates the number of tropical cyclones, in both hemispheres. ● HadGEM3-ORCA and IFS-FESOM simulates too many TCs in the Western Northern Pacific (WNP), South Pacific (SP), and South Indian (SI) basins, and too few in Eastern Northern Pacific (WNP) and North Atlantic basins (NATL). ● Overall, the biases in the coupled simulations are larger compared to the multi-model mean from coupled HIGHRESMIP/PRIMAVERA simulations. However, the small biases seen in the multi-model mean is only due to the inclusion of MPI-ESM and EC-Earth3P which significantly underestimates the track densities. ● All models overestimate the storm durations and underestimate the storm densities and intensification rates, with improvements seen with increase in resolution. ● HADGEM3-ORCA performs well in simulating stronger and more short-lived storms, given its coarse resolution (~20 km) compared to other coupled models. ICON simulates similar intensification rates as Obs, but this is likely to be due to extremely favourable but non-physical environments. Figure 25 shows the global annual TC activity detected in models, reanalysis, and observations. The total TC frequency increases with resolution for both the atm-only (HadGEM3-GC5) and coupled (HadGEM3-ORCA), as well as for ECMWF-IFS. The high-resolution members of HadGEM3 model tend to overestimate the total number of TCs compared to observations, while the proportion of TCs in the West North Pacific (WNP) decreases. The ICON-ESM model also significantly overestimates the total TC activity but underestimates the WNP activity. Overall, the medium-resolution members of HadGEM3 (60 km) and the high-resolution members of ECMWF-IFS (i.e. ECMWF-IFS and IFS-FESOM2) are more comparable with the observations and reanalysis. The mean bias in annual TC track density, which is defined as the mean number of tracks in a 4o cap, is shown in Figure 26. For TCs, the bias is computed with respect to IBTRACs (Obs). The HadGEM3-GC5 exhibit positive biases in the WNP basin, and negative biases in the ENP and NATL basins, all of which are enhanced in the coupled model (HadGEM3-ORCA). Although the increased resolution in HadGEM3-GC5 decreases the bias in the NATL and ENP basin, the positive biases in the other basins including the SP and SI are enhanced. The improvements are Page 40 of 76 less in the coupled model (HadGEM3-ORCA). Increasing resolution in atm-only ECMWF-IFS model also results in small improvements across all basins, but the biases are enhanced in the coupled IFS-FESOM2 model. The coupled ICON-ESM exhibits large positive biases across most of the basins, with the exception of WNP, where the biases are negative. Overall, the biases in atm-only models (except HadGEM3-GC5-HR) are comparable to the reanalysis as well as to the multi-model mutli-ensemble average of the HighResMIP/PRIMAVERA models. Figure 25: Annual tropical cyclone frequency in different basins from models, reanalysis and observations during historical period (1983-2013). The mean annual TC numbers are indicated in the center. The TC tracks in models and reanalysis are diagnosed using TempestExtremes. The top row indicates the atm-only EERIE simulations (highresSST-present) and bottom row indicates the coupled EERIE simulations (hist-1850 and hist-1950). The thickness of the doughnut is scaled to the total TC observed frequency [i.e., doughnuts thicker than in (k) indicate more TCs, and thinner than in (k) indicate fewer TCs]. The distribution of different TC characteristics such as TC lifetime or duration, intensity, latitude of maximum intensity, and intensification rate diagnosed in the EERIE models, reanalysis and observations are shown in Figure 27. Most of the EERIE models overestimate the TC duration compared to Obs but have similar TC durations compared to reanalysis. The notable exceptions Page 41 of 76 are the HadGEM3 models, which tend to produce more short-lived TCs. As expected, the high-resolution models tend to produce more intense TCs, with values of maximum 10-m winds (vmax) or minimum sea-level pressure (pmin) more comparable to Obs and JRA3Q. On the other hand, the low-resolution members in HadGEM3 produce similar intensity distributions as ERA5 reanalysis. Most models fail to simulate TCs stronger than Cat-4 (vmax >58 ms-1). Similarly to intensity, the high-resolution models also simulate greater intensification rates compared to low-resolution members; however, they still fail to simulate strongly intensifying storms, particularly those undergoing rapid intensification (with intensification rates > 15 ms-1/24 hr). The accumulated cyclone energy (ACE) index (Bell et al. 2000) is defined as the sum of the square of the maximum sustained 10m wind speed every 6 h while the cyclone is at least tropical storm strength (34 kt; 17.5 ms-1). However, since the model simulated TCs have lower wind speeds, the ACE is calculated for the entirety of the storm. The mean values of ACE are underestimated in all of the models compared to both Obs and