the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Apr 2026
Assessment of vegetation water dynamics by comparing microwave remote sensing signals from satellites and field-based GNSS reflectometry
Abstract. Monitoring plant water stress and biomass is limited by labor intensive measurements techniques. Observing plant water conditions more broadly is now enabled by microwave remote sensing. Specifically, satellite-based vegetation optical depth (VOD) provides daily observations of vegetation water volume at tens of kilometers. While satellite VOD has been used for many applications, VOD validations have rarely been carried out. A new method has enabled direct measurements of in-situ VOD, from Global Navigation Satellite Systems (GNSS). However, GNSS measurements have yet to be applied to more globally dominant grasslands and shrublands. Here, we explore how satellite-based VOD from SMAP and AMSR-2 compares with field-based microwave observations from 272 GNSS-based interferometric reflectometry (GNSS-IR) sites across the Western U.S as a part of the Plate Boundary Observatory (PBO) H20 network. These sensors measure a proxy for VOD at a scale of tens of meters, the normalized microwave reflectance index (NMRI). We find that satellite VOD generally positively correlates with GNSS NMRI with correlations between 0.2 to 0.6 across sites. These correlations increase to 0.3 to 0.7 when evaluating sites in regions with low spatial vegetation type heterogeneity, low tree cover, and large seasonal vegetation dynamics. The correlations are higher for X-band VOD, likely related to our finding that both X-band VOD and NMRI are both more sensitive to seasonal vegetation variations than C-band and L-band VOD products. These findings suggest that satellite VOD is capturing field-based GNSS signals in dryland ecosystems, and therefore that these sensors are a critical resource for validating satellite VOD at scale.
Competing interests: Andrew F. Feldman is currently an associate editor of Biogeosciences. The other authors declare that they have no conflict of interest.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.- Preprint
(4679 KB) - Metadata XML
-
Supplement
(3270 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-1759', Anonymous Referee #1, 11 May 2026
-
AC1: 'Reply on RC1', Andrew Feldman, 15 Jul 2026
Reviewer 1
Reviewer 1 Summary:
This study compares satellite-based vegetation optical depth (VOD) with field-based measurements of VOD using GNSS-IR techniques. There is a lot of information within this paper; different wavelengths and algorithms are used to calculate VOD. Overall, I think the manuscript is well-written, the presentation is clear, and the figures appropriate. This manuscript seems appropriate for Biogeosciences. This seems like a study where a lot of data are collected and analyzed and then one sees what comes out of it (this is not necessarily a bad starting point, and it is done well, but correlations of 0.2 or 0.3 are not great and some deeper understanding of why those particular sites have such low/poor correlations is needed). Not being an expert in satellite measurements, there are a lot of acronyms and terms which I was not readily familiar with. With that said, my comments (listed below) should probably be considered as comments from a "non-expert" so please take them (or leave them) as you like.Author Response:
We appreciate the thorough comments from the review. See our initial replies below.
Major Comments/Questions:
Reviewer 1 Comment 1:
1. There is a lot of emphasis on correlations. As a non-expert, I do not have a good feel for what these correlations actually look like (ie, I assume one can generate a x-y scatter plot between two different measurements). Using Fig. 1 as a specific example, it seems surprising how some sites have relationships with a strongly positive correlation and others have a negative one (ie, Fig. 1E). What does a scatter plot from strongly positive and negative correlations actually look like? Maybe I missed it, but is there something distinctive that leads to the negative correlations?Author Response:
This is a good point. We will add an example figure either in the main text or supplemental to show some example correlations, as well as the time series and scatterplots, at a few different sites.
The emphasis on correlation here is due to the satellite-based VOD metric and field-based GNSS NMRI metric not having a one-to-one correspondence. Therefore, comparisons of their means and standard deviations across other statistics are less meaningful. Rather than extend the analysis into interpretation of other detailed statistical comparison metrics, we’ve decided to comprehensively evaluate one covariation metric (correlation) along different measurement frequencies, data/satellite products, timescales, and land cover characteristics.
Reviewer 1 Comment 2:
- There is a lot of pre-processing that is described, but not shown. One that seems like it would be useful for the reader to see is the different seasonal time series for NMRI and VOD. As described in Sect. 2.3, this analysis is looking at differences from the seasonal cycle...this is fine, but it would be great to also see the actual seasonal cycle so one can see things like: how large the peak is relative to the rest of the year, how the timing change year-to-year,etc.
Author Response:
We will additionally provide a guiding example of the decomposition of the VOD and NMRI time series from its raw variations into the seasonal cycle and its higher frequency anomalies (raw time series minus seasonal cycle).
