the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 04 Apr 2024
Relative importance of uncertain model parameters driving water fluxes in a Land Surface Model
Abstract. We focus on the way temporal distributions of key components of the water cycle are influenced by typically uncertain parameters embedded in a Land Surface Model. We rest on a joint analysis of multiple global sensitivity metrics to provide a comprehensive assessment of the ranking of the relative importance of uncertain factors of various origins on the hydrological system response. The latter is rendered in terms of the temporal dynamics of transpiration, evaporation, and groundwater recharge. The NIHM (Normally Integrated Hydrological Model) modular Land Surface model is applied to simulate realistic field conditions (in terms of, e.g., climate, vegetation, and soil type) associated with two watersheds in the Vosges region (France) across a one-year period. These watersheds are characterized by similar climatic conditions while being associated with differing soil types and vegetation. Uncertain model parameters we consider comprise monthly values of albedo and leaf area index, vegetation-related parameters, as well as parameters related to the soil types associated with the litter layer and root zone. Four diverse sensitivity indices are used to quantify impacts of uncertain model parameters on the whole probability distribution or given statistical moments of the density function of model outputs. Our results document that the strength of the relative importance of model parameters depends on the statistical moment considered. Evaporation is directly influenced by the energy flow through the canopy and by the parameters associated with the top litter layer. As one could expect, transpiration appears as mainly influenced by the vegetation characteristics and by albedo that influences the incoming radiation. Groundwater recharge is influenced only by a very limited number of model parameters. It mainly depends on soil-related parameters and is unexpectedly not sensible to any of the vegetation parameters considered, except the root layer thickness and the intercept.
- Preprint
(3667 KB) - Metadata XML
-
Supplement
(365 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on hess-2024-73', Anonymous Referee #1, 08 May 2024
Review of the manuscript “Relative importance of uncertain model parameters driving water fluxes in a Land Surface Model”
GENERAL COMMENT:
This paper shows the results of a comprehensive sensitivity analysis for three key hydrological variables (evaporation, transpiration and recharge) using a complex land surface model, the NIHM modular LSM. The sensitivity analysis is based on three metrics that evaluate the relative weight of different model parameters on the probability distribution, expected value and variance of the chosen variables. In addition, an interesting feature of the analysis is to identify the temporal change in relative weight of the parameters. On the other hand, although the authors choose two different locations to perform the sensitivity analysis, the spatial dimension seems less important and does not seem to be addressed by the results.
I believe that the work of Luttenauer et al. shows interesting results that could help to propose improvements in LSMs with respect to parameter values. It could help to open new perspectives to correct some flaws that could be observed when evaluating an LSM against observed data. But also, I think the article could be improved, especially in the discussion part. In my opinion, the authors address the problems they point out in the introduction, i.e., uncertainty in the knowledge of parameter values, and the need to identify important parameters, but they do not present the limitations and differences of their results compared to other examples in the literature. Following the same line, the authors do not attempt to state some general conclusions that could be useful for other models and scales, although I understand the difficulties in reaching such conclusions. In conclusion, I think the manuscript could benefit from showing some additional results, so that the results are better understood, and from a longer discussion, so that the implications and perspectives of these results are clearly stated and compared with other efforts in the literature.
MAJOR COMMENTS
I would like to begin with two observations regarding the results presented.
First, LSMs differ from classical hydrological models mainly because they solve water and energy balances at the same time, estimating actual evapotranspiration fluxes as a function of available energy and water. Here I understand that the focus of the paper is water fluxes for a single pixel, so I will not consider any other energy-related variables, but I believe that at least one variable is missing in the analysis: surface runoff. Including surface runoff in Tables 3 and 4, and Figures 3 and 4 should be sufficient to better understand the response of the water system.
Also, if possible, I think the authors should include in the supplement a figure with the delimitation of the two river basins, and the pixels that were included in the analysis.