reanalysis, which can be attributed to the weaker TC intensities simulated in the models. The model TCs also tend to be located more poleward compared to observations at the time of maximum intensity, which is consistent with previous studies. Figure 26: Annual tropical cyclone density (storms transits per 4o cap) from IBTRACs (Obs), and track density bias in models during historical period (1983-2013). The top row indicates the atm-only EERIE simulations (highresSST-present) and middle row indicates the coupled EERIE simulations (hist-1850 and hist-1950). The two reanalysis products used for comparisons are ERA5 and JRA3Q, and the HighResMIP ensemble average includes the high-resolution members of the coupled models (hist-1950): CNRM-CM6-1, EC-Earth3P, ECMWF-IFS, HadGEM3-GC31, and MPI-ESM1-2. Page 48 of 76 Figure 32 : Composite means of sea surface temperature (SST), 10m wind speed, turbulent heat fluxes (positive out of the ocean), and total precipitation derived from daily mean ¼ degree data sampled for observed anticyclonic eddy locations from the Southern Ocean (25S-60S) calculated following Aengenheyster et al. (submitted). Before calculating composite means, each daily mean snapshot is rotated such that the large-scale winds are from left to right and resampled according to the identified eddy radius. Observational composite means are derived using ESA CCI SSTs (Embury et al., 2024), 10 m wind speed from merged scatterometer/ERA5 data (Copernicus Marine Service 2019), and precipitation from GPM-IMERG (Huffmann et al., 2019). Top four rows use the same 513397 eddy locations between 2010-2021. For high-resolution HadGEM3, 484837 eddy locations between 2000-2009 are used due to data availability. Page 49 of 76 Figure 33: As Figure 32, but for the difference between composites means of anticyclonic and cyclonic eddies. For top four rows, the same 513397 anticyclonic and 543972 cyclonic eddies are used, while for HadGEM3 484837 anticylonic and 518408 cyclonic eddies are used. Our estimate of the difference in precipitation between composites of anticyclonic and cyclonic eddies based on IMERG satellite data is approximately 0.2 mm day⁻¹ (Figure 33). This is smaller than the estimate reported by Frenger et al. (2013) but broadly consistent with IMERG-based Page 50 of 76 estimates from Liu et al. (2018). These differences between observation-based estimates could be a consequence of differences between satellite-based precipitation products (Lui et al. 2018), sensitivity to evaluation period, or other differences in the identification, scaling, and selection of eddies used for composite calculations. Lowerand higher-resolution Reference simulations from both IFS and HadGEM3 models qualitatively capture the observed composite-average precipitation and wind speed patterns associated with ocean eddies (Figure 32). These composite average patterns are qualitatively similar in lowerand higher-resolution simulations from the same model. The similarity between resolutions is stronger for IFS, while eddy-related features are generally weaker in low-resolution HadGEM3 simulations (figures 32 and 33), likely because the 60 km resolution is similar to the mean eddy radius in the Southern Ocean. However, there are some notable inter-model differences when comparing the atmospheric response to anticyclonic versus cyclonic eddies. All simulations tend to underestimate the imprint of eddies on 10 m wind speeds relative to observations, though this underestimate is smaller for higher-resolution simulations and smallest for the high-resolution HadGEM3 simulation. Conversely, the higher resolution HadGEM3 simulation has the strongest precipitation response relative to IMERG satellite estimates. The precipitation response is weakest (closest to observations) for the high-resolution IFS simulation. HadGEM3 also systematically displays weaker winds but stronger turbulent heat fluxes over eddies (Figure 32) which points to inherent model differences. The high-resolution HadGEM3 eddy composites were available for a different period than the remaining three models and observations (2000-2009 rather than 2010-2021). Judging from the low-resolution HadGEM3 composites being available for both periods, the eddy SST signatures and some responses appear slightly higher in the later period, suggesting that the high-resolution HadGEM3 response might be slightly underestimated in Figures 32. The atmospheric imprint of small-scale ocean eddies is effectively suppressed in the SmoothAnom experiments. This is clearly illustrated in Figure 34, which shows that the difference between anticyclonic and cyclonic eddies is strongly reduced in the SmoothAnom experiment with IFS Tco1279, with similar results obtained for other models (not shown). While our scale-selective filtering of extratropical SST anomalies does not completely eliminate the ocean eddy signature (figure 34), the composite-average impact on surface winds, turbulent heat fluxes, and precipitation is dramatically reduced compared to the corresponding Reference simulations (figure 33). From these results, we consider these experiments to be an appropriate dataset to evaluate the Page 51 of 76 large-scale atmospheric response to small-scale thermodynamic forcing from SST anomalies associated with extratropical ocean eddies. Figure 34: As Figure 33 but for the difference between composite means in the high-resolution IFS SmoothAnom experiment. 