Summary:

This paper presents a river discharge dataset for Chinese rivers. Discharge is estimated by exploiting the correlation between river width as an observable variable from satellite EO and discharge. The predictive relationships (rating curves) are estimated using in-situ river discharge observations from a national database. These approaches are well known and have been used in many previous studies, both at regional and global scale. Authors argue that in earlier studies, only few estimation points (or gauges) have been included from China and that their dataset increases data availability for China significantly. However, I believe this is mainly a consequence of access restrictions for in-situ river gauging data that China imposes on both international and national researchers. For this reason, I see the in-situ dataset as the main contribution of this paper. Authors must ensure that this dataset is accessible for everyone in the public domain before this paper can be published. This is the only way to ensure that other people can benefit from this work and that the authors’ work can be benchmarked and compared with alternative approaches.

Review comments:

• Authors have decided to use river width as the observable satellite EO proxy for river discharge. However, several other proxies have been used in the literature, most importantly water surface elevation from satellite altimetry. Can authors discuss the pros and cons of using river width as compared to the other proxies individually or combined in a data fusion approach, and motivate/justify their choice?

• The entire workflow hinges on the availability of in-situ discharge at stations. What authors present is not a truly contactless “satellite-only” approach. It is thus essential to share the dataset of in-situ discharge in the public domain to ensure that others can reproduce the work and benchmark it against their own approaches. Given the access limitations for Chinese in-situ discharge, publishing the in-situ data in itself has significant value. Authors refer to a website in the manuscript where the in-situ data supposedly can be downloaded. I made a few checks and tried to navigate this Chinese page, but was unable to actually access in-situ river discharge time series. Authors should either provide detailed guidance as to how the data can be accessed from the website or re-publish it in an open archive using science-friendly data formats (e.g. netcdf).

• The authors’ river discharge reconstruction resulted in a median KGE of 0.66. While this is useful, it is not stellar for a rating-curve based approach using in-situ data, especially for rivers with a highly seasonal flow regime such as in China. It would thus be interesting to benchmark the authors’ results using other discharge retrieval techniques developed by other groups.

Overall, this dataset fills an important gap in publicly accessible river discharge observations across China and will likely be of broad interest to the hydrological and remote sensing communities. My main concerns relate to the clarity and reproducibility of the methodology, the validation strategy, and the dataset uncertainty.

My major comments are as follows:

1. The in situ discharge observations used to establish the width-discharge relationships are primarily from 2014-2024, whereas the reconstructed discharge records span 1990-2024. This reconstruction implicitly assumes that the width-discharge relationship remains stable over the entire reconstruction period. However, channel morphology and hydraulic conditions may have changed substantially over the past three decades due to dam construction, reservoir regulation, channel engineering, sediment erosion/deposition, sand mining, or river course modifications. At a minimum, I recommend explicitly acknowledging this assumption and discussing its implications in the Limitations section. In addition, given that Landsat observations are available before 1990, could the authors clarify why 1990 was selected as the starting year of the reconstruction? Extending the dataset further back in time (e.g., to the 1980s, or earlier if feasible) could potentially provide additional value for long-term trend analyses.
2. The manuscript reports validation statistics based on concurrent width-discharge pairs. However, it is currently unclear whether the calibration and validation datasets are independent. Specifically, readers would benefit from knowing:
   • how many observations were used for model calibration versus validation;
   • whether an independent testing dataset, cross-validation, or leave-one-year-out validation was employed; or
   • whether the same width-discharge pairs were used for both model fitting and performance evaluation.
3. I think the dataset would benefit from a more comprehensive assessment of uncertainty. In particular:
   • Are uncertainty estimates or confidence intervals available for the reconstructed discharge values?
   • What is the uncertainty associated with the fitted AHG parameters?

Minor comments

• Line 143: Regarding Fig. 2d, could the authors explain why CDR2 contains substantially fewer gages in the Southwest Rivers Basin (Basin A) than the CHP dataset? Is this mainly due to the site-selection criteria adopted in this study?
• Lines 206–207: Since both Landsat and Sentinel were used, did the authors perform any cross-sensor bias correction or consistency assessment before combining observations from the two sensor systems?
• Lines 250–251: The site-selection procedure could be described more explicitly. It would be helpful to clearly state here that only 310 of the 401 candidate gages were ultimately retained because they satisfied the valid width-discharge pairing criteria.
• Line 349: “river dynamism” or “river dynamics”?
• Line 383: The font of “with Sen’s slope estimator” differs from the surrounding text. Similar formatting inconsistencies also appear around Lines 393 and 410.
• Lines 412–413: The manuscript refers to reconstructed “daily discharge” from 1990-2024. However, the reconstructed time series are not temporally continuous on a daily basis. I suggest clarifying this point in the manuscript to avoid potential misunderstanding.
• Line 439: I recommend adding a dedicated subsection discussing the limitations of the dataset and its appropriate scope of application.
• Line 479 (Section 4.1 and Fig. 9): The comparison with existing products should be interpreted with greater caution. CDR2 is derived from a screened subset of Chinese gages that satisfy the width-discharge relationship requirements, whereas the global datasets include a much broader range of river types, including hydraulically challenging sites. Therefore, statements such as “substantially outperforming” existing products may not represent a fully equivalent comparison and could be moderated accordingly.