The second observation concerns the relationship between LAI and albedo. The authors say that both parameters come from remotely sensed data, but also that this global albedo is split into soil and canopy albedos (section 2 of the supplement). Does this mean that albedo participates in the estimation of net radiation, but that both albedo and LAI participate in the partitioning of evapotranspiration between evaporation and transpiration? If so, that might explain why evaporation (which, I assume, always occurs under the canopy layer, i.e., the pixel does not have an area with bare soil fraction) is not sensitive to albedo values. If so, I would recommend the authors to clarify this relationship in section 2 (specifically line 333), and to consider it in the results section.
Also, I'm not sure if I missed this, but there is no value for bare soil albedo (in line 190 of the supplement, it is assumed that the soil albedo is known). Please put the value used in the supplement.
As for the conclusions, I have two main observations:
First, I believe that the authors have not sufficiently related the results to the climatic and local conditions of the two basins. For example, with respect to groundwater recharge, and its sensitivity mainly to soil-related parameters, one would think that the two basins have sufficient water for evapotranspiration (i.e., the basins are energy-limited), and therefore, vegetation does not play an important role in the recharge rate. This would also mean that the partitioning of runoff between surface runoff and baseflow is also controlled by soil parameters. This idea seems consistent with the precipitation rates in both catchments (903 and 2541 mm) and with a small sensitivity of recharge to LAI values between May and October. Also, with a slightly higher sensitivity in the Bruche basin for LAI.
Second, I think the manuscript needs an effort to further discuss the implications of these results. For example, what are the implications of the sensitivity analysis for other land surface models? Is it possible for other land surface models (such as those used in climate models) to guide a parameter calibration based on these results? Or is it still acceptable to use simplified, a priori parameter values? What are the implications of these results for a model that includes new processes, which could change the relative weight of some parameters? How could we address the complexification of LSMs and introduction of new features and processes, at least at the local scale used here? I understand that this LSM is run at a local scale, while other LSMs are run at regional and global scales, so there is a scale issue regarding this analysis. Also, that there are important differences in model structure and processes representation that could prevent to state general conclusions. But the authors could use the results to further discuss the prospects of defining parameter values more carefully, and using sensitivity analysis to reduce the uncertainty of key variables for complex models such as LSMs.
I add some references that may be interesting for this discussion. For parameter sensitivity at different scales:
(https://hess.copernicus.org/articles/24/3753/2020/) (https://onlinelibrary.wiley.com/doi/10.1029/2019WR026612)
An example of changes in the model output due to new processes or differences in the mathematical representation:
(https://gmd.copernicus.org/articles/17/2141/2024/) (http://doi.wiley.com/10.1002/2016JD025426)
Finally, I would suggest describing the mode setting in more detail. It is not clear if the model uses 8 km resolution like Safran, or if it is finer. Also, how does the model deal with land surface heterogeneity? On line 383 the authors say that only one vegetation type is used for Bruche, and two vegetation types for Doller. If the model uses an 8 km pixel, what is the implication of simplifying land surface heterogeneity for sensitivity analysis (if there is any implication)?
MINOR OBSERVATIONS
Some additional observations and suggestions:
Decharme et al., 2019 is not in the references section. On the other hand, Decharme et al., 2011 is in references but not in the text. They both refer to ISBA model, Decharme et al., 2011 is interested in the sensitivity of the model to pedo-transfer functions in a soil multilayer representation, and Decharme et al., 2019 presents the ISBA-CTRIP model. Please review the reference list, and clarify the reference within the text.
Also, here the authors are putting together regional and global scale LSMs. While essentially the same, regional models could represent in more details some processes, like horizontal groundwater flow, while global models try to parameterize the main processes in an idealized way. Two additional regional models: Catchment (Koster et al., 2000), LEAF2 (Fan and Miguez-Macho 2011), and one additional global model: ORCHIDEE (Krinner et al., 2005).
128: designed
132: “(a) 133 application of a unique model formulation across different soil and vegetation types is questionable” this phrase may be reformulated
334: I think that here, when you say “time” you refer to “moment”
384: not sure the word “exemplary” must be used here
519: It is difficult to say because curves are close between them, but it seems that unconditioned PDF and PDFs for higher albedo values are similar, so to say that higher transpiration values are controlled by higher albedo values is not straightforward for me.
