In this manuscript the authors aim to isolate the effects of vegetation greening on water yield in a large basin that serves as donor basin for a major water diversion project. To do so the authors designed a modelled-based scenario analysis. The overall finding of their analyses suggests that greening has the potential to considerably reduce basin water yield and thus supply for the water diversion project. While the experimental set-up is systematic and, in principle, logical and the manuscript is well written, I nevertheless have a number of serious concerns that need to be addressed and resolved in detail before this manuscript could be considered for publication.

(1) The experiment is designed, the results are interpreted and, as a consequence the manuscript is framed from a purely engineering perspective with focus on water yield available for the diversion project. As a consequence, the non-explicit message that is delivered between the lines here is the following: to secure water supply for the diversion project greening needs to be reduced. Or in other, more explicit words: stop afforestation - chop down the forest! I am not sure that this can and should be the message to be conveyed here for the simple reason that this analysis is not comprehensive enough to draw this conclusion. Little explicit consideration is given to potentially relevant feedback effects of greening the water cycle. This includes the potential for increased local precipitation recycling (P), reduced vapour pressure deficits (VPD), temperatures (T) reduced through increased latent heat flux, which in turn affect water partitioning and thus water yield. These points need at least to be discussed in substantial detail and the conclusion needs to explicitly state these limitations.

(2) Linked to (1) is the design of scenarios S2 and S3, which may indeed contain a fallacy. By fixing NDVI and land cover to the values of 2001 (S1) and 2018 (S2), respectively, the authors aim to isolate the “real” and potential effects of greening. This is in principle ok. BUT: it is not clear from the description of the experiment how the related feedbacks in P, T, VPD and even radiation (changes in albedo!) were accounted for. As far as I understood, the assumption for the 2 scenarios was that *only* NDVI and land cover is fixed to the values of the 2 individual years. If this is so, the authors overlook that the observed past P, T, VPD and radiation data already account for changes in NDVI and land cover. What are the effects of that? Do the results then really allow to isolate effects of greening? If I am mistaken here, then I would nevertheless ask the authors to makes this much clearer in the description of their experiment.

(3) Linked to (2), the description of the models and their actual implementation does not provide sufficient detail to allow the reader to meaningfully assess the results and interpretations. For example, it is completely unclear how NDVI, land cover but also soil data were used. The reader can only assume that these data are somehow used to estimate the variables PAR and FPAR (Eq.1). Even if this is described elsewhere in detail, it will be necessary to provide this crucial information here as well. In addition, it remains completely opaque, which parameters the two models required and which of those had to be calibrated and which were a priori fixed (e.g. from look-up tables). How were the models calibrated? Was the calibrated model tested on independent data (which should be a rather standard procedure in the year 2021)?

(4) Linked to (3), no attempt, whatsoever is made, to estimate the uncertainty around the models, their parameters and the associated results. Not even confidence intervals around the regressions (and the underlying parameters) are given. Quite frankly, I find this very surprising, as this should be part of any meaningful and serious scientific protocol.

Additional specific comments:

• 2, l.42: this is a completely non-sensical use of the term “drought”. The term “drought” is always refers to a negative anomaly with respect to a specific local reference value, defining a “normal”, typically a median. By convention, conditions below this normal are then defined as “drought”. By extension, there can then be no location with more “frequent” droughts than other locations, as drought is always the deviation from the local/regional normal.
• 2, l.50-51: if afforestation was meant to safeguard water availability, then this is in contradiction with l.55-56. Please rephrase.
• 2, l.56: greater leaf area in itself does of course not increase transpiration. Vegetation metabolic activity increases transpiration. Leaf area is merely an indicator for increased metabolic activity and thus transpiration.
• 2, l.61: should read as “…are not feasible…”
• 3, l.69: what are “hydrological entities”?
• 3, l.76: droughts are low frequency phenomena that require considerable time to develop and to recede. The 18 years of this study are thus likely not enough to make a meaningful statement about changes in drought regimes.
• 4, Figure 1: please also show the location of the reservoirs and hydroelectric facilities.
• 6, l.130-131: irrelevant, can be omitted
• 6, l.132-133: not clear what is meant here. Which other models?
• 6, l.138: more detail is needed for this choice here. Why 45%? How sensitive is the model to this choice?
• 6, l.139-140: how were PAR and FPAR determined?
• 6, l.139-150: Much more detailed is needed on which parameters these models feature and how the parameters were determined, including their prior distributions and the calibration strategy applied.
• 7, l.163: R2 and NSE have a very similar information content: NSE collapses to R2 in the absence of a bias. Thus, I am not sure of the added value of using R2 as performance metric here.
• 7, l.164: reliable? Many would argue otherwise (e.g. Schaefli and Gupta, 2007, HP). In addition, what does “reliable” actually mean here?
• 10, Figure 4: given that the model only provides monthly estimates of water yield, the model does not do a particularly good job in reproducing the observed water yield, in particular for the 2012-2014 period. What is the implication of this? What are the uncertainties around that? How does it affect the results and interpretation?
• 11, l.225-226: I am concerned that the change point analysis here is really very sensitive to the rather short time period considered and that the points identified here may be mere artefacts (e.g. Zhou et al., 2019; HSJ). I strongly suggest to omit this from the analysis.
• 11, l.220ff: am I right to assume that scenarios S1 and S2 are shown and discussed in this section? Please clarify and make this explicit.
• 12, 237ff: not clear what is considered here. Is it the difference between S1 and S2? If yes, I wonder how much of the correlation is spurious, as NDVI is kept constant, while still using observed T and VPD that are the result of a variable vegetation cover. This needs to be made much clearer.