Reviewer 1 Comment 3:
- Is the linear fit in Figs. 3B and C really significant? It seems like there is a LOT of scatter in these plots.
Author Response:
Yes, the Pearson’s correlations came back statistically significant (p<0.05). We will additionally test the Spearman’s correlation which should be more robust to outliers or nonlinearity.
We do note that the correlation is expected to be weak by design of the study and experiment. The variable on the x-axis is effectively a standard deviation metric of land cover types, which is inherently noisy. This is then correlated with a comparison statistic between two noisy observations (field and satellite vegetation water metrics) on the y-axis. Therefore, we don’t expect a tight relationship between the two variables. In this case, we are only looking to evaluate relatively (somewhat qualitatively) whether the correlation does tend to reduce when there is large spatial heterogeneity of land cover within the satellite footprint, or if the site is not in the dominant land cover within the pixel. We will clarify this point both in the methods and in the results.
Reviewer 1 Comment 4:
- For capturing the wetting and drying associated with individual rainfall events---if the objective is to capture anything like "interception on the leaf surfaces", the a 1-day temporal resolution is going to be an issue. Precipitation intercepted by the vegetation will be evaporated within a day. So, it seems like the daily resolution of this analysis will miss any shorter-term (ie, hourly scale) precipitation/evaporation effects. I realize this is likely known by the authors and part of the discussion in Sect. 3.5.2 about the lack of a clear "pulse" signal in Fig. 7.
Author Response:
This is true. The more likely scenario of a general increase after rainfall is going to be more related to rehydration or dehydration at daily timescales. The interception impact on NMRI may be significant over the days of the rainfall event, especially if the event is longer lasting. The integrated average of the microwave signal should increase due to interception. However, this effect is challenging to separate from increased water storage due to rehydration. We also will note that the NMRI time series that we used removed days when rainfall occurred, so these effects are likely further minimized here.
This point will be clarified in the methods and we will de-emphasize the point about interception in line 544.
Minor Comments:
Reviewer 1 Comment 5:
* l.50, provide the specific name of the indices being referred to?Author Response:
We will clarify that we are referring to, for example, normalized difference vegetation index (NDVI) and solar induced fluorescence (SIF) in line 50.
Reviewer 1 Comment 6:
* l.96, remove "Nevertheless"
Author Response:
We will remove this word here in line 96.
Reviewer 1 Comment 7:
* l.167, Eq.1 does the "max" refer to a max over a certain time period? Or, something else?
Author Response:
We will clarify here that the maximum is the maximum MP1rms in the time series, or the average of the 5th largest values in the time series. This is a method to normalize the metric by the lowest vegetation conditions when there is higher reflection of the signal from the bare soil surface. We’ll in-text reference the Larson and Small 2014 paper which discusses this point. This will be added to line 167.
Larson, K. M. and Small, E. E.: Normalized microwave reflection index: A vegetation measurement derived from GPS networks, IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 7, 1501–1511, https://doi.org/10.1109/JSTARS.2014.2300116, 2014.
Reviewer 1 Comment 8:
* l.204-205, how are "short statured" and "dense short statured" vegetation distinguished from each other? What is the height cut-off to make the vegetation "short"? Is a grassland considered a "cropland"? More clear definitions would be helpful.
Author Response:
We agree these descriptions are necessary and useful here. We will add these points to line 205. Specifically, within the Global Land Analysis and Discovery (GLAD) product, the short statured vegetation in general has a height of <3m with forested areas >3m. The dense designation for the short vegetation is if it also has >75% vegetation cover. Coverage less than this at a height of <3m is designated as “short vegetation” (without the “dense” designation). This includes grasslands, but not cropland necessarily. Cropland land use is determined from phenology metrics as well as machine learning training on high resolution imagery. We will reference the Hansen et al. 2022 manuscript that provides this information.
Hansen, M. C., Potapov, P. V., Pickens, A. H., Tyukavina, A., Hernandez-Serna, A., Zalles, V., Turubanova, S., Kommareddy, I., Stehman, S. V., & Song, X.-P. (2022). Global land use extent and dispersion within natural land cover using Landsat data. Environmental Research Letters, 17(3), Article 034050
Reviewer 1 Comment 9:
* l.253, why are you referring back to Section 2.2?
Author Response:
Yes, good catch. This is a mistake. We will remove mention of Section 2.2 in line 253.
Reviewer 1 Comment 10:
* l.255, is a 16-day value useful?
Author Response:
In this case, we are only evaluating the seasonal cycle with the NDVI 16-day time series. That time series should be sufficient for evaluating features of seasonality, which is effectively something that can be evaluated with a monthly scale process. We will clarify this point in Line 255.
