This is definitely an important study. I agree that Land Surface Model could be improved in the definition of soil depth. The impact of rooting depth in determining the active soil moisture and the amount of water that can transpire ultimately influence the occurrence and amplification of heat waves, an increasingly pressing issue in an era of climatic changes. The model development proposed in this study could improve climate services that are primary forms of climate adaptation in many sectors.

The manuscript is generally well-written and contains high-quality research. However, I was not familiar with the method used in this manuscript. This is reflected in some questions and comments that the author could consider.

The Memory Method is definitely very interesting. I believe it is reasonable enough to assume that local vegetation adapts the rooting depth according to the drought frequency. However, it's difficult to me to understand the way it is implemented in the model, i.e. varying the total soil depth in the model grid cells. I think it would have been more reasonable to change the rooting distribution Z from the model formulation instead of varying the bottom soil thickness. Afterall, in reality it is the vegetation adapting the roots, not the soil changing thickness. Why this approach was not followed?

The modelled approach assumes that the maximum holding capacity should be equal to Sr. If I understand well it should rather a minimum value corresponding to dry years right? For tall vegetation, the root depth is defined through the memory method as a function of the 40 years return period drought and 2 years for low vegetation. With the implementation considered in this study it seems like the soil cannot hold more moisture than the one available in dry years. Maybe this is the reason of the apparent systematic underestimation of Sr by the model.

Also, the definition od Sr for the observations is a linear superposition of different values that are obtained for high and low vegetation, suggesting the these values are different. Modelled Sr is defined as one value for all vegetation types. So not only the implementation is the model seem to assume that the soil cannot hold more water that than in dry years, but also that is does not depend on the vegetation type. Again, maybe a different approach is needed.

Finally, I think the modelled Sr could be computed exactly in the same manner as it is done for observations. I understand the author prefer to use a more physical definition, but probably computing it in the same way as in the observation would be a fairer comparison that could be used to better calibrate the modified model.

Minor comments

Equation 2 is formally incorrect. The time-dependency S(t) does not match the right-hand side where t is the integrating variable that should disappear after the integration is performed. The actual variable that survives should be related to t0 and t1, which are not defined either than in Appendix A. I think a subscript for the hydrological year should be preferred in that case. About the subscripts, here 'd' is used. 'r' in used elsewhere, why? I think ‘d’ stands for deficit and ‘r’ stand for roots. If so, is the right hand side of equation 7 there should be 'd', not 'r'.

I agree that considering constant ratio of actual vs. potential evapotranspiration is a crude approximation, especially in water limited regions. I also agree that the other factors mentioned by the authors (groundwater, irrigation) are important as well. These are difficult to improve in the model in short times. However, the author of implemented an iterative step in their approach to reduce strongly the uncertainty relate to the interannual variability of that ratio (Appendix A). Couldn't they use somehow the observed evaporation to further improve the estimation? I think this would improve the estimated Sr, especially regarding the intraseasonal variability, and eventually improve the modified model calibration.

To the uncertainties, I would add the model drainage rate that, in the current framework, it could as important as the rooting depth. The results obtained by the authors could be due to excessive retention (slow drainage) rather than too deep root sone. This is also supported by the fact that other models are rather augmenting the soil depth adding a groundwater layer instead of reducing it (e.g. CLM).

Equation 22: where is the equal sign?

Figure 6: Shouldn't Q + E equal P in the long period? If this should be true, the average anomalies of the modified model on the top panel should be equal the average anomalies of the bottom panels as precipitation does not change, but this is not true. E anomalies are much smaller, Why? Am I missing something?

see attached PDF

Review of ESD-2021-3

van Oorschot, F. et al., Climate controlled root zone parameters show potential to improve water flux simulations by land surface models

Review by Andrew J. Guswa, aguswa@smith.edu

In this work, the authors compare results from the HTESSEL land-surface model with vegetation root depth determined in two different ways.  The control case (CTR) comprises root depths (and, correspondingly, root-zone storage capacity, Sr) determined via soil depth; in another formulation (MD), Sr is determined via the memory method, which is based on the concept that plant roots adjust to mitigate droughts with certain return periods.  These models are implemented for 15 catchments in Australia with tropical, temperate, and Mediterranean climates.

Results reveal that the root-zone storage capacities determined via the memory method are shallower and more variable across the watersheds than those from the control cases.  This is in contrast to what others have suggested – that root depth in LSMs may be too small.  The changes in Sr (from the CRT to MD approach) manifest as improvements in the variability and seasonality of streamflow.  Long-term water balances and ET are relatively insensitive to changes in Sr.  The paper is well-written, and the methods and results are well-explained and valuable.  I offer a few suggestions and comments below, and I recommend publication with minor revision.

Low sensitivity of ET

The authors note that model estimates of ET do not appear very sensitive to the differences in Sr.  I agree with Hageman that the low sensitivity may be due in part to the fact that the values of Sr,MD are less than Sr,CTR.  The catchments in question are arid/water-limited, and ET dominates the water balance, with annual ET being 4-16 times the annual streamflow.  One of the proclaimed advantages of the memory method is that it facilitates deep and/or expansive roots so as to maintain ET during periods of drought.  In this work, it seems that the soil-based root depths (Sr,CRT) were already sufficiently large, such that the limits of the root zone storage were not being approached.  An analogy would be a water-supply reservoir that never fills – if you make it a little smaller, you do not affect the available supply, since you were starting with excess capacity.  The authors may wish to expand their discussion in the paper along these lines.

Additionally, the authors may also wish to acknowledge (again) the uncertainty in the ET observations.  As they point out in lines 110-115, the estimates from the water balances are lower than the estimates from the FLUXCOM data by 20%, and it may be worth reminding the reader of this in the discussion.

Minor comments

Lines 110-121 and Figure 1:  The authors may wish to address whether the lower estimates of ET via the water balance method might be affected by or attributable to deep groundwater drainage that does not get recorded by the stream gauges.

Lines 155-eq 7: Like the other reviewers, I am a bit confused by the approximation of “Catchment Sr,MM” by a weighted average of the Sr values for trees and grasses.  As drainage below the root zone is a non-linear process, this will affect the results (e.g., drainage below the grass portion may be non-zero, whereas drainage below an “average” root depth may be zero).  I understand that the model is limited in its resolution, and one has to make some concessions.  It may be worth an expanded comment, however.

Figure 6 is a nice figure with a high information density.  Like reviewer 1, however, I found the differences in y-axis scales to make interpretation challenging, e.g., the differences in the Temperate and Mediterranean Q appear (visually) to be much greater than those in E.  Perhaps it would be worth showing the six subplots, first with all the same y-scale, and to then provide second versions of 6b and 6c that are more zoomed in.  I think seeing how very small the runoff is from the Temperate site, before diving into the differences among the models and observations, would help the reader.

Lines 298-302: The authors make a point to acknowledge that Sr need not be synonymous with root depth, which is fair enough.  In this work, however, it IS synonymous, and I found this introduction to the discussion a bit odd.  I recommend that the second sentence of the first paragraph be dropped.