Zhang et al (2021) coupled the CCW and WaSSI models to study how vegetation greening impacted water yield of the Upper Han River Basin (UHRB). They first simulate water yield change from 2001-2018 to evaluate the model. Afterwards, they run two simulations to isolate the effect of vegetation on water yield and the effect on future potential water yield. Zhang et al (2021) show that vegetation greening significantly reduced water yield. The water yield reduction was stronger during warm or dry years. Furthermore, they show that greening could increase the number of droughts. They discuss their results in relation to the important role of the UHRB to provide water to other regions through a diversion project.

The study has an easy to understand set-up and addresses a relevant subject. The manuscript is clearly written. I listed some (major and minors) comments and suggestions below, both on the content and text.

• The authors show that vegetation greening significantly reduced water yield and streamflow during the last decades. The authors discuss the implications for the SNWDP and other Water Diversion Projects and state that (future) vegetation greening could potentially reduce the annual water yield supply by 7.3 km³. A few processes are missing in the manuscript that impact streamflow under changing vegetation. These processes could reduce the ‘negative’ effects of vegetation greening on water yield. First, the extra evaporated water will partly recycle back to the Earth’s surface and increase precipitation (P) (potentially within the UHRB catchment). This could have impacted your P during the studied years (therefore, the S2 and S3 scenarios are not entirely independent of vegetation status), and likely has an impact on future water yield. The study cannot separate this effect on increased P, but they could at least be included in the discussion of the manuscript. Second, the rising CO₂ concentrations are expected to increase the water use efficiency of vegetation, and this could reduce the ‘negative’ effects of future afforestation.
• L31: For example … hydrological services: this sentence should be rewritten
• L57: Consume instead of consumes
• L61: are instead of is
• Please adjust figure 1a (inset) to meet the HESS guidelines (remove the dashed line south of China to depoliticise the manuscript)
• L136: please specify that is the light use efficiency.
• L140: how are the values of the ‘environmental scalars’ determined? Are they independent of the vegetation data? And how is APAR determined? Is APAR also fixed under de S2 and S3 scenario?
• L161: the streamflow records of the reservoir ( / the Danjiangkou Reservoir)
• L173: dynamic greening effects instead of dynamics greening effects
• L179: The Mann-Kendall test is used for trend and change point detection. Could the authors elaborate on the change points you found? Why did they decided to use change-point detection analyses instead of trend analyses only? What extra information do these change-points add to the discussion or results of the manuscript?
• Fig 4a+b legend: km³ per year / mm per year.
• Fig 4a: the simulated WY seems to show a higher decreasing trend than the measured WY. Was there a negative trend in the measured WY?
  
  She et al, 2017 (fig 2a) (https://doi.org/10.1002/2016JD025702) fitted an increasing trend through WY at the Danjiangkou Reservoir between 2000 and 2010 (same data). How does this compare to your results?

• L237: why did WY increase due to vegetation greening in high elevation areas? Could it also be a climate-related effect in these high-elevation regions?
• Line 239-240: Did you mean to refer to fig. 5c instead of 5b?
• 8: How is the relative change calculated? Relative to the year 2010, or the S2 scenario? Why is the sign of the absolute WY change opposite of that of the relative WY change. What does this say about the effects of greening versus climate?
• L261: 2001-2018 instead of 2001~2018
• L307: ‘Unlike the Loess Plateau … but climate did’ seems to contradict with your results. How should this sentence be interpreted?

The manuscript simulated and analysed the effect of vegetation greening on water yield for the South-to-North diversion project (SNWDP). The manuscript is presented in a clear way and method and analyses are internally logical and consistent. However, the manuscript’s narrow scope and sole focus on ‘watershed management’ does not account for critically important connectivities in the water cycle over land, or discuss the overall sustainability benefits or trade-offs of land and water management options. While limitations in scope are necessary to all scientific studies, here, the limitations and the insufficient acknowledgements thereof have resulted in potentially misleading statements about the effect of vegetation greening on runoff and lead to misinterpretations in terms policy/management implications. 

