Technical note: Using long short-term memory models  to fill data gaps in hydrological monitoring networks

Ren, Huiying; Cromwell, Erol; Kravitz, Ben; Chen, Xingyuan

doi:https://doi.org/10.5194/hess-26-1727-2022

Articles | Volume 26, issue 7

https://doi.org/10.5194/hess-26-1727-2022

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

https://doi.org/10.5194/hess-26-1727-2022

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

Articles | Volume 26, issue 7

Technical note

|

05 Apr 2022

Technical note |

| 05 Apr 2022

Technical note: Using long short-term memory models to fill data gaps in hydrological monitoring networks

Huiying Ren, Erol Cromwell, Ben Kravitz, and Xingyuan Chen

Download

Final revised paper (published on 05 Apr 2022)
Supplement to the final revised paper
Preprint (discussion started on 16 May 2019)
Supplement to the preprint

Interactive discussion

Status: closed

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

- Printer-friendly version

- Supplement

SC1: 'Short comments', Haidong Pan, 21 Jun 2019
- AC1: 'Response to short comments', Xingyuan Chen, 04 Feb 2020
RC1: 'Interesting paper however there are some important questions', Gerald A Corzo P, 30 Aug 2019
- AC2: 'Response to Reviewer's comments', Xingyuan Chen, 04 Feb 2020
RC2: 'Review comments', Anonymous Referee #2, 04 Dec 2019
- AC3: 'Response to Reviewer's comments', Xingyuan Chen, 04 Feb 2020
EC1: 'Editor's Comment on hess-2019-196', Dimitri Solomatine, 09 Feb 2022

Peer-review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (15 Feb 2020) by Dimitri Solomatine

AR by Xingyuan Chen on behalf of the Authors (25 Jun 2020) Author's response

ED: Referee Nomination & Report Request started (13 Jul 2020) by Dimitri Solomatine

RR by Anonymous Referee #2 (14 Aug 2020)

RR by Anonymous Referee #1 (25 Oct 2020)

Suggestions for revision or reasons for rejection

Technical note: Using Deep Neural Network Models to Fill Spatio-Temporal Data Gaps in Hydrological Monitoring Networks

The updated paper is has improved, and I happy with the answers to the comments and most of the changes performed.

However, on the abstract mainly, I still suggest some minor updates to reflect clearly and better the work. In general, there are so many places where the author misleads readers with terms like dynamic data, dynamic parameter, dynamic high-frequency data, complex dynamic, highly challenging situations, the great advantages, multiple dominant modes and others. It is challenging to follow with all these overused terms.

The source of the dynamics that lead to the high variability in the groundwater levels is not known, and therefore, trying to fill in data is indeed a challenge for research. There is not needed to oversell the complexity, and instead of helping the paper, it makes it cumbersome to read and very conflictive for people in the area. Terminology should be correctly used in a scientific publication.

Minor comments are in the abstract

a. What are data set with high-frequency dynamics? I believe you want to highlight the dynamics of a phenomenon reflected in data? I am not sure to understand the high-frequency term here?
b. The sentence: "We selected a location at the U.S. Department of 5 Energy’s Hanford site to demonstrate and evaluate the new method" not sure which new method?
c. It is well known that wells data usually are spatially distributed (no two wells in the same place), so in many studies, the research address them as in a more precise way just saying temporal data from 42 wells.
d. The sentence: "that monitor the dynamic and heterogeneous hydrologic exchanges between ". I would suggest making it more clear. Same as in point "a" of this comments.
e. The sentence: "capturing nonlinear, dynamic patterns in wells that exhibit various dynamics signatures", it seems trying to oversell the work. Dynamics everywhere?
f. The sentence: "Although the ARIMA models yield better error statistics, they fail to capture abrupt changes or high-frequency (daily and subdaily) variations in system states that are typical characteristics of a complex dynamic system." I guess there was a mathematical way to verify this statement? Maybe would be nice to elaborate in this line if this was just visual, or what was used to assess the so-called high-frequency variations?

Hide

ED: Publish subject to revisions (further review by editor and referees) (04 Nov 2020) by Dimitri Solomatine

AR by Xingyuan Chen on behalf of the Authors (17 Apr 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (25 Apr 2021) by Dimitri Solomatine

RR by Anonymous Referee #2 (08 Jun 2021)

RR by Gerald A Corzo P (24 Jul 2021)

ED: Publish subject to revisions (further review by editor and referees) (28 Jul 2021) by Dimitri Solomatine

AR by Xingyuan Chen on behalf of the Authors (16 Sep 2021) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (04 Oct 2021) by Dimitri Solomatine

RR by Anonymous Referee #1 (31 Oct 2021)

RR by Anonymous Referee #3 (01 Dec 2021)

ED: Publish subject to revisions (further review by editor and referees) (10 Dec 2021) by Dimitri Solomatine

AR by Xingyuan Chen on behalf of the Authors (22 Jan 2022) Author's response Author's tracked changes Manuscript

ED: Referee Nomination & Report Request started (09 Feb 2022) by Dimitri Solomatine

ED: Publish as is (23 Feb 2022) by Dimitri Solomatine

AR by Xingyuan Chen on behalf of the Authors (05 Mar 2022) Manuscript

Short summary

We used a deep learning method called long short-term memory (LSTM) to fill gaps in data collected by hydrologic monitoring networks. LSTM accounted for correlations in space and time and nonlinear trends in data. Compared to a traditional regression-based time-series method, LSTM performed comparably when filling gaps in data with smooth patterns, while it better captured highly dynamic patterns in data. Capturing such dynamics is critical for understanding dynamic complex system behaviors.