Reviewer 1 Comment 11:
* l.371, can examples of the situation where the GNSS site is in a non-forested area, but the satellite pixel has a forest be explicitly explored/shown?
Author Response:
Yes, we can show an example of a site that has a high/dominant forest cover within the 9km satellite pixel. What we can do is show a supplemental figure with the land cover only within the 9km pixel and the location of the GNSS site (which would be outside of the forest). Such a figure may be placed in the supplemental information section.
Reviewer 1 Comment 12:
* l.523, is the peak in soil moisture expected to be before the peak of VOD?
Author Response:
In general, yes, the soil moisture peaks before the seasonality of VOD in many instances. This has at least been reported in former publications like Tian et al. 2018. However, the exact causes of the peak of soil moisture before VOD are under discussion still with many hypotheses that this is related to water availability leading the timing of peak biomass seasonality. We will provide a brief clarification in line 523 that this behavior is expected.
Tian, F., Wigneron, J. P., Ciais, P., Chave, J., Ogée, J., Peñuelas, J., ... & Fensholt, R. (2018). Coupling of ecosystem-scale plant water storage and leaf phenology observed by satellite. Nature ecology & evolution, 2(9), 1428-1435.
Reviewer 1 Comment 13:
* l.531, what do you mean by "physical representation differences"?
Author Response:
We will clarify in this sentence in line 531 that we are referring to differences in what NMRI electromagnetically represents (attenuation and roughness of the surface) versus VOD (attenuation of the microwave signal through the vegetation).
Reviewer 1 Comment 14:
* how dramatic are the seasonal peaks? Example time series that show the actual annual cycle?
Author Response:
We will provide an example seasonal cycle difference at one site for reference likely as an addition to this figure 6 or in the supplemental information.
Citation: https://doi.org/10.5194/egusphere-2026-1759-AC1
-
AC1: 'Reply on RC1', Andrew Feldman, 15 Jul 2026
-
RC2: 'Comment on egusphere-2026-1759', Anonymous Referee #2, 23 Jun 2026
This manuscript by Feldman et al. systematically compares the Normalized Microwave Reflection Index (NMRI), derived from pseudorange multipath at geodetic GNSS stations of the PBO H2O network using an interferometric reflectometry approach (GNSS-IR), against multiple satellite Vegetation Optical Depth (VOD) products across a range of land cover types, retrieval algorithms, and microwave frequencies. NMRI has been proposed as a cost-effective and scalable means to validate satellite VOD. Particularly significant is the application of GNSS-IR to short vegetation, biomes for which VOD validation is difficult with other approaches such as the related method of GNSS-Transmissometry. Given that satellite VOD validation remains an open challenge, this study represents an important step toward addressing it and developing a cross-biome VOD validation framework.
The manuscript is well-written and clearly structured. The main findings, for example that X-band VOD shows stronger agreement with NMRI than L-band, and that correlations are substantially higher at sites with low spatial heterogeneity, are clearly presented. I consider this a valuable contribution to the remote sensing community and hope the comments below are helpful pointers for the authors, focusing mostly on the definition of NMRI, its relation to vegetation metrics, and its consistency with the literature.
Major Comments:
1. P4 LL114.: A small note that might help unfamiliar readers: Chen et al. (2016) extracts the SNR multipath amplitude, the oscillation amplitude from the SNR interference pattern. This retrieval is distinct from the MP1rms (pseudorange multipath) approach underlying NMRI in the sense that, while in both cases GNSS-IR setups were employed, different GNSS observables and their subsequent multipath were used for downstream retrievals (SNR vs. pseudorange). The term 'reflectometry GNSS' as used in this line is therefore ambiguous, as it encompasses at least these two methodologically distinct GNSS-IR retrieval approaches, as well as GNSS-Reflectometry (commonly abbreviated as GNSS-R), which is introduced in line 120. With GNSS-T, GNSS-IR (and GNSS-R), a brief disambiguation of these configurations early on could be very helpful for readers not already familiar with the literature. Also, Chen et al. (2016) uses a horizontally polarized antenna, which is non-standard in typical geodetic GNSS applications, where RHCP antennas are standard.
2. P4 LL116: The claim that NMRI 'isolates the vegetation signal from the soil signal' might be worth revisiting. To my understanding, Small et al. (2010), Larson & Small (2014) and Small et al. (2014) describe NMRI to appear to exploit an empirical sensitivity difference, where soil moisture affects MP1rms (and subsequently NMRI, which in a first approximation contains a topography correction) far less than vegetation does, rather than achieving a formal isolation of the two contributions. The same applies to P4 LL132.
