Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what&rsquo;s the best way to leverage regionalised information?

Chidepudi, Sivarama Krishna Reddy; Massei, Nicolas; Jardani, Abderrahim; Dieppois, Bastien; Henriot, Abel; Fournier, Matthieu

doi:https://doi.org/10.5194/egusphere-2024-794

Preprints

https://doi.org/10.5194/egusphere-2024-794

Preprints

13 May 2024

| 13 May 2024

Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what’s the best way to leverage regionalised information?

Sivarama Krishna Reddy Chidepudi, Nicolas Massei, Abderrahim Jardani, Bastien Dieppois, Abel Henriot, and Matthieu Fournier

Abstract. In this study, we used deep learning models with recurrent structure neural networks to simulate large-scale groundwater level (GWL) fluctuations in northern France. We developed a multi-station collective training for GWL simulations, using both “dynamic” variables (i.e. climatic) and static aquifer characteristics. This large-scale approach offers the possibility of incorporating dynamic and static features to cover more reservoir heterogeneities in the study area. Further, we investigated the performance of relevant feature extraction techniques such as clustering and wavelet transform decomposition, intending to simplify network learning using regionalised information. Several modelling performance tests were conducted. Models specifically trained on different types of GWL, clustered based on the spectral properties of the data, performed significantly better than models trained on the whole dataset. Clustering-based modelling reduces complexity in the training data and targets relevant information more efficiently. Applying multi-station models without prior clustering can lead the models to learn the dominant station behavior preferentially, ignoring unique local variations. In this respect, wavelet pre-processing was found to partially compensate clustering, bringing out common temporal and spectral characteristics shared by all available time series even when these characteristics are “hidden” because of too small amplitude. When employed along with prior clustering, thanks to its capability of capturing essential features across all time scales (high and low), wavelet decomposition used as a pre-processing technique provided significant improvement in model performance, particularly for GWLs dominated by low-frequency variations. This study advances our understanding of GWL simulation using deep learning, highlighting the importance of different model training approaches, the potential of wavelet preprocessing, and the value of incorporating static attributes.

Received: 16 Mar 2024 – Discussion started: 13 May 2024

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 preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Preprint (PDF, 2818 KB)

Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Preprint (2818 KB)

Download & links

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Journal article(s) based on this preprint

18 Feb 2025

Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what is the best way to leverage regionalised information?

Sivarama Krishna Reddy Chidepudi, Nicolas Massei, Abderrahim Jardani, Bastien Dieppois, Abel Henriot, and Matthieu Fournier

Hydrol. Earth Syst. Sci., 29, 841–861, https://doi.org/10.5194/hess-29-841-2025,https://doi.org/10.5194/hess-29-841-2025, 2025

Short summary

Sivarama Krishna Reddy Chidepudi, Nicolas Massei, Abderrahim Jardani, Bastien Dieppois, Abel Henriot, and Matthieu Fournier

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-794', Anonymous Referee #1, 11 Jun 2024
Chidepudi et al. used deep learning approaches to simulate and predict groundwater level dynamics. Authors compared and discussed the performance of different approaches of different combinations, such as different DL models, different inputs (i.e., dynamic factors and static factors), wavelet decomposition of precipitation, one-hot encoding etc. Using deep learning approach to simulate and predict dynamic groundwater levels is challenging. This work is important and could be a good reference for the community. The paper is generally well organized but there are still a lot of details unclear. Major revision is needed for further review.
There are no clear introductions of model structures.

I didn’t find details of the model input or the structure of the input data. I especially wanted to know this in the multi-station approach

How did you choose the training and test sets?

I didn’t find how large your research area (only a figure). The resolution of ERA5 is low and the true variations of these hydrometeorological variables may not be accurately presented by the products

What do you think about the uncertainties of data products from ERA5

Did you only conduct the wavelet decomposition on precipitation or other variables also?

What is the resolution of the data products of static attributes?

What do you think the effects of hydraulic conductivity, elevation, slope etc. static attributes.

