We are grateful for the constructive feedback provided by all three reviewers. Below we describe how we will revise the manuscript and dataset to address the concerns. Reviewer comments below are italicized. 

Reviewer 1#

1. The manuscript needs to be seen by a native English speaker.

Reply: We will seek help from a native English speaker to check and polish the manuscript.

2. Abstract: as far as I know, the link to the dataset should be added here.

Reply: Yes, we will add it in the revised manuscript.

3. L12-13: correct would be "...there is a lack of a comprehensive..."

Reply: Thanks for your suggestion. We will revise it.

4. L26-27: I think soil erosion could be easily added here

Reply: Yes. Soil erosion is also influenced by streamflow regime. We will add it.

5. L67: “it is without doubt…” please, rephrase

Reply: Thank you. We will revise this phrase.

6. L187: please, explain why

Reply: 5% is a threshold that we defined as an acceptable missing value ratio for indices calculation. This number is referred to several publications (Gudmundsson et al., 2018; Sauquet et al., 2021; Tramblay et al., 2021) as a commonly acceptable value.

7. Fig. 5a. If you could make a zoom on Europe and then add it to the figure, that would be perfect. As many data overlap, would be nice if the reader could see the whole data only for Europe

Reply: Thanks for your advice. We will revise Fig. 5 to make it clearer and increase the readability.

8. Section 6 as a whole is more than welcome and could be further developed. 

L303: a few more details could be offered here, on the direction of future studies that could use this dataset to focus on more specific hydrological issues at a local to regional scale

Reply: Thanks for your kind suggestion. We will expand this section and offer more details about relevant future studies on specific hydrological issues based on our dataset. 

Reviewer 2#

1. The URL and DOI of the dataset should be provided at the end of the abstract.

Reply: Thanks for informing us. We will add it to the revised manuscript.

2. Ln 67-89, many references were cited to demonstrate the importance of the indices. These references were all cited as "who did …". It is recommended to further summarize the references and enrich the expression forms.

Reply: Thanks for your recommendation. We will polish these sentences to make them more concise and comprehensive with enriched sentence patterns.

3. Ln 159, the mean value of log(Q+0.01) and 6 times of the standard deviation can be added in Fig.2C. The examples of flood rather than outlier may be also pointed out in the example if possible. This may help readers to understand. For the assumption of 6 times of standard deviation, its applicability and possible impacts should be discussed.

Reply: Thanks for your advice. This approach is originally proposed by Tank et al. (2009) for evaluating temperature series. Gudmundsson et al. (2018) introduced the log(Q+0.01) and 6 times the standard deviation and used the method to do quality control of daily values for the construction of GSIM dataset. The applicability and possible impacts of this method actually have not been discussed by anyone yet according to our knowledge. However, the choice of 6 times the standard deviation is very conservative and only very extreme values like outliers can be detected. There is little possibility that floods are mistaken as outliers. We will assess this method in the revised manuscript, and if necessary, a manual check will be done to guarantee the high quality of our dataset. The mean value of log(Q+0.01) and 6 times the standard deviation will be added in Fig2.

4. Ln 303-320, the author briefly compared the dataset with existing data in the conclusion section. Comparative analysis is an important part of evaluating the characteristics and strengths of a dataset. It is recommended to add a separate section for comparative analysis. Quantitative analysis and examples may help to demonstrate the characteristics of the dataset.

Reply: Thanks for your suggestion. We will add a comparative analysis and supporting figures and tables in the revised manuscript to present the spatial distribution and statistics of the indices generated for quality check and comparison to other studies.

5. The tense used in the paper needs to be checked and revised. For example, the tenses of Ln 184-187 are inconsistent. There are many similar problems.

Reply: Thanks for pointing out these problems. We will carefully check and revise the tense of the whole manuscript.

Reviewer 3#

1. Databases. You merged the GRDC, WRIS, ArcticGRO, and CHY databases. Why did you not use many of the other sources that are available, such as USGS, HYDAT, ARCTICNET etc.? Your selection of databases resulted in a total of 9171 stations with daily data vs. 35002 stations in GSIM. On top of that, only 5548 time series could be used for the calculation of indices. That is a huge difference and analyses will produce different results on a global scale.

Reply: Thanks for your comment. At first, we intended to build an indices time series dataset based on the GRDC. However, we found that the coverage of data in GRDC was limited. In some regions like Europe, the stations were densely distributed. However, the stations were sparse in other regions like Asia. Therefore, we used databases of Asia, i.e., WRIS, ArcticGRO, and CHY, to supplement GRDC in order to make the spatial distribution of stations more even. The number of stations was not our concern since we thought the number of stations in GRDC was enough for analysis on a global scale. Currently, we will expand our dataset and incorporate other available databases like USGS, HYDAT, ARCTICNET etc. to support more studies.

2. Databases: merging and formatting. You do not describe how you merged and formatted the time series of the different databases. Did you check that there were no duplicate stations across databases? How did you handle different data formats, e.g., did some data come with flags and if yes did you include them? How did you check for and merge metadata, e.g., did you check gauge locations, was catchment information included?

