the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Numerical coupling of aerosol emissions, dry removal, and turbulent mixing in the E3SM Atmosphere Model version 1 (EAMv1), part I: dust budget analyses and the impacts of a revised coupling scheme
Abstract. An earlier study evaluating the dust life cycle in the Energy Exascale Earth System Model (E3SM) Atmosphere Model version 1 (EAMv1) has revealed that the simulated global mean dust lifetime is substantially shorter when higher vertical resolution is used, primarily due to significant strengthening of dust dry removal in source regions. This paper demonstrates that the sequential splitting of aerosol emissions, dry removal, and turbulent mixing in the model's time integration loop, especially the calculation of dry removal after surface emissions and before turbulent mixing, is the primary reason for the vertical resolution sensitivity reported in that earlier study. Based on this reasoning, we propose a simple revision to the numerical process coupling scheme, which moves the application of the surface emissions to after dry removal and before turbulent mixing. The revised scheme allows newly emitted particles to be transported aloft by turbulence before being removed from the atmosphere, and hence better resembles the dust life cycle in the real world.
Sensitivity experiments show that the revised process coupling substantially weakens dry removal and strengthens vertical mixing in dust source regions. It also strengthens the large-scale transport from source to non-source regions, strengthens dry removal outside the source regions, and strengthens wet removal and activation globally. In wind-nudged simulations of the year 2010 with 1-degree horizontal grid spacing and 72 layers, the revised process coupling leads to a 39 % increase in the global annual mean dust burden and an increase of dust lifetime from 1.9 days to 2.6 days when tuning parameters are kept unchanged.
The revised process coupling is implemented for all aerosol species in EAMv1. The same qualitative changes in process rates are seen in dust, sea salt, marine organic aerosols (MOA), black carbon (BC), and primary organic aerosols (POA), as these species have significant sources from surface emissions. Quantitatively, the changes are large for dust and sea salt but are considerably smaller for the predominantly submicron species (i.e., MOA, BC, and POA). The impacts on sulfate and secondary organic aerosols are very small, as these species have little or no surface emissions.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(0 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1330', Anonymous Referee #1, 22 Aug 2023
Wan et al. show that in an atmosphere model at a 1° horizontal resolution, a coupling of surface emissions to the planetary boundary layer turbulence parameterization becomes even more important with increasing vertical resolution. I find this result not only plausible, but expected. What seems surprising to me is rather that in standard EAMv1, the vertical mixing of emissions prior to dry deposition was entirely determined by the thickness of the lowest model layer. As far as I know, global atmosphere models (sometimes with a vertical resolution on the order of 100m near the surface) usually assume instantaneous vertical mixing of surface emissions extending at least across the height of the lowest model layer (perhaps excluding point sources in a few models). But up to now, I also thought that many models either already couple surface fluxes and surface emissions to the planetary boundary layer turbulence parameterization (for example similar to what is mentioned for the IMPACT model on page 6) or else call the turbulence parameterization right after the emissions or surface flux computation, as suggested in the present manuscript.
The problem with the instantaneous artificial vertical mixing across the height of the lowest model layer is that, unlike parameterized mixing, it is not based on physical arguments and independent of local stability and the actual boundary layer height, which varies in space and time including a dependence on surface type, latitude, and a prominent diurnal cycle over land. Although this artificial mixing in the lowest layer may to some extent mimic boundary layer mixing in low-resolution models, I think that it is definitely not desirable. This artificial mixing can however, be reduced by increasing vertical resolution as far as numerical stability permits. A proper coupling of surface fluxes to the planetary boundary layer turbulence parameterization seems advisable in any case, and becomes even more important as the vertical resolution near the surface increases. My guess is that even at very high resolution, the process splitting discussed here may still be important as long as vertical resolution near the surface decreases proportionally.
On the one hand, a large number of well known papers on process splitting exists especially in the chemistry transport community (see minor point #1 below for an example), and perhaps elsewhere. The treatment of surface emissions, especially from point sources, has long been a concern in coarse resolution chemistry transport and aerosol models as well. For point sources, various approaches for addressing the problem of finding the right emission height and also of instantaneous mixing because of finite grid resolution have been suggested and tested, such as plume-in-grid models. Furthermore, as mentioned above, some atmosphere models may already couple emissions to the planetary boundary layer turbulence parameterization or use the process splitting suggested here.
However, based on a quick literature search, it seems to me that the publication of the manuscript by Wan et al. would be important and timely. I did not find anything concrete on this topic (although I must admit that I spent a limited time on my literature research). Because of the major effect the process splitting has on the results, the importance of the results presented by Wan et al. for the general progress in the development of EAM seem obvious to me. I would also expect the general result to apply at least to some other models as well.
I find that the manuscript by Wan et al. is exceptionally well written, and I have only a few minor comments. While some modellers may not be overly surprised by the findings, I think that documenting this type of sensitivity and especially also suggesting a solution is extremely useful. Although I partially agree with the authors on their more cautious statements in the introduction, which suggest to me that one should think in some depth about these issues, the relative correctness of the solution suggested by the authors seems rather obvious to me.