Fig. 6 As Sobol and AMAV indices are similar (they both are based on variance), I would suggest to put them in the same column
621: maybe just put “by the drainage from the litter layer”
For supplementary, please numerate and put a caption to tables in section 3. Also, water content at saturation uses in line 156, and in table in section 3.
Citation: https://doi.org/10.5194/hess-2024-73-RC1 -
AC2: 'Reply on RC1', Philippe Ackerer, 18 Jun 2024
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2024-73/hess-2024-73-AC2-supplement.pdf
-
AC2: 'Reply on RC1', Philippe Ackerer, 18 Jun 2024
-
RC2: 'Comment on hess-2024-73', Anonymous Referee #2, 10 May 2024
This paper by Lutternauer et al. performed parameter sensitivity analysis of a land surface model at two watersheds in France. The authors employ various sensitivity analysis methods to assess an extensive set of model parameters, which is quite interesting. However, I do have some concerns that I feel need to be addressed:
- My major concern is on the water balance at those two catchments. The average evapotranspiration is only 38.6% and 11.6% of total precipitation in those two catchments, which means runoff (surface and subsurface) must be large. Does that align with the observations at those catchments? Are there observations available to evaluate the model performance?
I am concerned because I feel that the sensitivity analysis is most useful when the parameter values are sampled around their optimal values in the multi-dimensional parameter space. Otherwise, the analysis may not reflect the real parameter sensitivity. For example, in an extreme case, if the model simulates predominantly surface runoff with minimal evapotranspiration, parameters linked to evapotranspiration would exhibit weak sensitivity, which does not reflect reality. It would be helpful if the authors can show the observations of discharge, or evapotranspiration, if avaiable. - Related to the first point, I feel the paper can be strengthened if all components of the water balance can be included.
- The authors mentioned that the model was run in a distributed way, but only selected one grid for each catchment for analysis. I am wondering if the model is three-dimensional. From what I read, the model seems to be a one-dimensional grid model. If it is one-dimensional, running the other grids should not affect the results. Some clarification would be very helpful.
- The paper does not have a “discussion" section, which limits the paper's impact. I feel some discussion would strengthen the paper substantially. For example, how does the sensitivity analysis results compare with other studies? How do the four sensitivity analysis methods differ? Do the sensitivity analysis results reveal some important insights into the model mechanisms, or the hydrological conditions at those catchments?
- I feel it could be helpful to show the values of LAI and albedo in the manuscript, either using a figure or a table.
Specific comments:
- Units of field capacity and porosity in Table 2 are missing.
- Line 540: “ For example, the value of B for the evaporation at Bruche during the month of July associated with the LAI of January must be zero. However, inspection of Fig. 6a does not reflect this anticipated outcome. This apparent anomaly is attributed to a random noise…”
I don’t quite agree with the authors. The results from previous steps might affect future steps. I don’t think the B value for July evaporation associated with January LAI is necessarily zero. - Figures 6 to 11 are difficult to read. Can parameter symbols be used instead of parameter index in those figures? Or at least, put the parameter identification codes in the manuscript, instead of the supplemental materials?
- What are θL and θp in Equation 25 in the supplementary material?
- Can authors add how surface runoff is determined? I think that could help the readers understand some of the results.
Citation: https://doi.org/10.5194/hess-2024-73-RC2 -
AC1: 'Reply on RC2', Philippe Ackerer, 18 Jun 2024
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2024-73/hess-2024-73-AC1-supplement.pdf
- My major concern is on the water balance at those two catchments. The average evapotranspiration is only 38.6% and 11.6% of total precipitation in those two catchments, which means runoff (surface and subsurface) must be large. Does that align with the observations at those catchments? Are there observations available to evaluate the model performance?
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 318 | 80 | 27 | 425 | 45 | 25 | 36 |
- HTML: 318
- PDF: 80
- XML: 27
- Total: 425
- Supplement: 45
- BibTeX: 25
- EndNote: 36
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1