On the use of the GRACE normal equation of inter-satellite tracking data for estimation of soil moisture and groundwater in Australia

Tangdamrongsub, Natthachet; Han, Shin-Chan; Decker, Mark; Yeo, In-Young; Kim, Hyungjun

doi:https://doi.org/10.5194/hess-22-1811-2018

Articles | Volume 22, issue 3

https://doi.org/10.5194/hess-22-1811-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/hess-22-1811-2018

© Author(s) 2018. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 22, issue 3

Research article

|

13 Mar 2018

Research article |

| 13 Mar 2018

On the use of the GRACE normal equation of inter-satellite tracking data for estimation of soil moisture and groundwater in Australia

Natthachet Tangdamrongsub, Shin-Chan Han, Mark Decker, In-Young Yeo, and Hyungjun Kim

Download

Final revised paper (published on 13 Mar 2018)
Preprint (discussion started on 06 Jun 2017)

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

RC1: 'Reviewer Comments', Anonymous Referee #1, 04 Jul 2017
- AC1: 'Responses to reviewer 1', Natthachet Tangdamrongsub, 31 Aug 2017
RC2: 'Referee Comments', Anonymous Referee #2, 11 Jul 2017
- AC2: 'Responses to reviewer 2', Natthachet Tangdamrongsub, 31 Aug 2017
RC3: 'Review of ''hess-2017-318 On the use of GRACE intersatellite tracking data for improved estimation of soil moisture and groundwater in Australia''', Anonymous Referee #3, 26 Jul 2017
- AC3: 'Responses to reviewer 3', Natthachet Tangdamrongsub, 31 Aug 2017

Peer-review completion

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

ED: Reconsider after major revisions (further review by Editor and Referees) (10 Sep 2017) by Louise Slater

AR by Natthachet Tangdamrongsub on behalf of the Authors (17 Oct 2017) Author's response Manuscript

ED: Referee Nomination & Report Request started (20 Oct 2017) by Louise Slater

RR by Anonymous Referee #2 (16 Nov 2017)

RR by Anonymous Referee #3 (08 Dec 2017)

Suggestions for revision or reasons for rejection

This study introduces an inversion technique to use GRACE L1B data to improve the estimation of soil moisture and groundwater within Australia. As I mentioned in the previous round, I believe this line of research is interesting, however, I am not convinced that the proposed technique is well descried (with the current formulation it is not possible to re-do the work), and also it is not well justified (validation does not prove that the new method works any better than other available techniques). I recommend a reject / major review decision for this contribution.
Major Concerns:
A number of my previous comments are not adequately addressed
• The Methodology section needs to be specified, please add appendices to clearly how the equations are built. I cannot figure out how the normal equation is formulated, whether it includes KBRR and any orbital information? L120–> Please describe how the matrix A is derived and what are the entries. Similarly L128-L130 are unclear.

• The accuracy of recovery has not been justified  a synthetic study should be added to show the framework recovers the introduced gravity signals with a sufficient accuracy, see e.g., https://academic.oup.com/gji/article/158/3/813/2062077

• I cannot accept the following argument: “The approach optimally combines the GRACE’s least-squares normal equations with CABLE to improve ΔSM and ΔGWS estimates.” Therefore, the accuracy of the estimated result is only compared with the model-only result (please see Fig. 6). The accuracy of our TWS estimate can reach < 2 cm, which is in line with the GRACE accuracy of ~ 2 cm globally (Wahr et al., 2006).

A large part of the introduction is used to state other methods (e.g., assimilation and decomposition-based inversion) are erroneous. If this is true, reliable evidences must be provided to indicate the presented method works better than other technique

