Summary:

The manuscript represents updates of a South China Sea Operational Oceanography Forecasting System from version 1 to version2, which provide daily updated hydrodynamic forecasting in the South China Sea for the future 5 days. Comprehensive updates of model configuration and assimilation schemes have been mentioned. Among them, three major changes have been highlighted, namely the way of prescribing buoyancy flux, the tracer advection discrete scheme and data assimilation scheme. The model shows enhanced performance in the accuracy of the sea surface temperature and sea surface height.

Ocean states prediction has always been a challenging task and is of vital importance to the hazard prevention such as tropical cyclones and internal waves and so on. The South China Sea has unique monsoon system and topography and external Kuroshio forcing, which masks it a challenging region for ocean prediction. This paper is generally well written, and the way of improvement is well presented, which makes the comparison of results quite convincing. I believe this manuscript can be the basis of a useful publication after minor improvements.

1 There are too many acronyms and I sometimes have to go back to look for its meaning. I would recommend keeping some acronyms frequently used by other articles, such as SCS and OFS and SST, but don’t use abbreviation for only two words (such as RTOFS, PI) and abbreviation that is too hard to recall (such as RSUP3, U3H and C4V).

2 Line 1: The improvements to the regional South China Sea Operational Oceanography Forecasting System include all the previous versions. I suggest removing the acronym of “(SCSOFSv1)”.

3 Line 199: Please clarify the wards in this headline. What are the highlights and sensitive updates respectively.

4 Section 3.1, Line 220: The use of bulk formulation to calculate the buoyancy fluxes is reasonable, but it is not a real negative feedback because the atmospheric forcing, such as air temperature, relative humidity are prescribed, which are not adjusted from the modelled SST. Please clarify this sentence, e.g. how the SST is improved through the use of bulk formulation should be further elaborated.

5 Figure 3. For the SCSOFSv1, the area north of 24°N in the BulkFormula is even warmer than that in the no BulkFormula experiment. But the SST is much improved in the SCSOFSv2. Later results in Section 3.2 indicate that it may be related to the improved advection and mixing scheme. Please further explained this by providing more information.

6 Section 3.2: What about the temperature and salinity bias in the subsurface layer in the AAG scheme combination?

7 Line 335. Please explain the improvement of temperature and salinity with more detail. What is the diffusion term and advection term look like in the AGG scheme combination?

8 Line 360: Why do you set observational error for the SLA and SST as 0.09 cm and 0.5 °C? The along-track satellite data contains high-frequency noise, especially for the shallow area (Zhou et al., 2015). How do you filter out this noise?

Zhou, Xiao-Hui, Dong-Ping Wang, and Dake Chen. “Global wavenumber spectrum with corrections for altimeter high-frequency noise.” Journal of Physical Oceanography 45.2 (2015): 495-503.

9 Line 428: In the section 4, I suggest introducing why the SST, SLA and T/S profiles are used to validate the model. Are those element enough to represent the outputs from the model? In addition, you provide the importance of SST in Line 210 to connect with the sea surface atmospheric forcings, but what kind of validations is related to section 3.2 and 3.3.

10 Table 1: In Line 145 you mentioned the the new SODA 3.3.1 and 3.3.2 reanalysis were used, but in Table 1 you still mention the SODA 2.2.4, please check all the settings in this table.

11 Figure 13: Why the RMSE of temperature is suddenly large in June in Fig. 6b? In the paper of a recently published paper (Ding et al., 2021), there is also a similar large bias in June, can you provide some explanations to this?

Ding, R., Xuan, J., Zhang, T., Zhou, L., Zhou, F., Meng, Q., and Kang, I. (2021), Eddy-Induced Heat Transport in the South China Sea, Journal of Physical Oceanography, 51(7), 2329-2349, doi: 10.1175/JPO-D-20-0206.1.

12 Please check the fontsize in all the figures to make sure it is clear. Also add title for similar plots.

The authors describe the evolution of the South China Sea operational oceanography forecasting system (SCSOFS), which significantly improve the quality of the system. The manuscript provides a detailed description of different versions of the ROMS model which were used for SCSOFSv1 and SCSOFSv2 and compare the 11-year long free run of both models. The results show various improvements for the changes that have been made to the operational system. It provides useful suggestions for the community. My opinion is that the paper is worthy of publication in GMD after minor modifications. But several specific aspects of the paper detail and analysis should be considered first.

Specific comments:

The grammar and quality of the writing needs to be improved before publication: many sentences are over-complicated or confusing (Example lines: 113-114, 170-174, 341-343....). It would benefit from a thorough revision by someone with proficient written English skills.

Line 99: the section title is not clear, which datasets? Not all datasets used in the article are described here. 

Line 142: For what reasons are the initial conditions not deduced from soda 3.3 in correspondence with the boundary conditions?

Lines 180-184: Why does the inclusion of Guam at the open boundary have such a significant influence on the latitude of the NEC in the model? This need be explored or explained.

Lines 189-193: Why these choices as opposed to the ones used in v1? What is special about 3rd order and biharmonic?

Line 227: Any reference for OSTIA first appeared in the text? Is it provided to SCSOFS directly by the Metoffice or by other means?

Line 326: the sentence on the conclusion may come after the lines 327-337.

Line 355: Where do you get the data from in SCSOFSv2, is it still the same servers? Where do the ARGO observations come from?

Line 373: “around” is not an exact annotation, need to point out exactly.

Line 383: “for the calculation of innovations” change to “when calculating innovations”

Line 385: Lee et al., 22-25 Jun, format error!

Line 619: What are the known biases of the system? Some are described in the article; can you summarize them in the conclusion? Can you elaborate the summary in terms of what is the added value of this regional system compared to existing global systems?

Section 2: A table summarizing the v2 and v1 features would be useful.

Fig. 8: Why 7-day FGAT? Why not 3-days or 10-days? Can you explain more on this?