This paper summarizes the results of a global unstructured mesh configuration for the WW3 model that uses high resolution to resolve small islands in the tropical ocean. The model spans resolutions between 50km and 100m, and the paper demonstrates its ability to capture the impacts of distant-source swells on tropical island costs. This global island-resolving configuration is novel and results presented are impressive. Overall, the message of the manuscript is clear and the text is well organized and well written. Below, I have some comments and suggestions that I hope will strengthen and help clarify the paper prior to publication.

General Comments:

1. The resolution used around islands could be clarified a bit. The sentence starting in line 89 states: “Around the selected validation sites (see Sect.2.2.2), the mesh is further refined, reaching a resolution of 100m at La Reunion island where field observations are available close to shore.” Calling out La Reunion here makes it unclear whether the 100m resolution extends to all validation sites, or just La Reunion. From Figure 1, it looks like the validation sites have different minimum resolutions, this should be more explicitly stated and explained. Table 1 may be a good place to summarize the minimum resolutions around each validation location. Also, the refinement criteria would be good to discuss. For example, how is the resolution function based on distance to coast, depth dependent, bathymetric slope dependent, etc.
2. The authors mention islands smaller than 10km^2 were removed, which makes sense from a total node-count perspective. However, these could still be parameterized with the source term dissipation approach from Mentaschi et al. 2018. Was this considered?
3. In Figure 1, it would be more helpful to show the mesh resolution color contours in the global image rather than the bathymetry. The bathymetry is more helpful in the zoom boxes where the actual mesh is overlaid. Also, the global NRMSE results show large improvements in the areas around the Solomon Islands and French Polynesia. Presumably these are 1km resolution regions. It may be helpful to have a figure showing the mesh in these areas to give a sense of the resolution provided in those regions, since it clearly has a large contribution to reducing local errors.
4. I think the bias correction approach could be clarified starting in Line 108. Is the correction applied as a piecewise multiplication factor? Is this a correct way to express the correction: W = W for W<15m/s, W=1.1*W for 20m/s<W<25m/s, W = 1.15*W for W > 25m/s?. It may also be good to show a figure for the bias corrected winds similar to Figure 2.
5. Since the ERA5 bias correction strengthens the high winds, it seems counterintuitive that a positive bias is present for the uncorrected ERA5 SG model, while the ERA5 correction results in a negative bias for the SG model. Perhaps I am misinterpreting the bais values, but I’m used to thinking of a positive bias meaning the model is overpredicting observations (and a negative indicating it is underpredicting) Some explanation for this would be helpful.
6. In Figure 5, it is interesting that overall, the uncorrected SG has a lower absolute bias than either of the ERA5_QC models for the 10-12m wave bins. I would expect the ERA5 bias correction to have the greatest effect here.
7. The results in Section 3.2 are quite good, but I feel like they could be further strengthened and made more impactful by showing how the wave bulk parameters improve with resolution. For example, it would be really useful to see what 100m resolution buys you in terms of accuracy vs the 1km used for other islands. This would give the reader more guidance on the right level to use for their given application. I understand there may be restrictions here as it relates to resolving the actual observation location. I also understand this would mean developing a new mesh and running more simulations, which may not be practical. However, some commentary on whether 100m there is likely to be a difference between, say 100m and 200m resolution, would be helpful. This is another reason it would be good to be more clear about the local resolution used (see first comment) because that could provide these types of insights as well.
8. More could be said about the computational performance of the model. Line 277 states: “Indeed, for the same computational resources, a one year simulation with implicit scheme took 12 h.” What was the number of MPI ranks used for this to get this throughput and how does the parallel scalability of this configuration compare to Abdolali et al., 2020)?

Specific Comments:

1. The sentence starting in line 35 should better distinguish between the obstruction mask approach in Tolman 2003 and the approach in Mentaschi et al. 2018. The latter of these is based on a dissipation source term can be for both structured and unstructured grid models.
2. The latest release version of WW3 on the NOAA-EMC repo is 6.07.1, so more information should be provided about what the version 7.14 used in this study represents (and how to access it).
3. In the paragraph starting with line 65, triad interactions could be listed as an important source term for shallow water along with bottom friction and breaking-induced dissipation.
4. I would recommend adding the dates used to validate each site to Table 1.
5. The text in Section 3.1 was a little unclear to me at first reading (although certainly made sense when I read it more carefully). I would recommend a slight tweak to make this a little more clear: “Two different comparisons are performed in this section. The model validation in deep water is first performed *only* on structured grids to evaluate the effect of wind field correction. *Next,* the evaluation of the spatial discretisation is carried out comparing SG and UG approaches, both including the wind correction."
6. Even though the NRMSE and bias metrics are standard, I think they should be explicitly defined in the paper for completeness.
7. The overall mean NRMSE and NBIAS values in red on the plots in Figures 3 and 4 are difficult to read, larger text in black would be better.
8. I would recommend that the paragraph starting in Line 278 be moved ahead of the discussion of the two previous paragraphs on performance and numerics. I think this would flow better since now it shifts somewhat abruptly back to discussion of model results vs observations.
9. Line 285 that “The underestimation of the wave energy spectrum below 0.04 Hz will be discussed in the next section.” I didn’t see this discussion there. The next section seems to focus on the high frequency underestimation (Figure 13).

This paper presents a global unstructured mesh configuration of the WW3 model tailored for capturing distant-source swells in the tropical ocean, especially around small volcanic islands. The authors implemented the spectral wave model WAVEWATCH III over a variable-resolution grid, with finer mesh around island shorelines (as low as 100m), allowing for a more detailed representation of tropical islands. The model, forced by ERA5 wind fields and corrected for biases via satellite data, seeks to address limitations in previous models that relied on coarse, regular grids with obstruction masks, which introduced negative biases. The new approach demonstrates improved predictive accuracy for sea states in tropical areas, and results are compared with in-situ data nearshore at depths between 10-30m. However, several aspects need refinement, such as clarifying resolution differences around validation sites, examining triad interactions and partition comparison in shallow water, and addressing wind field temporal resolution. The manuscript’s clear and well-organized structure contributes valuable insights into the challenges and methodologies for wave modeling in complex island environments.

Major Comments:

1. Clarification of Novelty: The claim that this is the first study to directly compare with nearshore data in 10-30m depths is inaccurate; other global wave model studies have also achieved this. While this study is pioneering in certain aspects, the authors should revise the abstract and relevant sections to reflect this context accurately.
2. Wind Field Temporal Resolution: The choice of 3-hourly ERA5 data might be insufficient for fast-moving systems, such as hurricanes. Given that ERA5 offers hourly data, it would be valuable to understand why 3-hourly data was chosen. Further, suggestions on time interpolation techniques to better capture these conditions would be helpful.
3. Triad Interaction in Shallow Waters: While the study employs triad interactions, the impact on shallow water gauges isn’t clearly demonstrated. It would be beneficial to include an analysis of triad effects on gauge results in these areas.
4. Handling of Singularities with Regular Grid: If the regular grid configuration masked out the North Pole to avoid singularities, clarify at which latitude this masking was applied and what is the consequences on the results?
5. Quantifying Swells without Partitioning: The paper does not describe a method for separating wind-sea and swell partitions, which is essential for accurately quantifying swells (this is the main argument of this paper). Swell partitioning would allow better comparison between modeled and observed swells and enhance understanding of the model’s performance across different wave types. This would be beneficial, especially in coastal observations, to assess model accuracy across partitions.

Minor Comments:

1. Figure 5 Labeling Issue: Ensure the y-axis label in Figure 5 is fully visible.