In this manuscript, the authors assess the potential improvements or added value of MED-CORDEX RCSM relative to the GCM driving the RCSM downscaled results. The authors use a set of models coming from CMIP5 and CMIP6.

In the different sections of the manuscript, the authors run a complete set of statistical and spatial analysis of different SST and MHW metrics. This is a very comprehensive study, as it carries out an exhaustive analysis of all the models and the multi-model approach for various relevant metrics. The analyses are well-founded and describe the changes observed in each of the models, explaining the improvement or deterioration in the results in each case. It is interesting that the authors discuss both cases where the impact of using an RCSM is positive and those where it is negative. Furthermore, in some cases, possible causes of the change—whichever direction it takes—are indicated, depending on the models used.

The main conclusion from this manuscript is that the use of MED-CORDEX models, with higher spatial resolution, do not necessarily yield an improvement of SST and MHW analysis and metrics characterization. Whether the results show an improvement or a deterioration depends both on the metric analysed and the model used, as well as its specific configuration. Consequently, it cannot be stated that the higher spatial resolution of RCSM models leads to a general improvement in results. Although this may, on the face of it, seem a discouraging finding, it enables researchers in the field of modelling to identify areas for improvement in future generations of models. In this way, improvements can be incorporated into those model configurations that are yielding poorer results, whilst reinforcing those aspects or processes linked to their improvement.

Hence, my recommendation is to publish the article in its present form by clearly stating that it is an assessment that could improve the future generation of models. The authors should also address the questions below.

QUESTIONS:

CMIP5 and CMIP6 models are analysed together, is there an impact on the results because of comparing models coming from different CMIPs?

For the comparison of the GCM vs RCSM, the authors calculates metrics and statistics in coincident grid points and exclude “empty” points when only one of the GCM or RCSM has data. Have the authors analysed if there is a bias in the results because of excluding points with valid data? How many points are excluded? Is it only a small subset of data points?

Figure 1 (and others) shaded colours do not look like green, red, orange in my pdf.

The authors describe improvement in coastal areas for a good set of metrics and models. Usually coastal areas are “problematic” from the observational point of view at least from satellites. Is there the same problem in the oceanic model as the coastal areas are the model boundaries? Do the authors have a hypothesis why is there more improvement in the coastal areas and semi-enclosed seas?

review of "The added value of Med-CORDEX Coupled High-Resolution Regional Climate Models in representing Sea Surface Temperature and Marine Heatwaves in the Mediterranean Sea" by De Rovere et al

This study applies a series of diagnostics to the daily SST field from ~10 MedCordex simulations and their parent GCM, so as to diagnose when and where RCSMs adds value. The key result is that except for MHW duration and certain aspects of the PDF of SST variability where RCSMs clearly adds value, there is little evidence for systematic improvements across models away from coastlines or regions with topography. Further, rather than horizontal resolution adding substantial value, the authors seem to highlight vertical resolution in the mixed layer as a potentially more important model setting, though in an ensemble of opportunity as used here it is difficult to isolate processes.

This is a solid study that is nearly ready to be accepted. It also highlights the limitations of dynamical downscaling- this reviewer would have hoped that all of this CPU time and effort would have helped matters more. But there is one major issue: I expect a more careful assessment of the pros and cons of the diagnostics used. Specifically, they aren't using process-based diagnostics that attempt to follow the processes that drive SST variability, rather just examine SST itself. For example, we know that MHWs (and SST variability more generally) are driven by anomalies in evaporation, which in turn is driven by weak wind speed and enhanced atmospheric near-surface humidity, and also by enhanced surface downwelling radiation associated with fewer clouds (Bonino et al, already cited; Garfinkel et al 2026; GRL). I think the assessment of SST variability and MHWs might be more informative if the authors analyzed how these processes are represented in GCM vs. RCSMs. I would expect that evaporation would be better handled with enhanced horizontal resolution as fronts are now resolved, and the representation of clouds will also be very different in  a GCM vs. RCSMs especially if the RCSM turns off convection. But this needs to examined closely. The paper currently discussed how oceanic processes might differ in a GCM vs. a RCSMs, but atmospheric processes matter too (Bonino et al and Garfinkel et al 2026) and they are completely ignored. 

To be specific, I don't think the current study needs to include the analysis I'm suggesting above. Rather, a new paragraph in the Discussion would  suffice. This is mainly because at least four additional fields would be needed -  near-surface wind speed, surface downwelling shortwave, cloud fraction, specific humidity in the atmospheric surface layer -  and I don't know if this data is even available. But I certainly think that more could be done with the GCM and RCSMs used here to better understand why RCSMs don't help as much as this reviewer would have hoped. If this data is not available, then this could be a recommendation for the data request in future phases of the Cordex project. More generally, I suggest adding a sentence to the Discussion  that notes the limitations of dynamical downscaling; this reviewer would have hoped that all of this CPU time and effort would help matters more. 

specific comments:

1. Abstract: The role of vertical resolution in the upper ocean seems to be more important than that of horizontal resolution (based on the Discussion). This would seem to me to be an important point worth highlighting in the abstract.

2 line 82: "We focus on such period to maximizes overlap with" needs to be rewritten 

3. Figure 1 and similar: the red and green shading are not colorblind friendly. Please consider changing these colors to ones with more contrast. Also, why doesn't the orange overcompensation shading extend all the way to the green line for perfect error correction?

line 179, 202, 204, 229,230,272: My first language isn't English either, but I think "impoverished" is used incorrectly. Do you mean degraded?

Garfinkel, Chaim I., Sagi Knobler, Dan Liberzon, and Gil Rilov. "Precursors of marine heatwaves in the Eastern Mediterranean." Geophysical Research Letters 53, no. 8 (2026): e2025GL120652.