Evaluating Simplifications of Subsurface Process Representations for Field-scale Permafrost Hydrology Models
- Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA
- Environmental Sciences Division, Oak Ridge National Laboratory, Oak Ridge, Tennessee, USA
Abstract. Permafrost degradation within a warming climate poses a significant environmental threat through both the permafrost carbon feedback and damage to human communities and infrastructure. Understanding this threat relies on better understanding and numerical representation of thermo-hydrological permafrost processes, and the subsequent accurate prediction of permafrost dynamics. All models include simplified assumptions, implying a tradeoff between model complexity and prediction accuracy. The main purpose of this work is to investigate this tradeoff when applying the following commonly made assumptions: (1) assuming equal density of ice and liquid water in frozen soil; (2) neglecting the effect of cryosuction in unsaturated freezing soil; and (3) neglecting advective heat transport during soil freezing and thaw. This study designed a set of 62 numerical experiments using the Advanced Terrestrial Simulator (ATS v1.2) to evaluate the effects of these choices on permafrost hydrological outputs, including both integrated and pointwise quantities. Simulations were conducted under different climate conditions and soil properties from three different sites in both column- and hillslope-scale configurations. Results showed that amongst the three physical assumptions, soil cryosuction is the most crucial yet commonly ignored process. Neglecting cryosuction, on average, can cause 10 % ~ 20 % error in predicting evaporation, 50 % ~ 60 % error in discharge, 10 % ~ 30 % error in thaw depth, and 10 % ~ 30 % error in soil temperature at 1 m beneath surface. The prediction error for subsurface temperature and water saturation is more obvious at hillslope scales due to the presence of lateral flux. By comparison, using equal ice-liquid density has a minor impact on most hydrological variables, but significantly affects soil water saturation with an averaged 5 % ~ 15 % error. Neglecting advective heat transport presents the least error, 5 % or even much lower, in most variables for a general Arctic tundra system, and can decrease the simulation time at hillslope scales by 40 % ~ 80 %. By challenging these commonly made assumptions, this work provides permafrost hydrology modelers important context for better choosing the appropriate process representation for a given modeling experiment.
- Preprint
(3705 KB) -
Supplement
(13624 KB) - BibTeX
- EndNote
Bo Gao and Ethan T. Coon
Status: final response (author comments only)
-
RC1: 'Comment on tc-2021-362', Anonymous Referee #1, 18 Mar 2022
General Comments
The authors use their simulation code "Advanced Terrestrial Simulator" to simulate different scenarios switching certain processes (expansion of water during freezing, cryosuction, advective heat transport) on or off and investigate the extent of the changes to the simulation results. The authors draw conclusions on the relevance of the processes and try to address the question, if they can be safely ignored to simplify the model and save computation time. The scenarios make sense and try to give a representative sample. The paper is well written and the different scenarios are comprehensibly described. The figures are a bit small and overloaded, thus it is sometimes hard to see all the simulation results compressed into one figure.
In my opinion the key question is, if the content of the paper is of sufficient general interest and is really giving new insight. The title of the paper suggests that the results are generally relevant for "field-scale permafrost hydrology models". However, they will to some extent be influenced by the concrete modelling approach, discretization scheme, linear solver (the authors mention that the AMG-preconditioner they use is not well suited for advective transport)... Thus it is more a kind of sensibility analysis of the results produced by their code in different scenarios. The authors tend to not carefully distinguish between small differences in the model results produced by their code (assuming that the representation of the different processes in their code is correct) and a low relevance of the process in reality (or at least in modelling reality). However, this is not at all the same.
What is also missing is an analysis of the discretisation error associated with the different grids and the time step used (if the grids are too coarse, the results are not really realiable). Thus I think the paper can be published, but it is not an essential step forward.
Specific comments
Title: I would suggest to make the title a bit less general, e.g. "Evaluation the sensitivity of prediction results to process simplifications for the advanced terrestrial simulator". It is not clear, that the results obtained here, are really generalizable to other codes. Especially, as the change of the model results relative to the safed computation time is of interest.
line 191: The soil-freezing characteristic curve is usually used as a material property of a certain soil. Thus I find this term here rather confusing
figure 1: Too much information is packaged into too small figures here. It is very hard to see for example the rain precipitation at sage, because it is in the background of the other sites. Maybe you could make one set of plots pere site in a 3x3 matrix?
table 2: the van Genuchten-Mualem model can produce unphysical results for n values much smaler than 2 (which is true for all parameter sets here)
3.2 mesh design: why 78 cells? Have some tests be done, that this is a sufficient resolution to obtain grid convergence?
line 333-344: how was this column initialization transferred to the hillslope? Does this not produce an instable initial value for the hillslope?
line 374-376: might this averaging of local data not smooth the effect of neglecting processes? If you have only a local effect at one or two points, this could be greatly reduced by the averaging
figure 4 and 5: The figures are again rather small
figure 6, line 414: The concept of a "decrease percentage" is rather hard to understand (especially if it gets negative). Would it not be easier to understand, if you use the relative runtime? Which than would be either smaller than one (thus faster) or larger?
section 4.2: there is no information about the effect of neglecting cryosuction on the runtime. However, isn't that the main point of the paper (how much precision do you sacrifice for which acceleration?)
figure 13: I am not sure, how much this figure really helps in understanding. you need to read the text very carefully to only understand, what is represented here (not talking about what it means)
conclusions: as stated above: the results found here do not have to be representative for all "permafrost models" Thus conclusions like "Excluding soil cryosuction in permafrost models can..." or "Assuming equal ice-liquid densitiy will not result..." are a bit ambitious or even dangerous
line 494: "factitiously" is a very rare word. how about "artificially"?