I would very much appreciate if the authors would consider a follow-up study to check if and how much this revision affects ERFari+aci
Minor points:
1. I generally think that it is good practice to discuss the results of any given study in the context of existing literature. Therefore, at least at first, it seemed to me that the authors could have mentioned and perhaps included a brief survey of the existing literature on process (or operator) splitting in the introduction. Initially, I also thought it would be rather easy for me to suggest a few references that are of direct relevance to the issue at hand (such as the one mentioning the IMPACT model on page 6). However, after (admittedly a rather quick) literature search, I found myself mistaken. I did not find documentation of aspects that I more or less took for granted, at least not where I had expected to find it. In case the authors have more luck and/or patience, I think they could include a brief discussion of existing literature that is relevant to the topic beyond what is included in the method section the second paragraph of Section 2.2.1. My own brief literature research suggested to me that this manuscript is very timely and that it would require some effort to find studies that are directly relevant for this study should they exist. I do not think that it would be worthwhile to include a detailed discussion of other references only to show that some existing literature dealt with rather different aspects of process splitting in order to motivate this study. However, simply mentioning that some other studies have dealt with process splitting might be worthwhile. https://doi.org/10.1016/S0377-0427(99)00143-0 could perhaps serve as one starting point regarding the operator splitting literature, and there might be other good starting points. I also found https://doi.org/10.1029/2018MS001418 interesting, which the authors cited in a previous work on time step convergence, although it is not directly related to the topic of this study. Especially in case the authors do find papers which are more closely related than the ones suggested here, a brief discussion would of course be interesting.2. 2nd paragraph of Section 2.2.1 starting from "While ...": Perhaps this could be woven into the introduction?
3. e.g., Gong et al., 2003; Stier et al., 2005; Mann et al., 2010; Zhang et al., 2010 -> It would be interesting to know how process splitting was handled in these cases. In case the information is hard to find without consulting the codes, the authors could consider mentioning that they did not find the information. In the case of Zhang et al. the authors may be able to comment based on the code without first having to download it, but citing studies or documentation would be preferred.
4. First line of Sect. 2.3: I am not sure simply "timestep" is the right word here. The timestep for advection is 5 minutes. A 30 minutes time step at 1° horizontal resolution seems incompatible with the Courant–Friedrichs–Lewy condition.
5. I think that the long time step of 30 minutes for which dry removal is computed may affect the results. I thought that such long time steps were mainly used for radiation. Can you comment on this, perhaps somewhere in the discussion of your results? I do not expect additional sensitivity studies here. A very brief discussion in one or two sentences would suffice.
6. The decreasing sensitivity to the thickness of the lowest layer is briefly discussed in Sect. 4.3. I think that although the discussion is brief, it provides sufficient detail. In my opinion this result has important practical implications and also provides strong support and motivation for adopting the new coupling and/or for further improving on this simple approach. I would suggest to repeat this important result in Section 5 (Summary, conclusions, and outlook).
Other suggestions:
P. 1: to after dry removal and before turbulent mixing -> to before turbulent mixing and after dry removal (Please choose which version you find easier to read. I like the second version better because it emphasizes that in the revised scheme, the effect of turbulent mixing is computed prior to dry removal in the next time step.)
P. 9: I suggest to omit "stored in the Fortran variable cam_in%cflx" and also "those additional mixing ratio values were included in the output files under different variable names following the convention described in Wan et al. (2022)" on Page 10. This sounds like something that can better be included in a user manual or so.P. 13 To get an overview -> To obtain an overview
P. 13 to represent regions ... ( i.e. the remote regions) -> as an example of a region ... (i.e. a remote region)
P. 13: averaged over ... remote regions -> averaged over ... remote region
P. 16 mush -> much
P. 16: I suggest to remove or replace the references to cam_in%flx, on this page and also on the following pages.
P. 18: Are the authors aware of a model in which this approach has been taken? How did Zhang et al. (2012) handle this?
P. 18: Box 3 in Fig. 1 is moved to the location indicated by the pink box with dashed outline, after box 6 (deep convection) and before box 7 (in which turbulent mixing is calculated) -> perhaps just indicate the processes and point to Fig. 1. You mentioned the moving of the boxes in the end of Sect. 2.3, and this sentence struck me as a repetition.
P. 20: The global impacts -> Global impacts
P. 21: The first row ... <- sounds like a repetition of the figure caption. I think it can be omitted.
P. 21: "the location marked as 3" -> I think you could omit the first two sentences of Sect. 4.2 and instead simply say that the process splitting was modified not only for dust as described in Sects. 2.3, but also for other species.
P. 25: situation gets worse -> situation becomes worse Or: the problem is exacerbated
Figure 1: Perhaps add "see caption" to the box between 6 and 7. I first thought about suggesting to simplify this figure. But looking for concrete suggestions, I came to the conclusion that it is actually nice that the authors included several boxes for which the relevance is understandable on second thought.Figure 9d: Wouldn't it be better to use a more linear color scale in Figure 9d?
Citation: https://doi.org/10.5194/egusphere-2023-1330-RC1 -
AC1: 'Reply on RC1', Hui Wan, 27 Oct 2023
On behalf of all co-authors, I thank the referee for their appreciation of the value of the work and for the constructive comments that will help improve the manuscript. We will include a detailed point-by-point response and a list of the changes in the manuscript when submitting a revised manuscript. Here is a summary of our overall responses to the referee’s comments.