Location of the well, i.e., in confined or unconfined aquifers may also be important
Citation: https://doi.org/10.5194/egusphere-2024-794-RC1
- AC1: 'Reply on RC1', Sivarama Krishna Reddy Chidepudi, 20 Jun 2024
  
  Please find our point-by-point response to the comments in the attached documents.
  
  Citation: https://doi.org/10.5194/egusphere-2024-794-AC1
RC2:
'Comment on egusphere-2024-794', Anonymous Referee #2, 09 Jul 2024
Review comments on the manuscript: Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what’s the best way to leverage regionalised information? by Chidepudi et al.
The manuscript presents several different deep learning approaches to simulate groundwater levels. Dynamic as well as static variables are used to train deep learning models to represent fluctuations on a high temporal resolution (daily data) in northern France. These different deep learning models were combined with different sets of input data (including preprocessing) and training strategies. Overall, the work is timely and covers the important topic of data-driven approaches to simulate dynamic groundwater levels. However, the manuscript has several shortcomings which are listed below. Major revision is needed.
Main Comments
What is the best way to leverage regionalised information? - The authors raise this question in the manuscript title but in my opinion, they do not answer the question in a sufficient way. This has mainly two reasons:
The manuscript seems to be a combination of a technical note and a case study which leads to the result that a lot of essential information are missing. Reviewer 1 already pointed out several of the technical issues. In addition, a description of the data set is entirely missing. The only information available for the reader is the rough distance between the observation wells and the density in the region. Important information to understand the results and therefore the feasibility of the applied methodology is not supplied by the authors. For example: What is the distribution of static attributes in the different cluster groups? Looking at the attributes presented in Table 1, large differences between lithologies are to be expected (e.g karst vs. clay). Could it be that the annual group consist mainly of observation wells located in karstic/fractured areas and what would this tell us about the outcome of the study? Are these static attributes even presented/discussed in Chapter 4 (I assume that you can see them in Fig. 9 but they are not even named somewhere?

The presentation and discussion of the results lacks the already mentioned discussion of the regional context but also a discussion of the results in a broader context. For example, the authors write L398:“However, wavelet pre-processing shifts the importance towards dynamic components, reducing the contributions of static features or OHE. When clustering is combined with wavelet preprocessing, low-frequency precipitation components emerge as key contributors, improving model performance.

Does this mean that the importance of all dynamic components is higher by default, and we do not need to consider geological/hydrodynamic/topographic features? Does this apply to all kind of unconfined aquifer systems (shallow, deep, karstic…)? Here it would be interesting to combine/compare your results with/to other available publications considering static attributes on a regional scale (e.g. Heudorfer et al., 2024 or Haaf et al., 2023).

The quality in writing (language, clarity etc.) differs a lot throughout the manuscript. This makes it difficult to follow the central theme and therefore requires revision. Sometimes sentences reoccur, e.g. L73: DL models have proved effective on a local scale, and are also on a larger scale by collectively training a significant number of piezometers (Chidepudi et al., 2023b; Heudorfer et al., 2024) vs. L80: The DL models have proved effective at local scale and are also proving more effective on a larger scale. At the same time the introduction of terms and abbreviation is totally off, some examples: GWLs is first introduces in the Introduction and then again in line 185, 308, 378 and 436; SHAP is first introduced in line 231 and then again in 461; an introduction (even though they are quiet common) for AI/DL/KGE and NSE is entirely missing. Altogether it feels like sections/paragraphs of different origin were put together.