-
AC1: 'Reply on RC1', Bo Gao, 19 May 2022
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2021-362/tc-2021-362-AC1-supplement.pdf
-
AC1: 'Reply on RC1', Bo Gao, 19 May 2022
-
RC2: 'Comment on tc-2021-362', Anonymous Referee #2, 20 Apr 2022
This paper presents a quantitative evaluation, using numerical modelling, of the effects of making certain assumptions regarding relevant physical processes in the context of a coupled groundwater flow - permafrost (freeze/thaw) system. Specifically, three assumptions are tested - assuming a uniform water-ice density, neglecting cryo-suction and neglecting advective heat transport.
The analysis is conducted on two system scales -a simplified 3D hillslope domain and a 1D column. The ATS simulator is used for all simulations. The modelling approach is valid and the conclusions are logical.
There are a few weaknesses in the paper outlined below.
- The paper does a good job of presenting the quantitative numerical comparison of these assumptions, in terms of errors, but insights are lacking on explaining the reason for these differences. All results are shown as time series and error plots. More insight is needed into the actual physical processes and system behaviour, not just on 'dry' figures or plots showing errors. i.e. to answer WHY these processes are or are not important under these conditions. For example, there are no results shown in space of the flow system or temperature field of their 3D domain (maybe these are in the SI ?). This would at least help interpret what is happening in space, where flow is going, temperature gradients etc.
- I did not find the comparison of computational efficiency very relevant (ex lines 508-520). The authors seem to suggest if the computational cost of including advective heat transport is high, then it can be neglected. Computational cost should have little or no bearing on whether or not to include a process - if a process is important & relevant, it needs to be included, regardless of the computational cost. Most codes are efficient enough to include advective het transport even at large scales.
- I found the results and conclusions were cast too strongly as being definitive in the general context. These results are specific for the conditions assumed (geometry, flow system, etc.). For example, Line 502-503 states: ‘Therefore, for most Arctic systems at this scale, it is reasonable to neglect advective heat transport.’ Which is much too strong a statement and needs to be rewritten or deleted. There are many published cases showing that advective heat transport is indeed critical in many real-site cases, not just conceptual or simplified as in this case, where here it is cast as less relevant. I provide a few examples in the attached marked copy.
Specific comments:
- The paper refers a few times to 'a general Arctic system' (Line 27) or to '... a normal Arctic system' (Line 490) .... These should be replaced by, ex., 'a conceptual system'... or 'in these specific simplified cases'. There is no such thing as a 'general' or 'normal' Arctic system. Line 490 in particular reads like a general statement which is not true, you have to re-read it to finally understand it refers to these specific cases only. More acknowledgement is needed in general that these are specific results for these cases only, not generally-applicable conclusions.
- Line 111: The Nixon (1975) paper is much too old to use for justifying this statement that 'it is commonly recognized that heat conduction predominates ...'. This might have been the case in 1975, but not more recently in the past decade or so. So the entire viewpoint should be modified as well. i.e. that it has become recognized more recently that advection can be significant but is still often neglected ... Or perhaps that in some cases it is not known whether advection is important or not but it is neglected anyway …
- Line 156 - needs to be corrected to advection-dispersion (or advection-conduction). ('diffusion' is almost always used only in the context of mass transport). Same for Heading 2.3 (line 297).
- Table 6 – four significant digits is excessive here.
- See attached marked manuscript for all grammatical corrections and other comments.
-
AC2: 'Reply on RC2', Bo Gao, 19 May 2022
The comment was uploaded in the form of a supplement: https://tc.copernicus.org/preprints/tc-2021-362/tc-2021-362-AC2-supplement.pdf
Bo Gao and Ethan T. Coon
Data sets
Raw and processed forcing data, and simulated outputs Bo Gao https://github.com/gaobhub/data_for_paper_model_comparison
Model code and software
The Advanced Terrestrial Simulator (ATS) version 1.2 E. T. Coon, M. Berndt, A. Jan, D. Svyatsky, A. L. Atchley, E. Kikinzon, D. R. Harp, G. Manzini, E. Shelef, K. Lipnikov, R. Garimella, C. Xu, J. D. Moulton, S. Karra, S. L. Painter, E. Jafarov, and S. Molins https://github.com/amanzi/ats
Bo Gao and Ethan T. Coon
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
391 | 112 | 16 | 519 | 29 | 9 | 11 |
- HTML: 391
- PDF: 112
- XML: 16
- Total: 519
- Supplement: 29
- BibTeX: 9
- EndNote: 11
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1