Our impression of what exists and what is hard to find in the literature on process splitting is consistent with the referee’s comments. Within the atmosphere sciences, the chemistry transport community has paid a substantial amount of attention to the accuracy and stability of time integration methods, including process splitting. But in global weather, climate, and Earth system models (including aerosol-climate models), when it comes to the coupling among the various parameterized physics as well as the coupling of parameterizations with the resolved fluid dynamics, detailed documentation on the splitting methods being used and concrete explanations of the reasoning behind those choices are hard to find. Like pointed out by referee 2 and by the review paper of Gross et al. (2018, MWR, doi: 10.1175/MWR-D-17-0345.1), process splitting/coupling has been a largely overlooked topic. We fully agree with referee 1 that the results presented in the manuscript are not surprising, and we expect many readers will, like the referee, find the relative correctness of the revised coupling scheme rather obvious. On the other hand, the feedback we received from a number of aerosol physicists and modelers was that the numerical problem was not obvious to them until we explained the details. Our hope is that publishing concrete and easy-to-understand examples like ours in peer-reviewed journals will help remind climate model developers of the potentially dominant role of splitting errors and thereby help reduce the chance of “obviously” suboptimal coupling methods being used in the future, especially when new and/or more sophisticated process representations are brought in as new modules or packages. In the revised manuscript, we will add comments in the introduction section to point to work on process splitting in chemistry transport modeling and other communities and mention the general need for more attention to process splitting errors in climate models. In section 2, we will include comments on our understanding of how emissions and turbulent mixing are coupled in other aerosol-climate models and in the predecessors of EAMv1.
The referee expressed interest in a follow-up study to check if and how much the revise coupling scheme affects ERFari+aci. We have conducted pairs of simulations using the present-day and preindustrial emissions of anthropogenic aerosols and precursors, and we found the impacts on ERFari+aci to be very small. Since anthropogenic aerosols are mostly submicron species, and since our results presented in the manuscript show that the revision in process coupling has relatively small impacts on submicron species, we think the small responses in ERFari+aci are understandable.
The referee asked about the use of long timesteps of 30 minutes for aerosol dry removal in EAM. We note that the dry removal equations are numerically solved with a semi-Lagrangian scheme from Rasch and Lawrence (1998) to achieve reasonable stability and accuracy. Our study presented here focused on process coupling and did not investigate the impact of time integration methods used for individual processes. In EAMv1 (and EAMv2 as well as the candidate configuration for v3), not only dry removal, but also various other aerosol processes, including aerosol microphysics and its coupling to gas-phase chemistry, are integrated using 30-minute timesteps. The accuracies of time integration in these processes and their coupling are worth evaluating in the future. We plan to add these comments to the revised manuscript.
The referee had various specific suggestions on improving the wording and figures. We appreciate these suggestions and will revise the manuscript accordingly.
References
Gross, M., and Coauthors, 2018: Physics–Dynamics Coupling in Weather, Climate, and Earth System Models: Challenges and Recent Progress. Mon. Wea. Rev., 146, 3505–3544, https://doi.org/10.1175/MWR-D-17-0345.1
Rasch, P. J., and M. Lawrence, 1998: Recent development in transport methods at NCAR. In MPI Report No. 265, The MPI Workshop on Conservative Transport Schemes, pp. 65-75, Max-Planck-Inst. fuer Meteorol., Hamburg, Germany. https://pure.mpg.de/pubman/faces/ViewItemOverviewPage.jsp?itemId=item_3186487
Citation: https://doi.org/10.5194/egusphere-2023-1330-AC1
-
AC1: 'Reply on RC1', Hui Wan, 27 Oct 2023
-
RC2: 'Knowledge worth spreading', Anonymous Referee #2, 05 Sep 2023
General comments
Wan et al. describe how a reordering of aerosol processing computations in EAMv1 reduces the sensitivity of the model's global dust cycle to vertical resolution. Due to the small vertical extent of the lowermost model layer in the 72-level resolution in connection with the isolated treatment of the surface emissions flux, intra-timestep mixing ratio values become very large, and thus the whole budget strongly depends on the process that is computed next. In the current version, this is dry deposition, which is thus strongly overestimated in source regions, at the expense of other processes, specifically long-range transport and wet removal. Moving the emissions computation directly before the turbulent transport treatment, the authors obtain a dust budget that is less sensitive to the vertical extent of the lowermost model layer, and thus less sensitive to the vertical model resolution.
I agree with the first referee that the manuscript is very well written, and also find a lack of (quickly accessible) literature on the topic of "operator splitting" in the atmospheric aerosol modeling context. It is certainly worthwile to publish also such "overlooked" issues (as has recently also been done by Kawai et al., 10.1029/2022ms003128, for instance). In my opinion, the authors have done a very good job in thoroughly testing their suggested changes, and in documenting the tests and their results. For a publication in GMD, however, I recommend one of the following revisions.
When focusing purely on the vertical resolution (or lowermost layer thickness) sensitivity, i.e., on a rather technical problem and its solution, I suggest to considerably shorten the paper. In my opinion, it would be sufficient in this case to state and explain the expectations, and to provide no more than, say, 3-5 comparison figures and/or tables as proof. My feeling would be that it should be possible, without sacrificing scientific rigor, to condense the content for a target audience of aerosol modellers into a text that is at most half as long.
If the main goal is to describe an improvement of the model, I would still recommend some shortening (maybe even the same), but I would also request comparisons with observations, e.g., in analogy to the predecessor publication by Feng et al. (10.1029/2021MS002909). After all, an improvement in our simulation abilities can only be proven by better agreement with observations.
Specific comments
There are two questions that I would like to see addressed in addition to the current discussion (no more than a sentence or two necessary):
- Could the problem also be tackled with a reduction of the model time step, or sub-time steps for the aerosol processes?
- How do the differences between the simulations with the same vertical resolutions compare with interannual variability?
