Review of Heselschwerdt et al., egusphere-2025-5896

“Large impact of extreme precipitation on projected blue-green water shares“

This study investigates the partitioning of precipitation into runoff and transpiration, and its changes through this century. The authors introduce a new metric to describe this partitioning and use output of Earth system model simulations to determine the partitioning and its changes until the end of the century throughout the globe. They identify hot spot regions where precipitation is mainly partitioned into runoff or transpiration, respectively, and regions with strongest projected changes in the partitioning. Finally, they attribute the partitioning changes mainly to changes in extreme precipitation and vegetation and derive management implications.

-------------------

Recommendation:

We think the paper requires major revisions.

The topic of this study is interesting and timely. Ongoing global change affects climate and vegetation in various ways, and their complex interplay modulates the water cycle. This way, major water fluxes such as runoff and transpiration may change differently, and for partly different reasons. The precipitation partitioning metric introduced by the authors in this paper summarizes relevant water cycle changes beyond individual variables, including an attribution of the resulting spatial patterns of change. This analysis presents a relevant contribution because it helps to identify regions with relevant water cycle changes, and proposes related mechanisms that require additional attention in further research. 

However, before the paper can be published we would ask the authors to consider the concerns described below:

--------------------

General comments:

1. The introduction section falls short in introducing the knowledge gap that this study addresses. This comes in combination with an insufficient review of existing literature on the topic. The present text claims that not much research exists in this direction (lines 62-63) while there is actually quite a body of literature that jointly analyses runoff and (evapo)transpiration, such as for example the Orth & Destouni 2018 study which is cited later, among others.
   
   Related to this, lines 34-61 provide interesting mechanistic background on precipitation partitioning, which, however, is not connected to the rest of the introduction. So we suggest either to shift this to the discussion section, or connect it more to the introduction to motivate the relevance and knowledge gap in the topic. 
2. While the authors choose transpiration as green water flux, this choice is limiting the scope of this global analysis. For example, in arid areas transpiration is negligible while evaporation constitutes most of the surface water flux which is then ignored in the analysis. Additionally, soil moisture is referred to as green water and is modulated by transpiration but also evaporation.
3. A large part of the explanation of the results and the respective discussion, as well as attribution is based on changes in vegetation characteristics. However, at least one model has prescribed LAI dynamics (CNRM-ESM2-1, Seferian et al 2019, see page 4186), which means that its vegetation dynamics may be underestimated and related to this also its contribution to blue-green water partitioning and related trends. We suggest a review of the current model selection in this context.
   
   Furthermore, while extreme precipitation is identified as a main control of trends in blue-green water shares, this may not be well represented by Earth system models. For example, a recent study found that the current rather coarse resolution of data from climate and earth system models, as also used in this study, underestimates the impact of such extremes (Brunner et al. 2025). 
4. While we appreciate the comprehensive set of potentially influential variables included in the regression analysis, we would like to suggest some additions. In addition to RX5day, where the chosen time scale is actually not motivated, you could test RX1day to capture even more extreme precipitation. Moreover, the seasonality of precipitation and VPD (even mentioned in line 407, could be expressed as e.g. standard deviation across monthly values) could influence the partitioning which is expected to be different in regions with pronounced rainy seasons versus regions where precipitation is more equally distributed across the year. Also land use change could play a role in some regions. While this is indirectly captured in leaf area index it could be more explicitly included by considering the time-varying crop and forest cover fractions. Further, assessing the role of near-surface soil moisture in addition to that of total-column soil moisture may be insightful because they can be expected to be of different relevance to runoff and transpiration. In this context, it is unclear which “relative changes” of soil moisture (line 99) are evaluated instead of actual soil moisture output, and why. 
5. The validation of the modelled partitioning against reanalysis and observation-based products in Figure S1 is informative. At the same time you are studying changes in the partitioning over time which is different from the partitioning itself; for this reason you could consider comparing changes in the partitioning during the observational period between models and reference products. Further, Figure S1 is mentioned several times in section 3.1 such that it could be considered to move it to the main manuscript.
   
   Related to this, we appreciate the indication of model uncertainty in Figures 2-4. However, this is not really visible in the maps in Figures 2b-d and 4, and missing in Figure 1. Maybe consider to make this more prominent – or even to introduce a masking to avoid showing results in regions where (i) models disagree or (ii) models disagree with reference products in Figure S1, in order to avoid interpreting results in regions where uncertainties actually largely prevent this. This could also enhance the accuracy of the attribution analysis in Figure 3.

We do not wish to remain anonymous - joint review by Rene Orth and Josephin Kroll.

We have notified the editor of our collaborations with Peter Greve and Lan Wang-Erlandsson.

