Review of "Substantial root-zone water storage capacity observed by GRACE and GRACE/FO" by Zhao et al. . This paper describes the use of TWS estimates from the GRACE satellite project to estimate multi-year water storage changes.  Negative changes are used to estimate a lower-bound on root-zone water storage (Sr).  The estimates are compared to two alternative (Sr) methodologies, and all three Sr estimates are used to parameterize a hydrologic model.  The main result is that the authors' Sr is significantly larger than the previously described Sr estimates.  

Comments

General

I found the authors' use of GRACE data to be novel, and the results interesting.  The paper is well-written and generally clear.  As with other GRACE studies, the spatial resolution of the data is relatively coarse, so I suggest adding some discussion of how these results might be applied in the operational configuration of land models, which would typically have finer spatial resolution.  

Abstract

The maximum water held would be the difference from saturation to wilting point.  But saturated conditions are unlikely to occur at these spatial scales in many regions.

1st sentence defines Sr, and the next sentence discusses simulations.  Perhaps add a sentence indicating how Sr is used in a modeling context after the 1st sentence to provide context.

Line 15: to be clear, GRACE measures gravity and TWS is inferred from that, so the use of the word 'direct' can be problematic.  There are other geophysical processes that affect time-varying gravity.

Line 20: what does 'correlates realistically' mean?  Can you use a more specific or quantitative description?

Introduction

Line 26: 'plants can store during wet periods' should be 'plants can access'?  i.e. plants aren't storing the water, the soil is storing water.

Line 37: why would it overlook rock moisture and groundwater?  This sentence implies a different reason besides uncertainties in rooting depths or hydraulic properties, which are mentioned previously.

Line 49: again, the word 'direct' I find problematic.  If you wish to use this word, perhaps add a sentence explaining its use.

Methods

Line 61: clearly, 'root-zone' implies vegetated areas, but what might one learn from this method in more arid regions?

Line 68: typically P, ET, and R refer to fluxes.  To be more consistent with other literature, consider using rate or flux units consistently and include a summation symbol in equation 1.

Line 75: 'consumed' could be changed to 'transpired' or 'returned to the atmosphere'

Line 82: in areas in which widespread groundwater use is absent, how will this trend removal affect your results?  Is it likely to increase or decrease your Sr estimates for such areas?  Could you use maps of irrigated area, such as AQUASTAT, to confine this operation to areas where widespread irrigation occurs?

Line 91: how runoff is used here is not clear to me.  Is there a budget equation that could be shown?  What does 'surface water' encompass; rivers, lakes, reservoirs, ...?

Line 109: to what extent is Yang 2016 a model-based dataset versus an observational dataset?  

Line 111: how is water holding capacity defined?  Field capacity minus wilting point?

Line 132: why is this an approximation?  Are there other modeled water storage components in HydroModel that were ignored?

Line 141: 'ET anomalies'

Line 146: Does Xiong 2023 use GRACE water storage for their ET estimates?  If so, does that reduce its independence from your results?

Line 169: you say that you compare the two datasets, but you don't explicitly say what your hypothesized relationship between them is, so the justification here seems weak.  In areas that are not water limited, one could imagine that GPP would be high, but a deep root zone is not necessary.  Perhaps expand further on your reasoning in this paragraph.

Line 194: is this saying that the durations shown in 3c) and 3d) are often larger than that shown in 2b)?

Line 225: Do these patterns correlate with a particular land cover or vegetation type?

Line 271: plot d) is unclear to me.  You create an Sr estimate from Miguez-Macho 2021, but then plot it against transpiration instead of GPP; why is this done differently from a) - c)?

Figure 8: why are the x- and y-axis ranges different for plots a) - c)?  It is harder to compare the scatterplots because of this.

Discussion

Line 321: does root-accessible water require that the roots physically occupy the entire storage domain?  For example, as soils dry, upward moisture fluxes can occur which might replenish soil moisture deficits near roots.  Might this help explain the mismatch between observed rooting depths and the Sr estimates here?

Line 325: one could also interpret your Sr/WHC as simply the effective soil depth.  For land models that do not use an explicit Sr variable, this could indicate that models with a soil depth < 2m (i.e. some of the GLDAS models) are likely incapable of simulating these kinds of drawdowns, which would have implications for studies of groundwater that have used GLDAS to remove the soil moisture component of TWS.

Line 326: 'tap'

Figure A1: how does this result relate to the relationship between magnitude and duration?  Does it imply that during the largest drawdowns, there is also the largest 'net precipitation'?  That seems counterintuitive.

In this work, the authors use the Gravity Recovery and Climate Experiment (GRACE) and GRACE Follow-on (FO) to estimate root-zone storage capacity (Sr). They find estimates of Sr that are much larger than those using mass-balance approaches and rooting depth parameterizations. I found the work interesting, and the writing was succinct and clear. However, I had a difficult time understanding the assumptions and the implications of these assumptions to evaluate the results. I think the authors need to be much clearer about the implications of their assumptions.