Reply: Thanks for your comments. We will add a section to describe the merging and formatting in detail. In this dataset, GRDC is the main database and the other three databases are used to supplement GRDC. Therefore, we did not perform a specific examination to detect the duplicate stations. We will check it for our expanded dataset. The current dataset includes several basic and necessary fields from their sources, i.e., river name, station name, country, gauge location, catchment area, start and end year, and so on. Other metadata are not included to avoid inconsistence because the databases do not have the same metadata. As to the quality control flags, all the databases we merged did not have quality control flags. However, we performed a data quality control test on all the streamflow data; only data that passed the test were regard as reliable and used for indices calculation; data that failed to pass the test were declared suspect and removed. For the future revision, the dataset will include the catchment boundaries, catchment metadata, and basic metadata of GSIM corresponding to the stations in our expanded dataset to facilitate broader research.

3. Quality checks of the streamflow indices. Your quality control procedures are based on QC of the time series (which are the exact same filtering methods as GSIM used) and an assessment of record lengths and missing data. However, you did not perform (or did not describe in the paper) any quality checks on the indices time series themselves. Were there outliers within regions, and if yes can you explain them? Can you determine the reliability of the indices based on the quality of the underlying daily time series? What about abrupt shifts in the time series (i.e. from rating curve updates, instrumentation changes etc.)?

Reply: Thanks for your comments. We performed quality control on the streamflow data, and only qualified data in years with less than 5% missing data were included for indices calculation. It follows the recommendations of ECA&D as well as Gudmundsson et al. (2018). Therefore, the indices time series derived from controlled streamflow data should be reliable. We will add a comparative analysis in the revised manuscript to evaluating the dataset of indices. Whether there are outliers within regions or whether there are abrupt shifts in the time series is beyond the scope of the manuscript. There are increasing studies focused on the non-stationarity of streamflow regime time series, its spatial pattern, and attribution. Our dataset is a good material for these studies. In terms of shifts caused by changes of measures, corresponding correction should have been done by data providers in the phase of compilation of database as only they know the details and how to perform a correction, which is out of the scope of our work. We could guarantee that the reliability of the indices is determined by the quality of the underlying daily time series, but the quality of the underlying daily time series is only determined by the providers.

4. Indices based on baseflow estimates. The GSIM paper outlines certain issues relating to calculating indices based on baseflow, which made them decide to not include any. However, you provide recession indices without addressing any of the concerns outlined by GSIM.

Reply: Thanks for your comment. We do not find relevant sentences in both Do et al. (2018) and Gudmundsson et al. (2018), but find sentences in Gudmundsson et al. (2018) as follows: “Note also that index selection was limited to those that can be computed without a base period, which excludes many; examples include “the number of days in a year, or season, for which daily values exceed a time-of-year-dependent threshold” (Zhang et al., 2005), drought deficit volumes (Loon and Anne, 2015; Tallaksen et al., 1997) and anomalies with respect to a climatological normal (McKee et al., 1993; Shukla and Wood, 2008). There are two reasons for excluding these indices”. The term “base period” is not equivalent to baseflow.

5. Overview and presentation of global indices. The paper misses a section giving an overall summary of global statistics for the indices generated. For example, a table providing the mean, max, and min of each index. Such a summary is also important as a quality check and can be used to compare the results to other studies that have provided global streamflow statistics.

Reply: Thanks for your helpful advice. Multi-year streamflow indices of every hydrological station have been given in the “station_catalogue.csv” of the dataset. We will add a comparative analysis and several figures and tables in the revised manuscript to present spatial distribution and statistics of the indices generated for quality check and comparison to other studies.

6. Overall structure and presentation of the paper. Sections 3.2 and 4 are interesting but draw conclusions beyond the scope of this paper. For example, relating trends in streamflow to climatological drivers or land cover changes or other anthropogenic interference, without any robust analyses backing up these statements. Stick to a description of the dataset and a presentation of the data. Further on, the text contains many repetitions, use of casual language, grammar errors, and sentence structures that do not flow well. I recommend asking an external party to review your writing.

Reply: Thanks for your advice. We agree with you. We will replace the Section 4 with a comparative analysis mentioned in Reply #5. As to the Sections 3.2, we will retain and polish it since it gives an intuitive impression of our indices time series.

7. I have accessed the .csv files only, as I do not use Matlab myself. The data is easy to access and to download, the overall description and citation information is clear, and the metadata is easy to find and well described. However, since this is a global-scale dataset which will attract researchers interested in large-scale comparative analyses, I would strongly recommend merging the 5548 separate time-series files into one csv and providing this file additional to the separate location-specific csv’s. This way all information can be easily accessed using R or Python. Looking closely at a few of the individual files, they all start at the year 1806 and therefore contain much empty cells. I suggest removing the empty rows, merging the files and adding one column for station ID.

Reply: Thanks for your recommendation. We will adopt your suggestion and revise the dataset accordingly.

8. The metadata contains information about catchment area. Does this refer to catchment area of the entire river reach or the contributing area upstream of the gauge location? Please specify (also in the paper).

Reply: Catchment area refers to the catchment area of the contributing area upstream of the gauge location. We will specify it in the manuscript.

9. I mapped the MeanQ, Qmax, and Qmin and noticed a large region north-central Canada with no values, while they do contain values for other indices. This is a little suspicious to me. How can you calculate certain indices but not mean Q?

Reply: Thanks for informing us. This is due to the difference of algorithms for different multi-year indices. Some multi-year indices, for example multi-year Qmean, were calculated by taking the average of corresponding yearly Qmean, while other multi-year indices, for example multi-year Q50th, were calculated by taking the median value of the whole daily time series. When there are lots of missing data in every year, all the yearly Qmean will be set to missing value, and so will the multi-year Qmean. In contrast, for multi-year Q50th, the missing ratio has no influence on the calculation based on the whole daily time series. We will revise the algorithms to keep these indices consistent.