Vermunt et al. use non-destructive sap flow measurements to estimate the diurnal cycle of vegetation water content and then relate it to microwave radar backscatter. This paper is of high relevance to ongoing microwave vegetation measurements and answering large-scale ecosystems questions. I am in support of this work given the low amount of ground measurements and available techniques and consequently high uncertainty in microwave vegetation retrievals. It creatively uses a known application in plant physiology and ecohydrology for microwave remote sensing validation. I think the study is well done and is a great contribution. I ask that the authors consider some comments here before publication.

I do not wish to remain anonymous. -Andrew Feldman

Major comment

I think the methodology needs a clearer section or paragraph that explicitly outlines the method used here, its advantages and disadvantages, assumptions, and how the method can be modified in scenarios of different vegetation types (tree instead of corn). This could be a modification of Section 2. The sections afterward can expand on this as they currently do with section 3 and onward. While reading the methods, I felt as if I was finding out more components required for the method as it went along. It also seems like some steps are optional or can change for different types of vegetation (see my comments below). Be clearer earlier that sap flow sensors, destructive sampling, and weather stations are needed and that this approach is somewhat specific for corn or other herbaceous vegetation types. Lines 54-55 motivate the method as a standard approach used in previous studies, but this method seems different because a sap flow sensor could not be placed in the crown and transpiration needs to be modeled. Furthermore, destructive sampling needs to be used to constrain the VWC estimates (though I am not sure this is always needed; see below). If a different vegetation type other than corn is used, the method can become more reliable because one can use two sap flow sensors and not have to model transpiration (other than relying on an additional assumption about small leaf capacitance). Since this is in part a methods paper, a more organized overview of the method can make the method more reproduceable or easier to modify.

Line specific comments

Line 21: Here or further down, an explicit definition of how vegetation water content is traditionally defined is needed. “Water content” can be confusing because it could be a total water volume (as is the case traditionally with VWC) or could mean a ratio to the dry or total volume (as for soil moisture or soil water content). Therefore, a definition of kg/square meter or other used here would be helpful.

Line 47-49: This is an excellent introduction. The main thing I feel that is missing is I am wondering if the authors could be more descriptive here of the other VWC in-situ sample options, how prevalent they are, and why they didn’t choose them. A few things I am wondering: is the destructive sampling method the most common for radar validation? What specific destructive methods are used (oven drying leaves, branches, etc.)? Why not measure leaf/stem water potentials using automated psychrometers (Guo et al., 2019) since those sensors can provide rapid measurements (then mention why that does not directly provide water volume)? Have others used psychrometers or water potential measurements for radar validation? Another approach used in radiometry for VOD was to use water potential measurements and biomass to estimate VWC (Momen et al., 2017). Similarly, VOD was related to diurnal variations of leaf water potentials(Holtzman et al., 2021). I don’t think the authors need to provide a large description of this (or cite these papers for that matter). I think it may provide more context and perhaps strengthen the motivation to choose the sap flow method by contrasting with other known options.

Line 52: A more specific research question/objective could be helpful here. This objective has been broadly pursued before. The authors are specifically testing whether a non-destructive sap flow technique can measure VWC and thus be used to validate radar diurnal VWC measurements, which is a great endeavor that should be explicitly stated.

Lines 71-78: I became confused here because I thought in line 71 that this approach is applied here. Then I found out that it wasn’t in line 79. Please only mention the assumptions applicable here to a single sensor and estimated transpiration. Then give more detail about the assumptions. You could argue that this approach circumvents the first assumption which could be flawed; the first assumption I think suggests that capacitance is negligible in the leaves and is larger lower in the canopy (trunks and lower parts of branches). This may not always be true for succulents and large trees. With the second assumption and full approach here, I wonder whether day to day variations can still be measured with this approach if a storage term is estimated and stem flow measurements are consistent.

Line 109-113: This paragraph appears to give some extraneous information. It might be helpful to only mention the measurements relevant to this study. The authors are not trying to minimize day-to-day weather variations here.

Line 118: For clarity, one sensor is placed at the base of the plant for each plant (as suggested by lines 79-86)?

Line 144-146: Consider showing an equation of this here.

Line 169: What time of day are these samples from?

Line 180: Since modeling transpiration can be viewed as the largest uncertainty of the method, I would add more details about how P-M equation was used and choices made for certain parameters (like roughness height and others).

Line 184: Using the data to constrain and validate here becomes somewhat circular. I think the method is generally fine. However, I would note that I don’t think this step is entirely necessary – I think the authors can simply try to compare the temporal dynamics of the reconstructed VWC and measured VWC and not worry about correcting the bias too much.

