Monitoring vegetation condition using microwave remote sensing: the standardized vegetation optical depth index (SVODI)

Moesinger, Leander; Zotta, Ruxandra-Maria; van der Schalie, Robin; Scanlon, Tracy; de Jeu, Richard; Dorigo, Wouter

doi:https://doi.org/10.5194/bg-19-5107-2022

Articles | Volume 19, issue 21

https://doi.org/10.5194/bg-19-5107-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Special issue:

Microwave remote sensing for improved understanding of vegetation–water...

https://doi.org/10.5194/bg-19-5107-2022

© Author(s) 2022. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 19, issue 21

Research article

|

07 Nov 2022

Research article |

| 07 Nov 2022

Monitoring vegetation condition using microwave remote sensing: the standardized vegetation optical depth index (SVODI)

Leander Moesinger, Ruxandra-Maria Zotta, Robin van der Schalie, Tracy Scanlon, Richard de Jeu, and Wouter Dorigo

Download

Final revised paper (published on 07 Nov 2022)
Preprint (discussion started on 21 Jan 2022)

Interactive discussion

Status: closed

RC1:
'Comment on bg-2021-360', Anonymous Referee #1, 08 Feb 2022
This work presents an effective integration of different microwave remote sensing VOD into a novel vegetation index. The index is estimated from the different sensors with different Spatio-temporal coverage fitting the multivariate distribution, merging in a singular Spatio-temporal data set. This new index, SVODI, is demonstrated to be highly sensitive to the vegetation water content conditions.

The presented study makes a clear explanation of the process to integrate the different microwave data and presents sufficient proof of the sensitivity to the vegetation water content of its index comparing it with the VHI, VCI, TCI, and root moisture at different ground levels.

With a short battery of questions, I suggest this manuscript be accepted subject to a minor revision.

A few questions have to be answered:

For the estimation, only night or descent orbits have been used. Using also ascendent orbits can increase the Spatio-temporal coverage but probably introduce lower quality data. Can you make a short comment about if introducing extra orbital data will increase/decrease de quality of the index?

As you pointed out in line 206, long-term VOD trends are related to biomass changes. To extract vegetation structural changes, the data have been linearly detrended. Since the data set covers a long period, can rapid changes in biomass introduce variability into the index not related to the vegetation water content? Is the detrend enough to decouple both contributions, the biomass, and the vegetation water content? Can this mask the index sensitivity in regions with no water growth limitations as for example the peninsula of India? Make an extended comment on this.

Question two leads to this one: SVODI appears to be sensitive to vegetation water content in arid regions where the vegetation growth is water-limited. The correlation analysis with SOI and DMI shows this clearly. Is SVODI also sensitive to vegetation water content during a drought? Can capture as for example the 2010 Russian drought?

To estimate SVODI you integrate microwave data from C-, X-, and Ku- bands from different sensors. Since the last decade, there are other microwave sensors that integrate the L-band as SMOS and SMAP. L-band is sensitive to upper layer soil moisture variability but also can be used to extract VOD measures. Have you tried to integrate this sensor? It will be great to have a short discussion in the text to clarify the decision of not taking it into account.

Comparing SVODI with root moisture at different layer levels shows a good representation of ground physical processes. Can these results be reproduced using soil moisture from observational data as SMOS of SMAP upper layer soil moisture?

Other comments concerning the manuscript presentation:

line 51: Missing space --> "low.Some"

Figure 2: In the figure description: put a "∼" between "X1,X2" and "N(0,1)"

Figure 10 and 11: It will be useful to have in the legend the name of climate time series (SOI and DMI)
Citation: https://doi.org/10.5194/bg-2021-360-RC1
- AC1: 'Reply on RC1', Leander Moesinger, 11 Apr 2022
  
  We thank you for taking the time to write this constructive review!
  Your comments are in italics, our answers are plain text.
  Comment R1.1 For the estimation, only night or descent orbits have been used. Using also ascendent orbits can increase the Spatio-temporal coverage but probably introduce lower quality data. Can you make a short comment about if introducing extra orbital data will increase/decrease de quality of the index?
  
  Answer to R1.1 One of the main assumptions of LPRM, the retrieval algorithm, is a thermal equilibrium between surface and vegetation. Due to solar heating during the day, there likely is no thermal equilibrium. While technically we do have daytime LPRM VOD data (made with the assumption of a thermal equilibrium), it is still very experimental and the error magnitude unknown. We considered including the daytime observations with a corresponding quality flag and leave it to the users whether to use those observations or not. However, data sharing experiences of VODCA and our soil moisture products have taught us that even with extensive documentation, data are often not used and interpreted correctly.
  Comment R1.2 As you pointed out in line 206, long-term VOD trends are related to biomass changes. To extract vegetation structural changes, the data have been linearly detrended. Since the data set covers a long period, can rapid changes in biomass introduce variability into the index not related to the vegetation water content? Is the detrend enough to decouple both contributions, the biomass, and the vegetation water content? Can this mask the index sensitivity in regions with no water growth limitations as for example the peninsula of India? Make an extended comment on this.
  