4.3. Impact of extratropical ocean eddies on the atmospheric mean state 4.3.1. Assessment of basic impact To assess the climatological influence of ocean eddies, we focus on seasonal mean diagnostics from the 10-member ensemble of IFS Tco399 simulations (figure 35). The ensemble averaging reduces sampling uncertainty associated with year-to-year atmospheric variability, thereby isolating the response to mesoscale ocean forcing. Statistical robustness is assessed using a paired-sample t-test (p < 0.05) applied to ensemble mean differences between Reference and SmoothAnom experiments. Bootstrap-based significance testing gives almost identical results (not shown). In figure 35, hatching denotes regions where differences are statistically robust in the IFS Tco399 ensemble, while stippling indicates locations where the sign of the response is consistent across all four model/resolution combinations. Page 52 of 76 Figure 35: Global maps for Reference minus SmoothAnom DJF (left) and JJA (right) seasonal means for latent heat flux, daily maximum 2m temperature, total precipitation and total cloud cover in the IFS Tco399 10-member ensemble. Hatching indicates 5% significance considering the 29 seasonal ensemble means between 1985-2014 as independent samples. Stippling indicates regions where the four simulations (highand low-resolution IFS and HadGEM3) agree on the sign of the response. Seasonal-mean changes in surface latent fluxes reflect asymmetries in the atmospheric response to cyclonic versus anticyclonic eddies, highlighting nonlinearities in the response to positive and negative SST anomalies. Ocean eddies enhance latent heat release from the ocean to the atmosphere by approximately 5–10 W m⁻² in regions of intense eddy activity, including the Gulf Stream, Kuroshio Extension, and Agulhas Retroflection (figure 35). These signals are statistically Page 53 of 76 robust in the IFS Tco399 10-member ensemble, and in some cases (e.g. over the Agulhas Retroflection during JJA) are consistent across the single-member results for other models. Other significant changes to latent heat fluxes in IFS Tco399 experiments are evident in lower-latitude regions (e.g. tropical Atlantic during DJF), though these are not systematically reproduced across the other models and resolution combinations. Our experiments indicate that extratropical ocean eddies exert a limited influence on seasonal mean 2 m air temperature (T2m; not shown). However, they systematically affect the diurnal cycle such that all models show a robust increase of ~0.5 K in seasonal mean daily maximum T2m in eddy-rich regions during both DJF and JJA (figure 35). Remote signals associated with large-scale circulation changes (figure 35) are also evident in some regions. For example, IFS Tco399 shows wintertime warming over Greenland and temperature changes over Antarctica, though only the Antarctic response is consistent across models. Figure 36: Scatter plot of globally averaged JJA total cloud cover over a southern ocean box (see text) in coupled climate models versus their atmospheric horizontal resolution (km). The black dots are models from the CMIP multimodel ensemble (CMIP5, CMIP6 and HighResMIP). The coloured symbols are IFS-FESOM (purple square), IFS-NEMO (yellow triangle pointing down), ICON (blue diamond), HadGEM (green triangle pointing left) and ERA5 (red star). To the right is a histogram of the values for the CMIP multimodel data, with the coloured lines corresponding to the EERIE models and ERA5. The value of M is the mean across the CMIP multimodel ensemble. Page 54 of 76 In eddy-active regions, IFS Tco399 indicates a mean precipitation increase of ~0.1 mm day⁻¹ due to ocean eddies. However, the signal is noisy in single-member simulations, limiting multi-model consensus. An exception occurs in the Agulhas Retroflection during JJA, where regionally focused precipitation increases are accompanied by small but statistically robust (~1 %) reductions in total cloud cover across models. We speculate that the changes in cloud cover are a consequence of an eddy-induced shift between widespread stratiform clouds towards deeper more localized convection triggered by the enhanced latent heating associated with ocean eddies. However, we note that the coupled EERIE simulations show if anything an increase in cloud cover in this region relative to lower resolution CMIP models: see Figure 36. This suggests that the eddies exert a second order impact here, with the dominant role being played by the atmosphere. 