|This is the second time that I review this paper (which seems to have a somewhat longer history). In general, the paper has improved significantly, and the major coupling approach has become clearer. |
1. While the general approach is explained clearly in section 2.3, the normal naïve reader would not think that the approach works the way as described after reading the abstract and even the introduction. From the abstract, it should be clear that the groundwater storage is NOT taken out of mHm; instead fluxes into and out of the groundwater body are evaluated by mHm and used as Neumann boundary condition for the groundwater model. That is not stated explicitly.
2. Likewise, the introduction reads as if the approach was towards two-way coupling, which is not done in the presented method. While I agree that shallow groundwater tables influence actual ET in reality (pae 2, lines 8-12), this is not the case in the chosen approach where ET belongs to mHm which does not know the groundwater table. You really need to explain the problem that you DO solve by the approach, not the problems that you do not address. The motivation should be:
a. Conceptual hydrological models are good to get water balances right (essentially meeting hydrographs), they are quick to run and can be calibrated in stochastic frameworks. But these models cannot handle groundwater tables as that is not a state variable of the models.
b. Pde-based subsurface models are good to get hydraulic heads and travel-time distributions. In the unsaturated zone, they often fail getting quick flow components right because the relevant features are not resolved. Adding pde-based surface models is possible in general, but requires the resolution of fine-scaled features (which are sometimes not known), a tremendous number of uncertain parameters, and takes ages to run. Thus calibrating these models is cumbersome, and doing this in a stochastic framework is computationally unfeasible. Without pde-based surface models, the subsurface models still need boundary conditions at the surface.
c. The general idea of the paper is to perform the hydrological calculations with mHm, including the simplified linear groundwater storage, extract fluxes into- and out of groundwater from mHm, and use this as Neumann BC for a pde-based groundwater model using OGS. By this, mHm is “pimped up” to predict hydraulic heads in groundwater.
d. The model can thus NOT consider feedbacks between land-surface processes and the groundwater table. It is also clear that an aquifer system that inherently acts completely different than a linear storage cannot be simulated right. The approach builds on the assumption that the groundwater storage calibrated within mHm really represents groundwater and is not an alias of other storage mechanisms.
I highly recommend having a discussion on thinkable coupling strategies upfront and justifying the general choice already in the introduction. Do definitely not argue with features that you deem necessary but that you can’t deliver.
3. Running a groundwater model that has only Neumann boundary conditions (recharge, no flow, exchange with streams) can cause non-uniqueness problems. In the confined case, it is obvious that this is an ill-posed forward problem. In the phreatic case it’s somewhat better, but definitely not optimal (specifically not at steady state). I teach my students that they need at least one Dirichlet boundary to fix the head. This should be discussed somewhere. (In the previous version I thought that the rivers either form a fixed-head or leaky boundary conditions).
4. The Pearson correlation coefficient is NOT a metric for the goodness of a fit, as it tests linearity between two data sets rather than identity. The correct metric would be the root mean square error or the mean absolute error. Please exchange all correlation coefficients by RMSE or MAE.
1. Page 1, lines 5-6: “by which grid-based fluxes of groundwater recharge and river-groundwater exchange generated by mHm are converted to fixed-flux boundary conditions of the groundwater model OGS.”
2. Page 1, line 8: “Specifically, the grid-based vertical reservoirs in mHM, including a linear groundwater storage, are completely preserved”
3. Page 1, line 11: “using discharge for mHm and long-term mean of groundwater head measurements for OGS, respectively.”
4. Page 1, line 12 and several times thereafter: Eliminate the correlation coefficient as metric for the goodness of a model fit.
5. Page 1, line 14: “so that the one-way coupled model”
6. Page 1, line 16: “provide an adequate representation”
7. Page 2, lines 8ff.: While this is true, it is not resolved at all by the presented approach. The arguments given here speak against using the method presented.
8. Page 2, line 24: I have no clue what you mean by “parametrization of topographicsl features” in the context of a groundwater model.
9. Page 3, lines 3-5: Confusing arguments. What you really need to say here is that these models use pde's for both subsurface and surface flow, requiring a fine resolution and many parameters. There is also no “unsaturated groundwater”.
10. Page 3, line 6: replace “field-scale” by “hillslope-scale”
11. Page 3, line 7: Warning! Both Reed Maxwell and Ed Sudicky sell their pde-based models with applications on the continental scale. Whether the parameters in these models have any physical meaning is a different story, but you can’t say that the pde-based models are restricted to small-scale applications. Sabine Attinger is co-author on a paper submitted to GMD using TerrSysMP on the scale of the state of Baden-Württemberg.