In some places, I found the distinction between "source regions" and "non-source regions" a bit difficult, as I was unsure if or when the time dimension was included in this distinction, i.e., does a grid cell always belong to a source region, or only during time steps in which emissions actually occur?
Some even more specific comments as a list:
- Abstract: "The revised scheme [...] better resembles the dust life cycle in the real world." -> This confused me in the beginning, as to me the "natural" sequence would seem to be first emissions, then mixing, then removal. Only much later I guessed from Fig. 1 that moving mixing between emissions and removal would probably not be a simple task for coding reasons. This should be clarified here.
- Introduction: The authors cite very few "state-of-the-art" aerosol-climate models in a few places. I suggest to either extend the reference lists in these places or cite a more general publication, like a book, a review or a model intercomparison paper.
- p. 3: "the revised coupling provides better results" -> I suggest to be more specific about the "better" here. It may refer to better in a numerical sense, which the companion paper shall demonstrate. If it is intended to refer to agreement with observations, I would request some evidence for this assertion.
- p. 10: If nudging leads to a different surface wind climatology, is that not an argument _for_ the nudging, rather than against it? Furthermore, should the emissions not be driven by the "same" winds for this experiment?
- Fig. 3: A comment on the discontinuities in the lowest 2 - 3 layers should be added.
- Sect. 3.1:
- What is the motivation for the selection of the time range for the histogram? Is this representative of the whole year?
- What is the motivation for the distinction between the three "portions" of the histogram, and for the values of their borders?
- Fig. 12: The reasons for the reductions seen in parts of the zonal mean POA distribution should be explained.
In the interest of reproducibility, it might be worthwile to publish the scripts to create the figures and table data along with the model output.
Technical corrections
- p. 1: "tuning parameters" -> I assume this refers to parameters in the emissions computations. This should be stated explicitly, as "tuning" may otherwise be understood as tuning the radiation balance of the model.
- p. 7: The title of Sect. 2.2.3 could include "activation/resuspension".
- p. 9: "tracers (not including water vapor) stored in the Fortran variable cam_in%cflx" -> These should be listed or described, see also comments by the first referee.
- Table 3: I suggest to remove this table. If not, it might be useful to give the sequences of process names/abbreviations instead of, or in addition to, "Original" and "Revised".
- p. 10: "the Earth surface" -> "the Earth's surface" (again on p. 13)
- Fig. 3: As there is only one point with non-zero emissions, I suggest to remove the line "Emissions" frome the plots in panels (c) - (h).
- p. 13: "The upper row is the results" -> "The upper row shows the results"
- p. 13: "orders stronger" -> "orders of magnitude stronger"
- p. 15: "turbulence mixing" -> "turbulent mixing"
- Fig. 7, in my opinion, is redundant after Fig. 2.
- p. 18: "motivates" -> "motivate"
- Table 4:
- There may be an error in the units specification for the first four data rows. Probably, the given numbers refer to a flux per unit area? If so, I suggest to integrate them over the respective areas for a more intuitive presentation (and to adapt the caption and Table 5 accordingly).
- Suggestion: the emissions could also be included here for completeness, and for direct comparison.
- I stumbled upon "vertically integrated dry and wet removal". Is this the (net) removal from the lowermost layer, i.e., what actually leaves the model domain
- p. 23: "reference differences" -> "relative differences"
- Fig. A2: The portion of the histogram in Fig. 4 that corresponds to the data shown here should be specified.
- The bibliography should be groomed. Some of the links include a double "https://doi.org/", for instance, and I also noticed a "n/a".
- For future publications, I suggest a replacement of rainbow colors by a color scale that can be interpreted more easily by people with color vision deficiency, e.g., "cividis" (10.1371/journal.pone.0199239). (I noticed after writing this that Copernicus requested it already.)
Citation: https://doi.org/10.5194/egusphere-2023-1330-RC2 -
AC2: 'Reply on RC2', Hui Wan, 27 Oct 2023
We thank the referee for their careful and insightful review and the constructive comments that will help improve the manuscript. We will include a detailed point-by-point response and a list of the changes in the manuscript when submitting a revised manuscript. Here is a summary of our overall responses to the referee’s comments.
The major comment from the referee was to either focus on the numerical problem and significantly shorten the manuscript or to include comparisons with observations. We understand the referee’s point about being clear in scope and being concise in writing, hence will try to shorten the manuscript. We agree with both referees that the main value of the manuscript is to serve as a reminder of an overlooked issue. For the purpose of having a stronger reminder, we prefer to keep some of the contents that might seem redundant to some readers (e.g., annual and regional mean versus instantaneous budgets; Fig. 2 and Fig. 7) while we try to make the manuscript more compact.
We prefer to limit the scope of the manuscript to a discussion on numerical error without including comparisons with observations, because discrepancies between simulations and observations can result from multiple error sources including not only the numerical algorithms but also the formulation of the model equations, the values of the uncertain parameters, etc. The different types of errors might compensate each other, hence reducing one type of error might temporarily degrade the agreement with observations. We do note that the revised coupling scheme has been included in the candidate configuration of EAMv3; some of the colleagues on our author list will present comparisons with observations in a separate paper in the context of evaluating EAMv3.
The referee asked whether the numerical coupling problem could also be tackled with a reduction of the model time step or sub-time steps for the aerosol processes. Our answer is yes, if we reduce the overall model step or sub-cycle the aerosol processes together so that the different processes exchange information at shorter timesteps, then the splitting error will be reduced. However, both approaches have an important practical disadvantage, namely increased computational cost. We will add a comment on this in the revised manuscript.