• L16: (e.g., time-variable gravity fields, i.e., Level 2 data, and ...
• L21 remove ‘, not the post-processed Δ𝑇WS grid data.’
• L27: ....combination maximizing the strength of the model and observations while suppressing the weaknesses. The approach .....
This is not well justified, in other words, the authors do not show combining the final grace products and models is worse than the proposed joint-inversion.
• L33-34: ... estimates likely due to limitation of 33 GRACE’s temporal and spatial resolution... The way this has been validated in this study does not necessarily back this conclusion up.
• GRACE information rigorously and negate these limitations, this study uses the fundamental
• L48-L51, Also L55, 59,etc.: before the references add ‘e.g.,’ as these references are only examples of existing researches.
• L61: (Schumacher et al., 2016,2018; Tangdamrongsub et al., 2017 )
Schumacher, M, Kusche, J & Döll, P, 2016, A systematic impact assessment of GRACE error correlations on data assimilation in hydrological models. Journal of Geodesy, vol 90., pp. 537-559
Schumacher, M, Forootan, E, van Dijk, A, Schmied, HM, Crosbie, R, Kusche, J & Döll, P, 2018, Improving drought simulations within the Murray-Darling Basin by combined calibration/assimilation of GRACE data into the WaterGAP Global Hydrology Model. Remote Sensing of Environment, 204, pp212-228, doi:10.1016/j.rse.2017.10.029
• L80 ‘incorrect gravity information’ --> does it mean all previous researches are wrong?
• L81 ‘ began to employ’ --> have applied the L2's (Schumacher etl al., 2016, 2018; Khaki et al., 2017 a,b)
Khaki, M., Hoteit, I., Kuhn, M., Awange, J., Forootan, E., van Dijk, A., Schumacher, M., Pattiaratchi, C. (2017a), Assessing sequential data assimilation techniques for integrating GRACE data into a hydrological model. Advances in Water Resources, 107, pages 301-316, doi:10.1016/j.advwatres.2017.07.001
Khaki, M., Schumacher, M., Forootan, E., Kuhn, M., Awange, J., van Dijk, A. (2017b), Accounting for
spatial correlation errors in the assimilation of GRACE into hydrological models through localization.
Advances in Water Resources, 108, pages 99-112, doi:10.1016/j.advwatres.2017.07.024
• L83 ‘still affected by the post-processing filter’ --> This statement is very negative, and criticizes other processing strategies without providing a real measure. As far as I understand, in an assimilation formulation, incorporating the full co-variance matrix is already very important, useful and sufficient. Other post-processing steps have less impacts on the final assimilation results. See Schumacher et al 2018 and discussions for the Australian case study. If the authors suggest their approach is better than other formulations of signal separation, it should be validated and discussed.
• L86: It is not clear why this CABLE model has been selected
• After L92: (Dis-)Similarity of this work with previous studies that use inversion techniques might be addressed, see e.g., https://academic.oup.com/gji/article/158/3/813/2062077
• Please also report on the impact of choices, within the gravity inversion, on the final mass estimation see e.g., http://www.sciencedirect.com/science/article/pii/S0264370716301016?via%3Dihub
• After L98: You might add a discussion reflecting the fact GRACE data is sensitive to many processes. You select the signal over Australia to simplify the computation etc.
• L109: ‘from a model’ -->what does this mean? which model?
• Equation5 --> matrices should be shown, e.g., add a appendix
• L162: How is the contribution of the signal from outside removed or corrected?
• Equation11 --> Please discuss the condition of these matrices used in the inversion.
• Co-variance estimation of the solution and whether it is representative of all measurements errors and model errors should be added.
• 4.2 Model uncertainty --> From a hydrological point of view, model’s errors should contain those uncertainties related to parameters, forcing data and model structure. The current errors do not reflect all these three categories and even the assumptions, that are used to define the distribution of model parameters, are not that sophisticated. Therefore, the impact of over-/under-estimation of the model’s co-variance matrix on the final inversion results must be evaluated.

Hide

RR by Anonymous Referee #1 (11 Dec 2017)

Suggestions for revision or reasons for rejection

I read once more a version of the paper and also the rebuttal letter in response to my first review. However, I am not satisfied with the provided answers on two major points raised in my first review. I am therefore recommending to ask the authors for another revision in order to give them the chance to properly address those concerns.

--
R2: The computation of Delta TWS from ITSG L2 solutions (like other L2 solutions) is subject to the post-processing filters and often followed by the application of empirical scaling factors, obtained from the land surface models. Our study provides a more rigorous way of computing Delta TWS without going through such ad hoc procedures. GRGS and JPL mascon are internally or post-processed and it is not expected users to apply (subjective) post- processing. This is the primary reason we validate our results with GRGS and JPL mascon.
--

I disagree. The results of the new method needs to be compared exactly to conventionally post-processed gravity fields of the very same L2 data source in order to demonstrate the superiority of the proposed method. Conventional postprocessing at least implies restoring spectral deficits (degree 1, 2) and the removal of correlated errors (Kusche et al., 2009; or Swenson and Wahr, 2006). From my point of view, those are in no way just "ad hoc procedures".

--
R3: As the reviewer concerned, the GC approach is developed based on the least-square combination, which assumes the uncertainty following the normal distribution with zero mean and covariance C. The derivation and setting of model uncertainty under the given assumption (e.g., zero mean) and its limitation are clarified in Sect. 4.2. For clarity, we modify lines 298 – 301 of the revised manuscript as follows.
--

My original request was to provide evidence that the assumption of normal distribution is indeed justified. This was by no means attempted by the presented changes to the manuscript.

More detailed comments might be provided in case that the authors decide to submit a new version of the manuscript that reasonably adresses those two important points.

Hide

ED: Reconsider after major revisions (further review by editor and referees) (17 Dec 2017) by Louise Slater

AR by Natthachet Tangdamrongsub on behalf of the Authors (18 Jan 2018)

ED: Publish subject to minor revisions (review by editor) (31 Jan 2018) by Louise Slater

AR by Natthachet Tangdamrongsub on behalf of the Authors (01 Feb 2018) Author's response Manuscript

ED: Publish as is (02 Feb 2018) by Louise Slater

AR by Natthachet Tangdamrongsub on behalf of the Authors (04 Feb 2018)

Short summary

We present a new approach to improve the water storage estimate. Our approach combines GRACE's raw data (least-squares normal equation) with the results from the Community Atmosphere Land Exchange (CABLE) model. No post-processing filter is applied to GRACE data, and the full GRACE signal and error information are exploited. The approach is applied over 10 Australian river basins, and the evident improvement of the water storage estimate, particularly groundwater component, is clearly observed.