Summary

In this manuscript, the authors attempt to provide insights into possible future hydrologic changes over Western and Central Europe (WCE). To do so, they make use of historical and future (ssp585) projections from CMIP6 models. They cluster models into 8 groups based on the similarity of their hydrologic responses and show that there is no consistent response across models/clusters in future annual runoff changes over WCE, with half of clusters showing strong decreases and half showing no change or modest increases. They further show that even when clusters of models show similar runoff responses, they may do so for different reasons (e.g., divergent signs/strengths of precipitation and evapotranspiration trends). Additionally, they show that it is challenging to observationally constrain the model projections owing to observational uncertainty in components of the water budget (especially P and ET trends) and a weak linkage between historical and future changes in the models. Finally, they leverage additional CMIP6 experiments from C4MIP and LS3MIP to provide a detailed look at some of the biogeophysical mechanisms driving the models' hydrological responses and find a strong role for plant physiological responses to elevated CO₂ and a limited role for soil moisture-atmosphere feedbacks in the spread of model responses.

General comments

Overall, I found the questions the authors posed around future hydrologic changes over WCE both scientifically interesting, owing to the substantial uncertainties, and societally relevant. Their analysis was thorough and technically sound. I particularly appreciated the use of multiple sets of CMIP6 experiments to provide several lines of evidence. I only have two minor concerns that I would like the authors to address before I would consider this manuscript ready for publication.

Specific comments

My first comment pertains to the clustering algorithm used to group models. How was the number of clusters determined? The authors do not describe the dissimilarity measure shown on the x-axis in Figure 1 or how the dissimilarity threshold that determines the final clusters was chosen. I feel that these analytical choices do require some justification, and checking the sensitivity of the main results to the threshold/number of clusters would provide a valuable robustness check. 

My second comment relates to a point the authors flag in their discussion on ll. 399-403: the CMIP6 models display a wide range of climate sensitivities, and as such warm by very different amounts over WCE  in ssp585 by the end-of-century period the authors examine. Accordingly, it would make more analytical sense to normalize the hydrologic changes the authors examine by the temperature change over that period (i.e., %/K). This would help identify situations in which models might have a similar hydrologic response to warming, but warm different amounts, and to more effectively partition how much of the model spread is due to dynamic vs. thermodynamic differences. Additionally, this normalization would make for more interpretable expectations of future change that are benchmarked to warming levels and thus more scenario-independent.

Technical comments

I appreciated the authors' Table 2 and related discussion about the overlap in model components in their clusters. I'd encourage the authors to take a look at and cite the "model genealogy" literature (a few recommended papers below). to contextualize and strengthen this section. 

How much is the correlation between DJF SLP and precipitation change shown in Figure 6c driven by the outliers in C8 and TaiESM1? Given that all models show increased DJF precipitation regardless of the sign of SLP change, I'm not sure what conclusions to make about the importance of circulation changes for wintertime precipitation changes.

General Comment:

This manuscript provides a comprehensive analysis of future runoff changes over western and central Europe using the CMIP6 multi-model ensemble under a high-end emissions scenario. The hierarchical classification approach used to categorize models is helpful to identify the sources of inter-model uncertainty. I also appreciate the authors investigate multiple mechanisms using other experiments such as LS3MIP and C4MIP.

Specific comments:

1. Regarding the BGC simulations shown in Fig. 9, the authors did not discuss runoff results. Unlike ET and P, runoff does not show a statistically significant correlation between ALL and BGC scenarios. It also shows varied signs of change between ALL and BGC, e.g., runoff in BGC is negative while ALL is positive, whereas ET and P show more consistent signs of change. It would be valuable to discuss these discrepancies explicitly. This reference may be relevant: Lesk, Corey S., Jonathan M. Winter, and Justin S. Mankin. "Projected runoff declines from plant physiological effects on precipitation." Nature Water(2025): 1-11.
2. Is there a particular reason the authors chose a 20-year climatology period (2081-2100 vs. 1995-2014)? The standard practice is typically using a 30-year climatology, as used in the given reference above.
3. One finding of the authors is that multi-model mean and inter-model standard deviation could be uninformative. It would be helpful if the authors suggest alternative metrics or practices for better representing model agreement or uncertainty in hydrological projections.
4. Because the mechanisms investigated are focused on particular seasons, e.g., DJF for large-scale circulation and JJA for others, in the abstract the seasons should be explicitly mentioned to avoid confusion or overly generalized claims.
5. The current phrasing of "extreme hydrological changes" in the abstract could lead to misunderstandings. I though the authors suggest a focus on extremes (such as the 99th percentile runoff values), but later it becomes clear that the authors refer to models showing unusually high responses compared to other models. Clarifying this phrasing early on in the abstract would enhance reader understanding.

Technical corrections:

1. Line 97: ERA5-Land should be described as a reanalysis dataset rather than an observational dataset.
2. Line 94: Please clarify the temporal resolution of the model output used. Does the daily mean temperature imply PET is calculated at a daily timestep? Please specify the resolution of other variables as well.
3. Table 2: The caption mentions “the colors of the first column”, but the first column is entirely black. It is also unclear why the cluster is not in a sorted order. It only becomes clear in Fig. 4 that sorting is based on annual runoff.
4. Fig. 2: The caption mentions the "multi-model ensemble mean," but line 157 says one member per model was used. Please clarify why all members per model were not averaged or specify the criteria for selecting the single member.
5. Line 159-160: This methodological explanation should be moved to the methods section. Additionally, please briefly define the "linearly detrended modern period," and clarify the period used to calculate the standard deviation. Is sigma-1yr calculated from the multi-model ensemble mean or individually from each member? This clarification also applies to lines 175-180.
6. Line 166: How do you define significant? Is it referring to statistical significance?
7. Fig. 5: Figures 5d and 5i seem to include only GLEAM as a reference dataset, although ERA5-Land also provides the variable "Evaporation from Vegetation Transpiration." Please clarify why ERA5-Land was not included.
8. Line 182: Section 2b should be Section 2.2.
9. Line 260: Please check and correct the ordering of letters corresponding to the correct sub-figures.
10. Lines 253-256: There seems to be some redundancy.
11. Line 242: Please be aware of the difference of observations and reanalysis. Here the authors use “observations” to refer to GLEAM and ERA5-Land, but in line 245, they mention GLEAM has a lack of direct observations for transpiration.
12. Line 250: Please provide relevant citations to support the statement.
13. Line 269: Cite Fig. 4b as well since annual precipitation is also mentioned.
14. Fig. 7: Consider adding observational data to help assess which LS3MIP experiment better aligns with observed conditions.
15. Lines 298-299: The text refers to differences in ET and P between two LS3MIP experiments, but Fig. 8 shows differences between historical and SSP585 simulations.
16. Lines 321-322: The authors mention “ALL” and “BGC”, but in the figure they didn’t use the same notation.
17. Fig. 9: The caption does not follow the sub-figures being shown. The y-axis and x-axis are also not consistent with the description.