  Answer to R1.2 Yes, we mention that this might be a problem (line 208) and also that more powerful methods bear the high risk of removing actual vegetation condition fluctuations. But the description is indeed a bit short. Therefore, we will expand the discussion and mention that short-term biomass-fluctuations (e.g., harvest) would lead to an anomaly if they occur outside the (climatologically) expected time of year. However, the low resolution mitigates the effect of small-scale changes and therefore anomalies only become visible when they occur at large scales.
  Comment R1.3 Question two leads to this one: SVODI appears to be sensitive to vegetation water content in arid regions where the vegetation growth is water-limited. The correlation analysis with SOI and DMI shows this clearly. Is SVODI also sensitive to vegetation water content during a drought? Can capture as for example the 2010 Russian drought?
  
  Answer to R1.3 Yes, is it also sensitive to vegetation water content in response to drought. Fig R1 shows SVODI in August 2010, at the peak of the Russian wildfires. Reaching values of less than -2 in Western Russia, SVODI suggests that the vegetation was indeed in an exceptionally poor condition. While this might make for another interesting case study, we will not add it to the paper because, as Reviewer # 2 noted, our paper lacks a bit of focus which another case study would further decrease.
  Figure R1: Mean SVODI during August 2010
  Comment R1.4 To estimate SVODI you integrate microwave data from C-, X-, and Ku- bands from different sensors. Since the last decade, there are other microwave sensors that integrate the L-band as SMOS and SMAP. L-band is sensitive to upper layer soil moisture variability but also can be used to extract VOD measures. Have you tried to integrate this sensor? It will be great to have a short discussion in the text to clarify the decision of not taking it into account.
  
  Answer to R1.4 L-VOD exhibits completely different temporal characteristics than C-, X- and Ku-VOD. It is mostly susceptible to slow structural changes, which, by design, are not shown with SVODI. This susceptibility to slow changes is also one of the main reasons why most studies aggregate L-Band VOD to a very low temporal resolution (e.g yearly means) as daily L-VOD changes are mostly noise. We expanded our reasoning for not using L-VOD .
  Comment R1.5 Comparing SVODI with root moisture at different layer levels shows a good representation of ground physical processes. Can these results be reproduced using soil moisture from observational data as SMOS of SMAP upper layer soil moisture?
  
  Answer to R1.5 In theory, this would be possible but we regarded them as inappropriate to validate VOD products for two reasons: 1) SMOS and SMAP soil moisture is also based on microwave sensors, so they share similar errors as the VOD products in our study. 2) Such satellite products only provide surface soil moisture estimates while we wanted to differentiate the responses with respect to different rooting depths. For these two reasons, we chose to use ERA-5.
  Thank you for bringing the minor mistakes to our attention, we will fix them!
  
  Citation: https://doi.org/10.5194/bg-2021-360-AC1
RC2:
'Comment on bg-2021-360', Anonymous Referee #2, 14 Mar 2022
This paper describes the development of a vegetation index that combines VOD data from multiple sensors. The work is relevant to the journal. The evaluation strategy and the discussion of results could use some improvements.

Main comments:

The paper should justify the strategy of rescaling all products to AMSR-E. Why not use the newer sensor AMSR2 as the reference? This rescaling approach will smooth out the contribution of each band, which as noted in the manuscript has different sensitivity to different parts of the vegetation. The implication of potential loss of information should be discussed.

The paper compares data an older sensor (AVHRR) as analogs for optical data. I strongly suggest the use TCI and VCI from more modern sensor such as MODIS.

The patterns of improvements (Figure 7) is not consistent with prior studies, as claimed in the article. In Figure 4a of Moesinger et al. 2020, the correlation pattern is very different. For example, the correlations are strong in the eastern US and weak in the west. Similarly correlations are strong in vegetated areas like Amazon and Congo. Here, because SVODI is an anomaly product, the semi-arid areas stand out more?

Similarly, the patterns in Figure 8 backup the statement that correlations are strongest in places where vegetation growth is limited by water availability. For example, over the agricultural areas in North America degradations are seen (see Kumar et al. 2020 ; https://hess.copernicus.org/articles/24/3431/2020/). Is that because ERA5 doesn’t get the soil moisture patterns over agricultural areas, but SVODI do? You can also see similar features over Eastern China and Indus(?)

