the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Sep 2024
Comparison of different ensemble assimilation methods in a modular hydrological model dedicated to water quality management
Abstract. Hydrological models are valuable tools for understanding the movement of water and contaminants in agricultural catchments. They are particularly useful for assessing the impact of landscape organization on pesticide transfers and for developing effective mitigation strategies. However, using these models in an operational context requires reducing uncertainties in their outputs, which can be achieved through data assimilation methods. In this study, we aim to integrate surface moisture images into the PESHMELBA water and pesticide transfer model using data assimilation techniques. Twin experiments were conducted on a virtual catchment consisting of vineyard plots and vegetative filter strips. We compared the performance of the Ensemble Kalman Filter (EnKF), the Ensemble Smoother with Multiple Data Assimilation (ES-MDA), and the iterative Ensemble Kalman Smoother (iEnKS) in jointly estimating vertical moisture profiles and certain input parameters. Results indicate that ES-MDA performs the best in estimating surface moisture and related input parameters, while all methods show similar results for subsurface moisture variables and parameters. Furthermore, we examined the sensitivity of the methods to observation error magnitude, observation frequency, and ensemble size to establish an effective assimilation setup. This study paves the way for future operational applications of data assimilation in PESHMELBA.
- Preprint
(2275 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on hess-2024-219', Benjamin Mary, 03 Oct 2024
# Comparison of different ensemble assimilation methods in a modular hydrological model dedicated to water quality management
Emilie Rouzies1, Claire Lauvernet1, and Arthur Vidard2I have greatly appreciated the opportunity to read the article by Rouzies et al. The paper presents a valuable synthetic study, effectively illustrating the strengths and weaknesses of the EnKF method, both sequentially and with smoothing over observations, which in this case mimic remote sensing of soil water content. The article is well-written, with clear explanations of the methodology and model used.
While I have some reservations about relying on Soil Moisture Remote Sensing products, this study clearly highlights the challenges associated with using such data to calibrate subsoil and plant parameters. Despite this, I believe the overall objectives of the study are achieved, particularly in terms of selecting the most suitable data assimilation (DA) scheme. However, I think a more detailed analysis of the DA results would strengthen the paper. For example, it would be beneficial to explore the following points:
The choice of DA localization schemes, either through local domain DA or covariance localization, particularly given the variability in soil units (see, for instance, https://www.frontiersin.org/journals/applied-mathematics-and-statistics/articles/10.3389/fams.2019.00003/full).
The distinction between DA with only state updates versus DA that incorporates a training and validation phase (see Botto et al., 2018, www.hydrol-earth-syst-sci.net/22/4251/2018/).
Another important consideration is the number of parameters being updated. I understand that a previous sensitivity analysis was conducted on the same site. Would it have been advantageous to reduce the number of parameters updated based on the results of this prior analysis? (For reference: Global sensitivity analysis of the dynamics of a distributed hydrological model at the catchment scale).Overall, I believe the manuscript holds strong potential for publication in HESS, pending major or minor revisions and further clarifications.
## SPECIFIC COMMENTS
# ---------------------------------
The title could be more specific. I suggest rephrasing "modular hydrological model dedicated to water quality management," which is somewhat broad, to something more aligned with the core focus of the study, such as "Twin Experiments on a Virtual Catchment with Vegetative Filter Strips" or "Hydrology of Agricultural Catchments." This would better target the intended audience and improve the visibility of the paper in bibliographic searches.## Abstract
L8: "some input parameters" specify them if possible
L9: "related input parameters" again specify (I suppose mainly VGP and plant traits?)## Introduction
L14-19: add references
L17: it is not clear what the authors means by "often simulate several physical processes" please rephrase
L18: "They need large sets of input parameters" specify i.e. hydraulic conductivities, soil and vegetation physical properties, atmospheric boundary conditions, ..
L21: rephrase " .. from observations distributed in time and space and PREDICTIONS from a numerical model". In DA the model is used for predicting.Â
L24: I would not say in geophysics but more in Earth Sciences (originally used for ocean modeling and weather forecasting)
L24: Please rephrase "They consist of Monte Carlo algorithms and linear solutions of the estimation problem"
L25: I suggest rephrasing: an ensemble of realizations (instead of vectors) and adding i.e. by approximating the state by a state mean and covariance matrix. (if you finally want to keep the terminology "vectors" provide a reference to understand where it is from).Â
L27: as they come -replace by-> sequentially (?)