Secondary Comments
L85: sensitivity to human activities - I do not really understand why this is an additional challenge compared to runoff data. Does it mean runoff data are not sensitive to human activities (e.g. river straightening, dam construction etc.)?
L121: their application to GWL simulation is still questionable. – Do you really mean questionable?
L141: We refer to (Beven and Young, 2013), for differences in the use of the terms simulation and forecasting. - I do not see the connection between the sentence and the rest of the paragraph. Maybe a few more words are needed?
L164: Although they seem somehow redundant, they are expected to provide complimentary information about the hydrogeological nature of the hydrosystems – This could and should be tested at one point (which does not mean that you have to add it here).
L167/ L173/180/323: Baulon et al., 2022a/b?
L187: Bidirectional LSTM - I would be good to provide a reference especially since you write in L192: BiLSTM […] are particularly good at identifying various patterns in data sequences, making them ideal for simulating GWLs, that change over time. or is this already a result of your study?
L304: Further explanation needed. The figure does not provide any details, especially no comparison, as written by the authors.
L355: This is an information you expect earlier in the manuscript.
L372: Why do you formulate “new research questions” here, is this necessary?
L425: No_ohe_no_stat approach?
References: Nourani, V., Alami, M. T., & Vousoughi, F. D. (2015). - I do not find a citation of this paper.
References:
Heudorfer, B., Liesch, T., & Broda, S. (2024). On the challenges of global entity-aware deep learning models for groundwater level prediction. Hydrol. Earth Syst. Sci, 28, 525–543. https://doi.org/10.5194/hess-28-525-2024

Haaf, E., Giese, M., Reimann, T., & Barthel, R. (2023). Data-driven estimation of groundwater level time-series at unmonitored sites using comparative regional analysis. Water Resources Research, 59, e2022WR033470. https://doi.org/10.1029/2022WR033470
Citation: https://doi.org/10.5194/egusphere-2024-794-RC2
- AC2: 'Reply on RC2', Sivarama Krishna Reddy Chidepudi, 25 Jul 2024
  
  Please find our response to the comments in the attached file.
  Thanks
  Sivarama Krishna Reddy Chidepudi
  
  Citation: https://doi.org/10.5194/egusphere-2024-794-AC2

Interactive discussion

Status: closed

RC1:
'Comment on egusphere-2024-794', Anonymous Referee #1, 11 Jun 2024
Chidepudi et al. used deep learning approaches to simulate and predict groundwater level dynamics. Authors compared and discussed the performance of different approaches of different combinations, such as different DL models, different inputs (i.e., dynamic factors and static factors), wavelet decomposition of precipitation, one-hot encoding etc. Using deep learning approach to simulate and predict dynamic groundwater levels is challenging. This work is important and could be a good reference for the community. The paper is generally well organized but there are still a lot of details unclear. Major revision is needed for further review.
There are no clear introductions of model structures.

I didn’t find details of the model input or the structure of the input data. I especially wanted to know this in the multi-station approach

How did you choose the training and test sets?

I didn’t find how large your research area (only a figure). The resolution of ERA5 is low and the true variations of these hydrometeorological variables may not be accurately presented by the products

What do you think about the uncertainties of data products from ERA5

Did you only conduct the wavelet decomposition on precipitation or other variables also?

What is the resolution of the data products of static attributes?

What do you think the effects of hydraulic conductivity, elevation, slope etc. static attributes.

Location of the well, i.e., in confined or unconfined aquifers may also be important
Citation: https://doi.org/10.5194/egusphere-2024-794-RC1
- AC1: 'Reply on RC1', Sivarama Krishna Reddy Chidepudi, 20 Jun 2024
  
  Please find our point-by-point response to the comments in the attached documents.
  
  Citation: https://doi.org/10.5194/egusphere-2024-794-AC1
RC2:
'Comment on egusphere-2024-794', Anonymous Referee #2, 09 Jul 2024
Review comments on the manuscript: Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what’s the best way to leverage regionalised information? by Chidepudi et al.
The manuscript presents several different deep learning approaches to simulate groundwater levels. Dynamic as well as static variables are used to train deep learning models to represent fluctuations on a high temporal resolution (daily data) in northern France. These different deep learning models were combined with different sets of input data (including preprocessing) and training strategies. Overall, the work is timely and covers the important topic of data-driven approaches to simulate dynamic groundwater levels. However, the manuscript has several shortcomings which are listed below. Major revision is needed.
Main Comments
What is the best way to leverage regionalised information? - The authors raise this question in the manuscript title but in my opinion, they do not answer the question in a sufficient way. This has mainly two reasons:
The manuscript seems to be a combination of a technical note and a case study which leads to the result that a lot of essential information are missing. Reviewer 1 already pointed out several of the technical issues. In addition, a description of the data set is entirely missing. The only information available for the reader is the rough distance between the observation wells and the density in the region. Important information to understand the results and therefore the feasibility of the applied methodology is not supplied by the authors. For example: What is the distribution of static attributes in the different cluster groups? Looking at the attributes presented in Table 1, large differences between lithologies are to be expected (e.g karst vs. clay). Could it be that the annual group consist mainly of observation wells located in karstic/fractured areas and what would this tell us about the outcome of the study? Are these static attributes even presented/discussed in Chapter 4 (I assume that you can see them in Fig. 9 but they are not even named somewhere?

