Li et al. 2026 presented a temperature-salinity dataset with a 4D-MGA method using WOD in weekly 0.25-deg resolutions in the global oceans from 2005 to 2023, compared with datasets from ARMOR3D and GLORYS. The analyzed temperature and salinity distributions shows reasonable magnitude of geostrophic currents at western boundaries. The datasets should be helpful to the community and could be published in ESSD. However, I am unable to accept the manuscript in the present form due to numerous questions when I go through the manuscript. My major concerns are (1) OISST dataset is an observation-based analysis filled to the global ocean, which is not suitable to ingest into the 4D-MGA as observations, (2) AVISO SLA is ingested into 4D-MGA while is compared with as independent data, (3) Comparisons with ARMOR3D and GLORYS may also be problematic since it is not mentioned whether they also ingest AVISO and OISST datasets, (4) Comparison against WOA23 appears independent, but actually WOA23 and WOD23 are not independent, (5) Comparisons with selected months and years may misleading, and should be done over the entire period of 2005-2023, (6) Many statements should be carefully clarified as detailed below.

Abstract: It should be stated that 4D-MGA uses WOD24.

L14, delete “http”

L14-15, add descriptions of the linear trend of WBCs

L15-18, it is better to be removed in “Abstract”

In introduction section, should the authors describe the 4D-Var that is popular in data assimilation?

L86-90, it is problematic to include DOISSTv2.1 and AVISO as “observations” while they are actually observation-based analyses, not real observations.

L112, what is “current grid layer”? Is it the vertical layers of the 4D-MGA system? How many vertical layers has the 4D-MGA, and what is the thickness for each layer?

Figure 1, what is exactly the “adjustment for SST” at different layers?

L156, what is the “reference observation”, and where has it been described? Note ARMOR3D and GLORYS in section 2.1.3 can only be used for intercomparison, not as reference observation.

L166, “WOD23 observations as the reference” cannot be used for Validation since it has been used in 4d-MGA as described in L78.

Figure 2. Are the RMSEs averaged from the surface to the bottom of the ocean?

L168-175, the low RMSEs in 4D-MGA should directly result from using the WOD23 since WOD23 is not independent from 4D-MGA, while ARMOR3D and GLORYS may not use the same WOD23 data or with different QC procedures or with different analysis resolutions horizontally and vertically.

L180, why WOA23 is compared in vertical in Figure 3 but not in Figure 2? Why the RMSEs and biases in WOA23 are much larger than 4D-MGA? “WOA” in Figure legend should be WOA23.

Figure 4, the “SLA observations” is not a correct term, the AVISO SLA analysis or observation-based AVISO SLA analysis should be used.

Figure 4, Could the low RMSEs and higher ACCs in 4D-MGA could directly result from the ingestion of DOISST product, while others are not?

L227-229, what are the references for these ACCs and RMSEs?

Figure 5, I doubt it is proper to compare with AVISO, which has been ingested into the 4D-MGA.

Figure 6 (i,j), What does “WBCs” represent, EAC and BMC? Or includes those in Figure 5? Is “BMC” the Brazil Current, and what does “M” mean?

L230, “respectively” in Kuroshio, Gulf Stream, AC, EAC, and BC. It would be better if these currents can be compared with those derived from other products ARMOR3D and GLORYS.

L280, why these month and year are selected?

L289, “randomly sampling 30% and 60%”, how about the OISST used in 4D-MGA? Should it also be reduced by 30% and 60%?

Figure 8, please clarify the statement “the linear correlation coefficient

between the linear trend and the mean anomaly”. The abbreviations of KOE, ARC, and BMC should be revised throughout the manuscript as KE, AC, and BC.

Figure 9, (1) quality/resolution is bad, (2) what is “background”, (3) what is the reference? Are 4D-MGA compared with independent non-sampled/ingested data? (4) It should be very helpful if the same comparisons are done for ARMOR3D and GLORYS. (5) “shaded area represents ±1 standard deviation” of 30%, 60%, or background? I guess there are three different shadings but are hard to identify. (6) Circles, triangles, and squares are not exactly on the time axis of Jan, Apr, Jul, and Oct, is this intentional designed? (7) What do the error bars represent for 2005, 2014, or 2023?

L299-301, very interesting. Generally speaking, an analysis system can filter out noises and random errors in observations, and the comparison of analysis against those noisy observations may not be problematic. A better way of evaluation is to filter out noises in observations such that the STDs in filtered observations are equivalent to that in analysis (Huang, B., X. Yin, T. Boyer, C. Liu, M. Menne, Y. D. Rao, T. Smith, R. Vose, H.-M. Zhang, 2025: Extended Reconstructed Sea Surface Temperature Version 6 (ERSSTv6): Part II. Upgrades on Quality Control and Large-scale Filter. J. Climate, 38, 1123-1136. https://doi.org/10.1175/JCLI-D-24-0185.1).

L310, what are “temporal grid layers”? are they same as “temporal layers” in section 2.3?

This is one of many papers that use data from various repositories to provide 'intermediate' products for other products. Its strength lies in its attempt to provide a global data set with weekly temporal resolution. Its weakness lies in its failure to clearly state the product's limitations.

Based on the authors' presentation, the method appears to work reasonably well in cases where geostrophy plays a predominant role (for example, in the case of along-fluxes - western boundaries currents). The analysis is centered on two equations (1) and (2) which (in my opinion) significantly smooth the data and therefore lose much of the non-geostrophic signal. The results are integrated with dynamic heights which (again) are based on geostrophy. Figures 4 effectively confirm this refutation of mine.

In any case, the fields obtained with the method at ¼ degree resolution are always smoothed, and this is not stated in the paper.

In verifying the results, the authors use data sets that are not independent, having already been ingested in analyses/models used for comparison (section 3, Validation). This is a very significant weakness and makes the work not yet suitable for publication.