3. P4 L135: Small et al. (2014) [https://doi.org/10.1109/JSTARS.2014.2320597] explicitly found no clear relationship between NMRI and vegetation height in natural grasslands in Montana, but Small et al. (2010) [https://doi.org/10.1029/2010GL042951] does for the agricultural ecosystems studied there. The citation may therefore be worth revisiting for the height claim specifically.
Related, P20 L506 & P28 L659 the authors further propose a potential sensitivity of NMRI to biomass, yet according to Small et al. (2014) only two of the studied grassland sites showed weak correlations between NMRI and biomass. Maybe I am unaware of a more recent study showing such relationships, but otherwise a clarification in the manuscript would be beneficial.
4. P17 LL458: Small et al. (2014) discussed the impact of ‘hot spots’ within the NMRI footprint on the retrieved signal, even though NMRI does try to correct for topography. Without detailed knowledge on the local topography, which is likely not available for all sites a./o. at sufficiently high resolution, the NMRI signal may not be assumed to be representative of the entire area surrounding the antenna. I could imagine such effects further impacting the comparability of NMRI and VOD.
5. P29 L684: The manuscript characterizes NMRI as ‘a measure of roughness of the surface due to vegetation cover.’ This conflicts with the other NMRI literature: Small et al. (2014) explicitly define it as 'a measure of vegetation water content', and Jones et al. (2014) describe it as 'sensitive to daily vegetation water content changes'.
Minor Comments:
6. P1 LL27: Maybe mention the full name of the method, GNSS-T?
7. P1 L31: H20 instead of H2O (same here: P5 L162, P15 L407, P28 L666) and as H20 here: P30 L721 and P31 L744
8. P4 LL113: The authors argue that GNSS-T cannot be used to evaluate non-forested biomes with shorter vegetation. Yet, Zribi et al. (2017) [https://doi.org/10.1155/2017/6941739] successfully deployed a GNSS-T setup in a sunflower field over one vegetation period, with the sunflowers growing up to 140cm in height. The transmissivity was calculated from the GPS L1 signal. It seems there might exist a range of vegetation that may be probed with both GNSS-T and GNSS-IR, delivering potentially invaluable new information for validating VOD products. The authors should adapt the statement accordingly, that both methods are complementary to a certain degree, rather than mutually exclusive. A similar statement is found in P29 L695.
9. P4 L137: Do the authors maybe have a hardware recommendation for the newer options? This could be interesting for potential new users, even though this is not the scope of this manuscript per se.
10. P5 L169: L1 is at 1.575GHz, it would be more appropriate to round to 1.6GHz than 1.5GHz therefore.
11. P6 LL184: How were the 9km products upscaled? The chosen strategy may impact the final product and thus could be interesting to the reader.
12. P6 L204: Forest LC class: To my knowledge, no in situ study has validated the MP1rms/NMRI retrieval at forested sites (Larson 2016 used SNR-based retrievals for snow depth analysis in a forest-adjacent meadow clearing, but this is a methodologically distinct case). Further, the literature advises against GNSS-IR deployment in forests due to direct-signal obstruction. I do believe it is genuinely interesting to retain the Forest LC analysis in the manuscript, but a brief note acknowledging that this is the first time NMRI has been tested at forest-classified sites would be an important addition.
13. P10 LL332: The manuscript describes NMRI as 'capturing vegetation attenuation of incoming signals from satellites' and cites Humphrey and Frankenberg (2023), a GNSS-T study. This definition of NMRI reads practically identical to the one of GNSS-T-retrieved VOD, even though both approaches are fundamentally different in geometry and retrieval. A small side note clarifying this could help readers unfamiliar with the different in situ GNSS methods to not conflate them. A similar situation exists in l. 544, where sensitivity of GNSS-IR NMRI to rainfall interception on vegetation is discussed, but a GNSS-T study (Schellenberg et al. (2024)) is cited.
14. P14 L387: The 30m LC product pixel size is potentially smaller than the GNSS-IR footprint [P14 L393 states ~100m scale]. Thus, GNSS-IR footprints may extend beyond the chosen LC pixel, even if a perfectly centered co-location is assumed. Did the authors verify for edge cases, where between neighboring LC pixels within the same GNSS-IR footprint a drastic change in LC type occurred, that may have impacted the NMRI?
15. P22 L541: Parentheses around the two citations
16. P28 L668 & P29 LL681: And GNSS-IR captures signal that passed the vegetation layer twice, if reflected back from the ground; unlike GNSS-T or satellite radiometers.
