1. The contributions of subsoil layers to the water supply of forests has been documented since the mid-1990s. What was originally interpreted as a rare and unusual phenomenon is now recognized as essential to the ecohydrology of many terrestrial biomes.  There have been calls to implement a broader definition of the rhizosphere in global vegetation models to include subsoil strata. Apart from Jimenez-Rodriguez et al (2022) cited by the authors, the present manuscript may be the only other attempt towards this goal. Thus, the paper is timely and important.

2.  The authors present a novel tool for predicting the influence of plant-available subsoil water on vegetation composition and the hydrological cycle. This involves a restructuring of the hydrological scheme in the DGVM LPJ-GUESS, an age-structured plant functional type model constructed to predict global biome distributions through an optimization process that maximized NPP. The modification was introduced as having two independent components (Fig. 3): the increase in storage by incorporating the subsoil and the increase in subsoil recharge by modifying runoff prediction. As it turns out, are both needed for the best fit. 

3.  The substantial result is that root access to subsoil ('rock moisture') AND greater partitioning of precipitation into the subsoil are essential to more accurately predicting tree LAI and summer ET across the continental United States (and in the case study).  

4.  The manuscript is very clear explaining the hydrological modification that affects runoff generation (Q_surf v Q_baseflow). However, I missed an equally clear explanation of the way in which root uptake of water is regulated in the model and how that scheme was adapted to the restructuring of subsurface storage. If that part of the LPJ-GUESS hasn’t changed from earlier versions, (Haxeltine & Prentice 1996; BIOME3), water supply is downregulated through a simple linear correlation with the amount of plant-available soil moisture remaining in two soil layers. Furthermore, tree and grass PFTs are distinguished by their relative access to these pools (e.g., grasses = 90% roots in the to 50 cm, versus trees 33%). I would urge the authors to expand on this aspect of the model description. Note that on page 24 lines 25-29 they are actually addressing the issue of root distribution, so it only makes sense to talk about this up-front. I would be curious to know if grasses had access to the second layer (the subsoil) per default parameters…. Seems that they do per line 32 on the same page. The BIOME3 model might not assign detailed strategies on root profiles, but it does have a simple and essential one when it comes to distinguishing trees and grasses competing for two soil water pools. One has to look and see if that makes still sense after the redefinition of the two pools.

5. Related to the omission of addressing plant interactions with water storage pools, page 23, lines 16 – 26 was a bit undeveloped. Seems to me, one should always be able to find out why a model acts in a certain way. Furthermore, I don’t see why residual storage water at the end of summer is necessarily a problem. One would in reality not expect all water to be used by the end of summer every summer. From page 11, line 1, I gather that the simulations were run for the period from 1981-2021, and so I assume that model output is composed of multi-year averages, suggesting that on average there should be positive residual moisture. 

Furthermore, given that Fig. 4 was a model prediction, the fact that ‘rate limitation from photosynthetic pathways are still not fully understood’ was sort of beside the point. Perhaps the point can be made in relation to Figure 6, though. However, I am not convinced that the problem lies necessarily in the physiology: It is very possible that assumed water and root distribution could affect the calculation of optimal LAIs in the LPJ-GUESS model. Really, the model probably needs three layers to maintain the partitioning of soil moisture between grasses and trees. At least for the case study, the authors could have tried to optimize the assumed root distributions, or at least do a sensitivity analysis to investigate the question of root distribution as another source of uncertainty. In my opinion this would add a useful message, rather than dismissing the topic out of hand.  

6. It would have been helpful to have more information on the implementation of the model.  From page 11, line 1, I gather that the simulations were run for the period from 1981-2021, and so I assume that model output is composed of multi-year averages and that tree and grass LAIs were optimized over the same period. But this has not been explicitly stated in the manuscript.