The referee asked how the differences between the simulations with the same vertical resolution compare with interannual variability. We have conducted multi-year simulations and will briefly mention the results in the revised manuscript.
The referee suggested replacing rainbow colormaps. We admit that we have only very recently started to appreciate both the importance of CVD-friendly colormaps and the challenges associated with them. For the revised manuscript, we plan to use CVD-friendly colormaps for all contour plots and use different line and mark styles in line plots to help make the figures more perceptible by readers with CVD.
The referee also posed additional specific questions or suggestions, which we will address during the revision of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2023-1330-AC2
Interactive discussion
Status: closed
-
RC1: 'Comment on egusphere-2023-1330', Anonymous Referee #1, 22 Aug 2023
Wan et al. show that in an atmosphere model at a 1° horizontal resolution, a coupling of surface emissions to the planetary boundary layer turbulence parameterization becomes even more important with increasing vertical resolution. I find this result not only plausible, but expected. What seems surprising to me is rather that in standard EAMv1, the vertical mixing of emissions prior to dry deposition was entirely determined by the thickness of the lowest model layer. As far as I know, global atmosphere models (sometimes with a vertical resolution on the order of 100m near the surface) usually assume instantaneous vertical mixing of surface emissions extending at least across the height of the lowest model layer (perhaps excluding point sources in a few models). But up to now, I also thought that many models either already couple surface fluxes and surface emissions to the planetary boundary layer turbulence parameterization (for example similar to what is mentioned for the IMPACT model on page 6) or else call the turbulence parameterization right after the emissions or surface flux computation, as suggested in the present manuscript.
The problem with the instantaneous artificial vertical mixing across the height of the lowest model layer is that, unlike parameterized mixing, it is not based on physical arguments and independent of local stability and the actual boundary layer height, which varies in space and time including a dependence on surface type, latitude, and a prominent diurnal cycle over land. Although this artificial mixing in the lowest layer may to some extent mimic boundary layer mixing in low-resolution models, I think that it is definitely not desirable. This artificial mixing can however, be reduced by increasing vertical resolution as far as numerical stability permits. A proper coupling of surface fluxes to the planetary boundary layer turbulence parameterization seems advisable in any case, and becomes even more important as the vertical resolution near the surface increases. My guess is that even at very high resolution, the process splitting discussed here may still be important as long as vertical resolution near the surface decreases proportionally.
On the one hand, a large number of well known papers on process splitting exists especially in the chemistry transport community (see minor point #1 below for an example), and perhaps elsewhere. The treatment of surface emissions, especially from point sources, has long been a concern in coarse resolution chemistry transport and aerosol models as well. For point sources, various approaches for addressing the problem of finding the right emission height and also of instantaneous mixing because of finite grid resolution have been suggested and tested, such as plume-in-grid models. Furthermore, as mentioned above, some atmosphere models may already couple emissions to the planetary boundary layer turbulence parameterization or use the process splitting suggested here.
However, based on a quick literature search, it seems to me that the publication of the manuscript by Wan et al. would be important and timely. I did not find anything concrete on this topic (although I must admit that I spent a limited time on my literature research). Because of the major effect the process splitting has on the results, the importance of the results presented by Wan et al. for the general progress in the development of EAM seem obvious to me. I would also expect the general result to apply at least to some other models as well.
I find that the manuscript by Wan et al. is exceptionally well written, and I have only a few minor comments. While some modellers may not be overly surprised by the findings, I think that documenting this type of sensitivity and especially also suggesting a solution is extremely useful. Although I partially agree with the authors on their more cautious statements in the introduction, which suggest to me that one should think in some depth about these issues, the relative correctness of the solution suggested by the authors seems rather obvious to me.
I would very much appreciate if the authors would consider a follow-up study to check if and how much this revision affects ERFari+aci
Minor points:
1. I generally think that it is good practice to discuss the results of any given study in the context of existing literature. Therefore, at least at first, it seemed to me that the authors could have mentioned and perhaps included a brief survey of the existing literature on process (or operator) splitting in the introduction. Initially, I also thought it would be rather easy for me to suggest a few references that are of direct relevance to the issue at hand (such as the one mentioning the IMPACT model on page 6). However, after (admittedly a rather quick) literature search, I found myself mistaken. I did not find documentation of aspects that I more or less took for granted, at least not where I had expected to find it. In case the authors have more luck and/or patience, I think they could include a brief discussion of existing literature that is relevant to the topic beyond what is included in the method section the second paragraph of Section 2.2.1. My own brief literature research suggested to me that this manuscript is very timely and that it would require some effort to find studies that are directly relevant for this study should they exist. I do not think that it would be worthwhile to include a detailed discussion of other references only to show that some existing literature dealt with rather different aspects of process splitting in order to motivate this study. However, simply mentioning that some other studies have dealt with process splitting might be worthwhile. https://doi.org/10.1016/S0377-0427(99)00143-0 could perhaps serve as one starting point regarding the operator splitting literature, and there might be other good starting points. I also found https://doi.org/10.1029/2018MS001418 interesting, which the authors cited in a previous work on time step convergence, although it is not directly related to the topic of this study. Especially in case the authors do find papers which are more closely related than the ones suggested here, a brief discussion would of course be interesting.2. 2nd paragraph of Section 2.2.1 starting from "While ...": Perhaps this could be woven into the introduction?