The presentation and discussion of the results lacks the already mentioned discussion of the regional context but also a discussion of the results in a broader context. For example, the authors write L398:“However, wavelet pre-processing shifts the importance towards dynamic components, reducing the contributions of static features or OHE. When clustering is combined with wavelet preprocessing, low-frequency precipitation components emerge as key contributors, improving model performance.

Does this mean that the importance of all dynamic components is higher by default, and we do not need to consider geological/hydrodynamic/topographic features? Does this apply to all kind of unconfined aquifer systems (shallow, deep, karstic…)? Here it would be interesting to combine/compare your results with/to other available publications considering static attributes on a regional scale (e.g. Heudorfer et al., 2024 or Haaf et al., 2023).

The quality in writing (language, clarity etc.) differs a lot throughout the manuscript. This makes it difficult to follow the central theme and therefore requires revision. Sometimes sentences reoccur, e.g. L73: DL models have proved effective on a local scale, and are also on a larger scale by collectively training a significant number of piezometers (Chidepudi et al., 2023b; Heudorfer et al., 2024) vs. L80: The DL models have proved effective at local scale and are also proving more effective on a larger scale. At the same time the introduction of terms and abbreviation is totally off, some examples: GWLs is first introduces in the Introduction and then again in line 185, 308, 378 and 436; SHAP is first introduced in line 231 and then again in 461; an introduction (even though they are quiet common) for AI/DL/KGE and NSE is entirely missing. Altogether it feels like sections/paragraphs of different origin were put together.

Secondary Comments
L85: sensitivity to human activities - I do not really understand why this is an additional challenge compared to runoff data. Does it mean runoff data are not sensitive to human activities (e.g. river straightening, dam construction etc.)?
L121: their application to GWL simulation is still questionable. – Do you really mean questionable?
L141: We refer to (Beven and Young, 2013), for differences in the use of the terms simulation and forecasting. - I do not see the connection between the sentence and the rest of the paragraph. Maybe a few more words are needed?
L164: Although they seem somehow redundant, they are expected to provide complimentary information about the hydrogeological nature of the hydrosystems – This could and should be tested at one point (which does not mean that you have to add it here).
L167/ L173/180/323: Baulon et al., 2022a/b?
L187: Bidirectional LSTM - I would be good to provide a reference especially since you write in L192: BiLSTM […] are particularly good at identifying various patterns in data sequences, making them ideal for simulating GWLs, that change over time. or is this already a result of your study?
L304: Further explanation needed. The figure does not provide any details, especially no comparison, as written by the authors.
L355: This is an information you expect earlier in the manuscript.
L372: Why do you formulate “new research questions” here, is this necessary?
L425: No_ohe_no_stat approach?
References: Nourani, V., Alami, M. T., & Vousoughi, F. D. (2015). - I do not find a citation of this paper.
References:
Heudorfer, B., Liesch, T., & Broda, S. (2024). On the challenges of global entity-aware deep learning models for groundwater level prediction. Hydrol. Earth Syst. Sci, 28, 525–543. https://doi.org/10.5194/hess-28-525-2024

Haaf, E., Giese, M., Reimann, T., & Barthel, R. (2023). Data-driven estimation of groundwater level time-series at unmonitored sites using comparative regional analysis. Water Resources Research, 59, e2022WR033470. https://doi.org/10.1029/2022WR033470
Citation: https://doi.org/10.5194/egusphere-2024-794-RC2
- AC2: 'Reply on RC2', Sivarama Krishna Reddy Chidepudi, 25 Jul 2024
  
