the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 21 May 2026
Century-long kilometre-scale Ocean eddy-rich global climate simulation with the coupled IFS CY48R1 – FESOM 2.5 model
Abstract. We present novel centennial-scale global climate simulations at kilometre-scale resolution utilizing the coupled IFS-FESOM model, featuring a 9 km atmosphere and a minimal 5 km ocean. Following the HighResMIP protocol, a 50-year high-resolution coupled spin-up was conducted, which was followed by a 65-year historical simulation (1950–2014) and a scenario simulation (SSP2-4.5, 2015-2050). This was accompanied by a 100-year control simulation (1950–2050) employing the 1950 radiative forcing. These simulations explicitly resolve ocean mesoscale eddies within a long-term climate context. Overall, the model demonstrates an improved mean climate state compared to CMIP6 models, with a notable reduction in persistent model biases, except for the polar regions. Performance metrics reveal reduced global errors in surface temperature, winds, and cloud formations. The very high-resolution ocean captures eddy-rich dynamics and realistic boundary current variability, contributing to an improved sea surface salinity patterns and a strengthened Atlantic Meridional Overturning Circulation (peak ~20 Sv). The simulation also reproduces internal climate variability with high fidelity, notably a realistic El Niño–Southern Oscillation with the desired quasi-periodicity (~4–5 years) and realistic winter teleconnection patterns. Sea ice and high-latitude biases have been identified as the primary remaining challenges: the model overestimates the extent of Arctic sea ice, resulting in a cold bias in the Northern high latitudes, while an initialization error in Antarctic snow cover induces a warm bias over Antarctica. Furthermore, there is a warm bias over the Weddell Sea with high ocean mix layer depth, associated with a winter devoid of sea ice. Despite persistent sea-ice and high-latitude biases, the coupled system remains stable over centennial time scales with minimal long-term drift. These results demonstrate the feasibility and scientific value of global coupled climate simulations operating in the ocean eddy-rich regime at sub-10 km resolution. The IFS–FESOM kilometre-scale configuration thus represents a significant step forward in the development of next-generation Earth system models that robustly bridge global climate dynamics and regional-scale processes over multi-decadal to centennial periods.
- Preprint
(2553 KB) - Metadata XML
-
Supplement
(4337 KB) - BibTeX
- EndNote
Status: open (until 16 Jul 2026)
- RC1: 'Comment on egusphere-2026-1289', Anonymous Referee #1, 23 May 2026 reply
-
RC2: 'Comment on egusphere-2026-1289', Pablo Ortega, 01 Jul 2026
reply
I would like to congratulate the authors for the excellent work and commendable dedication in preparing the model, performing the simulations, and demonstrating their added value. Papers like this are very important for advancing km-scale modeling for climate-scale applications. I fully support its publication and provide below a list of suggestions for the authors to address. All the comments below are minor and intended to help clarify and strengthen the manuscript.
Specific Comments
Line 25: The “minimal 5 km ocean” description is a bit deceptive, as it suggests this is the main resolution when in fact you have an unstructured mesh with adaptive resolution, where many regions have 10 km grid spacing. I suggest a more accurate description of the mesh.
Lines 35–36: In this sentence, what does the strengthened AMOC relates to?
Line 37: To what simulations do you refer to? The control? Or all of them?
Line 61: remove a from “advent of a more powerful…”
Line 65: There are two phases of HighResMIP, but both share the same goal. The main difference is that the first phase (Haarsma et al., 2016) mostly included eddy-parameterized and eddy-present configurations, with only a few models reaching eddy-rich resolutions (HadGEM3-GC31-HH, MPI-ESM1.2-ER, EC-Earth3P-VHR, CESM1-CAM5-SE-HR), while the second phase has put more emphasis on eddy-rich scales, allowing models to contribute with only one resolution (Roberts et al., 2025). If you don’t intend to distinguish between the two phases, I would avoid specifying one in particular. The paragraph also fails to mention that 10 km simulations were already performed during the first phase, which helped establish their added value and motivated the greater focus on eddy-rich resolutions in phase 2.
Line 66: Rather than “the effects of high resolution,” I would refer to “the added value of high resolution.”
Line 67: Who is “they”?
Lines 104–107: The 1/12° configuration already resolves the largest eddies. You would need to clarify that this further refinement allows the smaller-scale eddies to be simulated as well.
Lines 124–126: What do you mean by “km-scale simulations”? As mentioned above, HighResMIP phase 1 already included simulations with eddy-rich oceans and highly resolved atmospheres: 25 km for CESM1-CAM5-SE-HR (Chang et al., 2020) and 16 km for EC-Earth3P-VHR (Moreno-Chamarro et al., 2025). If you are referring to atmospheric scales at which storms and convection are resolved, this should be stated more explicitly. Such resolutions, however, are not the ones used in the simulations presented here, so the authors may be referring to something else. In this context, it would be worth mentioning the Climate Digital Twins from Destination Earth, which have underpinned the development and use of km-scale configurations for climate applications.