Some of the discussions around the Figures is pretty minimal and doesn’t go into any depth. For example, for Figure 11 – there is no discussion of the middle and the right columns. Why have them? Similarly, Section 4.2.5 and Figure 12 provide little added information to the paper. I encourage the authors to remove extraneous and distracting results and focus on tightening the key contributions of the paper.

Other specific comments:

Page 2, para 2: A more comprehensive review of the vegetation remote sensing can be found at: Houborg, R., Fisher, J. B., and Skidmore, A. K.: Advances

in remote sensing of vegetation function and

traits, Int. J. Appl. Earth Obs. Geoinformation, 43, 1–6,

https://doi.org/10.1016/j.jag.2015.06.001, 2015

Line 39: A standalone sentence as a paragraph?

Line 48: The wording needs to be more precise. By ‘normal’, I assume that the authors are talking about deviations in a normalized distribution? Here ‘normal conditions’ sounds like the long-term average described in the previous para.

Line 51. Add a space between ‘low.Some’

Line 64-65: The lit review needs to broader. There are lots of other work evaluating the use of VOD as an above ground biomass analog. Some are listed below:

Konings, A. G., Rao, K., and Steele-Dunne, S. C.: Macro

to micro: microwave remote sensing of plant water content

for physiology and ecology, New Phytol., 223, 1166–1172,

https://doi.org/10.1111/nph.15808, 2019.

Teubner, I. E., Forkel, M., Camps-Valls, G., Jung, M., Miralles,

G., Tramontana, G., van der Schalie, R., Vreugdenhil, M.,

Mâsinger, L., and Dorigo, W. A.: A carbon sink-driven approach

to estimate gross primary production from microwave

satellite observations, Remote Sens. Environ., 229, 100–113,

https://doi.org/10.1016/j.rse.2019.04.022, 2019.

Line 68-70: Have you established the issue of heteroscedasticity with VODCA data? It will be important to show an example and how the use of SVODI helps to reduce the noise to backup this claim.

Section 2.1.2: I suggest describing the sensors in chronological order, starting with the earliest (SSMI, TMI, WindSat, AMSR-E, AMSR2)

Figure 1: There are hardly any warm colors visible in this plot. I suggest revising the colorbar to be from 0 to 1, so that it has a better contrast.

Line 186: Revise to ‘… are drawn and p is calculated for them, …’

Line 218: What is the rationale for 16 days?

Line 240: Change to ‘latter is explained in more detail below.’

Section 3.2.1: What is the rationale for examining the temporal shifts? What are you expecting to find?

Section 3.2.2: Similar to the previous section, please explain what the objective of examining the extreme values is. A reader will not know what the ‘plots in Van Der Schrier et al (2013)’are.

Line 301: Is the positive trend statistically significant?

Line 310: Figure 7a shows SVODI and VCI correlation (and not VHI)

Line 311: This figure is not in line with previous studies as claimed here.

Line 365: needs a closing bracket.
Citation: https://doi.org/10.5194/bg-2021-360-RC2
- AC3: 'Reply on RC2', Leander Moesinger, 11 Apr 2022
  
  We thank you for taking the time to write this constructive review!
  
  Your comments are in italics, our answers are plain text.
  
  Comment R2.1 The paper should justify the strategy of rescaling all products to AMSR-E. Why not use the newer sensor AMSR2 as the reference? This rescaling approach will smooth out the contribution of each band, which as noted in the manuscript has different sensitivity to different parts of the vegetation. The implication of potential loss of information should be discussed.
  
  Answer to R2.1 AMSR-E is used due to its temporal overlap with most other sensors, allowing for a direct rescaling using concurrent observations. AMSR2, while being newer, only overlaps with TMI. Also note that at its base, the bias correction to AMSR-E is just a piece-wise linear scaling, therefore the dynamics of each sensor are not altered. We will expand the corresponding section to elaborate our choice.
  
  Comment R2.2 The paper compares data an older sensor (AVHRR) as analogs for optical data. I strongly suggest the use TCI and VCI from more modern sensor such as MODIS.
  
  Answer to R2.2 Generally the main advantages of MODIS to AVHRR are both a higher spectral and spatial resolution. For our application however neither the higher spectral resolution (not relevant for VCI calculation) nor the higher spatial resolution (AVHRR resolution is already much higher than our 0.25 degree grid) are of any benefit. MODIS and AVHRR NDVI correlate very strongly with each other [1, 2, 3], so the results would not get much more accurate. However, AVHRR has the practical benefit of being available for the whole duration of SVODI, while MODIS is only available since 2000, which means that AVHRR allows for a more robust analysis.
  