Main comments:

• The proposed method is quite different from previous work because it directly uses total water storage (TWS) from GRACE. However, GRACE measures a combination of surface water, groundwater, soil moisture, snow and ice. You explain how you remove the streamflow and snow/ice…but how do you remove the effect of groundwater? Are you assuming that groundwater is part of Sr? In some cases, as water table becomes more shallow, conditions become anoxic for plants…wouldn’t this decrease Sr? The role of gw in Sr calculations must be better explained and the assumptions clearly laid out.
• The proposed method is also quite different from previous methods in the spatial and temporal scale. You are looking at monthly data at 3x3 degrees. This would include several ecosystems that behave very differently. It also includes multi-year droughts…whereas other calculations would account for periods of deficit (calculated at the daily timescale) with a certain return period. This is a completely different metric…is it really appropriate to compare these?
• I am having a difficult time understanding physically what it means to calculate deltaTWS as the difference between TWS anomalies. Are you assuming that the soil will be at saturation at the beginning of the drawdawn, but will never reach saturation throughout the drawdawn period? Is this an appropriate assumption?
• I don’t think you should use GRACE to evaluate the performance of HydrModel that includes GRACE information. You state that this is not circular…but it is. Another metric could be streamflow, it would be independent.
• The part about linking Sr to vegetation growth was not very convincing. I think you are comparing maximum GPP to the point of saturation…so if I understand correctly what you are showing is that vegetation activity is enhanced when there is enough water. I don’t think this argument is necessary for your paper.

Title: Substantial root-zone water storage capacity observed by GRACE and GRACE/FO

Author(s): Meng Zhao et al.

MS No.: egusphere-2024-1939

The manuscript derives “root water storage capacity” (Sr) from GRACE and GRACE-FO observations of terrestrial water storage (TWS), along with uncertainty estimates.   The GRACE-based Sr estimates are compared to Sr estimates derived (i) from soil parameters (soil depth and soil water holding capacity) and (ii) water balance estimates (using precipitation and evapotranspiration [ET] observations).  The authors find that the GRACE-based Sr estimates are 50% larger than those derived from water balance estimates and 380% than those derived from soil parameters.  The different Sr estimates are further used to parameterize a USGS “bucket model”, with TWS and ET output from the model validated against GRACE TWS observations and ET estimates from a water balance approach.  Finally, the authors find that their GRACE-based Sr estimates correlate “realistically” with vegetation productivity data.

The authors address a clear need for accurate estimates of root zone water storage capacity, a topic of interest to HESS readers.  However, the findings of the manuscript are not supported with independent observations and are largely circular.  It is no surprise that the GRACE-based Sr estimates have a relatively lower error against GRACE-based TWS observations.  Specifically, the GRACE-based Sr estimates essentially reflect the range of the GRACE TWS observations, and the NSE metrics primarily measures skill in terms of the mean-square error (MSE).  Additionally, it remains unclear to me how the authors remove the groundwater signal from the TWS observations.  I recommend that the manuscript be rejected. 

Major comments:

• The validation approach is circular (contrary to the statement in Lines 137-140). The GRACE-based Sr estimates reflect, by construction, approximately the dynamic range of the validating GRACE TWS observations (as shown in Figure 1).  The surface meteorological forcing inputs to the USGS model are the same for all three simulations, and the only difference between the USGS model configurations is in the Sr parameters.  The simulated TWS and ET will therefore have very similar *standardized* anomalies (Z-scores), and the key determinant of the NSE metric will be whether the dynamic range of the simulated TWS anomalies matches that of the verifying observations.  The latter were used to determine the GRACE-based Sr, thereby essentially guaranteeing a lower MSE and higher NSE for the simulation with the GRACE-based Sr relative to the other simulations.  (As an aside, Line 226 refers to “performance in simulating TWS temporal dynamics”.  This is a bit of an overstatement given the fact that the experiment design primarily measures how well the estimated Sr reflects the dynamic range of the TWS observations.  “Temporal dynamics” suggests skill differences in seasonal and interannual variations, which are not explicitly examined and which are likely to be small, given the experiment setup.)
• The ET estimates used to validate the USGS model simulations are based on water balance estimates derived from precipitation and water storage change datasets, which is similarly circular when it comes to validating the model output from the simulations that use Sr estimates based on GRACE observations or water balance estimates.
• The definition of Sr as “root zone storage capacity” seems inconsistent the derivation from GRACE TWS observations. The authors explain how they remove the snow signal and anthropogenic groundwater signals from the TWS observations when they derive the GRACE-based Sr estimates. However, it remains unclear how natural groundwater fluctuations are handled.  TWS observations include natural variations in groundwater levels that are not related to water storage in what would usually be considered the “root zone” (e.g., in grasslands).   Perhaps it is intentional that such fluctuations are included, but then the derived parameter is then no longer a “root zone storage capacity” in the sense that the control volume is no longer what is commonly understood to be the “root zone”. 
• It is highly concerning that no model attains positive NSE values for 40% of the global *vegetated* domain (Lines 216-217). This area includes most of the subtropics and Southern Hemisphere!  If the model is so poor that for nearly half of the domain of interest a time-invariant constant would be a better estimator, what does it say about the skill of the model in the other half of the domain?  And what does it mean for the Sr estimates in the nearly half of the domain of interest where NSE is negative for all three model simulations?   

Minor comments:

1. The heading of section 3 should probably be “Results”
2. The caption of Figure 3 does not clearly state the base for the “percentage changes”. This can only be understood from the text.
3. Line 208: Be more specific about the “drier climates and lower-biomass regions”