Line 127: Change “assessing” to “achieving,” and note that this introduces spurious drifts that can compromise the climate signal.
Lines 131–132: Rather than emphasizing that these are “one of the first” simulations (which is vague), I would simply state that this is a contribution to HighResMIP phase 2 / EERIE with IFS-FESOM at 5/10 km resolution.
Line 142: I suggest rephrasing “A brief description of the model components is as follows” as “In the following subsections, the different model components are briefly described.”
Lines 150–152: It is more accurate to say that this IFS version is an evolution of the version used at ECMWF for operational weather predictions (Rackow et al 2025).
Line 213: Remove space after “.”
Line 261: “Here we follow the HighResMIP protocol in performing the EERIE Phase 1 simulations.” This phrasing assumes readers already know what these simulations are. I suggest rephrasing as: “This study presents the phase 1 simulations performed within the EERIE project, which follow the HighResMIP protocol.”
Lines 276–278: I would suggest a different phrasing for this sentence in which you use the control simulation to characterize the drift and potentially correct it from the transient historical and scenario simulations.
Lines 282–283: Aerosol emissions peaked during the 1970s, which implies that the back extension should lead to an overestimation of the actual aerosol forcing during the 1950s and 1960s in the simulations. This should be acknowledged.
Lines 291–293: In IFS-NEMO’s version of the ClimateDT the aerosol loadings in the projections are not from MACv2-SP (See https://platform.destine.eu/services/documents-and-api/doc/?service_name=climate-dt-user-guide). They are the same as those used in CONFESS to avoid inconsistencies, but their future evolution is scaled based on MACv2-SP projections. I imagine that it’s the same for your version of IFS-FESOM.
Line 303: Change “or those originating” to “from those originating.”
Lines 310–311: You could specify that this stability/minimal drift is necessary to be able to interpret the forced and internally-driven variability while avoiding the effects of spurious signals.
Line 318: It’s important to clarify that this correspondence refers to long-term trends, since decadal-scale variability is considered to be largely of internal origin and is therefore not expected to be reproduced by the transient simulations.
Lines 345–347: Indeed, the control simulation can be used to diagnose and “correct” the spurious drift in the transient simulations.
Lines 348–359: Note that, by construction, the anomalies in Figure 3 cannot be interpreted as biases, since they are not computed relative to any observational benchmark but to the initial simulation state. Without further evidence, it is also possible that the initial state itself is too warm due to insufficient spin-up, and that this apparent warm bias disappears as the model approaches its own attractor. The same applies to the subsurface temperature and salinity biases discussed later. A supplementary figure showing the anomaly of the control simulation relative to 1950 EN4 values would help support or refute this interpretation. In addition, it would be useful to quantify the drift as a trend, and to show how much of the annual-scale variance is explained by this trend.
Lines 355–359: It is expected that the control run exhibits a smaller drift (trend) than the OMIP runs, since the latter are subject to the forced historical evolution from reanalysed surface atmospheric fluxes.
Line 385: Are you referring to the Southern Hemisphere summer here?
Lines 398–402: I would expect observational uncertainty at the ocean subsurface to be particularly large in the Arctic, which raises the question of whether these are true model biases.
Lines 421–422: Note that the mixed layer depth (mlotst) can vary considerably across reanalyses (see e.g. Toyoda et al., 2017), which raises the question of whether these biases, estimated against a single reanalysis, are reliable.
Line 469: Change “temperature bias” to “temperature biases.”
Figure 9: It would be useful to add statistical significance markers (as in Figure 11) to identify which aspects of the regression patterns are robust and therefore worth discussing. One example is the negative SST anomalies in the equatorial Atlantic, which are relatively small in magnitude yet discussed in the text.
Line 702: Change “association” to “regression.”
Lines 700–734: Some of the differences might arise from different sampling of ENSO events between the simulations and the observations over the 36-year period considered, given the critical role that the most intense El Niño/La Niña events (whose sampling may differ largely between the model and the observations) play in shaping ENSO climate impacts.
Figure 11: The stippling is hard to see in some of the more saturated color areas. You could adjust the colorbar to reduce saturation and improve visibility.
Lines 759–776: You interpret the deviation of IFS-FESOM from the observed location of the centers of action as a result of model SST biases. However, this could also be explained by differences in simulated versus observed internal variability. Figure 7 of Carmo-Costa et al. (2025) shows, for a selection of historical ensembles, that the location of the centers of action (defined by the maximum of the high-pressure system) can vary substantially across ensemble members for most models, indicating a strong influence of internal variability. The fact that the AMIP simulations reproduce the correct location does not invalidate this interpretation, since by construction the AMIP simulations use observed SST patterns and therefore sample the same internal variability as the observations.
