|This is an interesting study, showing a significant effort to organize groundwater models at the national scale in France. I found this paper version generally clear and easy to follow. |
I have only a few suggestions detailed below. Since there are many illustrations, I made suggestions to remove some of them. I suggest publication after minor revision.
1. P1, L32: "Nash-Sutcliffe"
2. P3, L13: “State scale”
3. P4, L9-25: I found this part not well placed in the introduction, since it starts describing in details the AquiFR platform, which is the objective of the subsequent section (section 2), with some overlap between these two parts. I suggest moving this part at the beginning of section 2 and removing the redundant information with this section.
4. P4, L13: “and contain”
5. P5, L32: Remove “Erreur…introuvable”
6. P6, L4: Remove “Erreur…introuvable”
7. P6, 26: “assigned to”
8. P7, L13: “is accounted for”
9. P7, L26: “dimensional”
10. P9, L13: “99 km²”
11. P9, L21-22: Is this use of average pumping estimates realistic? Was this mean cycle calculated over all pumping data available? Would not it be more realistic to take only the first years of the available records and to extend this over the past years? For aquifers where water abstractions have much increased, taking a mean over all the available pumping data may overestimate the abstractions over the most remote parts of the evaluation period. Could the author add a few words on this?
12. P9, L26: “do not cover”
13. P10, L3: “new fluxes”
14. P10, L4: “will be soon updated”: any reference to this ongoing work?
15. P10, L5: I found it difficult to check all the cited references to get the information on the periods used for calibration. Could the authors add a column in Table 1 to give the period used for calibration in each case? (for each line of the table, if different calibration periods are considered, maybe show the longest period used in each case). This would help the reader assess the level of overlap between the simulation period used in this study and the calibration period.
16. P10, L10: Similarly to the previous comment, it would be useful for the reader if the authors could add an extra column in Table 1 giving the type of criteria used for calibration (and possibly the various variables – piezometric level and/or flows). This would show the various strategies used in these past works.
17. P10, L29: “relative”: the formulation is not expressed in relative but absolute terms
18. P10, L31: In the equations, the summation symbol refers to i index but then the variable X is referring to t. Maybe use instead Q(i) to be more consistent in notations. Change notations accordingly for the other equations. Maybe remind that BIAS has the same unit as X and that the perfect value is 0, with negative values corresponding to overestimation and positive values corresponding to underestimation by the model (or maybe there is an error in Eq.1, see also comment # 25 on this issue).
19. P11, top: Please give a few words on how the values of the NRMSE_BE criterion should be interpreted (what could be considered a good NMRSE_BE value?).
20. P11, L11: I found this sentence unclear. What do you mean?
21. P12, L8-21: This part is not well placed in the Results section, since it presents the data used. I suggest moving it to a new subsection on data at the end of section 3.
22. P12, L8-21: I did not find information on the way model warm-up is made at the beginning of the simulation period. I guess this can be an issue and a potential source of error. Could the authors give some information on this?
23. P12, L12: “a few measurements”
24. P12, L20: “not yet fully available”
25. P12, L23-24: From equation (1), a positive value would mean an underestimation.
26. P12, L26: instead of “accumulated distribution”, use “cumulative distribution”. Change accordingly elsewhere in the text and figures (graphs and captions in Fig. 6, 7, 15)
27. P12, L29-30: How these values can be interpreted? (see comment #19)
28. P13, L6-7: The observations show annual cycles starting from 2008. Is there any change in the measurement device?
29. P13, L8 and 11: “Helloin” is written with uppercase elsewhere in the article.
30. P13, L12: This period seems also badly simulated for the Omiécourt and Farceaux case studies, which are all in the same region. Is it just a coincidence or were there specific climatic conditions on this period, which are not well captured by the models?
31. P13, L24-26: I am not sure these two figures are really useful. Very little is said on these figures and they do not seem to convey different information. I would remove the figure on correlation.
32. P14, L29: “Kling-Gupta efficiency”
33. P15, L13: “negative”
34. P15, L19: Not sure this comment is fully consistent with what is said on top of page 13. Please clarify this.
35. P16, L18: “in section 3.2”
36. P16, L17-19: Generally, testing a model outside calibration range is considered a good way to evaluate its extrapolation capacity and robustness. Here it is difficult to evaluate this since the calibration and evaluation periods overlap. The authors could discuss this point. Is there some perspective to more thoroughly evaluate the robustness of model proposed in AquiFR?
37. Fig. 6: Could the cumulative distribution give also the value for an absolute bias lower than 1, to make a better correspondence with the scale of the map above (in fig. a).
38. Fig. 12, caption: “for June 2016 for”
39. Fig. 13: It seems that the observations are not available on the whole period in each case. But it is a bit difficult to distinguish between the case where observed and simulated values perfectly overlap from the case where there are gaps in the observations. I suggest adding the information on periods of gaps in the observations, e.g. by putting a grey background on the graphs for the corresponding periods.
40. Fig. 13: in the caption: “dashed blue”
41. Fig. 14: Same comments as #39 and 40
42. Table 4: The information contained in the table could be directly introduced in Fig. 14 (in brackets after the name of the catchment, as done in Fig. 13). This would reduce the number of tables.