12. Page 3, line 9: Sorry, but it’s conceptual models that neglect the momentum balance altogether. Darcy’s law is derived from a momentum balance, and the shallow-water equations even more so.
13. Page 3, lines 23-24: Yes, but you can of course use more complex formulations for land-surface processes in pde-based models (see TerrSysMP coupling CLM to PARFLOW). The true difficulty is that, unless you resolve all relevant fine-scale features, you don't get quick flow components right with pde-based approaches. This is an inherent scaling problem. The conceptual hydrological models overcome this difficulty by introducing quick flow out of the blue, which is conceptually impossible in the pde-based approaches.
14. Page 3, line 30: Remove “therefore”
15. Page 3, lines 31-32: If the overall aim was to model regional-scale groundwater flow dynamics, you would choose a stand-alone groundwater model. So that cannot be the overall goal. What you really want is to extend a regional-scale conceptual hydrological model to also predict hydraulic heads (and eventually groundwater transport and a proceeding step).
16. Page 4, line 25: eliminate “by applying the MPR methodology”.
17. Page 4, line 31: “we only listed the equations needed for the coupling”
18. All explanations related to equations: In the mHm part, the variable symbols are followed by untis (e.g., “[mm]”), in the OGS part they are followed by dimensions (e.g., “[L]”). Either way is OK, but it should be consistent. (I’d prefer the dimensions over the units)
19. Page 5, line 11: beta_5 is not a rate, it’s a rate coefficient. And it has a dimension (of an inverse time, which explains why it is NOT a rate but a rate coefficient).
20. Page 5, line 14: same arguments for beta_6
21. Page 6, line 25: In which direction is z oriented? According to equation (10), where z is subtratct from the pressure head, it must be downwards. Please specify.
22. Page 6, lines 26 & 27: “the OGS source code” and “the OGS mesh”
23. Pahe 8, line 1: “OGS is used to simulate groundwater flow for prescribed fluxes at all boundaries” rather than “hydrological processes”
24. Page 8, line 14: “mHM is run independently of OGS to calculate land surface fluxes including exchange fluxes of the groundwater storage”
25. Page 8, line 21: “is transformed to distributed river discharges along the streams and spatially distributed exchange rates between streams and groundwater needed in OGS”
26. Equation 11: What happens if the stream nodes are irregularly spaced?
27. Page 9, line 15: “For each node” rather than “each of the nodes”
28. Equations (12) and (13). This is still not clear. I understand equation (13), provided that W_i(x) is the FEM weighting function. But what exactly is done in equation (12)? W_j(x) seems to be a shape function of the mHM grid, but that is not stated. Actually, my understanding is that the index j relates to an mHM grid cell and that W_j(x) is unity within the mHM cell and zero elsewhere. But that is not written in the text accompanying the equation.
29. Page 9, line 30: “the groundwater model simulates groundwater flow to obtain hydraulic heads. The groundwater model may also be used to compute travel times and solute transport within the groundwater domain, requiring additional boundary conditions; but this is not described in the present paper.”
30. Page 11, line 7: “3.2 Meteorological and surface properties”
31. Page 11, line 12: “kriged onto”
32. Page 11, lines 12ff.: “Potential ET was estimated by the method of Hargreaves and Samani (1985). Other datasets used in mHM are the digital elevantion data ... soil and geological maps, and derived properties such as sand and clay contents, bulk density, CORINE land-cover information…” (geology appeared twice, I have no clue what meta-data you meant)
33. Page 11, section 3.3: It’s not “stratified aquifer model”, it’s “stratigraphic” (as you relate aquifer properties to stratigraphic units, which need not be stratified).
34. Section 3.6 and also 4.1, 4.3: See my remarks related to the Pearson correlation coefficient. Calculate and report RMSE or MAE instead.
35. Page 14, line 25: “the wells” rather than “those wells”
36. Page 15, line 1: “which is due to the unknown local or eve sub-grid-scale properties”.
37. Page 16, line 3: “where the Muschelkalk crops out”
38. Page 16, line 9: Which figure is “the figure”
39. Page 20, line 16: The model was not calibrated in the transient mode. Thus, you did not calibrate the specific yield and the specific storativity, which largely determine the dynamic behavior of hydraulic head.
40. Page 20, lines 30-31: This again is the trap of arguing with something that you do not resolve at all. Your hydrological model does not consider feedbacks of groundwater to surface hydrology at all.
41. Page 22, line 16: Double “derived recharge rates”
42. Page 22, line 25: “capability” rather than “ability”
43. Page 22, line 29: “computational effort” rather than “consumption”
44. Page 22, line 34: “can explicitly be modeled” (word order)