4.3.3. Gulf Stream precipitation For a detailed investigation over a specific region, we focus on mean and extreme precipitation over the Gulf Stream region, where a consistent precipitation response was identified (Figure 35), and where the coupled EERIE simulations also show an increase in these quantities relative to lower resolution models (Figure 17). Figure 37 shows the precipitation impact over the extended winter season NDJFM for the four experiments, highlighting the Pacific and Atlantic regions. Figure 38 shows precipitation statistics for the Gulf Stream region (35-45N, 70-50W). The AMIP simulations have mean Gulf Stream precipitation generally larger than observations, with higher values for high-resolution runs, and HadGEM higher than IFS. IFS Tco399 tends to underestimate the magnitude of Gulf Stream precipitation extremes compared to observations and the other models. Mesoscale eddies have a consistent enhancing impact on both mean and extreme precipitation in the Gulf Stream region, as demonstrated by Reference simulations showing larger magnitudes than SmoothAnom. The exception is HadGEM3 for DJF means, and high-resolution HadGEM3 for DJF extremes. In JJA, the impact is stronger, and all simulations agree on the sign of the response. The precipitation response is largely due to convective precipitation (not shown) likely caused by convection induced by mesoscale SST anomalies. Page 55 of 76 Figure 37: Reference minus SmoothAnom NDJFM precipitation over the Pacific and Atlantic from the four atmosphere-only simulations (10 member ensemble for IFS-AMIP tco399). A box highlights the Western US region from Figure 1 of Ma et al., 2015. Note that the difference in precipitation extremes between the Reference and SmoothAnom simulations is, for the high resolution configurations, around 1 mm/day. Examination of the linear relationship between resolution and extremes in Figure 17b suggests this is about equivalent to the increase expected when moving from a 25km atmosphere and ¼ degree ocean to a 10km atmosphere and 1/12th degree ocean. Thus it is likely that the presence of ocean eddies is contributing to the increased precipitation extremes in the EERIE coupled simulations. However, the intermodel spread in the average magnitude of the extremes is larger than this (cf. the coloured symbols in figure 17b), suggesting that the atmospheric model (resolution and choice of convection parameterisations) plays a larger role in setting this value. Page 56 of 76 Figure 38: Precipitation statistics over the Gulf Stream (35-45N, 70-50W) from atmosphere-only simulations. (a-b) Seasonal mean, (c-d) 99th percentile of daily precipitation over DJF (left) and JJA (right) are computed point-wise based on data regridded bilinearly to a 1 degree regular grid. Each panel shows Reference simulations and GPM-IMERG on the left and SmoothAnom simulations on the right. Methodology and time period (2000-2015) is chosen to align with Figure 17. 4.3.3. Remote changes and dynamical response to eddies We also find some evidence that ocean eddies can modulate remote precipitation. In particular, all models show a ~10 % reduction in DJF precipitation over the northeastern Pacific and the adjacent North American coast (figure 36 and 37). This signal aligns with previous regional modeling studies, which attributed similar wintertime anomalies to changes in moist baroclinic instability and a northward shift of the North Pacific storm track induced by Kuroshio eddies (Ma et al., 2015, 2017; Liu et al., 2021). There is also evidence for changes in tropical precipitation, which may reflect large-scale adjustments in the atmospheric overturning circulation and associated shifts in the intertropical convergence zone in response to changes in extratropical latent heating. However, these signals are not generally consistent across models indicating that the tropical response to extratropical ocean eddy forcing remains uncertain. Page 57 of 76 Figure 39: As figure 35, but for 500 hPa geopotential height and 300 hPa zonal and meridional wind. Hatching indicates 5% significance considering the 29 seasonal ensemble means between 1985-2014 as independent samples. Stippling indicates regions where the four simulations (highand low-resolution IFS and HadGEM3) agree on the sign of the response. The most compelling evidence for a large-scale dynamical response to ocean eddies emerges in the Northern Hemisphere winter (DJF; figures 36 and 37). In IFS Tco399, the North Pacific jet exhibits a statistically robust northward shift at its exit region, evident as ~1 m s⁻¹ changes in zonal mean zonal wind at 300 hPa. This is accompanied by an intensification of the climatological stationary waves in meridional winds and geopotential height (figure 39). The spatial pattern of these changes in the North Pacific is consistent across models and also with the previous regional modelling studies that have linked modulation of precipitation in the northeastern Pacific with eddy-forced changes in the North Pacific storm track. In the wintertime North Atlantic (figures 39 and 37) there is some evidence that ocean mesoscale eddies can induce a tripole-like response in zonal winds at 300 hPa, which manifests as a ~0.5 m s⁻¹ weakening in the climatological jet exit region, with strengthening to the north and south. Page 64 of 76 SI3 to HadGEM3 using conductive fluxes. Geoscientific Model Development, 17(17), 6799–6817. https://doi.org/10.5194/gmd-17-6799-2024 Bourdin, S., Fromang, S., Caubel, A., Ghattas, J., Meurdesoif, Y., Dubos, T. (2024). Tropical cyclones in global high-resolution simulations using the IPSL model. Clim Dyn 62, 4343–4368. https://doi.org/10.1007/s00382-024-07138-w Bromwich, D. H., Ensign, A., Wang, S.