  Please find our response to the comments in the attached file.
  Thanks
  Sivarama Krishna Reddy Chidepudi
  
  Citation: https://doi.org/10.5194/egusphere-2024-794-AC2

Peer review completion

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

ED: Reconsider after major revisions (further review by editor and referees) (02 Aug 2024) by Monica Riva

AR by Sivarama Krishna Reddy Chidepudi on behalf of the Authors (15 Aug 2024) Author's response Author's tracked changes Manuscript

ED: Publish subject to minor revisions (further review by editor) (28 Oct 2024) by Monica Riva

The authors have considered the main critical issues raised by the reviewers during the revision process. For this reason, while in my previous assessment I was inclined in requiring a further revision process, after a thorough analysis of the manuscript I decided not to go for a second round of review.

This been said, upon a careful review of the manuscript, I must highlight that it still contains a significant number of typos and awkwardly constructed sentences that require the attention of the authors. While I understand the time constraints, especially with one of the authors being a Ph.D. student nearing graduation, it is crucial to uphold the quality of the manuscript for the benefit of both the journal and yourselves for the purpose of effectively communicating your research. Below, I have highlighted few (non-exhaustive) examples of suggested changes to be made.

Line 25 “ …to learn the dominant station …. preferentially…”, I suggest to rephrase the sentence as “…to preferentially learn…”

Line 37 “…for grasping a more global view of water reserves..” Here the adjective “more” is not appropriate. I suggest rewriting as “for grasping a global view of water reserves”

Line 43 “Building the large -scale model…...” should be “Building a large -scale model…”

Line 46 “However, the numerical, physics-based representation of all the physical processes occurring during the hydrological cycle …..” The sentence is awkward; consider rewriting it, for example as “However, the numerical, physics-based, representation of all processes occurring during the hydrological cycle ….

Lines 160-161 and lines 168. The repetition of the sentence “All the wells considered in the study are in unconfined aquifers” should be rectified.

Line 188 “Gualtieri (2022) highlighted that ERA5 uncertainties were greater in mountainous and particularly in coastal locations ….”. Consider rewriting this sentence, for example as: “Gualtieri (2022) highlighted that ERA5 uncertainties are greatest at mountain and coastal locations …..

Line 193 “However, for our study area, we have been evaluating different potential alternative reanalysis products, such as….”. Consider rewriting this sentence, for example as “For our study area, we evaluated several potential alternative reanalysis products, such as …..”

Line 205 “BDLISA was originally designed at a 25km scale and later upscaled to larger scales. For our study, we kept information coming from BDLISA at its original scale (25km), which means aquifer static attributes have a resolution of 25km”.
Since you are using the original scale (without upscaling), I would remove the sentence “For our study…”, to avoid confusing the reader.

Line 235 “… we have to make sure” replace by “…we need to ensure…”

Page 9. Remove the Figure without number and caption, it is already included in Fig. 1.

Line 238 “Also, for some attributes like hydraulic conductivity, it might not be straightforward to get the most relevant resolution, which is needed to account for the most appropriate characteristic describing the well…”. Please consider rephrasing this sentence, did you mean “describing the aquifer”?

Line 272 “Details of range of hyperparameters used are shown in Table 1”. The correct reference should be Table 2

Line 379 “….Furthermore, to facilitate hyperparameter tuning, the last 20% of the training data was used as a validation set” and line 385 “This split corresponds to approximately 80% of 385 the data for training and 20% for testing.” These two sentences convey the same information. Please rewrite.

Line 445 – 446.Clarify the basis for stating “While single station models perform best” as it is not evident from Fig. 8-9

Line 470. “In particular…”.Consider rephrasing this sentence.

Lines 552-560. The following sentence is unclear “In the framework of our study, we decided to exclude some relevant characteristics such as vadose or saturated zone thickness: even when averaged over quite long periods (several years), these values actually represent GWL (the target variable)….. These authors concluded that the models did not show any entity awareness and eventually utilized static attributes as simple identifiers (almost similar to the OHE approach presented herein), meaning that the models did not make use of the relevant (hydro)geological information.” Please rewrite.