Section 4.6: You state several times that the IFS-FESOM simulation has reduced biases relative to the CMIP6 multi-model mean without providing supporting evidence or citing a relevant reference. It may suffice to refer to the results of the performance index table.
Line 956: Change “linse” to “lines.”
Figure 17: This plot appears noticeably blurrier than the others.
Lines 1029–1030: Providing trends for the RAPID period, in both the model and the observations, would help support this statement.
Line 1031: You have not actually demonstrated that AMOC sensitivity increases with resolution. Also, to further validate the realism of the AMOC in the model, it would also be useful to show the vertical climatological profile in both the model and the observations.
Figure 15c: I would recommend specifying what is meant by “winter” here, as you do in lines 908–909, to avoid confusion, since this refers to different months in the Northern and Southern Hemispheres.
Discussion section: This section reads more like a Results/Conclusions section, with little discussion of the results in the context of past literature. It may make more sense to merge it with the Conclusions section, which is comparatively small and repeats several of the messages.
References
Carmo-Costa, T., Bilbao, R., Robson, J., Teles-Machado, A., and Ortega, P.: A multi-model analysis of the decadal prediction skill for the North Atlantic ocean heat content, Earth Syst. Dynam., 16, 1001–1028, https://doi.org/10.5194/esd-16-1001-2025, 2025.
Chang, P., Zhang, S., Danabasoglu, G., Yeager, S. G., Fu, H., Wang, H., and et al.: An unprecedented set of high-resolution earth system simulations for understanding multiscale interactions in climate variability and change, J. Adv. Model. Earth Syst., 12, e2020MS002298, https://doi.org/10.1029/2020MS002298, 2020.
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., and von Storch, J.-S.: 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, 2016.
Moreno-Chamarro, E., Arsouze, T., Acosta, M., Bretonnière, P.-A., Castrillo, M., Ferrer, E., Frigola, A., Kuznetsova, D., Martin-Martinez, E., Ortega, P., and Palomas, S.: The very-high resolution configuration of the EC-Earth global model for HighResMIP, Geosci. Model Dev., 18, 461–482, https://doi.org/10.5194/gmd-18-461-2025, 2025.
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., Jung, T., Keeley, S., Kluft, L., Koldunov, N., Koldunov, A., Kölling, T., Kousal, J., Kühnlein, C., Maciel, P., Mogensen, K., Quintino, T., Polichtchouk, I., Reuter, B., Sármány, D., Scholz, P., Sidorenko, D., Streffing, J., Sützl, B., Takasuka, D., Tietsche, S., Valentini, M., Vannière, B., Wedi, N., Zampieri, L., and Ziemen, F.: Multi-year simulations at kilometre scale with the Integrated Forecasting System coupled to FESOM2.5 and NEMOv3.4, Geosci. Model Dev., 18, 33–69, https://doi.org/10.5194/gmd-18-33-2025, 2025.
Toyoda, T., Fujii, Y., Kuragano, T., Kamachi, M., Ishikawa, Y., Masuda, S., Sato, K., Awaji, T., Hernandez, F., Ferry, N., Guinehut, S., Martin, M., Peterson, K. A., Good, S. A., Valdivieso, M., Haines, K., Storto, A., Masina, S., Köhl, A., Zuo, H., Balmaseda, M., Yan, C., Fujii, Y., Toyama, K., Tsujino, H., Matsumoto, S., Sakamoto, K., Murakami, S., and Usui, N.: Intercomparison and validation of the mixed layer depth fields of global ocean syntheses, Clim. Dynam., 49, 753–773, https://doi.org/10.1007/s00382-015-2637-7, 2017
Citation: https://doi.org/10.5194/egusphere-2026-1289-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 246 | 83 | 17 | 346 | 26 | 13 | 15 |
- HTML: 246
- PDF: 83
- XML: 17
- Total: 346
- Supplement: 26
- BibTeX: 13
- EndNote: 15
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This is a major achievement, congratulations!
What I miss is mainly more context and some key details:
Could you devote a paragraph other groups with similar projects? ICON is one, but there are probably also efforts in Japan, the US and China.
SST is clearly improved compared to the 'standard' 1 deg. setups, but with precip. I am not quite sure. Could you add one extra panel for SST and precip each, that shows the IFS-FESOM biases in a coarser set-up, like 1/2 or 1 degree?
It appears as if the model is simply switched on and then analyzed, but it probably required years of engineering to get there. For the non-modeller,
could you describe some of the key steps?
The FESOM grid promised a better Gulfstream separation. Has this been achieved and could you be explicit about it?
There appears to be no AABW in the Atlantic. Could you show the global MOC as well? And how strong is the ACC?
In all studies that I know, the tropical precip biases are similar in coupled and uncoupled mode, whereas in your model, the uncoupled IFS is almost bias free. Is this the result of skilled tuning, or are there key processes that only exist in IFS but not in other models?