Citation: https://doi.org/10.5194/egusphere-2026-1759-RC2 -
AC2: 'Reply on RC2', Andrew Feldman, 15 Jul 2026
Reviewer 2Reviewer 2 Summary:
This manuscript by Feldman et al. systematically compares the Normalized Microwave Reflection Index (NMRI), derived from pseudorange multipath at geodetic GNSS stations of the PBO H2O network using an interferometric reflectometry approach (GNSS-IR), against multiple satellite Vegetation Optical Depth (VOD) products across a range of land cover types, retrieval algorithms, and microwave frequencies. NMRI has been proposed as a cost-effective and scalable means to validate satellite VOD. Particularly significant is the application of GNSS-IR to short vegetation, biomes for which VOD validation is difficult with other approaches such as the related method of GNSS-Transmissometry. Given that satellite VOD validation remains an open challenge, this study represents an important step toward addressing it and developing a cross-biome VOD validation framework.
The manuscript is well-written and clearly structured. The main findings, for example that X-band VOD shows stronger agreement with NMRI than L-band, and that correlations are substantially higher at sites with low spatial heterogeneity, are clearly presented. I consider this a valuable contribution to the remote sensing community and hope the comments below are helpful pointers for the authors, focusing mostly on the definition of NMRI, its relation to vegetation metrics, and its consistency with the literature.
Author Response:
We appreciate the many constructive suggestions by the reviewer. Please see our responses below.
Major Comments:
Reviewer 2 Comment 1:
- P4 LL114.: A small note that might help unfamiliar readers: Chen et al. (2016) extracts the SNR multipath amplitude, the oscillation amplitude from the SNR interference pattern. This retrieval is distinct from the MP1rms (pseudorange multipath) approach underlying NMRI in the sense that, while in both cases GNSS-IR setups were employed, different GNSS observables and their subsequent multipath were used for downstream retrievals (SNR vs. pseudorange). The term 'reflectometry GNSS' as used in this line is therefore ambiguous, as it encompasses at least these two methodologically distinct GNSS-IR retrieval approaches, as well as GNSS-Reflectometry (commonly abbreviated as GNSS-R), which is introduced in line 120. With GNSS-T, GNSS-IR (and GNSS-R), a brief disambiguation of these configurations early on could be very helpful for readers not already familiar with the literature. Also, Chen et al. (2016) uses a horizontally polarized antenna, which is non-standard in typical geodetic GNSS applications, where RHCP antennas are standard.
Author Response:
This is a fair point and we agree it is critical to point out the differences of the terminology early. We will rewrite lines 114-122 to clarify the difference between GNSS-IR and GNSS-R by:
Describing the difference between the SNR and pseudorange approach for GNSS-IR.
Introducing Chen et al. 2016 along with the SNR approach and Larson and Small (2014) with the pseudorange approach. We’ll make it clear here the focus of the PBO network is on the GNSS-IR pseudorange approach as well in the methods.
Discussing the use of GNSS-R mainly for current satellite applications.
Reviewer 2 Comment 2:
- P4 LL116: The claim that NMRI 'isolates the vegetation signal from the soil signal' might be worth revisiting. To my understanding, Small et al. (2010), Larson & Small (2014) and Small et al. (2014) describe NMRI to appear to exploit an empirical sensitivity difference, where soil moisture affects MP1rms (and subsequently NMRI, which in a first approximation contains a topography correction) far less than vegetation does, rather than achieving a formal isolation of the two contributions. The same applies to P4 LL132.
Author Response:
The statement in line 117 might simplify things a bit in using the term “isolate.” The nuance in the language used by the reviewer is more appropriate here and we agree it would be useful to establish this point early. We will revise this point in lines 117 and 132 that NMRI contains a normalization that attempts to correct for topography and soil moisture effects. Later in the methods, more detail will be added describing that MP1rms attempts to remove some impacts of topography under assumptions that the topography is relatively static in referencing the dry conditions. Additionally, we will describe how MP1rms is more sensitive to vegetation than soil moisture.
Reviewer 2 Comment 3:
- P4 L135: Small et al. (2014) [https://doi.org/10.1109/JSTARS.2014.2320597] explicitly found no clear relationship between NMRI and vegetation height in natural grasslands in Montana, but Small et al. (2010) [https://doi.org/10.1029/2010GL042951] does for the agricultural ecosystems studied there. The citation may therefore be worth revisiting for the height claim specifically.
Related, P20 L506 & P28 L659 the authors further propose a potential sensitivity of NMRI to biomass, yet according to Small et al. (2014) only two of the studied grassland sites showed weak correlations between NMRI and biomass. Maybe I am unaware of a more recent study showing such relationships, but otherwise a clarification in the manuscript would be beneficial.
Author Response:
This is correct. We will point out the nuance that while VWC and NMRI have a closer linear relationship, there is some disagreement on the relationship between vegetation height and NMRI between the Small et al. 2014 and Small et al. 2010 findings.