Line 200: Is it true that the VWCt0 reference is needed to get the day to day dynamics right while the VWCt (= sap flux – transpiration) term is all that is needed to explain the backscatter diurnal variations within a day? If so, I would be more explicit about this. This can be seen where, in eq. 3, the constant VWCt0 term would mostly get lumped into the y intercept term. VWCt0 are mostly a magnitude scaling and won’t change the relationship between the VWCt (= sap flux – transpiration) term and backscatter within a day. The VWCt0 is essentially picking up on the biomass and total water storage changes day to day. VWCt is effectively the storage anomaly which is all that is needed to evaluate the backscatter anomaly. The consequence is that if one is only interested in the subdaily variations, the destructively sampled VWCt0 reference used to scale the VWC is not necessarily needed and is an extraneous step (this can be seen with using a panel regression in place of eq. 3 where the eq. 3 regression is effectively applied separately to each day’s diurnal variations). If true, I think this idea should be mentioned. Perhaps the extra step to use destructive VWC sampling each day is to evaluate day to day changes in VWC. The point is that I think one can test the time dynamics of backscatter at large spatial scales using only sap flux and transpiration estimation without needing labor intensive, destructive methods to constrain the magnitude of VWC. If I am wrong, consider clarifying the issue in the text.

Line 235: I was worried about using P-M equation and CDF matching to sap flow to estimate transpiration because transpiration is very hard to estimate/measure. However, Fig. 5 shows this generally works well. It is stated somewhat indirectly, but I would emphasize clearly here or elsewhere that Fig. 5 shows that while modeling transpiration is a major drawback of the method, it works generally well in representing the VWC diurnal cycle.

Line 245-247: Does this mean there is evidence that full rehydration does not take place overnight every day and that capacitance is large enough to have some storage deficit carry-over from day to day? And that the assumption to use sum of sap flux over the day does not hold (lines 144-146)?

Line 266: I think Fig. 8 is somewhat of a disservice to the authors and their nice results. The approach is well set up for sub-daily sampling, but ad-hoc modifications like CDF matching and VWC scaling are needed to represent day to day variations. I would say that the method is strong and well-developed for evaluating sub-daily VWC variations and a bit weaker for evaluating daily variations. In Fig. 8, my eyes are drawn more to the daily than diurnal changes which is not the focus of the paper and section heading. Consider showing diurnal variations individually for a few days (by segmenting individual days) to emphasize the results if possible.

Line 292: Is this unexpected that VV is more sensitive than cross-pol to vegetation?

Line 300: Dew is receiving increased interested in its impact on diurnal observations of microwave emission and backscatter. Can the authors contextualize the dew results in the table a bit more? It is hard to tell if “c” is a large or small contribution to the signal compared to “b” without knowing typical dew variations in kg/m2. Maybe a variance-explained or normalized slope metric can help readers determine how much dew and internal water content relatively influence each backscatter signal. Only comparing the absolute slopes here does not fully show the relative contribution to the signals. It seems the authors are exhibiting less confidence in the dew results (i.e., lines 371-377) and it is not clear why (while the result in lines 366-370 are very interesting!).

Line 308: Arguably, the destructive samples may be optional here, especially for sub-daily variations, which strengthens the results.

Line 400: I think it is worth mentioning the sapfluxnet project and how one can use those data (along with station or flux tower data) to validate time dynamics of VWC seen by passive and active satellites at large scales.

I recommend commenting on whether such an approach here can be used to evaluate day to day variations in VWC. Are there ways in which the transpiration = sap flow scaling in the early morning can be reIaxed such that total storage over a day can be computed and evaluated day to day (line 147-149 start to get at this)? I know this becomes uncertain due to phase lags between transpiration and sap flux caused by the capacitance that the method is trying to measure. But I think this study is a nice step towards that and the authors recommendations for how that can be done or recommendations against it could be helpful for the active and passive microwave vegetation community. If the authors feel that is off topic, feel free to ignore.

I also recommend commenting somewhere on how one could use such a method to validate satellite observations. What would be required in this case to estimate a diurnal VWC cycle at the footprint scale? It seems like a few meteorological measurements and sap flux sensors would suffice to at least understand the diurnal cycle.

