While the authors do acknowledge that model bathymetry remains a limiting error in prior hydrodynamic solutions, they state that it is still the "overwhelming priority" and "enormous efforts have been dedicated to this" in the FES2014 configuration. However, more specifics on these efforts and their limitations would strengthen the paper. Additionally, the composite nature of the bathymetry, drawing from various sources with potentially varying accuracies and resolutions, introduces inherent uncertainties that are not fully addressed.

The authors mention using 19 regions for bathymetry perturbations but do not provide a map or a detailed description of these regions. Understanding the spatial distribution of these perturbations is crucial for assessing the robustness of the ensemble approach and the potential influence of localized bathymetry errors on the overall solution.

The justification for using FES99-derived LSA for some components and FES2012-derived LSA for others lacks clarity. A more detailed explanation of the physical reasoning behind these choices and the potential implications of using different LSA sources would be beneficial.

The rationale for data decimation in the assimilation process, especially at high latitudes, requires more clarification. The trade-off between computational cost and data coverage needs to be explicitly addressed. It is unclear how the authors determined the optimal balance between these competing factors.

The discussion of M4 tidal constituent processing highlights challenges in separating the signal from other oceanographic signals. More details on the specific methodologies employed and their effectiveness would provide valuable context for interpreting the accuracy of M4 predictions in the FES2014 atlas.

The validation is largely based on comparisons with other global tide models, some of which (like TPXO9v2) were released *after* FES2014. While comparison with TPXO8 would have been more contemporaneous, the authors justifiedly use TPXO9v2 due to its superior accuracy. However, using newer models in validation can make the assessed model look better than it actually was at the time of its creation, since other model advancements have since occurred. More emphasis on independent validation with in-situ data, beyond the Australian ADCP network, would enhance the validation.