In terms of biomass, we’d like to clarify that VWC is inherently a function of dry biomass where VWC can be decomposed into relative water content and dry biomass (see Momen et al. 2017). This point will be made earlier in the introduction that while there have been challenges relating biomass to NMRI, that the relation to VWC does inherently include some sensitivity to biomass.
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. Journal of Geophysical Research: Biogeosciences, 122(11), 3031–3046. https://doi.org/10.1002/2017JG004145
Reviewer 2 Comment 4:
- P17 LL458: Small et al. (2014) discussed the impact of ‘hot spots’ within the NMRI footprint on the retrieved signal, even though NMRI does try to correct for topography. Without detailed knowledge on the local topography, which is likely not available for all sites a./o. at sufficiently high resolution, the NMRI signal may not be assumed to be representative of the entire area surrounding the antenna. I could imagine such effects further impacting the comparability of NMRI and VOD.
Author Response:
Yes, this is correct as an additional point of difference that we will point out in the revised manuscript. Around line 458, we will add the idea of hotspots and, in general, variations of the footprint area that contributes to the signal depending on the topography. For this revision, we will reference the Devine et al. 2026 manuscript that worked with GNSS-IR and investigated these different footprint impacts as well as the Jones et al. 2014 paper.
Devine, C. J., Scott, R. L., Biederman, J. A., Du, J., Moore, D. J. P., Feldman, A. F., Guo, J. S., Adams, D. K., and Smith, W. K.: Proximal measurements of microwave reflectance using GNSS-IR track semi-arid grassland vegetation dynamics during greening and browning phases, Agric. For. Meteorol., 383, 111136, https://doi.org/10.1016/j.agrformet.2026.111136, 2026.
Jones, M.O., Kimball, J.S., Small, E.E. et al. Comparing land surface phenology derived from satellite and GPS network microwave remote sensing. Int J Biometeorol 58, 1305–1315 (2014). https://doi.org/10.1007/s00484-013-0726-z
Reviewer 2 Comment 5:
- P29 L684: The manuscript characterizes NMRI as ‘a measure of roughness of the surface due to vegetation cover.’ This conflicts with the other NMRI literature: Small et al. (2014) explicitly define it as 'a measure of vegetation water content', and Jones et al. (2014) describe it as 'sensitive to daily vegetation water content changes'.
Author Response:
Yes, this is true. We will clarify this in Line 648 and elsewhere that the NMRI metric is primarily sensitive to the attenuation through the vegetation (and thus vegetation water content variations) and that the surface roughness points are normalized based on the max MP1rms normalization.
Minor Comments:
Reviewer 2 Comment 6:
- P1 LL27: Maybe mention the full name of the method, GNSS-T?
Author Response:
At this point in the abstract, it might not make sense to designate the GNSS as GNSS-T since we are mainly discussing GNSS broadly and then narrow into the GNSS-IR approach.
Reviewer 2 Comment 7:
- P1 L31: H20 instead of H2O (same here: P5 L162, P15 L407, P28 L666) and as H20 here: P30 L721 and P31 L744
Author Response:
Good catch on this mistake throughout. We will change all occurrences to H2O, which is the official acronym of the PBO H2O network.
Reviewer 2 Comment 8:
- P4 LL113: The authors argue that GNSS-T cannot be used to evaluate non-forested biomes with shorter vegetation. Yet, Zribi et al. (2017) [https://doi.org/10.1155/2017/6941739] successfully deployed a GNSS-T setup in a sunflower field over one vegetation period, with the sunflowers growing up to 140cm in height. The transmissivity was calculated from the GPS L1 signal. It seems there might exist a range of vegetation that may be probed with both GNSS-T and GNSS-IR, delivering potentially invaluable new information for validating VOD products. The authors should adapt the statement accordingly, that both methods are complementary to a certain degree, rather than mutually exclusive. A similar statement is found in P29 L695.
Author Response:
Yes, this is a great point. We did not necessarily intend to remove GNSS-T from consideration for shorter statured biomes. We will temper this statement here to be more specific that GNSS-T are often less capable to evaluate all shorter statured vegetation given the constraints of the receiver and satellite viewing angle. We will later add a point in the discussion that GNSS-T methods can be explored to determine at which vegetation height or geometry that GNSS-IR should be used, referencing Zribi et al. 2017 as a success point for some herbaceous vegetation cases.
We referred to Zribi et al. 2017 in a former version of the paper, but it looks like it did not remain in the final version. We will reference that paper here.
