Information-based uncertainty decomposition in dual-channel microwave remote sensing of soil moisture

Li, Bonan; Good, Stephen P.

doi:https://doi.org/10.5194/hess-25-5029-2021

Articles | Volume 25, issue 9

https://doi.org/10.5194/hess-25-5029-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/hess-25-5029-2021

© Author(s) 2021. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 25, issue 9

Research article

|

17 Sep 2021

Research article |

| 17 Sep 2021

Information-based uncertainty decomposition in dual-channel microwave remote sensing of soil moisture

Bonan Li and Stephen P. Good

Download

Final revised paper (published on 17 Sep 2021)
Supplement to the final revised paper
Preprint (discussion started on 28 Oct 2020)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'Comment on hess-2020-534 paper', Anonymous Referee #1, 01 Dec 2020
- AC1: 'Reply on RC1', LI BONAN, 21 Jan 2021
RC2: 'Comments on Information – based uncertainty decomposition in dual channel microwave remote sensing of soil moisture B', Anonymous Referee #2, 07 Dec 2020
- AC2: 'Reply on RC2', LI BONAN, 21 Jan 2021
RC3: 'Review of Li and Good information-based uncertainty decomposition paper', Anonymous Referee #3, 09 Dec 2020
- AC3: 'Reply on RC3', LI BONAN, 21 Jan 2021

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

ED: Reconsider after major revisions (further review by editor and referees) (03 Feb 2021) by Roger Moussa

AR by LI BONAN on behalf of the Authors (23 Feb 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (09 Mar 2021) by Roger Moussa

RR by Anonymous Referee #1 (14 Apr 2021)

Suggestions for revision or reasons for rejection

Compared to the previous version, the methodology is presented in a clearer way.
The fact that the results are treated by type of surface makes it possible to enrich the discussion. However, these groupings do not respond to the suggestion made, i.e. to treat all the surfaces together for the calculation of entropy in order to properly reflect the model's capacity to treat different surfaces. By calculating the parameters site by site and comparing the results to local statistics one can arrive at a wrong interpretation of the statistics. I find it counter-intuitive that the best results are obtained when the information is redundant rather than synergistic. One explanation for this is that the best correlation is obtained on surfaces where the vegetation cover is less important (with larger range of value and little vegegetation effects). This is also the situation where Tbh and TBv are the most correlated, hence the high R. To say that R can be a good indicator to evaluate a model is a bit hazardous in this case. A comment on this point should be made in the discussion.
Detailed comments
Comment est calculé teff (a mettre dans la section 2 plutôt que 4)
L241 coarse
L244 …more sensitive …. In situ soil moisture : I don’t undersdand
L244 you can consider the following papers dedicated to sampling depth
M.-J. Escorihuela, A. Chanzy, J.-P. Wigneron, et Y. H. Kerr, « On the effective soil moisture depth of L-band radiometry : a case study », Remote Sensing of Environment, vol. 114, p. 995 1001, 2010, doi: doi:10.1016/j.rse.2009.12.011.
S. Raju, A. Chanzy, J. P. Wigneron, J. C. Calvet, Y. Kerr, et L. Laguerre, « Soil moisture and temperature profile effects on microwave emission at low frequencies », Remote Sensing of Environment (NLD), vol. 54, no 2, p. 85 97, 1995.

L246 is it really an overestimate
L294 what you mean by high quality data set ?
L305 : This indicates ...: see my general comment. The DCA is designed to take profit of TBh and the difference between TBh and TBv. Physically this statement is wrong and probably requires deeper discussion
L320'321 How can you say that?
L333 why schrubland are less sensitive to water availability.
L374 Tbh and TBv are correlated but Tbv-Tbh give an orthogonal information linked to the roughness and the vegetation. This is the essence of DCA. Taking care of sweeping the physic like this on the basis of your statistic that might be not well posed.

For the originality of the approach, the quality of presentation this paper deserve publication but apparent conflict between the physic and the results need a bit more discussion and some final statement might a bit tempered.

Hide

RR by Anonymous Referee #2 (23 Apr 2021)

Suggestions for revision or reasons for rejection

The manuscript has improved significantly in readability, the methodology explanation is much clearer in the new version. The analysis per land cover classes is an interesting addition. Using the SMAP data set which is provided in a 36 km, closer to the instrumental resolution, is a good choice as well. A plot was provided in the authors answer to my previous comments comparing the results using the "36 km" or the "9 km" SMAP data sets, showing no significant differences. This is expected as providing a 50 km resolution data set in a 9 km grid or in a 36 km grid should not affect the results of this study.