-H., & Zou, X. (2024). Major artifacts in ERA5 2-m air temperature trends over Antarctica prior to and during the modern satellite era. Geophysical Research Letters, 51, e2024GL111907. https://doi.org/10.1029/2024GL111907 Chassignet, E. and Marshall, D. (2008). Gulf Stream separation in numerical ocean models. Geophysical Monograph Series, 177. https://doi.org/10.1029/177GM05 Chelton, D. B., Schlax, M. G., Freilich, M. H., & Milliff, R. F. (2004). Satellite Measurements Reveal Persistent Small-Scale Features in Ocean Winds. Science, 303(5660), 978–983. https://doi.org/10.1126/science.1091901 Chelton, D., & Xie, S.-P. (2010). Coupled Ocean-Atmosphere Interaction at Oceanic Mesoscales. Oceanography, 23(4), 52–69. https://doi.org/10.5670/oceanog.2010.05 Chen, D., Dai, A., & Hall, A. (2021). The convective-to-total precipitation ratio and the “drizzling” bias in climate models. Journal of Geophysical Research: Atmospheres, 126, e2020JD034198. https://doi.org/10.1029/2020JD034198 Copernicus Marine Service. (2019). Global Ocean Hourly Reprocessed Sea Surface Wind and Stress from Scatterometer and Model [Dataset]. Mercator Ocean International. https://doi.org/10.48670/MOI-00185 Crueger, T., Giorgetta, M. A., Brokopf, R., Esch, M., Fiedler, S., Hohenegger, C., et al. (2018). ICON-A, the atmosphere component of the ICON Earth system model: II. Model evaluation. Journal of Advances in Modeling Earth Systems, 10, 1638–1662. https://doi.org/10.1029/2017MS001233 Danilov, S., Wang, Q., Timmermann, R., Iakovlev, N., Sidorenko, D., Kimmritz, M., Jung, T., & Schröter, J. (2015). Finite-Element Sea Ice Model (FESIM), version 2. Geoscientific Model Development, 8(6), 1747–1761. https://doi.org/10.5194/gmd-8-1747-2015 Page 65 of 76 Dong, E., Song, F., Wu, L., Dong, L., Wang, S., Liu, F., & Wang, H. (2025). The process-oriented understanding on the reduced double-ITCZ bias in the high-resolution CESM1. Geophysical Research Letters, 52, e2024GL112087. https://doi.org/10.1029/2024GL112087 Embury, O., Merchant, C. J., Good, S. A., Rayner, N. A., Høyer, J. L., Atkinson, C., Block, T., Alerskans, E., Pearson, K. J., Worsfold, M., McCarroll, N., & Donlon, C. (2024). Satellite-based time-series of sea-surface temperature since 1980 for climate applications. Scientific Data, 11(1), 326. https://doi.org/10.1038/s41597-024-03147-w Eyring, V., Bony, S., Meehl, G.A., Senior, C.A., Stevens, B., Stouffer, R.J. and Taylor, K.E. (2016). Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geoscientific Model Development, 9(5), 1937-1958. https://doi.org/10.5194/gmd-9-1937-2016 Feng, K., Rao, J., Chen, H., Ren R., & Guo D. (2024). Improvement of the simulated southern hemisphere stratospheric polar vortex across series of CMIPs. Climate Dynamics, 62, 5605–5621. https://doi.org/10.1007/s00382-024-07250-x Foussard, A., Lapeyre, G., & Plougonven, R. (2019). Storm Track Response to Oceanic Eddies in Idealized Atmospheric Simulations. Journal of Climate, 32(2), 445–463. https://doi.org/10.1175/JCLI-D-18-0415.1 Frenger, I., Gruber, N., Knutti, R., & Münnich, M. (2013). Imprint of Southern Ocean eddies on winds, clouds and rainfall. Nature Geoscience, 6(8), 608–612. https://doi.org/10.1038/ngeo1863 Giorgetta, M. A., Brokopf, R., Crueger, T., Esch, M., Fiedler, S., Helmert, J., et al. (2018). ICON-A, the atmosphere component of the ICON Earth system model: I. Model description. Journal of Advances in Modeling Earth Systems, 10, 1613–1637. https://doi.org/10.1029/2017MS001242 Guiavarc’h, C., Storkey, D., Blaker, A. T., Blockley, E., Megann, A., Hewitt, H. T., Bell, M. J., Calvert, D., Copsey, D., Sinha, B., Moreton, S., Mathiot, P., & An, B. (2024). GOSI9: UK Global Ocean and Sea Ice configurations. EGUsphere, 2024, 1–38. https://doi.org/10.5194/egusphere-2024-805 Haarsma, R.J., Roberts, M.J., Vidale, P.L., Senior, C.A., Bellucci, A., Bao, Q., Chang, P., Corti, S., Fučkar, N.S., Guemas, V. and Von Hardenberg, J. (2016). High resolution model intercomparison project (HighResMIP v1. 0) for CMIP6. Geoscientific Model Development, 9(11), 4185-4208. https://doi.org/10.5194/gmd-9-4185-2016 Page 66 of 76 Haarsma, R.J., Roberts, M.J., Vidale, P.L., Senior, C.A., Bellucci, A., Bao, Q., Chang, P., Corti, S., Fučkar, N.S., Guemas, V., Von Hardenberg, J., Hazeleger, W., Kodama, C., Koenigk, T., Leung, L.R., Lu, J., Luo, J.