Guo, J.S., Hultine, K.R., Koch, G.W., Kropp, H., Ogle, K., 2019. Temporal shifts in iso/anisohydry revealed from daily observations of plant water potential in a dominant desert shrub. New Phytol. https://doi.org/10.1111/nph.16196

Holtzman, N., Anderegg, L., Kraatz, S., Mavrovic, A., Sonnentag, O., Pappas, C., Cosh, M., Langlois, A., Lakhankar, T., Tesser, D., Steiner, N., Colliander, A., Roy, A., Konings, A., 2021. L-band vegetation optical depth as an indicator of plant water potential in a temperate deciduous forest stand. Biogeosciences 18, 739–753. https://doi.org/10.5194/bg-2020-373

Momen, M., Wood, J.D., Novick, K.A., Pangle, R., Pockman, W.T., McDowell, N.G., Konings, A.G., 2017. Interacting Effects of Leaf Water Potential and Biomass on Vegetation Optical Depth. J. Geophys. Res. Biogeosciences 122, 3031–3046. https://doi.org/10.1002/2017JG004145

This research demonstrates the estimation of continuous vegetation water content (VWC) in corn crops at two research sites by adapting an existing method for measuring internal VWC in trees. Sub-daily VWC was succesfully calculated based on the difference between modelled transpiration and sap flow rates at the base of corn stems and constrained and validated with destructive sampling. Second, the research demonstrates the effect of diurnal variations of VWC and dew on radar backscatter. The study is innovative and is a valuable contribution to the field as itprovides new methods and insight in current questions in microwave remote sensing, such as the effect of internal VWC and surface canopy water on the radar signal. I highly recommend to publish the paper, but I have some minor comments. 

I believe the data and methods can be described a bit better. If I understood correctly, backscatter data is only available for the 2018 campaign in Florida, but sub-daily destructive samples are only available for the 2019 campaign in the Netherlands. So in short, the method to calculate sub-daily VWC is developed and validated on the 2019 data and then applied to the 2018 data to assess the effect of VWC on sub-daily backscatter variations. I think the paper will be easier to follow if this is stated clearly in section 3. I would even suggest to split data and methods in different sections for clarity. 

Figure 2: make the colors more intuitive, either by making the colors an indicator for drought stress (e.g. brown/red), or otherwise a colormap according to date to make it easier to interpret. 

Section 3.2.1: It is unclear on which days the destructive samples were taken. It can be seen in figure 2 (but here there are only 7 days, whereas section 3.2.1 states 14 days?), but I think this information should be mentioned already here. To me it led to some confusion at line 230 in combination with section 3.2.2 where it states that because of power issues when measuring sapflow only three days have all data needed to estimate and validate VWC: July 25, Aug. 23 and 28. A table with an overview of days with destructive samples and sapflow data yes/no would be informative.

Line 243: On July 25 all available data for the CDF-matching were used. Why? What is the difference with the other days?

Line 277: "A sharp backscatter increase after rainfall was observed in all polarizations". Yes this seem true for those rainfall events where soil moisture is also increasing strongly. The event on June 12th seems different, whehre CW increases significantly, but soil moisture shows a very small response. Here VV, HH and crosspol backscatter drop strongly, and then go back to the level before the event, or get slightly higher. Can you explain what is happening here? 

Figure 8: maybe only show those days you actually used for the fit?

Figure 9: for the fit you consider the VWC of June 5 and 6 not reliable enough. But for fig 9 a and d are aggregated over 9 days. This means that you did use june 5 and 6 for figure 9, is this correct?  

Line 285: delete "the"

Line 286: During the last four aggregated acquisitions... which are these?

Line 295 and onward: where do the values for changes in soil moisture, VWC and SCW come from for typical dry days? 

Also the multiple linear regression to assess the effect of moisture stores on backscatter is somewhat unclear. I think it might be more sophisticated to do this calculation with all units in mm? It needs an assumption on soil depth and penetration depth, but it should be possible. If not, I think the statements in line 295 and onward are confusing, since sensitivity and mentioning e.g. "three times more sensitive" is not really the right term here since they units are not the same. Maybe change to something like: Note that the coefficients from soil and vegetation water stores (Table 1) have non-homogeneous physical units. Nonetheless, it shows us that for a typical dry day during the campaign of 2018, e.g. such as June 9th, soil moisture reduced with 0.02 m3m-3 and that this translates to a -0.5, -0.8 and -0.8 dB change in VV, HH and cross-polarized backscatter. During the same day VWC changed with 0.5 kg m-2, which would translate to a change of 1.5, 1.1 and 1.2 dB. This shows us that typical diurnal variation in VWC leads to a three times higher change in VV-polarized backscatter than a typical diurnal change in soil moisture.  