------------------

Specific comments:

Line 4: given that this is the abstract, maybe replace blue/green water shares with a more generally understood terminology

Line 4: Climate simulations → Earth system model simulations (this also applies to other mentions of climate simulations throughout the manuscript)

Line 8: according to the results in Figure 3 it is leaf area index and water use efficiency individually rather than their interaction that influences the partitioning  

Line 12-16: this motivation of the relevance of precipitation partitioning is very brief, would expand this

Line 21: “stabilises Earth system” is unclear

Line 23: “climate regulation” is unclear

Line 27: "critical water functions” is unclear

Line 54: "vegetation activity” is unclear

Line 57: "runoff-consuming sinks” is unclear

Line 80: required variables are unclear at this point

Line 114: Good point about the non-closure of the water balance. See our recent paper Huang et al. for this. (sorry for the self-promotion)

Line 120: The term ‘climate indices’ does not fit the chosen variables/indices. For example, (i) RX5day is not calculated over a period that would refer to climate (~30 years), (ii) VPD is not an index, and also not referred to over a period considered as climate, (iii) WUE is not representing climate, but rather vegetation characteristics. An alternative could be ‘hydroecological variables’ as used in the manuscript before

Line 163: Explain more clearly in this section that this attribution analysis is done for the spatial patterns rather than for the temporal changes in each grid cell or region. Also emphasize this in the discussion of Figure 3 that you are not directly attributing the temporal changes of the partitioning shown in Figure 2 but merely the spatial patterns therein, where determined drivers are not necessarily the same for both. 

Line 170: Would start this section with this paragraph.

Line 192: “destroyed” → ”removed”

Line 192/193: not sure we get the point here

Line 195: Can you comment on the consideration of lateral flow in the models, and its potential role for the partitioning?

Line 199-201 & 326 & elsewhere: The results section incorporates interpretation/discussion already such that it would be more clear to adapt the title to ‘Results and Discussion’ and remove ‘Discussion and’ from the title of section 4.

Line 202: unclear what’s meant with absolute BGWS values and whether it is referring to model-mean data or observation-based results

Line 207: why interpret regions with large baseline biases at all?

Line 209: Would start the section with this paragraph.

Line 215: Unclear to which average ‘below-average’ is referring - global?

Lines 231/232: There are no visible patterns in the Sahara.

Line 233: unclear what’s the difference between saturation excess vs. infiltration excess

Line 244: Can/Should you assume broadly similar soil properties across rainforests?

Line 257: section 3.2 seems a bit convoluted and was harder to follow: suggestion to rather summarize the main possible mechanisms and then mention example regions instead of selecting regions and describing those 

Line 265: Fig. 2: place legend in panel a on top or below as for the other figures; brings space for spatial average subpanel

Line 268: ‘strongly linked’ suggests causality, which cannot be derived from covariance across space

Line 342: increases instead of ‘is increasing’

Line 351: blue-to-green shift is of similar magnitude, why is this not covered in the text?

Line 352: Actually this section is about implications rather than management. There is little discussion about land or water management measures and strategies. Either revise this section, or drop the management from the title here, and also adapt the wording in the abstract in line 9 (“impact-relevant”, “actionable”).

Line 355: One could ask why it matters where the next rain drop goes in the presence of little to no rain in some regions

Line 368: insert Δ before changes as done in the lines below

Line 385: I wonder about the causal direction here: are transpiration increases supporting vegetation productivity or does increased vegetation productivity induce/come with increased transpiration?

Line 387: Figure 5: this is rather a table than it is a figure. For a figure, I would suggest to make two x-y graphics, having dR or dT on the y-axis and dBGWS on the x-axis. Insert text in the respective 4 panels of the graph

Line 392: ‘sizeable areas’ is unclear

Line 398: ‘environmental flows’ is unclear

– Figure 1: The content of section 3.1 mainly describes and discusses results in the light of the i) water-energy-limitation framework as well as ii) the role of extreme precipitation. Hence, Figure 1 should include maps illustrating the three BGWS features - energy-limited partitioning, water-limited partitioning, precipitation-intensity related partitioning - and the ratio of RX5day/mean precipitation (so fig. S5a). This would make the figure more in line with the title of the manuscript, which emphasizes the role of extreme precipitation and follows the explanation used in the text. Finally, it would be good to cite some more related literature such as Seneviratne et al., 2010 and Denissen et al. 2022.

– Figures 1 and 2: Adjust color bar range across panels b–d in both figures.

– Figure 3: Do the boxes in panels a and b show the spread of the attribution results for the individual Earth system models? Also, panels c and d are not described and discussed in the text. For example, (i) y-axis unclear, response to what?, and (ii) what is the meaning of the lines connecting ‘end-of-century response’ across predictors, especially as the y-axis scale changes. Instead, you could use boxplots and depict individual models by differently colored dots or differently shaped points; +/- dBGWS could be indicated by outline of each dot/symbol or by making two boxplots per predictor.