3. e.g., Gong et al., 2003; Stier et al., 2005; Mann et al., 2010; Zhang et al., 2010 -> It would be interesting to know how process splitting was handled in these cases. In case the information is hard to find without consulting the codes, the authors could consider mentioning that they did not find the information. In the case of Zhang et al. the authors may be able to comment based on the code without first having to download it, but citing studies or documentation would be preferred.
4. First line of Sect. 2.3: I am not sure simply "timestep" is the right word here. The timestep for advection is 5 minutes. A 30 minutes time step at 1° horizontal resolution seems incompatible with the Courant–Friedrichs–Lewy condition.
5. I think that the long time step of 30 minutes for which dry removal is computed may affect the results. I thought that such long time steps were mainly used for radiation. Can you comment on this, perhaps somewhere in the discussion of your results? I do not expect additional sensitivity studies here. A very brief discussion in one or two sentences would suffice.
6. The decreasing sensitivity to the thickness of the lowest layer is briefly discussed in Sect. 4.3. I think that although the discussion is brief, it provides sufficient detail. In my opinion this result has important practical implications and also provides strong support and motivation for adopting the new coupling and/or for further improving on this simple approach. I would suggest to repeat this important result in Section 5 (Summary, conclusions, and outlook).
Other suggestions:
P. 1: to after dry removal and before turbulent mixing -> to before turbulent mixing and after dry removal (Please choose which version you find easier to read. I like the second version better because it emphasizes that in the revised scheme, the effect of turbulent mixing is computed prior to dry removal in the next time step.)
P. 9: I suggest to omit "stored in the Fortran variable cam_in%cflx" and also "those additional mixing ratio values were included in the output files under different variable names following the convention described in Wan et al. (2022)" on Page 10. This sounds like something that can better be included in a user manual or so.P. 13 To get an overview -> To obtain an overview
P. 13 to represent regions ... ( i.e. the remote regions) -> as an example of a region ... (i.e. a remote region)
P. 13: averaged over ... remote regions -> averaged over ... remote region
P. 16 mush -> much
P. 16: I suggest to remove or replace the references to cam_in%flx, on this page and also on the following pages.
P. 18: Are the authors aware of a model in which this approach has been taken? How did Zhang et al. (2012) handle this?
P. 18: Box 3 in Fig. 1 is moved to the location indicated by the pink box with dashed outline, after box 6 (deep convection) and before box 7 (in which turbulent mixing is calculated) -> perhaps just indicate the processes and point to Fig. 1. You mentioned the moving of the boxes in the end of Sect. 2.3, and this sentence struck me as a repetition.
P. 20: The global impacts -> Global impacts
P. 21: The first row ... <- sounds like a repetition of the figure caption. I think it can be omitted.
P. 21: "the location marked as 3" -> I think you could omit the first two sentences of Sect. 4.2 and instead simply say that the process splitting was modified not only for dust as described in Sects. 2.3, but also for other species.
P. 25: situation gets worse -> situation becomes worse Or: the problem is exacerbated
Figure 1: Perhaps add "see caption" to the box between 6 and 7. I first thought about suggesting to simplify this figure. But looking for concrete suggestions, I came to the conclusion that it is actually nice that the authors included several boxes for which the relevance is understandable on second thought.Figure 9d: Wouldn't it be better to use a more linear color scale in Figure 9d?
Citation: https://doi.org/10.5194/egusphere-2023-1330-RC1 -
AC1: 'Reply on RC1', Hui Wan, 27 Oct 2023
On behalf of all co-authors, I thank the referee for their appreciation of the value of the work and for the constructive comments that will help improve the manuscript. We will include a detailed point-by-point response and a list of the changes in the manuscript when submitting a revised manuscript. Here is a summary of our overall responses to the referee’s comments.
Our impression of what exists and what is hard to find in the literature on process splitting is consistent with the referee’s comments. Within the atmosphere sciences, the chemistry transport community has paid a substantial amount of attention to the accuracy and stability of time integration methods, including process splitting. But in global weather, climate, and Earth system models (including aerosol-climate models), when it comes to the coupling among the various parameterized physics as well as the coupling of parameterizations with the resolved fluid dynamics, detailed documentation on the splitting methods being used and concrete explanations of the reasoning behind those choices are hard to find. Like pointed out by referee 2 and by the review paper of Gross et al. (2018, MWR, doi: 10.1175/MWR-D-17-0345.1), process splitting/coupling has been a largely overlooked topic. We fully agree with referee 1 that the results presented in the manuscript are not surprising, and we expect many readers will, like the referee, find the relative correctness of the revised coupling scheme rather obvious. On the other hand, the feedback we received from a number of aerosol physicists and modelers was that the numerical problem was not obvious to them until we explained the details. Our hope is that publishing concrete and easy-to-understand examples like ours in peer-reviewed journals will help remind climate model developers of the potentially dominant role of splitting errors and thereby help reduce the chance of “obviously” suboptimal coupling methods being used in the future, especially when new and/or more sophisticated process representations are brought in as new modules or packages. In the revised manuscript, we will add comments in the introduction section to point to work on process splitting in chemistry transport modeling and other communities and mention the general need for more attention to process splitting errors in climate models. In section 2, we will include comments on our understanding of how emissions and turbulent mixing are coupled in other aerosol-climate models and in the predecessors of EAMv1.
The referee expressed interest in a follow-up study to check if and how much the revise coupling scheme affects ERFari+aci. We have conducted pairs of simulations using the present-day and preindustrial emissions of anthropogenic aerosols and precursors, and we found the impacts on ERFari+aci to be very small. Since anthropogenic aerosols are mostly submicron species, and since our results presented in the manuscript show that the revision in process coupling has relatively small impacts on submicron species, we think the small responses in ERFari+aci are understandable.