Reviewer 2 Comment 9:
- P4 L137: Do the authors maybe have a hardware recommendation for the newer options? This could be interesting for potential new users, even though this is not the scope of this manuscript per se.
Author Response:
Given some government-based authors on the paper, we want to refrain from recommending specific hardware to the community. We will evaluate if there are several options that can be mentioned here that are not a recommendation of a specific company.
Reviewer 2 Comment 10:
- P5 L169: L1 is at 1.575GHz, it would be more appropriate to round to 1.6GHz than 1.5GHz therefore.
Author Response:
Yes, good catch. We will update this number to be accurate at the 1.575GHz rather than rounding it in line 169.
Reviewer 2 Comment 11:
- P6 LL184: How were the 9km products upscaled? The chosen strategy may impact the final product and thus could be interesting to the reader.
Author Response:
Linear averaging of the 9km pixels within the 36km EASE2 grid pixels were used here. There are 16, 9km EASE2 grid pixels within the 36km EASE grid pixel. Therefore, these 16 pixels were averaged linearly. We will add this detail to line 184 as requested.
Reviewer 2 Comment 12:
- P6 L204: Forest LC class: To my knowledge, no in situ study has validated the MP1rms/NMRI retrieval at forested sites (Larson 2016 used SNR-based retrievals for snow depth analysis in a forest-adjacent meadow clearing, but this is a methodologically distinct case). Further, the literature advises against GNSS-IR deployment in forests due to direct-signal obstruction. I do believe it is genuinely interesting to retain the Forest LC analysis in the manuscript, but a brief note acknowledging that this is the first time NMRI has been tested at forest-classified sites would be an important addition.
Author Response:
We need to make a distinction here and throughout the manuscript that, for the forest cases, the GNSS-IR sites were not necessarily installed within forests. Rather, the sites were likely installed in a location that is in clearings nearby taller vegetation of >3m. We will add a sentence that makes the distinction that these forest-classified sites (1) do not necessarily have the GNSS-IR sensor within a forest, (2) that the vegetation might still be of relatively short stature (but still >3m) and may not be a traditionally recognized forest, (3) and the possibility that while there is dense vegetation, the pixel might have been incorrectly classified as forest (since land cover products do have associated error). Finally, we note that the forest classification in some cases is for the full satellite pixel; while the site is not in a forest, the full satellite pixel may be observing abundant forest cover. We will clarify these points in the methods around line 204 and again in the discussion.
Therefore, we do not necessarily think that this manuscript qualifies as an investigation of forested sites.
Reviewer 2 Comment 13:
- P10 LL332: The manuscript describes NMRI as 'capturing vegetation attenuation of incoming signals from satellites' and cites Humphrey and Frankenberg (2023), a GNSS-T study. This definition of NMRI reads practically identical to the one of GNSS-T-retrieved VOD, even though both approaches are fundamentally different in geometry and retrieval. A small side note clarifying this could help readers unfamiliar with the different in situ GNSS methods to not conflate them. A similar situation exists in l. 544, where sensitivity of GNSS-IR NMRI to rainfall interception on vegetation is discussed, but a GNSS-T study (Schellenberg et al. (2024)) is cited.
Author Response:
In the case of the line 544 Schellenberg et al. 2024 reference, we will indeed add a phrase distinguishing that such a finding was pointed out in that 2024 reference for GNSS-T and that further investigation is needed to translate such a finding to NMRI for GNSS-IR. We also will note that the PBO H2O NMRI time series removed days when rainfall occurred, so these effects are likely further minimized here. This point will be clarified in the methods.
In the case of the line 332 reference to Humphrey and Frankenberg (2023), this appears to be a mistake and an incorrect reference here for this sentence and we will reference Small and Larson (2014) instead.
We will also add the point about differences in GNSS-IR and GNSS-T in the discussion as addressed below in response to Reviewer 2 Comment 16.
Reviewer 2 Comment 14:
- P14 L387: The 30m LC product pixel size is potentially smaller than the GNSS-IR footprint [P14 L393 states ~100m scale]. Thus, GNSS-IR footprints may extend beyond the chosen LC pixel, even if a perfectly centered co-location is assumed. Did the authors verify for edge cases, where between neighboring LC pixels within the same GNSS-IR footprint a drastic change in LC type occurred, that may have impacted the NMRI?
Author Response:
This is an interesting question. One issue is the spatial scale can vary from 10s to 100s of meters depending on the topography and satellite viewing angle, among other factors. However, in this case, we will evaluate the land cover surrounding each site in the 3x3 grid (or 90m x 90m) grid around the site to evaluate if there may be influencing heterogeneity surrounding each site (a 5x5 or 150m x 150m grid might be too large of an area). Based on the findings, we may create a new metric which quantifies how much vegetation type variability there is locally around each site and if this has an impact on the findings.