  [1] https://doi.org/10.5589/m06-001
  
  [2] https://doi.org/10.1016/j.rse.2005.08.014
  
  [3] https://doi.org/10.3390/rs5083918
  
  Comment R2.3 The patterns of improvements (Figure 7) is not consistent with prior studies, as claimed in the article. In Figure 4a of Moesinger et al. 2020, the correlation pattern is very different. For example, the correlations are strong in the eastern US and weak in the west. Similarly correlations are strong in vegetated areas like Amazon and Congo. Here, because SVODI is an anomaly product, the semi-arid areas stand out more?
  
  Answer to R2.3 Figure 4a in Moesinger et al. 2020 shows mean VOD-C and is thus not related to figure 7 of this paper. The equivalent figure in Moesinger et al. 2020 are figures 11b, 11d and 11f which show the correlation between VODCA anomalies and MODIS LAI anomalies and exhibit the same pattern, both globally and in sub regions such as the US. Also both papers agree that the strongest correlations are found in grasslandssemi arid areas.
  
  Comment R2.4 Similarly, the patterns in Figure 8 backup the statement that correlations are strongest in places where vegetation growth is limited by water availability. For example, over the agricultural areas in North America degradations are seen (see Kumar et al. 2020 ; https: // hess. copernicus. org/ articles/ 24/ 3431/ 2020/ ). Is that because ERA5 doesn’t get the soil moisture patterns over agricultural areas, but SVODI do? You can also see similar features over Eastern China and Indus(?)
  
  Answer to R2.4 We understand the issue as to why the correlations between ERA5 soil moisture anomalies and SVODI in the Eastern US, Eastern China and the Indus are negative. In this light, indeed, the correlations are generally highest in water-limited areas. The Eastern US and Eastern China however are mostly limited by radiance [4]. The Indus is generally water-limited [4] and shows near-zero correlation coefficients for surface soil moisture and positive correlations for deeper soil moisture. Therefore in all these regions the correlation coefficients are not unexpected with respect to the main climatic constraints of vegetation growth.
  
  [4] https://www.science.org/doi/10.1126/science.1082750
  
  Comment R2.5 Some of the discussions around the Figures is pretty minimal and doesn’t go into any depth. For example, for Figure 11 – there is no discussion of the middle and the right columns. Why have them? Similarly, Section 4.2.5 and Figure 12 provide little added information to the paper. I encourage the authors to remove extraneous and distracting results and focus on tightening the key contributions of the paper.
  
  Answer to R2.5 This is a very good point. Figure 12 was supposed to replace the center and right columns of figure 11, but we mistakenly left those in. We will remove the middle and right column of figure 11. As SOI/DMI are traditionally used in the analysis of global extreme events, there is merit to analyze their relationship to SVODI, whose purpose is also to monitor extreme events. Therefore we propose to keep figure 12 in. We will generally also flesh out some of the figure descriptions and their interpretation.
  
  Thank you a lot for the very thorough list of minor corrections and improvements, we will fix them!
  
  Citation: https://doi.org/10.5194/bg-2021-360-AC3

Peer review completion

AR: Author's response | RR: Referee report | ED: Editor decision | EF: Editorial file upload

ED: Reconsider after major revisions (21 Apr 2022) by Jean-Christophe Calvet

ED: Reconsider after major revisions (26 Apr 2022) by Trevor Keenan (Co-editor-in-chief)

AR by Leander Moesinger on behalf of the Authors (15 Jun 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (16 Jun 2022) by Jean-Christophe Calvet

RR by Nataniel Holtzman (31 Jul 2022)

ED: Publish subject to minor revisions (review by editor) (01 Aug 2022) by Jean-Christophe Calvet

ED: Publish subject to minor revisions (review by editor) (09 Aug 2022) by Trevor Keenan (Co-editor-in-chief)

AR by Leander Moesinger on behalf of the Authors (02 Sep 2022) Author's response Author's tracked changes Manuscript

ED: Publish as is (13 Sep 2022) by Jean-Christophe Calvet

ED: Publish as is (13 Sep 2022) by Trevor Keenan (Co-editor-in-chief)

AR by Leander Moesinger on behalf of the Authors (03 Oct 2022) Author's response Manuscript

Short summary

The standardized vegetation optical depth index (SVODI) can be used to monitor the vegetation condition, such as whether the vegetation is unusually dry or wet. SVODI has global coverage, spans the past 3 decades and is derived from multiple spaceborne passive microwave sensors of that period. SVODI is based on a new probabilistic merging method that allows the merging of normally distributed data even if the data are not gap-free.