-J., Mao, J., Mizielinski, M.S., Mizuta, R., Nobre, P., Satoh, M., Scoccimarro, E., Semmler, T., Small, J., Von Storch, J.-S. (2016). High Resolution Model Intercomparison Project (HighResMIP v1.0) for CMIP6. Geosci. Model Dev. 9, 4185–4208. https://doi.org/10.5194/gmd-9-4185-2016 Hartmann, D. L., Ramanathan, V., Berroir, A., & Hunt, G. E. (1986). Earth Radiation Budget data and climate research. Reviews of Geophysics, 24(2), 439–468. https://doi.org/10.1029/RG024i002p00439 Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz‐Sabater, J., Nicolas, J., Peubey, C., Radu, R., Schepers, D. and Simmons, A. (2020). The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730), 1999-2049. https://doi.org/10.1002/qj.3803 Hodnebrog, Ø., Myhre, G., Jouan, C., Andrews, T., Forster, P.M., Jia, H., Loeb, N.G., Olivié, D.J., Paynter, D., Quaas, J. and Raghuraman, S.P. (2024). Recent reductions in aerosol emissions have increased Earth’s energy imbalance. Communications Earth & Environment, 5(1), 166. https://doi.org/10.1038/s43247-024-01324-8 Hohenegger, C., Korn, P., Linardakis, L., Redler, R., Schnur, R., Adamidis, P., Bao, J., Bastin, S., Behravesh, M., Bergemann, M., Biercamp, J., Bockelmann, H., Brokopf, R., Brüggemann, N., Casaroli, L., Chegini, F., Datseris, G., Esch, M., George, G., … Stevens, B. (2023). ICON-Sapphire: Simulating the components of the Earth system and their interactions at kilometer and subkilometer scales. Geoscientific Model Development, 16(2), 779–811. https://doi.org/10.5194/gmd-16-779-2023 Huffman, G. J., Stocker, E. F., Bolvin, D. T., Nelkin, E. J., & Jackson, T. (2019). GPM IMERG Final Precipitation L3 Half Hourly 0.1 degree x 0.1 degree (Version V06B) [Dataset]. NASA Goddard Earth Sciences Data and Information Services Center (GES DISC). https://doi.org/10.5067/GPM/IMERG/3B-HH/06 Huffman, G.J., Bolvin, D.T., Braithwaite, D., Hsu, K., Joyce, R., Xie, P. and Yoo, S.H. (2015). NASA global precipitation measurement (GPM) integrated multi-satellite retrievals for GPM (IMERG). Algorithm theoretical basis document (ATBD) version, 4(26), 30. https://gpm.nasa.gov/resources/documents/gpm-integrated-multi-satellite-retrievals-gpm-imerg-al gorithm-theoretical-basis- Page 67 of 76 John, V.O. and Soden, B.J., 2007. Temperature and humidity biases in global climate models and their impact on climate feedbacks. Geophysical Research Letters, 34(18). https://doi.org/10.1029/2007GL030429 Jungclaus, J. H., Lorenz, S. J., Schmidt, H., Brovkin, V., Brüggemann, N., Chegini, F., Crüger, T., De-Vrese, P., Gayler, V., Giorgetta, M. A., Gutjahr, O., Haak, H., Hagemann, S., Hanke, M., Ilyina, T., Korn, P., Kröger, J., Linardakis, L., Mehlmann, C., … Claussen, M. (2022). The ICON Earth System Model Version 1.0. Journal of Advances in Modeling Earth Systems, 14(4), e2021MS002813. https://doi.org/10.1029/2021MS002813 Knapp, K. R., Diamond, H. J., Kossin, J. P., Kruk, M. C., & Schreck, C. J. I. (2018). International Best Track Archive for Climate Stewardship (IBTrACS) project, version 4, national centers for environmental information. National Oceanographic and Atmospheric Administration. Korn, P. (2017). Formulation of an unstructured grid model for global ocean dynamics. Journal of Computational Physics, 339, 525–552. https://doi.org/10.1016/j.jcp.2017.03.009 Korn, P., Brüggemann, N., Jungclaus, J. H., Lorenz, S. J., Gutjahr, O., Haak, H., Linardakis, L., Mehlmann, C., Mikolajewicz, U., Notz, D., Putrasahan, D. A., Singh, V., von Storch, J.-S., Zhu, X., & Marotzke, J. (2022). ICON-O: The Ocean Component of the ICON Earth System Model—Global Simulation Characteristics and Local Telescoping Capability. Journal of Advances in Modeling Earth Systems, 14(10). https://doi.org/10.1029/2021MS002952 Liu, W., Lu, J., Xie, S.-P., & Fedorov, A. (2018). Southern Ocean Heat Uptake, Redistribution, and Storage in a Warming Climate: The Role of Meridional Overturning Circulation. Journal of Climate, 31(12), 4727–4743. https://doi.org/10.1175/JCLI-D-17-0761.1 Liu, X., Ma, X., Chang, P., Jia, Y., Fu, D., Xu, G., Wu, L., Saravanan, R., & Patricola, C. M. (2021). Ocean fronts and eddies force atmospheric rivers and heavy precipitation in western North America. Nature Communications, 12(1), 1268. https://doi.org/10.1038/s41467-021-21504-w Lockwood, J.F., Athanasiadis, P., Baker, A.J., Hodges, K., Priestley, M.D.K., Roberts, M., Scaife, A.A., Vidale, P.-L., Zappa, G. (2025). The effect of increasing model resolution on the Northern Hemisphere winter mid-latitude storm track: An equatorward shift due to contraction of the Hadley cell. Journal of Climate. https://doi.org/10.1175/JCLI-D-24-0414.1 Page 68 of 76 Loeb, N.G., Doelling, D.R., Wang, H., Su, W., Nguyen, C., Corbett, J.G., Liang, L., Mitrescu, C., Rose, F.G. and Kato, S. (2018). Clouds and the earth’s radiant energy system (CERES) energy balanced and filled (EBAF) top-of-atmosphere (TOA) edition-4.0 data product. Journal of climate, 31(2), 895-918. https://doi.org/10.1175/JCLI-D-17-0208.1 Ma, X., Chang, P., Saravanan, R., Montuoro, R., Hsieh, J.