Review of "An assessment of equatorial Atlantic interannual variability in OMIP simulations" by A. Prigent and R. Farneti.

This paper focuses on the evaluation of the realism of the seasonal and interannual variabilities in the Atlantic equatorial band (3°S-3°N) as simulated by some global ocean models in the context of the Ocean Model Intercomparison Project Phases 1 (OMIP1) and 2 (OMIP2). The two exercises differ in the surface forcing, i.e. CORE-II for OMPI1 and JRA55-do for OMIP2. Ensemble means are computed using 6 models for OMIP1 and 7 models for OMIP2, and analyses are performed over a 20-year period (1985-2004). The authors report classical biases in the ocean mean state for OMIP1 and OMIP2 and highlight a drastically reduced interannual variability in OMIP2 (compared to OMIP1) in SSH, SST, and subsurface temperature. Using model experiments with the GFDL-MOM5 model, they attribute the differences between OMIP1 and OMIP2 interannual variability to surface wind forcing.

General comments:

This paper is useful for the modeling community and for the improvement of ocean models. The figures are of good quality and the writing is good. However, the paper could be significantly improved. In particular:

- The introduction needs to be entirely revised. The actual introduction is based on the analysis of 4 figures (Figures 1 and 2, and Figures S1 and S2) that are already part of the paper’s results. On the other hand, the forced and coupled dynamics of the equatorial Atlantic are hardly explained. One can also wonder why it is important to document the equatorial Atlantic interannual variability. Specific questions seem to be thrown at the end of the introduction. 1) Why analyzing the seasonal cycle, knowing that the paper focuses on the interannual variability? 2) Analyzing the difference in interannual variability between OMIP1 and OMIP2: we already know that OMIP2 lacks variability, it has been diagnosed in Figures 2, S1, and S2. 3) Does the interannual variability depend on the atmospheric forcing used?: This is a rhetorical question because OMIP1 and OMIP2 differ only in their atmospheric forcing (see your own comment at line 425).

- It would be very nice if the authors could use the data from the PIRATA buoy network to assess the monthly climatological state of the ocean models. Depending on the availability of observations, the authors could also assess the realism of the interannual temperature variability in OMIP1 and OMIP2 using PIRATA data.

- The model experiments carried out with the GFDL-MOM5 model (Section 5) are not very informative, knowing that the seasonal and interannual variability in the equatorial Atlantic is mostly linear. If the model uses classical bulk formulations (this information is not given in the manuscript), then the prescribed surface winds control many aspects of the surface forcing (wind stress, latent, and sensible heat, evaporation). In particular, the model sensitivity experiment (MOM-LR-winds) designed to analyze the role of the surface winds on the interannual variability does not allow to disentangle the momentum forcing from the heat and freshwater forcing, which is a weakness for the interpretation of the results. Furthermore, the use of bulk formulae to estimate the surface wind stress is accompanied by a drastic dependence of the wind stress amplitude on the climatological SST (see how the drag coefficient is estimated in the model), which again limits the interpretation of the difference between MOM-LR and MOM-LR-winds. An additional experiment could be run with the GFDL-MOM5 model to analyze the effect of changes in the mean state on the interannual variability. I suggest running MOM-LR forced by climatological winds / wind stress from CORE-II and the anomalies from JRA55-do. Or test the role of the forcing off the equatorial band, as compared to the local equatorial forcing.

- Note that the seasonal cycle is the seasonal deviation relative to the ocean mean state. For this study, the authors have to (estimate and) refer to the monthly climatology, which, in contrast, does contain the long-term mean.

Specific comments:

1. Introduction:

Fig.1: Caption mentions anomalies, are these interannual anomalies? If yes, improve the caption and clearly state that you are describing interannual variability in Lines 17-20. Note that in many studies such as in M. Martìn-Rey’s work, they do not only remove the linear trend, but they remove the 7-yr low-frequency component (using fft).

Fig.1: The boxes can be removed. Also, NINO3.4 is not used in the article.