Figure 10 is not discussed much in the text. It shortly states that the effect of SCW on backscatter is underestimated based on Fig. 10, but more explanation here would be good. 

I suggest to split section 5.1 in two at line 330. One deals with the development and validation of the method with in situ data. The second part is applying the method to a longer period and a different region. 

Line 310: what about August 23rd? 

Line 352: Also here, i think using "1.5 to 3 times more sensitive" is not the right wording.  

Line 366 and onward: make more clear in the text what the results are from your study. Now it is hard to discern if these results are from another study or yours. 

This paper provides an update on a previous analysis (Vermunt et al. 2020) of microwave radar data taken in Florida in 2018 over a corn field. In this previous paper, the authors have identified a diurnal cycle in backscatter which may be related to changes in vegetation water content (VWC). However, validating this hypothesis requires sub-daily measurements of VWC changes which are notoriously hard to obtain. The authors thus present a technique to reconstruct daily changes in VWC from a combination of sapflow measurements and weather-station based estimates of evapotranspiration. They evaluate this technique against a set of sub-daily destructive VWC samples taken in another location. The technique is then applied to the 2018 Florida data and used to demonstrate that sub-daily changes in backscatter are consistent with the reconstructed diurnal variability in VWC (in addition to surface canopy water and soil moisture).

Considering what the authors aim to achieve, the study set-up and the available measurements are not 100% ideal. The absence of more reliable ET data (i.e. from a flux tower) is a bit unfortunate, as is the fact that only few days have all types of measurements available. Contrary to what may be thought from the title, the proposed technique is not able to entirely reconstruct VWC variability, rather it can be used to extrapolate sub-daily VWC behavior from a single measurement (made daily, for example in the morning). Still I believe this to be a very useful attempt, especially if one focuses on sub-daily variability alone, and it may guide future similar research. There is certainly an interest in reconstructing sub-daily VWC from fewer of the time-consuming destructive samples.

I have a few comments below which I think need to be considered, followed by some more minor comments and suggestions.

Major comments ---

Figure 10. The presentation of this figure is a bit misleading. If I understood correctly, the regression only attempts to predict intra-day variability in backscatter (Eq. 3). The initial backscatter value for each day is not reconstructed, but taken from the measurement directly. This is why there is a perfect match between ‘observed’ and ‘calculated’ at the start of each day. This should be made much clearer so as to not give the impression that the substantial inter-day variability in Fig. 10 can be explained from the regression. In fact, the quality of the regression for intra-day variability remains to be demonstrated as the authors do not report it (neither do they report if the coefficients of the regression are statistically significant).

In view of this, it’s hard to tell if the regression is actually reliable, especially since much of the sub-daily variability in backscatter doesn’t seem to be well predicted in Figure 10 (but it’s hard to evaluate). Showing a scatter plot of the measured vs predicted sub-daily variations would be more informative in that respect.

One could also make it clear which points are the ones that are used as the “anchor points” at t_0, for instance by giving them a different symbol color or shape.

Also the data in Figure 9 d-e-f provides the opportunity to better illustrate the modeled diurnal impact on backscatter (and compare it against the data in panels a-c). The contributions of all variables are mixed up in Figure 10, so it’s difficult to learn much from that figure alone.

Section 3.2.3 is a bit difficult to read because the purpose or context of some new methods that are explained there only becomes apparent or fully understandable later in the paper. Maybe there is potential to reorganize this section a bit and potentially already illustrate the different approaches with a figure (Figure 4 provides some of that but too late for the reader). In general, the methods (when they document a new approach) seem a bit excised from the rest of the text. It wouldn’t hurt to give a bit more meat to it, for instance by providing a figure to explain the reconstruction method in 3.3. as well (for instance, Figure 4 does that well for CDF-matching).

Are there any downsides to CDF-matching? You force the T rates to follow the same distribution as the sap-flow rates. Is there any evidence that this is may or may not be true in papers comparing transpiration and sap flow measurements? I think it’s fine to test this method, but the implications and plausibility should be better discussed. For instance, there is a physical rationale for having a long-term balance between sap flow and T rates that justifies the 24-hour (or more) sum approach.

Minor comments ---

Title: because the proposed method still requires some daily VWC measurements as constraints. I wonder if “Extrapolating continuous vegetation water content …” would be more appropriate and a better description of the paper’s contribution. Alternatively, you could put the emphasis on sub-daily ("Reconstructing diurnal vegetation water content..."), which does not need daily VWC measurement as constraint if one focuses on anomalies.

