In their manuscript, He et al. present a steady-state groundwater model forced by outputs of the global hydrological model H08.

In general, this study is very timely, and it is nice that H08 is also approaching this difficult task of including a gradient-based groundwater component. However, there are also several areas where this manuscript needs improvement.

1) As already mentioned in an early community comment, some studies provide some more context for the finding presented here.

2) Instead of the global maps, a direct comparison to existing published results would significantly improve the scientific value of this study. I made a concrete suggestion further below. Parts section 3.5 adds not much and should either be extended or removed.

3) The sensitivity analysis is currently a manual calibration and not a classical sensitivity analysis. This needs either to be named manual calibration or the one traditional goal of sensitivity analysis needs to be achieved (see also more details below.

3) And finally, I suggest also reconsidering the framing of the study because what the authors present is not yet a coupled model but rather a global steady-state groundwater model forced by aggregated inputs from H08. Either the authors need to add at least ideas on how the complete two-way coupling can be implemented, or they need to adapt how the results are presented.

With regards,

Robert Reinecke

    

Detailed comments:

16: but then it is a manual calibration, not a sensitivity analysis

18: Two previous studies? There is much more than that. Also, it is unclear what this refers to here. Studies on H08 or more generally global groundwater modeling

17: Does this refer to WTD or head? This needs to be clarified because I suspect it is head.

20: Did you mean shallow?

35: Groundwater is also in itself an ecosystem Sacco et al (2023).

40: You don't really model salinity here. I suggest removing this. Even if intrusion and discharge are critical processes, this deviates too much from the story you want to tell here

51 and following: I don't know if this focus on these specific models is necessary. An increasing number of models that represent the terrestrial global water cycle are starting to include groundwater as an explicit component. There is a plethora of models that we could call global water models (https://wires.onlinelibrary.wiley.com/doi/full/10.1002/wat2.70025). Especially since you then also discuss Parflow. A more general framing of the idea that there is an interest in better representing the global water cycle is better than trying to categorise H08 with two other models as the only global hydrological models out there, which is also not true.

82: Reinecke instead of Reneicke

93: Does this refer to a 41-year mean of conditions that were used to force a steady-state model where time was removed from the model formulation to reach an equilibrium, or did you use a transient formulation to reach a defined steady-state? This is unclear here. Also, why this period and not, e.g., 1901, which is often the starting point for ISIMIP simulations?

Also, the rest of the sentence is unclear - included in what?

112: This is a limitation that should be mentioned in the abstract.

175: Why this exact period and not, e.g., 1901 or a mean of 100 years?

176? Why are the aggregated to monthly when you are running at monthly time steps (line 175)? Or does that mean you calculated the arithmetic mean over this period? Ah, it is described in the following sentence. Please be sure to improve the clarity in your manuscript.

203: This is not correct. It has not been proven in any scientific way other than that it provides decent results, however, not better than in models that don't use this approach. Essentially, it is a calibration of transmissivity to observed heads. I am not saying it is wrong, but it should be introduced correctly.

Fig. 3a Please use appropriate colors - also in other figures. https://hess.copernicus.org/articles/25/4549/2021/hess-25-4549-2021.html is a good read concerning this. Also, the figure text does not include the units shown.

218: Is there a reason why you chose to not use the permafrost categorization to limit conductivity in the North?

230: Which is consistent with https://hess.copernicus.org/articles/23/4561/2019/ What does "1d" refer to?

Figure 4: Bad figure quality. The subfigure references have different sizes, and sometimes they are in the figure. C is missing a legend altogether.

250: The DEM is at a higher resolution than 5-arcmin. How did you determine the river elevation exactly?

280: While it is great that the authors did this. It is also important to frame it correctly. This is a local one-at-a-time SA which does not account for interactions. And since you don't achieve any of the traditional goals of SA, i.e., screening, ranking, or mapping https://www.sciencedirect.com/science/article/pii/S1364815216300287, I would say it is rather a manual calibration than a SA.

282: see also Reinecke et al. (2021) for recharge, (2019) for other parameters, (2020) for spatial resolution, and (2024) regarding simulated WTD

285: RRB is upper case in the text and r_rb in the table. Please be consistent. Also, what does "ref" mean? What is the baseline value for this?

304: I would disagree with this. The head is what the model actually simulates, but what is relevant for many applications, e.g., in determining whether groundwater is available to humans or ecosystems, is WTD. Furthermore, calculating statistical metrics such as an error metric on the head is likely very biased by the topographic influence that is encoded in the head distribution. Scatterplots look very different depending on whether head or WTD is shown. Please show 1) scatterplots of simulated head vs. observed and scatterplots of simulated vs. observed WTD. See also Reinecke (2020) for a related discussion. Furthermore, I suggest comparing your outputs to other existing steady-state simulations. You could even use the ensemble published with Reinecke et al. (2024).

310: Again, this is an artifact of using head instead of WTD and suggests a much better model performance than actually is the case. Since the model is likely performing very well in shallow aquifers but much worse in deeper aquifers.

370: What is the impact of comparing human-impacted observation to simulation based only on a natural run? What deviation can be explained by this, and which are the model limitations?

Fig. 5, 7: Please also adjust the color here.

Fig.5: The small maps are not very helpful. How about showing the results of the manual calibration in terms of different error metrics (bias, root-mean squared, max deviation) and scatterplots of head and WTD here. This would be much more informative.

Fig. 7: Instead of showing a direct comparison, e.g., as a difference map or scatter plots, to the existing results would be more informative than a global map. Consider comparing the functional relationships to the slope we propose in 2024 as well.