L27: You could replace Dakar with Senegal to have two country names.

L28: “Discrepancies” should be replaced by differences.

L45: ENSO acronym has already been defined (and is used only twice in the paper).

L55: There is an unnecessary closing bracket.

L64: “was comparable”, do you mean that the magnitude was comparable?

L67: OMIP acronym has already been defined.

L70: CMIP acronym has already been defined.

1. Data

Table 1: The ocean resolution column is not a resolution but a number of points. What are the criteria that make you choose these specific models? Are these all available models with a resolution lower or equal to 1°x1°? Why did you choose an unequal number of models between OMIP1 and OMIP2? I notice that some of the models are identical between the two phases 0-9, 2-10, 4-11, and 5-12. Why can’t you use the same model ensemble for both phases?

L119: The 55km zonal resolution is only the resolution at the equator.

L129: What is the criterium to choose 18 CMIP6 models?

L130: What is rli1p1f1?

L133: Add modeling to “We conducted several experiments”.

L134: What does z* mean?

L136: What does nominal mean?

L140: What is the bulk formula used for the estimation of momentum/heat/freshwater fluxes? What about the rivers, is there a relaxation to climatological SSS or runoffs?

L142: Specify where the 10m-winds are used in the surface forcing estimation (wind stress, latent and sensible heat, evaporation).

L147: Specify if the prescribed longwave is the longwave_in or the sum of longwave_in and longwave_out (that depends on SST**4).

L163: Is this potential density?

L167: What is the expected influence of a change in the thermocline tilt? Modal dispersion?

L170: The feedbacks could be explained in the introduction, along with the impacts of changes in certain components.

1. Comparison of the monthly climatologies

L177: See my general comment on the definition of seasonal cycle vs. monthly climatology. Also introduce this section, because it is not obvious to all readers why it is important to evaluate the realism of the ocean mean state and its seasonal variations and what are the implications of biases on the interannual variability.

L185: The seasonal cycle of SLA is driven by resonance modes (Brandt et al., 2016) associated with baroclinic modes 2 (at semiannual frequency) and 4 (at annual frequency).

Brandt, P., Claus, M., Greatbatch, R. J., Kopte, R., Toole, J. M., Johns, W. E., and Böning, C. W.: Annual and semiannual cycle of equatorial Atlantic circulation associated with basin-mode resonance, J. Phys. Oceanogr., 46, 3011–3029, https://doi.org/10.1175/Jpo-D-15-0248.1, 2016.

L195: Can you comment on the eastern part of the basin, which is more important for the Bjerknes feedbacks.

L214: What about the stratification (you could use the 24°C isotherm for the calculation).

L198-233: Summarize all the estimated values in a table. The text is too technical to grasp the main message.

Figure 3: On the right, you should add 3 curves for ATL4 or ATL3 averaged values. The y-axis labels should be centered between ticks positioned at the beginning and end of the month. Currently, half a month is missing at the beginning of January and half a month is missing at the end of December. Furthermore, the figure caption can be reduced. Sentences are too repetitive.

Figure 4: For a better comparison, can you align subplot a) with subplots c) and g), and align subplot b), with subplots e) and i). Can you plot ORAS5 vertical velocity?

1. Comparison of the interannual variability

L254: Imbol Koungue et al (2017) is not an appropriate reference, this study is not about equatorial waves as it focuses on Benguela Niño/Niña events.

L259: Why don’t you compare OMIPs to ORAS5.

Figure 5: On the right, you should add 3 curves for ATL4 or ATL3 averaged values. Caption: What do the horizontal lines highlight? Specify that vertical lines denote the ATL4/ATL3 regions. The caption could be drastically reduced: “Same as Figure 3 but for the monthly climatological standard deviation of interannual anomalies.”

Figure 6: Does BF1 have some meaning in the case of a forced simulation?

L270: Mention (here or in the introduction) that the peaks of variability correspond to the classical Atlantic Niños/Niña events phase-locked to boreal spring/summer and the Atlantic Niño II in November-December (Okumura and Xie, 2006).