The authors write for example: “Overall, our study suggests that afforestation could potentially reduce local WY, thus weakening the capacity of the water supply to SNWDP.” and “Our study suggests that improved watershed management (e.g., forest management and reducing water use) is needed to address the effects of vegetation greening and climate change on water supply capacity in watersheds serving as water sources for large water diversion projects.” The authors might not mean this, but it is easy to interpret this as an argument for limiting re-greening and reducing vegetation. The slight absurdity in such sentence formulations can perhaps be illustrated by applying the same logic to the model simulation results of (Kleidon, Fraedrich, and Heimann 2000), who found that a global ‘desert world’ yields 37 000 km³ per year runoff whereas a ‘maximal green world’ yields 28 000 km³ per year of runoff. Of course, Kleidon et al., (2020) also noted that both precipitation over land and total evaporation from land were substantially higher in the ‘green world’ scenario. However, with the authors' logic and narrow focus on ‘water yield’, would they have stated that ‘the presence of terrestrial vegetation potentially weakens the capacity of the water supply’? 

I recommend the authors to (1) either expand their scope (to test how the results would be affected by accounting for moisture recycling including greening in upwind areas, and/or CO₂ fertilization under different assumptions), or (2) substantially revise the framing and conclusions of the paper. To test the sensitivity of the results to moisture recycling and greening in upwind moisture supply areas, the authors could for example make use of publicly available data of atmospheric moisture flows (Tuinenburg and Staal 2020; Tuinenburg, Theeuwen, and Staal 2020; Link et al. 2020). Sensitivities to CO₂ fertilization could potentially be investigated by testing different parameterizations in the models.

If option 1 is considered out of the scope, I recommend the authors to revise the title, the abstract, discussion, and conclusions so it is among others clear that i. the vegetation change considered are only within the basin; ii. that key processes and feedbacks such as moisture recycling are missing from the simulations which are likely to reduce the water yields risks reported (see for example Weng et al. (2019) that shows that strategic location of reforestation in upwind areas can in fact help support water use demands, and Wang-Erlandsson et al. (2018) that shows that irrigation in India and other countries contributes to precipitation over China by increased moisture supply); and iii. that overall ecosystem services and trade-offs (e.g., Onaindia et al. 2013) provided by reforestation or restoration projects have not been considered herein (please consider discussing these). I find the authors’ current recommendations to be cautious of over-reliance on the water supply of the SNWDP project under future greening scenarios to be motivated and relevant. The authors could also elaborate on what they mean by consideration of ‘forest management’, for example referring to examples of reforestation approaches that provide relatively high ecosystem service benefits with low evaporation rates. Elaborating on these points could help make the paper more nuanced and insightful, and less prone to being mis-interpreted. 

References

Kleidon, Axel, Klaus Fraedrich, and Martin Heimann. 2000. “A Green Planet Versus a Desert World: Estimating the Maximum Effect of Vegetation on the Land Surface Climate.” Climatic Change 44 (4): 471–93.

Link, Andreas, Ruud van der Ent, Markus Berger, Stephanie Eisner, and Matthias Finkbeiner. 2020. “The Fate of Land Evaporation – a Global Dataset.” Earth System Science Data. https://doi.org/10.5194/essd-12-1897-2020.

Onaindia, Miren, Beatriz Fernández de Manuel, Iosu Madariaga, and Gloria Rodríguez-Loinaz. 2013. “Co-Benefits and Trade-Offs between Biodiversity, Carbon Storage and Water Flow Regulation.” Forest Ecology and Management 289 (February): 1–9.

Tuinenburg, Obbe A., and Arie Staal. 2020. “Tracking the Global Flows of Atmospheric Moisture and Associated Uncertainties.” Hydrology and Earth System Sciences 24 (5): 2419–35.

Tuinenburg, Obbe A., Jolanda J. E. Theeuwen, and Arie Staal. 2020. “High-Resolution Global Atmospheric Moisture Connections from Evaporation to Precipitation.” Earth System Science Data 12 (4): 3177–88.

Wang-Erlandsson, L., Ingo Fetzer, Patrick W. Keys, Ruud J. van der Ent, Hubert H. G. Savenije, and Line J. Gordon. 2018. “Remote Land Use Impacts on River Flows through Atmospheric Teleconnections.” Hydrology and Earth System Sciences 22 (August): 4311–28.

Weng, Wei, Luís Costa, Matthias K. B. Lüdeke, and Delphine C. Zemp. 2019. “Aerial River Management by Smart Cross-Border Reforestation.” Land Use Policy.