-S., Wu, D., Lin, X., Wu, L., & Jing, Z. (2015). Distant Influence of Kuroshio Eddies on North Pacific Weather Patterns? Scientific Reports, 5(1), 17785. https://doi.org/10.1038/srep17785 Ma, X., Chang, P., Saravanan, R., Montuoro, R., Nakamura, H., Wu, D., Lin, X., & Wu, L. (2017). Importance of Resolving Kuroshio Front and Eddy Influence in Simulating the North Pacific Storm Track. Journal of Climate, 30(5), 1861–1880. https://doi.org/10.1175/JCLI-D-16-0154.1 Moreno-Chamarro, E., Caron, L.P., Loosveldt Tomas, S., Gutjahr, O., Moine, M.P., Putrasahan, D., Roberts, C.D., Roberts, M.J., Senan, R., Terray, L. and Tourigny, E. (2021). Impact of increased resolution on long-standing biases in HighResMIP-PRIMAVERA climate models. Geoscientific Model Development Discussions, 2021, 1-33. https://doi.org/10.5194/gmd-15-269-2022 Notz, D. and Community, S.I.M.I.P. (2020). Arctic sea ice in CMIP6. Geophysical Research Letters, 47(10), p.e2019GL086749. https://doi.org/10.1029/2019GL086749 Olonscheck, D. and Rugenstein, M. (2024). Coupled climate models systematically underestimate radiation response to surface warming. Geophysical Research Letters, 51(6), p.e2023GL106909. https://doi.org/10.1029/2023GL106909 Piazza, M., Terray, L., Boé, J., Maisonnave, E., & Sanchez-Gomez, E. (2016). Influence of small-scale North Atlantic sea surface temperature patterns on the marine boundary layer and free troposphere: A study using the atmospheric ARPEGE model. Climate Dynamics, 46(5–6), 1699–1717. https://doi.org/10.1007/s00382-015-2669-z Priestley, M.D.K. and Catto, J.L. (2022). Improved Representation of Extratropical Cyclone Structure in HighResMIP Models, Geophysical Research Letters, 49(5), p. e2021GL096708. https://doi.org/10.1029/2021GL096708 Priestley, M.D.K., Ackerley, D., Catto, J.L., Hodges, K.I., McDonald, R.E., Lee, R.W. (2020). An Overview of the Extratropical Storm Tracks in CMIP6 Historical Simulations. Journal of Climate 33, 6315–6343. https://doi.org/10.1175/JCLI-D-19-0928.1 Page 69 of 76 Rackow, T., Pedruzo-Bagazgoitia, X., Becker, T., Milinski, S., Sandu, I., Aguridan, R., Bechtold, P., Beyer, S., Bidlot, J., Boussetta, S., Deconinck, W., Diamantakis, M., Dueben, P., Dutra, E., Forbes, R., Ghosh, R., Goessling, H. F., Hadade, I., Hegewald, J., … Ziemen, F. (2025). Multi-year simulations at kilometre scale with the Integrated Forecasting System coupled to FESOM2.5 and NEMOv3.4. Geoscientific Model Development, 18(1), 33–69. https://doi.org/10.5194/gmd-18-33-2025 Rao, J., Garfinkel, C., Wu, T., Lu, Y. and Chu, M. (2022). Mean State of the Northern Hemisphere Stratospheric Polar Vortex in Three Generations of CMIP Models. Journal of Climate, 35(14), 4603-4625. https://doi.org/10.1175/JCLI-D-21-0694.1 Ren, Q., Hodges, K.I., Schiemann, R., Dai, Y., Jiang, X., Yang, S. (2023). Seasonal Effects of the Tibetan Plateau on Cyclonic Transient Eddies: A System-Centered View. Journal of Climate 36, 6007–6020. https://doi.org/10.1175/JCLI-D-23-0067.1 Reynolds, C. A., Crawford, W., Thoppil, P. G., Rushley, S. S., Janiga, M. A., McLay, J. J., & Barton, N. P. (2025). Impact of Ocean Resolution on Navy ESPC Forecast Skill. Weather and Forecasting, 40(8), 1463–1479. https://doi.org/10.1175/WAF-D-24-0241.1 Roach, L.A., Dörr, J., Holmes, C.R., Massonnet, F., Blockley, E.W., Notz, D., Rackow, T., Raphael, M.N., O'Farrell, S.P., Bailey, D.A. and Bitz, C.M. (2020). Antarctic sea ice area in CMIP6. Geophysical Research Letters, 47(9), p.e2019GL086729. https://doi.org/10.1029/2019GL086729 Roberts, C. D., Balmaseda, M. A., Tietsche, S., & Vitart, F. (2022). Sensitivity of ECMWF coupled forecasts to improved initialization of the ocean mesoscale. Quarterly Journal of the Royal Meteorological Society, 148(749), 3694–3714. https://doi.org/10.1002/qj.4383 Roberts, M.J., Camp, J., Seddon, J., Vidale, P.L., Hodges, K., Vanniere, B., Mecking, J., Haarsma, R., Bellucci, A., Scoccimarro, E., Caron, L.-P., Chauvin, F., Terray, L., Valcke, S., Moine, M.-P., Putrasahan, D., Roberts, C., Senan, R., Zarzycki, C., Ullrich, P. (2020). Impact of Model Resolution on Tropical Cyclone Simulation Using the HighResMIP–PRIMAVERA Multimodel Ensemble. Journal of Climate 33, 2557–2583. https://doi.org/10.1175/JCLI-D-19-0639.1 Seo, H., O’Neill, L. W., Bourassa, M. A., Czaja, A., Drushka, K., Edson, J. B., Fox-Kemper, B., Frenger, I., Gille, S. T., Kirtman, B. P., Minobe, S., Pendergrass, A. G., Renault, L., Roberts, M. J., Schneider, N., Small, R. J., Stoffelen, A., & Wang, Q. (2023). Ocean Mesoscale and Frontal-Scale Page 70 of 76 Ocean–Atmosphere Interactions and Influence on Large-Scale Climate: A Review. Journal of Climate, 36(7), 1981–2013. https://doi.org/10.1175/JCLI-D-21-0982.1 Small, R. J., deSzoeke, S. P., Xie, S. P., O’Neill, L., Seo, H., Song, Q., Cornillon, P., Spall, M., & Minobe, S. (2008). Air–sea interaction over ocean fronts and eddies. Dynamics of Atmospheres and Oceans, 45(3–4), 274–319. https://doi.org/10.1016/j.dynatmoce.2008.01.001 