I have a number of comments:

"Higher redundant information provided by Tbh and Tbv tends to be found in land covers with less woody components" . This is surprising, the effet of those woody components is to create a depolarisation making TbH and Tbv more similar. In this sense, I would expect that they are more redundant when the vegetation cover is denser. The manuscript will still improve if such affirmations are interpreted in relation with our physical understanding of the signal. My feeling is that the analysis in terms of "redundancy" is only showing that the for the easiest sites (those with more homogeneous land covers, topography, roughness, meteorological conditions,..., those for which remote sensing estimations agree the best with in situ measurement), the results are good using either TbH or Tbv. This does not imply that "redundancy" is the reason of the better results. Regarding the affirmation "The informational redundancy between these remotely sensed observation can be used as independent metric to assess the retrieval quality of the algorithms". I do not agree until the physical insight mentioned above is included.

Same for Figure 7. The higher correlations of DCA SM and in situ measurements would be found when the unique information of TBv is the lowest, when the unique information of TBh is the lowest and when the synergistic information is the lowest... could the physical mechanisms be explained ?

Regarding physical insight, Fig 4b could be interesting, but it is not commented in the text. The mutual information of TbH, TbV and Teff with respect to in situ do show a correlation with the entropy of the in situ measurements, in contrast to I(DCA, in situ). Could this be interpreted in terms of dynamics of the time series ? More dynamics, more entropy, more information content of TbH, TbV and Teff with respect to in situ ?

What Konings et al (2017) actually shows is that not because there are two measurements one can retrieve simultaneously two variables with good accuracy (SM and VOD). SMOS can retrieve both SM and VOD because there are tens of Tb measurements for different incidence angles. This could have been mentioned explicitly in the introduction, and this could justify using DCA instead of MDCA. However, I think that in addition to Tbh, Tbv and Teff, the DCA algorithm use NDVI as input, in contrast to MDCA. What would be the implications for this work of not taking NDVI into account?

Abstract Line 23-25: "Quality": please say explicitly which is the quality metric used. A few lines afterwards "Pearson correlations" are mentioned but again nothing is said of the variables used to compute that correlation.

Line 28: "... than FOR other land covers"
Line 31: "redundancy"

Line 135: It would be good to add the list of the stations used to ensure the reproductivility of the results

Line 146: please give explicitly the quality flags and/or thresholds used to filter the data
Line 147: "data points" here refer to time samples or in situ sites ?

Lines 270-276: Could you add an intuitive explanation on why the mutual information between model outputs and in situ observations cab never exceed the entropy of in situ observations ? In terms of Eq 7 this would mean that H_CN(DCA)-H_CN(DCA, insitu) is necessarily negative. But personally, from the information given in the manuscript I still do not understand why.

Line 354: "The sensing depth is more of imperfection L band sensor". I am afraid I do not agree. The representativeness of the two measurements (in situ or remote sensing) in depth is conceptually of the same characteristics of the spatial representativeness of both measurements. Both in depth or in space, in situ sensors or remote sensor simply measure different things.

Figure 7 x label of panel c: correct "inforamtion"

Hide

ED: Reconsider after major revisions (further review by editor and referees) (29 Apr 2021) by Roger Moussa

AR by LI BONAN on behalf of the Authors (09 Jun 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (09 Jun 2021) by Roger Moussa

RR by Anonymous Referee #2 (19 Jul 2021)

ED: Publish subject to minor revisions (review by editor) (10 Aug 2021) by Roger Moussa

AR by LI BONAN on behalf of the Authors (11 Aug 2021) Author's response Author's tracked changes Manuscript

ED: Publish as is (12 Aug 2021) by Roger Moussa

AR by LI BONAN on behalf of the Authors (17 Aug 2021) Manuscript

Short summary

We found that satellite retrieved soil moisture has large uncertainty, with uncertainty caused by the algorithm being closely related to the satellite soil moisture quality. The information provided by the two main inputs is mainly redundant. Such redundant components and synergy components provided by two main inputs to the satellite soil moisture are related to how the satellite algorithm performs. The satellite remote sensing algorithms may be improved by performing such analysis.