The referee asked about the use of long timesteps of 30 minutes for aerosol dry removal in EAM. We note that the dry removal equations are numerically solved with a semi-Lagrangian scheme from Rasch and Lawrence (1998) to achieve reasonable stability and accuracy. Our study presented here focused on process coupling and did not investigate the impact of time integration methods used for individual processes. In EAMv1 (and EAMv2 as well as the candidate configuration for v3), not only dry removal, but also various other aerosol processes, including aerosol microphysics and its coupling to gas-phase chemistry, are integrated using 30-minute timesteps. The accuracies of time integration in these processes and their coupling are worth evaluating in the future. We plan to add these comments to the revised manuscript.
The referee had various specific suggestions on improving the wording and figures. We appreciate these suggestions and will revise the manuscript accordingly.
References
Gross, M., and Coauthors, 2018: Physics–Dynamics Coupling in Weather, Climate, and Earth System Models: Challenges and Recent Progress. Mon. Wea. Rev., 146, 3505–3544, https://doi.org/10.1175/MWR-D-17-0345.1
Rasch, P. J., and M. Lawrence, 1998: Recent development in transport methods at NCAR. In MPI Report No. 265, The MPI Workshop on Conservative Transport Schemes, pp. 65-75, Max-Planck-Inst. fuer Meteorol., Hamburg, Germany. https://pure.mpg.de/pubman/faces/ViewItemOverviewPage.jsp?itemId=item_3186487
Citation: https://doi.org/10.5194/egusphere-2023-1330-AC1
-
AC1: 'Reply on RC1', Hui Wan, 27 Oct 2023
-
RC2: 'Knowledge worth spreading', Anonymous Referee #2, 05 Sep 2023
General comments
Wan et al. describe how a reordering of aerosol processing computations in EAMv1 reduces the sensitivity of the model's global dust cycle to vertical resolution. Due to the small vertical extent of the lowermost model layer in the 72-level resolution in connection with the isolated treatment of the surface emissions flux, intra-timestep mixing ratio values become very large, and thus the whole budget strongly depends on the process that is computed next. In the current version, this is dry deposition, which is thus strongly overestimated in source regions, at the expense of other processes, specifically long-range transport and wet removal. Moving the emissions computation directly before the turbulent transport treatment, the authors obtain a dust budget that is less sensitive to the vertical extent of the lowermost model layer, and thus less sensitive to the vertical model resolution.
I agree with the first referee that the manuscript is very well written, and also find a lack of (quickly accessible) literature on the topic of "operator splitting" in the atmospheric aerosol modeling context. It is certainly worthwile to publish also such "overlooked" issues (as has recently also been done by Kawai et al., 10.1029/2022ms003128, for instance). In my opinion, the authors have done a very good job in thoroughly testing their suggested changes, and in documenting the tests and their results. For a publication in GMD, however, I recommend one of the following revisions.
When focusing purely on the vertical resolution (or lowermost layer thickness) sensitivity, i.e., on a rather technical problem and its solution, I suggest to considerably shorten the paper. In my opinion, it would be sufficient in this case to state and explain the expectations, and to provide no more than, say, 3-5 comparison figures and/or tables as proof. My feeling would be that it should be possible, without sacrificing scientific rigor, to condense the content for a target audience of aerosol modellers into a text that is at most half as long.
If the main goal is to describe an improvement of the model, I would still recommend some shortening (maybe even the same), but I would also request comparisons with observations, e.g., in analogy to the predecessor publication by Feng et al. (10.1029/2021MS002909). After all, an improvement in our simulation abilities can only be proven by better agreement with observations.
Specific comments
There are two questions that I would like to see addressed in addition to the current discussion (no more than a sentence or two necessary):
- Could the problem also be tackled with a reduction of the model time step, or sub-time steps for the aerosol processes?
- How do the differences between the simulations with the same vertical resolutions compare with interannual variability?
In some places, I found the distinction between "source regions" and "non-source regions" a bit difficult, as I was unsure if or when the time dimension was included in this distinction, i.e., does a grid cell always belong to a source region, or only during time steps in which emissions actually occur?
Some even more specific comments as a list:
- Abstract: "The revised scheme [...] better resembles the dust life cycle in the real world." -> This confused me in the beginning, as to me the "natural" sequence would seem to be first emissions, then mixing, then removal. Only much later I guessed from Fig. 1 that moving mixing between emissions and removal would probably not be a simple task for coding reasons. This should be clarified here.
- Introduction: The authors cite very few "state-of-the-art" aerosol-climate models in a few places. I suggest to either extend the reference lists in these places or cite a more general publication, like a book, a review or a model intercomparison paper.
- p. 3: "the revised coupling provides better results" -> I suggest to be more specific about the "better" here. It may refer to better in a numerical sense, which the companion paper shall demonstrate. If it is intended to refer to agreement with observations, I would request some evidence for this assertion.
- p. 10: If nudging leads to a different surface wind climatology, is that not an argument _for_ the nudging, rather than against it? Furthermore, should the emissions not be driven by the "same" winds for this experiment?
- Fig. 3: A comment on the discontinuities in the lowest 2 - 3 layers should be added.
- Sect. 3.1:
- What is the motivation for the selection of the time range for the histogram? Is this representative of the whole year?
- What is the motivation for the distinction between the three "portions" of the histogram, and for the values of their borders?