Okumura, Y., and S. Xie, 2006: Some Overlooked Features of Tropical Atlantic Climate Leading to a New Niño-Like Phenomenon. J. Climate, 19, 5859–5874, https://doi.org/10.1175/JCLI3928.1.

L279: Replace the word disparities with biases.

L300: I guess that the plus/minus 10 meters has been chosen quite arbitrarily?

L302: How does the thermocline depth influence the MLD. The MLD is controlled by momentum stress, isn’t it?

L303: The thermocline is not that close to the MLD, maybe the word “closer” is better here.

L307-L310: Quantify by how much the subsurface temperature anomalies have been reduced compared to ORAS5 (or from OMIP1 to OMIP2).

Figure 7: On top of each panel, you could plot the interannual SSH variability (STD), which should mirror the subsurface temperature variability.

1. Sensitivity tests on the wind forcing

L328-329: In forced ocean models/simulations, the surface forcing controls the mean state, the seasonal cycle, and the variability. This statement is quite empty here (same as question 3 at the end of the introduction). What could be important to test is the effect of the forcing away from the Atlantic equatorial band as opposed to the local equatorial forcing.

L334-351: I do not see the purpose of comparing CORE-II and JRA55-do to other surface wind products. Can you please introduce this paragraph with your objectives?

L351: Have these simulations/model configurations been validated?

L352: I do not get the implication of the “consequently”.

L372: “We have demonstrated” is a very strong statement. In the equatorial Atlantic, the ocean dynamics is mostly linear (see work by P. Brandt, S. Illig, or others), so there is no surprise here. That is why the shift of one month in the wind forcing causes the shift of one month in SSH variability (Line 366-368).

Figure 9: The subplots c) and d) are mistakenly referred to as a) and b).

Figure 10: On the right side of the plot, you should add ATL3 curves for both SSH (top panels) and SST (bottom panels).

1. Conclusion and Discussion:

L389-391: this statement seems out of context.

L394-396: This can be proven with model experiments (see my general comment).

L407: This can be associated with the estimation of the drag coefficient.

L424: The fact that models in OMIP1 and OMIP2 use the same model physics should be said in section 2.1.2. This echoes with my previous question: why do you use different models for OMIP1 and OMIP2 ensemble means?

L425: see my general comment.

Figure 11: The point associated with MOM-LR-winds could be blue because it is closer to OMIP1 protocol (shown with blue numbers).

This study compares tropical Atlantic variability among forced ocean simulations (CORE-I and CORE-II) and a subset of CMIP6 models and identifies a diffusive thermocline bias among models. 

My primary concern with this study is that the model representation is biased towards Eulerian vertical coordinate models such as MOM5 . NorESM is the only isopycnal coordinate configuration , however it is using a high background vertical diffusivity ( nominally 1-1.5e-5 m2 s-1). Near-equatorial background levels are reduced in several CMIP configurations, notably NOAA/GFDL-CM2G (https://doi.org/10.1175/2008JPO3708.1) ,which is a quasi-Isopycnal coordinate model, similar to NorESM.  

Model sensitivity results suggest that increasing model resolution slightly reduces the diffuse thermocline bias (MOM5-HR).  This is not discussed further and deserves further attention. Would an implication be that additional high resolution studies are needed to assess to what degree stratification bias can be reduced by increasing horizontal resolution?  To what extent could improved representation result from numerics (e.g. Lagrangian coordinate models)?  Including an isopycnal with low equatorial diffusivities (CM2G) would help to address this question.

Figure quality is good. In Figures 3 and 4 (and perhaps 5), it would be helpful to show anomalies for all fields, with respect to ORA-S5.

Did the authors consider analyzing mean and time-varying contributions to the upwelling heat budget, i.e. how much of the variability is related to changes in the background stratification/upwelling versus eddy contributions?  This could be helpful for the disussion, however, the existing figures reasonably convey the point of the dominance of vertical processes in this region.