Reviewer 2 Comment 15:
- P22 L541: Parentheses around the two citations
Author Response:
Good catch. We will correct these parentheses in line 541.
Reviewer 2 Comment 16:
- P28 L668 & P29 LL681: And GNSS-IR captures signal that passed the vegetation layer twice, if reflected back from the ground; unlike GNSS-T or satellite radiometers.
Author Response:
Yes, we will add that clarification in the discussion around line 668. Similar to our response to Reviewer 2, Comment 5, we will provide a more detailed distinction that the NMRI metric is primarily sensitive to the attenuation through the vegetation (and thus vegetation water content variations). It differs from GNSS-T based VOD in having potentially two passages through the canopy (incoming and after reflection) and, a more minor point, some topography influence if the maxMP1rms method normalization is insufficient for dynamic or highly undulated cases.
Citation: https://doi.org/10.5194/egusphere-2026-1759-AC2
-
AC2: 'Reply on RC2', Andrew Feldman, 15 Jul 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 539 | 186 | 25 | 750 | 34 | 20 | 19 |
- HTML: 539
- PDF: 186
- XML: 25
- Total: 750
- Supplement: 34
- BibTeX: 20
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Title: Assessment of vegetation water dynamics by comparing microwave
remote sensing signals from satellites and field-based GNSS
reflectometry
Authors: Feldman, A. F., et al.
Summary:
This study compares satellite-based vegetation optical depth (VOD)
with field-based measurements of VOD using GNSS-IR techniques. There
is a lot of information within this paper; different wavelengths and
algorithms are used to calculate VOD. Overall, I think the manuscript
is well-written, the presentation is clear, and the figures
appropriate. This manuscript seems appropriate for Biogeosciences.
This seems like a study where a lot of data are collected and analyzed
and then one sees what comes out of it (this is not necessarily a bad
starting point, and it is done well, but correlations of 0.2 or 0.3
are not great and some deeper understanding of why those particular
sites have such low/poor correlations is needed). Not being an expert
in satellite measurements, there are a lot of acronyms and terms which
I was not readily familiar with. With that said, my comments (listed
below) should probably be considered as comments from a "non-expert"
so please take them (or leave them) as you like.
Major Comments/Questions:
1. There is a lot of emphasis on correlations. As a non-expert, I do
not have a good feel for what these correlations actually look like
(ie, I assume one can generate a x-y scatter plot between two
different measurements). Using Fig. 1 as a specific example, it seems
surprising how some sites have relationships with a strongly positive
correlation and others have a negative one (ie, Fig. 1E). What does a
scatter plot from strongly positive and negative correlations actually
look like? Maybe I missed it, but is there something distinctive that
leads to the negative correlations?
2. There is a lot of pre-processing that is described, but not shown.
One that seems like it would be useful for the reader to see is the
different seasonal time series for NMRI and VOD. As described in
Sect. 2.3, this analysis is looking at differences from the seasonal
cycle...this is fine, but it would be great to also see the actual
seasonal cycle so one can see things like: how large the peak is
relative to the rest of the year, how the timing change year-to-year,
etc.
3. Is the linear fit in Figs. 3B and C really significant? It seems
like there is a LOT of scatter in these plots.
4. For capturing the wetting and drying associated with individual
rainfall events---if the objective is to capture anything like
"interception on the leaf surfaces", the a 1-day temporal resolution
is going to be an issue. Precipitation intercepted by the vegetation
will be evaporated within a day. So, it seems like the daily
resolution of this analysis will miss any shorter-term (ie, hourly
scale) precipitation/evaporation effects. I realize this is likely
known by the authors and part of the discussion in Sect. 3.5.2 about
the lack of a clear "pulse" signal in Fig. 7.
Minor Comments:
* l.50, provide the specific name of the indices being referred to?
* l.96, remove "Nevertheless"
* l.167, Eq.1 does the "max" refer to a max over a certain time
period? Or, something else?
* l.204-205, how are "short statured" and "dense short statured"
vegetation distinguished from each other? What is the height
cut-off to make the vegetation "short"? Is a grassland considered a
"cropland"? More clear definitions would be helpful.
* l.253, why are you referring back to Section 2.2?
* l.255, is a 16-day value useful?
* l.371, can examples of the situation where the GNSS site is in a
non-forested area, but the satellite pixel has a forest be
explicitly explored/shown?
* l.523, is the peak in soil moisture expected to be before the peak
of VOD?
* l.531, what do you mean by "physical representation differences"?
* how dramatic are the seasonal peaks? Example time series that show
the actual annual cycle?