Dear Daniel and co-authors,

As discussed earlier, your manuscript remains an edge case for ESSD, but we agreed that we wanted to constitute an example for synthetic data outside more common domains (e.g. remote sensing) with the intention to demonstrate its capability, requirements, and corroboration with real data and transparent models. After the original referees not being available for a second round of reviews (noreply to nomination and reminder), I decided to involve another topical editor for the assessment and guidance. We have re-evaluated your manuscript independently, discussed our findings, and come to the following critiques and options:

In general terms, a more topical manuscript about assessing the generation of such a synthetic dataset still appears feasible (e.g. with HESS). For a data paper, usability, reproducibility, and evaluation are the essential core of the publication. If we want to publish this in ESSD, we really have to convince with the current version that this data product is really something users can trust and something we can and should use. If you rather seek to discuss the difficulties with synthetic data, we are under the impression that this actually requires the topical discourse, which is not in the scope of ESSD. Hence, it is not the question if ESSD is only open for measured data but rather that ESSD papers cannot simply trust on disclaimers in the manuscript for promoted data being correctly used.

Dataset and data basis
As we discussed based on your initial submission, to not just ignore the growing body of synthetic data in earth system sciences or leaving it to domains like remote sensing or spectrometry, where such tools are more established, we think that your work can become an interesting and valuable contribution to shape the landscape. Plus, you have confidence that these kinds of datasets will grow in number and application. However, we also discussed that we need to clarify why this requires the published dataset and not a proper publication of the model. Moreover, we have to convince with the data foundation for evaluation, reproducibility, and further developments.

As the referees highlighted too, expanding the Blöschl et al. 2017/2019 analyses is a great showcase for the streamflow reconstruction with LSTMs. Especially providing an evaluation of the streamflow characteristics above NSE (or other global evaluation criteria) is very interesting. Also, the reference to mHM is a great asset.
For the general assessment of your model, you have to be more transparent about the selection of the validation and testing gauges. Comparing Fig 4b and Fig 1, I suspect some sort of stratified random sampling nested by country? Since you have the catchment attributes, maybe this is actually informed by several attribute classes? Since almost the entire eastern part of the catchments is ungauged (or lacks reference in your data basis), this might be a specific weak spot, which could be eased when there are arguments beyond spatial proximity (i.e. coverage of respective catchment attributes) or even improved overall performance?
You are comparing your results based on the Bloeschl et al. maps. Did you ask the authors for their data? And also with the limited data they provide (https://github.com/tuwhydro/europe_floods, https://www.tuwien.at/cee/wih/hydro/downloads-1), an assessment beyond visual patterns should be feasible.
Although you argue and we understand that you are intending to contribute to living data sets, for the sake of reproducibility and direct usage, we think that a snapshot of the state of your data would be quite important to complement table B1 with these sources/links/data changing. Maybe including the mHM data too, would make it even more interesting (and giving Rohini Kumar co-authorship).

Ultimately, the data will be used when it is published and only a few will really care about any disclaimers or your quite critical assessment. First of all, you have to clarify why this dataset deserves publication and description in a data paper in ESSD. If you would not recommend using it or if after all the insight comes from the workflow building and evaluating the LSTM, then we should opt for other outlets to publish and discuss your work. This decision is not about the result not being "good enough" but about where your work can become shining and inspiring. If you still think that your dataset and EARLS remain worth pushing forward, you have to work out more clearly why spatial resolution still is an advantage, why you need to publish the data and not your LSTM, why the dataset is needed and what it can be used for in its entirety and with the claimed advantages above E-OBS and EStreams.

Manuscript
Again, you have been warned about the difficult terrain you have approached. Since you chose to contribute a foundation case for emerging synthetic catchment data, we have to make sure that this is really done in a rigorous manner. You are comparing with CAMELS and similar large sample hydrology references, but the promised advances over "sampling gaps", the "high-quality runoff observations", and more broad use cases while keeping it as simple as daily specific discharge is not really worked out. Following the data philosophy of mHM, soilgrids, HydroSheds, etc., a gridded netCDF or similar could be such a respective data set. Its evaluation would take your approaches but specifically check towards the claimed advances for the ungauged basins and the higher density.

We are still not really happy that your LSTM is not published along with the data and that many of the claims and ideas are rather left for future additions. We see this as an argument for including as much data as possible in a structured way to help people easily plug their rainfall runoff model and run all the various evaluations you did and the referees/we suggested. If we could advance to such a high level of transparency, interoperability, and reproducibility, this would also help to discuss the matter of open and synthetic hydrological data within hydrology and with other earth system science domains.

Above our general assessment, a couple of typos (e.g. L75, BAFU in Table B1) and an overly self-confident language (e.g. L8, L77, L105) is something which also does not ease acceptance of the work. The C in CAMELS stands for catchment, not for basin (L39).

How to proceed
We think we need your decision whether to really push for becoming the promised great case in ESSD or to rethink and reallocate resources, maybe for a discussion as a more topical paper. Re-reading the assessment from the referees actually points into the same direction. While you did address many of the aspects in the revised version, it somehow did overlook the big picture?

We hope you understand the delay it took to come to this assessment. We needed to do a couple of internal discussions about your manuscript, and I am still hoping for an outstanding demonstration in ESSD after all.

Kind regards and thank you for going this difficult path with us.
Best. Conrad

Dear Daniel Klotz and co-authors,
many thanks for the revision and our online meeting to clarify the open requests for the final publication of your manuscript.
We have agreed that you do the following modifications to the paper and data:

1) please add an explanation about the reasons why you cannot publish your training/ ground-truth data (the gauge data retrieved from federal and state agencies)

2) please make sure that you clearly differentiate the region where you have ground truth data (i.e. Western and Northern Europe) and where you show extrapolated model results (Eastern and Southeastern Europe). It would be good to have separate sections in the manuscript for (1) model results that are validated by ground-truth data and (2) extrapolation of the model to areas in which you don't have ground-truth. I understand your motivation to show a similar extend like the Blöschl Model, but it would be good to clearly mention the extrapolated part also for other readers

(3) please add the data underlying your Figure A5 " Comparison of the trend analysis for flood peaks from 1960 to 2010 derived from different datasets..." to the Zenodo data publication

After that, everything is ok, I think

Best regards,
Kirsten Elger

Dear Daniel Klotz et al,
many thanks for the revision, I am accepting your manuscript for final publication after addressing two minor / technical points:

line 250, please cite the Zenodo DOI for the Code after the DOI number as Klotz et al. (2026) and inclulde the following reference to the "References" section: Klotz, D., Miersch, P., Nascimento, T., Fenicia, F., Gauch, M., and Zscheischler, J.: EARLS: Code and Weights, https://doi.org/10.5281/ZENODO.19107555, 19 March 2026.

Please add the information that you have added the data underlying Figure D5 (the comparison between your data and Blöschl et al) to the figure caption with citation of the respective data or code publication

many thanks and best regards,
Kirsten

Dear Daniel et al,
many thanks for this additional adjustments, I am accepting your paper for final publication with ESSD.

Best regards,
Kirsten