7. The authors do a good job presenting their work in the context of prior contributions.

8. The title is fine.

9. The abstract is also fine.

10. In general, it would have been better to have fewer and/or less complex figures. The figure content was excessively comprehensive, given that the main results were quite straight forward. For example, after the first few results, it is quite evident that the second storage pool needs the enhanced recharge to have the desired effect on ET. Once this is established (and the most fitting place to establish this in the case study, e.g. Fig. 10 is really good in this respect), it is perhaps enough to contrast only the default model, the fully modified model and the ET data product. Perhaps consider Figs 5 and 8 for supplementary data. 

11. The language is fine.

12. The symbols are consistently used and have correct units.

13. See comments above: more should be said about the way in which LPJ-GUESS predicts functional type composition and how plants interact with storage, i.e., how transpiration is constrained by supply not demand. 

14. The references are appropriate. 

15. I recommend expanding the supplementary information section, in exchange for striving for greater synthesis in the results figures.   

The study of Lapides et al. investigates the inclusion of rock moisture into a global dynamic vegetation model. To test the approach the team tests modifications to the storage compartments of the model and compare the model results to two study sites and available data for the continental US.

Overall the authors present a very strong research paper, which unfortunately currently lacks clarity and a discussion of uncertainty. Due to imprecise language and a lot of figures it is sometimes hard to follow the authors in their conclusions. Moving 1-3 figures to a supplement or appendix would help to tell a clearer story.

A lot of figures spaced out over short and precise explanations. Moving ~3 Figures to the supplement would help clarity a lot. Currently very hard to read. Figure 12 seems not to be referenced and explained anywhere.

I would also appreciate if the authors could comment on how the data uncertainty influences the results. As the authors note data in the US is relatively good compared to global soil and bedrock datasets. It would greatly improve the study if the authors could provide an insight of how their results are impacted by data uncertainty in the presented study regions but also what this possibly entails for the global scale.

Linked to the comment on uncertainty the authors should provide a clearer link to how the two sites are representative for the US and globally. What other regions should future research investigate to figure out on how to represent rock water on a global scale in these models? The authors now have modified mainly the storage component of the model. Would it be beneficial to also provide a lateral groundwater connection inside the model?

Additional detailed comments:

P2 11: You motivate your paper with the Mediterranean and then evaluate it for the US, why?

P3 1: Again unclear.

P4 10: You lost me. How do these places relate to the MED issues you highlighted? How much will they be transferable?

Table 2: Nice but very difficult to read. Could you move the justification just to text and pivot the table?

P8 1-4: This is unnecessary. Cite one key paper and be done with it. This seems more like self-advertisement than actual scientific proof

P8 6: More recent versions than what? The one used in this paper? If so this doesn't matter then. Or if it matters explain why.

P11 23: In the model or in the real-world site?

P11 27: For what time frame?

P11 28: First reference to PML-V2 is that a dataset or a model? Why do you use it as benchmark?

29: Why did you not use a hydrological model which might compute runoff instead?

P12 8: Unclear and confusing sentence. The mass balance of what? What spatial distribution metric and what is it used for?

10: This is a result and belongs into section 3.

Fig 3: Missing section reference. This does not need to be an extra figure, e.g., add as small legend to Fig. 5.

Fig 5: Please explain first what a-d show. It is unclear whether a and c both show results for the whole CONUS. It is currently easy to get lost in the information.

P15 18: I assume T stands for transpiration? Make it explicit also in Figure 6

Fig. 6: The effect seems strongest if storage capacity is large but what is the explanation for transpiration underestimation in small capacity sites?

PML should be a different color since you already use black to indicate change. Also, the colored arrows make the plot hard to read and may be confused with datapoints. For all figs: the annotation is helpful but it should be very clear that it is an annotation. Should also be added to legend in all plots.

Fig 7 and 8: Maybe these two could be combined? Because they do not show that much new content and you already have a lot of complex figures and in the text and you jump between these two a lot. Maybe some of this could be moved to the supplement?

P16 6: Indeed, but where are the reductions the lowest and why?

P22 4: Is this supposed to be Fig 12?