Fig.8: Log scale of the river bed conductance? And where exactly is this zoom in from?

Section 3.5.The title mentions implications for megacities, but the section only discusses lateral fluxes computed by the model. This doesn't add much to the paper. Either the discussion of relevance to megacities needs to be addressed in much more detail, which would turn this into a completely different paper. Or I would remove this and write a specific paper about this another time - which would be great because the representation of megacities in global hydrological modeling is a topic we should talk about more. Also showing the later fluxes makes for interesting maps but currently provides no scientific insights. Either this needs a direct comparison to deGraaf and Stahl (2022) and others, such as agupubs.onlinelibrary.wiley.com/doi/10.1029/2024WR038523 or I would remove this as well.

Review Report

Thank you for this great contribution. The paper presents a global groundwater modeling framework (H08-GM) that couples the H08 global hydrological model with MODFLOW version 6 at 5 arcmin resolution. The authors focus on steady-state simulations under natural conditions. The study includes sensitivity analysis on aquifer parameters, validates simulated groundwater heads against Fan et al. (2013) water table depth dataset, and produces global maps of groundwater table depth and lateral flow.

This study addresses the growing demand for a better groundwater representation in global hydrological modelling. I see it as a valuable and well-executed newly developed model with parameter sensitivity analysis. It provides useful global visualizations and emphasizes the importance of improving subsurface data and lateral flow.

Overall, I recommend Minor revisions, mainly to improve clarity and figure presentation.

Specific comments

1. The study is described as a coupled framework, but only a one-way coupling is implemented (H08 to MODFLOW). This limitation matters because groundwater feedback to surface processes is not represented. I suggest the authors be more explicit about this limitation and what it means for interpreting their results.
2. For the Aquifer thickness section (lines 196–210): I found this part hard to follow. The motivation for introducing aquifer thickness is not clearly connected to the model setup. Since the model simulates an unconfined aquifer, the authors should make the link between the aquifer thickness map, bottom elevation, and groundwater head more explicit.
3. Figures could be improved with a few adjustments:
   
   Many maps use red/blue scales that are not colorblind-friendly.
   
   b. Some captions are overly dense and read like mini-methods sections (e.g., Figure 2), while others don’t give enough description or citations (e.g., Figure 3). Captions should primarily tell the reader what the figure shows; technical details can stay in the text.
   
   c. Units: Please make sure all colorbars explicitly show units.
4. The manuscript is too wordy in several places, with very long sentences that were difficult to follow. For example, lines 54–56, 85–90, 365–370. Breaking these into shorter sentences would make the paper easier to read.
5. A few minor comments related to typos, clarity, and style (these are just examples, not a complete list):

• Line 20: Typo — “hallow WTD” → “shallow WTD.”
• Line 28: “…compared to the two previous studies” — unclear which studies are meant; please name them explicitly.
• Lines 54–56: Sentence too long. Suggest splitting into two: one on natural hydrology (supply), one on human use (demand).
• Line 67: ParFlow is also a groundwater model, not a land surface model.
• Lines 85–90: Break into two sentences.
• Lines 140–144: Too wordy and dense; hard to follow.
• Lines 175–176: “…monthly timestep … aggregated to a monthly step” — I assume this is a typo; should be daily timestep aggregated to monthly.
• Lines 196–210 (Aquifer thickness): Hard to follow.
• Several inconsistent citations.

Final comment

This paper has strong potential. With clearer writing, stronger justification for the aquifer thickness part and the one-way limitation, and improved figures/captions, I believe it will be a valuable contribution to the global hydrology community.

This is an ambitious and timely global MODFLOW implementation that advances the representation of lateral groundwater flow. The manuscript is well organized and thoughtfully executed; addressing the points below would further strengthen its physical clarity and practical utility.

Major comments

1. H08 runs on a geographic grid, while MODFLOW uses a rectilinear grid in metric units. Please state if all areal and volumetric fluxes were converted.

1. Treating abstraction as a simple subtraction from net recharge neglects the spatial propagation of drawdown (cones of depression) and can bias heads even under steady state. Could you clarify this limitation? Do you intend to use the same approach in the transient runs? If so, wouldn’t that undercut a key advantage of replacing a bucket model with an explicit groundwater model—namely, resolving spatially distributed drawdown and capture?
2. Discuss biases expected where deep confined systems exist (e.g. North China Plain, Central Valley): vertical gradients, leakage from over/under-lying units, and coastal interfaces.
3. For the net lateral groundwater flux, the current explanation is long and switches polarity to compare with other studies. consider adopt one sign convention throughout (and in captions) and move any polarity flips to the supplement.
4. Improve readability with a colorblind-safe palette. Consider annotating break values on the colorbar.Replace “aquifer conductivity” with “hydraulic conductivity (K)” throughout to avoid confusion.

Minor comments

• L20: Typo hallow-> shallow
• L51–54: The list of global models is illustrative, not exhaustive. Rephrase to avoid implying exclusivity.
• L111–114: The absence of two-way coupling here stems from the steady-state design, not chiefly computational burden.
• L124–125:  how to explain the reason behind? is it indicating that when the resolution is high enough, subgrid variation is not an important factor anymore?
• L141–142: Statement is unclear. I didn’t understand.
• L173: For near-surface air temperature downscaling, consider a lapse-rate correction, which is preferable to purely linear interpolation in high-relief regions.

• L494: The claim about evaluating city-scale groundwater inflow/outflow from a global 5′ (~10 km) model feels too strong. At this resolution—and given uncertainties in K, recharge, riverbed properties, and boundaries—such budgets are not robust. City-level assessments typically require carefully delineated regional/local models.