Strommen, K., Michel, S. L. L. and Christensen, H. M. (2025). The Link between Gulf Stream Precipitation Extremes and European Blocking in General Circulation Models and the Role of Horizontal Resolution. J. Climate, 38, 3763–3786. https://doi.org/10.1175/JCLI-D-24-0311.1 Su, Z., Xie, Y., Huang, J., Wu, G., Liu, Y., Guan, X. (2024). Impact of the Tibetan Plateau on Global High-Frequency Temperature Variability. Journal of Climate 37, 4347–4365. https://doi.org/10.1175/JCLI-D-23-0271.1 Taylor, K.E., Stouffer, R.J. and Meehl, G.A. (2012). An overview of CMIP5 and the experiment design. Bulletin of the American Meteorological Society, 93(4), 485-498. https://doi.org/10.1175/BAMS-D-11-00094.1 Tian, B. and Dong, X. (2020). The double‐ITCZ bias in CMIP3, CMIP5, and CMIP6 models based on annual mean precipitation. Geophysical Research Letters, 47(8), p.e2020GL087232. https://doi.org/10.1029/2020GL087232 Tian, T., Yang, S., Høyer, J.L. et al. (2024). Cooler Arctic surface temperatures simulated by climate models are closer to satellite-based data than the ERA5 reanalysis. Commun Earth Environ 5, 111. https://doi.org/10.1038/s43247-024-01276-z Trenberth, K.E., Smith, L., Qian, T., Dai, A. and Fasullo, J. (2007). Estimates of the global water budget and its annual cycle using observational and model data. Journal of Hydrometeorology, 8(4), 758-769. https://doi.org/10.1175/JHM600.1 Tsartsali, E.E., Haarsma, R.J., Athanasiadis, P.J., Bellucci, A., de Vries, H., Drijfhout, S., De Vries, I.E., Putrahasan, D., Roberts, M.J., Sanchez–Gomez, E. and Roberts, C.D. (2022). Impact of resolution on the atmosphere–ocean coupling along the Gulf Stream in global high resolution models. Climate Dynamics, 58(11), 3317-3333. https://doi.org/10.1007/s00382-021-06098-9 Page 71 of 76 Ullrich, P.A. and Zarzycki, C.M. (2021). TempestExtremes v2.1: a community framework for feature detection, tracking, and analysis in large datasets, Geoscientific Model Development, 14(8), 5023–5048. https://doi.org/10.5194/gmd-14-5023-2021 Wild, M. (2020). The global energy balance as represented in CMIP6 climate models. Climate Dynamics, 55(3), 553-577. https://doi.org/10.1007/s00382-020-05282-7 Willison, J., Robinson, W. A., & Lackmann, G. M. (2013). The Importance of Resolving Mesoscale Latent Heating in the North Atlantic Storm Track. Journal of the Atmospheric Sciences, 70(7), 2234–2250. https://doi.org/10.1175/JAS-D-12-0226.1 Xavier, P., Willett, M., Graham, T., Earnshaw, P., Copsey, D., Narayan, N., Marzin, C., Sellar, A., Ackerley, D., Blockley, E., Bodas-Salcedo, A., Bushell, A., Chua, X. R., Guiavarc’h, C., Hassim, M., Heming, J., Hudson, D., Ineson, S., Jones, C., Keane, R., Kuhlbrodt, T., Martin, G., Mccabe, A., Ridley, J., Roberts, L., Schiemann, R., Storkey, D., Tennant, W., Tomassini, L., Tsushima, Y., West, A., Wheeler, M., Zhu, H., Blaker, A. T., Jones, A., Megann, A., Regayre, L., and Williams, K. (2024). GC5 documentation paper, in preparation. Xie, S.-P. (2004). Satellite Observations of Cool Ocean–Atmosphere Interaction. Bulletin of the American Meteorological Society, 85(2), 195–208. https://doi.org/10.1175/BAMS-85-2-195 Zarzycki, C.M., Ullrich, P.A. (2017). Assessing sensitivities in algorithmic detection of tropical cyclones in climate data. Geophysical Research Letters 44, 1141–1149. https://doi.org/10.1002/2016GL071606 Zhang, Q., Liu, B., Li, S., & Zhou, T. (2023). Understanding models' global sea surface temperature bias in mean state: From CMIP5 to CMIP6. Geophysical Research Letters, 50, e2022GL100888. https://doi.org/10.1029/2022GL100888 Zhang, X., Liu, H. and Zhang, M. (2015). Double ITCZ in coupled ocean‐atmosphere models: From CMIP3 to CMIP5. Geophysical Research Letters, 42(20), 8651-8659. https://doi.org/10.1002/2015GL065973 Zhou, W., L. R. Leung, and J. Lu. (2022). Linking Large-Scale Double-ITCZ Bias to Local-Scale Drizzling Bias in Climate Models. J. Climate, 35, 7965–7979, https://doi.org/10.1175/JCLI-D-22-0336.1. Page 72 of 76 Supplementary Material Table 1. CMIP6 models used in this paper, along with our crude estimate of the nominal atmospheric resolution (km). Model name Ensemble members Nominal resolution (km) ACCESS-CM2 r1i1p1f1 140 AWI-ESM-1-1-LR r1i1p1f1 250 BCC-CSM2 r1i1p1f1 100 CanESM5 r1i1p1f1 250 CESM2-FV2 r1i1p1f1 250 CESM2 r1i1p1f1 100 CESM2-WACCM-FV2 r1i1p1f1 250 CESM2-WACCM r1i1p1f1 100 CNRM-CM6-1 r1i1p1f2 140 CNRM-CM6-1-HR r1i1p1f2 50 CNRM-ESM2 r1i1p1f2 140 EC-Earth3 r1i1p1f1 80 FGOALS-g3 r1i1p1f1 190 FGOALS-f3-K r1i1p1f1 100 GFDL-CM4 r1i1p1f1 100 GISS-E2-1-G r1i1p1f1 250 HadGEM3-GC31-L r1i1p1f3 140 HadGEM3-GC31-MM r1i1p1f3 100 INM-CM4-8 r1i1p1f1 150 INM-CM5-0 r1i1p1f1 150 IPSL-CM6A-LR r1i1p1f1 160 MIROC6 r1i1p1f1 120 MPI-ESM1-2-HR r1i1p1f1 80 MPI-ESM1-2-LR r1i1p1f1 170 MRI-ESM2-0 r1i1p1f1 100 Continued on next page Page 73 of 76 Continued from above page Model name Ensemble members Nominal resolution (km) NorESM2-LM r1i1p1f1 190 NorESM2-MM r1i1p1f1 100 TaiESM1 r1i1p1f1 100 UKESM1-0-LL r1i1p1f2 140