L28: Â specify where the Gaussian assumption applies: on the assumption of Gaussian statistics (i.e. with forecast and measurement error distributions to be Gaussian). Reference: https://doi.org/10.1016/j.advwatres.2012.06.009
L27-28: I suggest specifying that integrated hydrological models based on the Richards equation still represent a challenge, due to strong nonlinearities that may significantly affect the filter performance.L75: Please provide a reference that shows the following statement: "a filter, a hybrid variational/ensemble smoother that is efficient over short data assimilation windows and an ensemble smoother that is efficient over long data assimilation windows". Or maybe this is your assumption? by the way what timeline is consider long/short windows i.e days, years?? In any case explain. Thanks!
## 2.1 Model description
L97 to 110: very interesting thanks for the detailed explanation of the model.Â## 2.2 Data assimilation methods
L120: replace/specify which parameters: "some input parameters" --> VGP see Appendix A.
L135: can the author specify a reference for the inclusion of "an evolution law for the estimated parameters in addition to the state dynamical evolution".## 2.3 Case study
- Size, superficie of the catchement?
- Figure 4: are there important differences in surface/subsurface hydrology between the different Soil Units (SU)?Â
If so can you add another SU to fig 4.
If not could you mention it in the text?
In my perception, it is important to understand SU unit dynamics for the DA. If they are very different from each other then DA with localization (see comment below). Â## 2.4 DA setup
L273: I understand the idea of the TWIN experiment and using the True model to generate the observations. Something I'm not sure to understand is the spatial distribution of the observations. Are there several/one observations for each vineyard plot and VFS in the catchment? Are those gridded regularly? do you pick the mean for each zone?Have you thought about Localization using the Local Analysis (LA) scheme for the different SU? The idea is to perform by spatially limiting the assimilation process within a certain distance from a grid point. read for instance: https://doi.org/10.1016/j.advwatres.2020.103813
In any case, it would be interesting to analyze/discuss it in the text.ÂL284: Have you considered the mutual correlation between the Van Genuchten parameter? (according to Carsel and Parrish (1988), who described their statistics and transformed them into normally distributed variables via the Johnson system (Johnson, 1970)? https://doi.org/10.1029/WR024i005p00755
### 3.1.1 Performances on moisture variable correction
Fig 5. I wonder if the 2nd part works better because the rain events are stronger or just because the parameters were already calibrated for a certain period and the model is thus already calibrated.Â
There are two ways of testing it:Â
- run without parameters update
- run with DA until time 1000h for instance and then let the system free
In any case, it would be interesting to analyze/discuss it in the text.Â### 3.1.3 Computational cost
L372: for how many ensembles that hCPU where calculated?Â
Table 2: would it be possible to differentiate those numbers between soil and vegetation parameters? I'm curious to know for instance how root depth parameters are affected by DA.## 4.3 On the limitations of the methods
L444: From where those correlations are calculated/derived? is this somehow related to the state covariance matrix?
L444: As the state is not perturbated initially, I'm curious to know what the correlations look like at time 0; How about showing the correlation at time 0 to see the evolution with time (in appendix?).Â### Code availability
I appreciate seeing that the study can be reproducible with data accessible and open-source codes. Thanks to the authors for this effort.### Appendix A
- Please explain the nomenclature in the Pdf column: what are N, TN, and LN (Normal, Log-Normal, ...)
- isnt a rooting depth of 0.9m as a nominal value too high for grassland? (Vegetation parameters for VFZ)Best wishes,
Benjamin Mary
ICA-CSIC MadridÂ
Citation: https://doi.org/10.5194/hess-2024-219-RC1 -
RC2: 'Comment on hess-2024-219', Anonymous Referee #2, 04 Oct 2024
Authors presented a comparison of different DA methods in the context of modular hydrological model for water quality management. The paper is well-written and looks like very comprehensive. I have a couple of comments:
1) In the literature, a few papers about the comparison of DA methods have been published in the field of hydrogeology. Also, those methods are well established, and the disadvantages and advantages are well-known. Authors highlighted the modular hydrological model in this study instead of many studies using numerical models. if so, authors should clarify why there are differences using different physical models for DA, not only from the results of DA experiments, but from the methodology. Fundamentally, DA methods such as EnKF can be coupled with any transfer functions.
2) in line 273, the true value comes from perturbation from Gaussian noises. does this mean that your ground truth has a Gaussian distribution. How is this close to the real data? Does the real data follow Gaussian distribution? If it has a non-Gaussian distribution, how does those DA methods perform?
3) As we know, those DA methods are impacted by the ensemble size. Have you considered to implement some localizations to constrain the covariance so that the filter inbreeding issue could be reduced? In figure 11, it looks like that, if ensemble size is increased from 50 to 200, the performance of DA gets worse. This does not make sense. Â
Â
Â
Citation: https://doi.org/10.5194/hess-2024-219-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 143 | 34 | 77 | 254 | 7 | 4 |
- HTML: 143
- PDF: 34
- XML: 77
- Total: 254
- BibTeX: 7
- EndNote: 4
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1