L49: « unavoidable » suggest to replace with « acceptable »

L83: Was a bit hard to get on first read. Maybe modify the sentence into: “… lag between transpiration and upper sap flow, compared to the lag with basal sap flow, …”.

L145-155: It may be useful to provide an illustration of the time series (before and after correcting ET with these different processing options) as a supplementary figure. Right now, it is a bit difficult to visualize what is happening to the ET time series.

By the way, even if P-M ET was a perfect method and produced close to truth ET time series, you’d still need to separate the plant transpiration part from the soil evaporation part. My point is that the “correction” actually also serves to do that operation.

L153: I thought on first reading that CDF matching was done with the daily totals (not the sub-daily time steps). This may need to be mentioned here.

L155: It could be useful to give a final high level summary of what happens here. For instance: “information on the diurnal shape of ET is entirely derived from Penman-Monteith, but the ET daily totals are scaled so that T estimates that are consistent with sap flow over long periods of time”.

Equation 2: I think the notation is not appropriate (or at least it is very unclear to me). I think I understand what you did in the end, but the equation does not reflect it:

• Is “k=15 minutes” meaningful here? The lower position should indicate the starting point (i.e. k = t_0, or k = t_0+15 minutes), check for instance: http://www.columbia.edu/itc/sipa/math/summation.html
• In Fk and Tk, does k denote the start or the end of the 15 minute time period?
• Why multiply (Fk – Tk) by delta_t, if Fk and Tk are already expressed in per 15 minute rates? (I assume delta_t would equal 15 minutes, since t and t_0 are indicated to be expressed in minutes).

L188: Why these 10 days in particular?

192: “did not overlap”. I don’t understand what this means. Do you simply mean, if they are not equal to each other?

L200: So this expression allows for an investigation of the sub-daily dynamics and basically removes the potential inter-day differences (since all data is relative to t_0). Maybe this should be stated more explicitly

L219: It is unclear what is meant by “the linear estimate”. I guess this means the scaling to match the 24-hr totals. Maybe section 3.2.3 needs to be better structured. You could potentially make a quick list of the different methods which you are testing and comparing.

L227: “observed [on that day] from”

Figure 4. It is assumed that ET estimates need correction to maintain some balance between transpiration and sap flow, but what about biases in sap flow measurements for high rates of flow? Are they possible and how big could they be?

L242: “An exception to this rule was July 25, when all available data for the CDF-matching were used.” I don’t understand why this is an exception, which sample was used as a constrain there then?

Figure 5. In each time series, it would be useful to show with a different symbol the one sample VWC that was used as constrain.

Figure 5. This Figure shows well how the 24-hour method does not allow for a difference between the start and end-of-day VWC. Could be mentioned.

Figure 5. Unlike the other days, Aug 23 had a lot of dew, so it could be that the VWC measurements were biased up because of that (one can remove dew with paper towels only on the accessible parts of the plant). This would explain why the reconstruction has a hard time for that day.

L264: “see fig 4d”. It’s hard to understand how this relates to what is being said. This could be better explained.

Figure 9. It would be useful to show some +/- 1 std deviation error bars (or envelopes) around the averaged data.

L285 typo

L298: Is it 3 times more if the units are dB ? (and same later)

L301: I don’t understand why (where?) Fig. 10 would show that. Please indicate what you mean about Fig. 10 more clearly.

Table1: Was the significance of the coefficients tested? Please report if they are statistically significant, their confidence interval, and what is the overall performance of the regression.

L310: “of dew” => “that dew”

L317: “This is comparable to estimated dew evaporation in this period, which was 0.09 kg m−2”. Can you explain where this estimate comes from?

Does the temperature of the canopy water or of the soil water have any possible impact on backscatter and if yes, could it explain some of the diurnal variability?

L340-345: Yes I think most of the re-scaling approaches you presented here would still be potentially needed to get from measured ET to T.

L350-355: This is based on the fitted coefficients but it’s not clear if these are actually significant.

L357: I agree that it is a credible interpretation of Figure 9, however, I think it would be more convincing if a physical model of backscatter was there to demonstrate that both effects are indeed of similar magnitude and can cancel each other. But I guess this would also mean adding a whole new section to the paper..

The conclusion makes a good summary and some good points on why the research is relevant, good job! It would also be interesting to read the authors’ perspective on what type of future work would be needed to achieve better comparability between in-situ microwave data and eco-hydrological observations.

In particular, it seems that when it boils down to sub-daily variability only, the time lag between sap flow and the transpiration estimate will control most of the VWC cycle. If it’s really the case, the authors may provide some recommendations on the needed temporal resolution (already touched on L335, but could deserve more space).