Dear editor,

Thank you very much for inviting me to review this manuscript. In this paper, the authors produce a dataset on global annual mean surface runoff and another dataset on global annual mean river flow rate. The topic is interesting and important, and the method used in this study have been explained detailedly. Nonetheless, in my opinion, the manuscript should be further improved before it can be finally accepted, and I am a bit worry about the novelty of this study.

1）As resolutions of all input data (land cover, soil group & precipitation) for driving the CN approach is larger than 250 m, why the runoff is produced at 50 m? Why not produce runoff data at 250m, 100 m, or 10 m? In addition, for many large rivers, the width of the river channel can be hundreds or thousands of meters. A high resolution, like 50 m, is necessary only when you are going to calculate the water flow in small rivers. However, in your approach, the small rivers in each of the small level-12 catchment have been aggregated. For a global river flow database, is it necessary to produce river discharge data at a resolution of 50 m?

To my understanding, the only outstanding point of the runoff and river flow datasets produced in this study, compared with previous datasets, is the high spatial resolution (50 m). However, as I have expressed above, a spatial resolution of 50 m is not necessary for larger rivers. It can be very useful to calculate the river flows in small rivers. However, the authors have aggregated all small rivers in each small catchment.

2） I would suggest the authors to further evaluate the interannual variation of the simulated river flows in this study using the observations. For such a dataset, it will be great if the scheme used in this study can capture the change trend of river flow. Otherwise, the datasets produced in this study seems to be not that useful.

3） Is it possible for the authors to compare the accuracy of river flows simulated in this study with existing global datasets of river flow? Are the results of this study more accurate than existing datasets?

4） I did not find any discussion on the uncertainties in the datasets produced in this study. The authors have calculated the surface runoff and river flow using the simple empirical CN equations, and the CN equations only consider land cover and soil type. However, many other factors, such as topography, dam/reservoir, irrigation, underground drainage and temperature, can also strongly affect river flow. I would suggest the authors to add some discussion on these factors.

5） In Figure 1, the authors have provided a nice flowchart to show the approach for calculating surface runoff. I would suggest the authors to add a detailed flowchart to show the approach for calculating the river flow based on the surface runoff, HydroBASINS and HydroRIVERS.

Please further polish the English of this paper.

Minor comments:

L15: ** to create a global gridded dataset of annual mean surface runoff at a spatial resolution of 50 meters.

L160: how the monthly precipitation data is converted into daily precipitation rates? Did you assume the precipitation is evenly distributed in each month?

L225-280: Based on the description in sections 2.5, 2.6, surface runoff has been calculated for each day using Eq. 2. Why not produce the river flow rate for each day or each month?

Fig. 2: the unit of surface runoff in panel (a) should be mm d^-1, and the unit of water flow in panel (b) should be changed to m³ s^-1

Please check the whole manuscript to make sure the format of units is consistent through the whole paper.

OVERVIEW

The paper describes the development of global dataset of mean annual flow obtained by using the Curve Number method.

GENERAL COMMENTS

The paper is fairly well written and well structured. The topic is of interest for the readership of ESSD as a 50-m resolution dataset of mean annual flow (MAF) could have high impact in the hydrological community.

I read the paper with interest but I have contrasting feeling. The dataset, mainly due to its very high spatial resolution, might be relevant, but the method used for its development has several issues that must be carefully addressed for having the paper published. I believe that the actual spatial resolution is much coarser and that the usability of the datasets in several areas is limited.

To clarify, I have listed below the major issues with the indication of their relevance.

• MAJOR: The main problem is related to the spatial resolution, or better sampling, of the developed dataset. The dataset is distributed at 50 m resolution but the forcing, i.e., precipitation, and mainly the method is not appropriate at such resolution. The method does not consider human impact on streamflow, such as reservoirs, wetlands, water diversions for agricultural, civil and industrial water uses. Other missing processes are related to high altitude regions, such as snow melting, glacier processes. If we want to deliver a MAF dataset at 50 m resolution, these processes must be included. Otherwise, in many areas, the MAF dataset here developed has very little reliability and applicability. This must be discussed, and better it should be fixed.

• MAJOR: A second major problem is related to the choices made for the application of the method. The method is applied by using monthly precipitation data disaggregated (simply the monthly values divided by the number of days) at daily resolution and it is not appropriate. Several global scale datasets of precipitation are currently available. The choice of the 1 km climatological dataset is questionable. CN values for intermediate antecedent moisture conditions seem to be used, and it has no sense in many areas of the world. It is not clear how the time of concentration at basin scale is used, and similarly the travel time for the river. Actually, I believe that the way the method was used in the paper to develop the dataset is not appropriate. It must be clarified and discussed in details.

• MODERATE: The developed dataset is compared with data over US, and GRDC data globally. Currently, 30-year reanalysis datasets developed everywhere in the world are available. Such datasets should be used for the assessment of the developed dataset.

• MAJOR: The high values of R-square obtained in the paper are mostly related to the difference in river discharge from basins of different size. Larger basins have larger river discharge than smaller ones. A more robust assessment should be carried out by computing the mean annual flow normalized by the basin area. Moreover, section 3.3 should be strongly summarized. The assessment in terms of RMSE has little assessment for basins of different size. Normalized scores should be used (as RSR, but its formula should be given). The gauges used for the comparison should be shown in a map. This section should be significantly revised.

A number of specific comments should be also addressed, but I believe the text should be significantly revised and technical corrections are not relevant at this stage.

SPECIFIC COMMENTS (L: line or lines)

L69: River discharge is generated by surface, subsurface and groundwater runoff, all components should be considered to obtain a physically reasonable approach.

L211-221: The comparison with GCN250 dataset is not clear. The comparison of maps with the computation of BIAS or RMSE should be carried out. It would be clearer.

In Figure 2a it is hardly possible to distinguish the different classes. To be improved.

L384: typo.

L428: What is the number of adjusted gauges? It should be reported to GRDC to correct the database.

L523: typo.

RECOMMENDATION

On this basis, I suggest a major revision to carefully address the raised comments.