Lines 542-545 “… for alluvial aquifers with probably quite high hydraulic conductivity overall” Clarify the meaning of “probably” here.

Conclusions, line 595-600 “In this article, we introduced the following question: “What’s the best way to leverage regionalised information?”. In light of our results, it then seems like this is highly dependent on the amount and types of static attributes. It is generally expected that a much higher number of static attribute types would allow for a much better improvement of the multi-station simulation approach. However, Heudorfer et al. (2024) found no improvements using around 28 static features (including 18 environmental and ten time series-based). Also, as pointed out by these authors, employing static attributes for model training might be more relevant in applications on larger scales and/or larger datasets”.
Please consider rewriting these sentences. As written, it seems that your results are in contrast with the findings of Heudorfer et al. (2024). On the other hand, my understanding is that static attributes can be used for model training on large scales, while they are not particularly useful on small-moderate scales (as such investigated by Heudorfer et al., 2024). Please also include the magnitude levels for both “large” and “small” (or moderate scales.

Table 4. Fix the typo (“Lattitude”)

Additionally, I recommend enhancing the quality of Figures 1-2,8-13, several labels are blurred or not visible.

Hide

AR by Sivarama Krishna Reddy Chidepudi on behalf of the Authors (29 Nov 2024) Author's response Author's tracked changes Manuscript

ED: Publish as is (13 Dec 2024) by Monica Riva

AR by Sivarama Krishna Reddy Chidepudi on behalf of the Authors (24 Dec 2024) Manuscript

Journal article(s) based on this preprint

18 Feb 2025

Training deep learning models with a multi-station approach and static aquifer attributes for groundwater level simulation: what is the best way to leverage regionalised information?

Sivarama Krishna Reddy Chidepudi, Nicolas Massei, Abderrahim Jardani, Bastien Dieppois, Abel Henriot, and Matthieu Fournier

Hydrol. Earth Syst. Sci., 29, 841–861, https://doi.org/10.5194/hess-29-841-2025,https://doi.org/10.5194/hess-29-841-2025, 2025

Short summary

Sivarama Krishna Reddy Chidepudi, Nicolas Massei, Abderrahim Jardani, Bastien Dieppois, Abel Henriot, and Matthieu Fournier

Viewed

Total article views: 810 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
546	227	37	810	38	36

HTML: 546
PDF: 227
XML: 37
Total: 810
BibTeX: 38
EndNote: 36

Views and downloads (calculated since 13 May 2024)

Month	HTML	PDF	XML	Total
May 2024	190	81	7	278
Jun 2024	115	32	9	156
Jul 2024	90	28	11	129
Aug 2024	32	8	4	44
Sep 2024	28	22	0	50
Oct 2024	9	3	0	12
Nov 2024	13	12	0	25
Dec 2024	18	9	0	27
Jan 2025	24	21	3	48
Feb 2025	27	11	3	41

Cumulative views and downloads (calculated since 13 May 2024)

Month	HTML	PDF	XML	Total
May 2024	190	81	7	278
Jun 2024	115	32	9	156
Jul 2024	90	28	11	129
Aug 2024	32	8	4	44
Sep 2024	28	22	0	50
Oct 2024	9	3	0	12
Nov 2024	13	12	0	25
Dec 2024	18	9	0	27
Jan 2025	24	21	3	48
Feb 2025	27	11	3	41

Viewed (geographical distribution)

Total article views: 811 (including HTML, PDF, and XML) Thereof 811 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 18 Feb 2025

Download

The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.

Preprint (2818 KB)
Metadata XML

Short summary

This study explores how deep learning can improve our understanding of groundwater levels, using an approach that combines climate data and physical characteristics of aquifers. By focusing on different types of groundwater levels and employing techniques like clustering and wavelet transform, the study highlights the importance of targeting relevant information. This research not only advances groundwater simulation but also emphasizes the benefits of different modelling approaches.


Total:	0
HTML:	0
PDF:	0
XML:	0