– Figure S1: Labelling of panels in the caption is not in line with the figure.

– You could consider adding summary statistics to the maps in the figures as also stated in the text.

References:

Orth, R. and G. Destouni 2018: Drought reduces blue-water fluxes more strongly than green-water fluxes in Europe. https://doi.org/10.1038/s41467-018-06013-7

Huang, H. et al. State-of-the-art hydrological datasets exhibit low water balance consistency globally, doi: 10.5194/essd-2025-376

Séférian et al. 2019: Evaluation of CNRM Earth System Model, CNRM-ESM2-1: Role of Earth System Processes in Present-Day and Future Climate. https://doi.org/10.1029/2019MS001791

Seneviratne et al. 2010: Investigating soil moisture–climate interactions in a changing climate: A review. https://doi.org/10.1016/j.earscirev.2010.02.004

Denissen et al. 2022: Widespread shift from ecosystem energy to water limitation with climate change. https://www.nature.com/articles/s41558-022-01403-8

Brunner et al. 2025: A global perspective on the spatial representation of climate extremes from km-scale models. https://doi.org/10.1088/1748-9326/ade1ef 

The authors use the results of 11 Earth System Models to estimate the current partitioning of precipitation into runoff and transpiration and how this partitioning changes under the SSP3-RCP7.0 combined socioeconomic scenario. They also determine which climatic and ecophysiological factors could be responsible for any changes using multiple linear regression methods with analyses of predictor importance (size of regression coefficient and randomnization of predictors). This is an interesting study that deserves publication. I do have some major and some minor issues:

Major issues

1. The use of green and blue water.The BGWS metric is more a measure of process than a measure of quantity fir for management decisions (see hereafter). As such I do not see why one should use the terms green and blue water that are from sustainability water science, rather than from hydrology, which is about process. It would have been much easier to look at a hydrological measure such as the runoff coefficient RC and its change.  1 -RC is then the change in evaporation which is directly related to transpiration as well.
2. If water management is the main target, why not look at absolute changes in runoff (Q) and transpiration (T). If I live in a water-stressed region, I would be happy if both Q and T increase, regardless of whether or the other changes more to lead to a change in the BGWS.In section 3.4 the authors themselves show that for interpretation for water management, one requires to take along changes in Q (floods) or T (drought). This seems overly cumbersome if we can look at changes in P, Q and T directly.
3. I wonder also how much of the changes in BGWS should be attributed to land cover change. If the ESMs are driven by combined SSP-RCP scenarios, exogeneous land cover change is likely included in the ESM land modules, which has a direct impact on evaporation and runoff (see e.g. Bosmans et. al, 2017: https://hess.copernicus.org/articles/21/5603/2017/). Also, human water use change such as the extension of in irrigated areas, when included in the land modules, will play a role.
4. What is missing from the possible important factors is the impact of climate seasonality. For regions that have constant rainfall throughout the year and a clear summer season with more radiation (growing season determined by energy limitation), an equally proportional increase in temperature and precipitation will likely leadto a higher runoff coefficient, because almost all the increased precipitation will runoff in the winter and the increased temperature will not lead to a proportional increase in evaporation because of warm season water limitation. This is even more so the case in areas where precipitation falls predominantly in winter.

Minor issues

1. Global climate models and likely earth system models are often suffering from permanent drizzle, underestimating precipitation intensity. This could lead to an underestimation of the runoff coefficient. Also, the hydrology does not always contain a sub-grid parameterization that results in the correct saturation excess runoff, making runoff overly flashy. If this is a potential issue, this should be discussed.
2. On the other side of the spectrum: it could be that the limited representation of groundwater convergence in ESMs underestimates the fact that shallow groundwater will support evaporation in drylands or during dry periods. This may partly explain the over-estimation of the BGWS as seen in Figure S1. See e.g. Maxwell and Condon (2016). https://www.science.org/doi/abs/10.1126/science.aaf7891
3. Line 109 and line 113: GLEAM includes a soil moisture accounting model to reduce potential evaporation. As such it is a modelled product itself. It is better to compare evaporation of the ESMs with in-situ laten heat fluxes from e.g. FLUXNET (https://fluxnet.org/).
4. Lines 212-213: Here, part of the BGWS is explained by energy limitation and Budyko. Then why not add a map of the aridity index to corroborate this and also use this as a predictor?
5. Lines 219-220: In high altitudes it is also likely that runoff is overestimated due to the underestimation of snow sublimation in ESM: see e.g.Stigter et al. (2018). (https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2018.00108/full
6. Lines 248-253: Here the differences in BGWS between Southern China, Japan and Florida are discussed. Is topography a factor to consider here?