- Fig. 12: The reasons for the reductions seen in parts of the zonal mean POA distribution should be explained.
In the interest of reproducibility, it might be worthwile to publish the scripts to create the figures and table data along with the model output.
Technical corrections
- p. 1: "tuning parameters" -> I assume this refers to parameters in the emissions computations. This should be stated explicitly, as "tuning" may otherwise be understood as tuning the radiation balance of the model.
- p. 7: The title of Sect. 2.2.3 could include "activation/resuspension".
- p. 9: "tracers (not including water vapor) stored in the Fortran variable cam_in%cflx" -> These should be listed or described, see also comments by the first referee.
- Table 3: I suggest to remove this table. If not, it might be useful to give the sequences of process names/abbreviations instead of, or in addition to, "Original" and "Revised".
- p. 10: "the Earth surface" -> "the Earth's surface" (again on p. 13)
- Fig. 3: As there is only one point with non-zero emissions, I suggest to remove the line "Emissions" frome the plots in panels (c) - (h).
- p. 13: "The upper row is the results" -> "The upper row shows the results"
- p. 13: "orders stronger" -> "orders of magnitude stronger"
- p. 15: "turbulence mixing" -> "turbulent mixing"
- Fig. 7, in my opinion, is redundant after Fig. 2.
- p. 18: "motivates" -> "motivate"
- Table 4:
- There may be an error in the units specification for the first four data rows. Probably, the given numbers refer to a flux per unit area? If so, I suggest to integrate them over the respective areas for a more intuitive presentation (and to adapt the caption and Table 5 accordingly).
- Suggestion: the emissions could also be included here for completeness, and for direct comparison.
- I stumbled upon "vertically integrated dry and wet removal". Is this the (net) removal from the lowermost layer, i.e., what actually leaves the model domain
- p. 23: "reference differences" -> "relative differences"
- Fig. A2: The portion of the histogram in Fig. 4 that corresponds to the data shown here should be specified.
- The bibliography should be groomed. Some of the links include a double "https://doi.org/", for instance, and I also noticed a "n/a".
- For future publications, I suggest a replacement of rainbow colors by a color scale that can be interpreted more easily by people with color vision deficiency, e.g., "cividis" (10.1371/journal.pone.0199239). (I noticed after writing this that Copernicus requested it already.)
Citation: https://doi.org/10.5194/egusphere-2023-1330-RC2 -
AC2: 'Reply on RC2', Hui Wan, 27 Oct 2023
We thank the referee for their careful and insightful review and the constructive comments that will help improve the manuscript. We will include a detailed point-by-point response and a list of the changes in the manuscript when submitting a revised manuscript. Here is a summary of our overall responses to the referee’s comments.
The major comment from the referee was to either focus on the numerical problem and significantly shorten the manuscript or to include comparisons with observations. We understand the referee’s point about being clear in scope and being concise in writing, hence will try to shorten the manuscript. We agree with both referees that the main value of the manuscript is to serve as a reminder of an overlooked issue. For the purpose of having a stronger reminder, we prefer to keep some of the contents that might seem redundant to some readers (e.g., annual and regional mean versus instantaneous budgets; Fig. 2 and Fig. 7) while we try to make the manuscript more compact.
We prefer to limit the scope of the manuscript to a discussion on numerical error without including comparisons with observations, because discrepancies between simulations and observations can result from multiple error sources including not only the numerical algorithms but also the formulation of the model equations, the values of the uncertain parameters, etc. The different types of errors might compensate each other, hence reducing one type of error might temporarily degrade the agreement with observations. We do note that the revised coupling scheme has been included in the candidate configuration of EAMv3; some of the colleagues on our author list will present comparisons with observations in a separate paper in the context of evaluating EAMv3.
The referee asked whether the numerical coupling problem could also be tackled with a reduction of the model time step or sub-time steps for the aerosol processes. Our answer is yes, if we reduce the overall model step or sub-cycle the aerosol processes together so that the different processes exchange information at shorter timesteps, then the splitting error will be reduced. However, both approaches have an important practical disadvantage, namely increased computational cost. We will add a comment on this in the revised manuscript.
The referee asked how the differences between the simulations with the same vertical resolution compare with interannual variability. We have conducted multi-year simulations and will briefly mention the results in the revised manuscript.
The referee suggested replacing rainbow colormaps. We admit that we have only very recently started to appreciate both the importance of CVD-friendly colormaps and the challenges associated with them. For the revised manuscript, we plan to use CVD-friendly colormaps for all contour plots and use different line and mark styles in line plots to help make the figures more perceptible by readers with CVD.
The referee also posed additional specific questions or suggestions, which we will address during the revision of the manuscript.
Citation: https://doi.org/10.5194/egusphere-2023-1330-AC2
Peer review completion
Journal article(s) based on this preprint
Data sets
EAMv1 output from simulations using tag v1_cflx_2021: annual averages Hui Wan and Kai Zhang https://doi.org/10.5281/zenodo.7996742
EAMv1 output from simulations using tag v1_cflx_2021: instantaneous values Hui Wan and Kai Zhang https://doi.org/10.5281/zenodo.8000745
Model code and software
EAMv1 code with revised aerosol process coupling (tag v1_cflx_2021) Hui Wan https://doi.org/10.5281/zenodo.7995850
Viewed
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 178 | 0 | 0 | 178 | 2 | 1 |
- HTML: 178
- PDF: 0
- XML: 0
- Total: 178
- BibTeX: 2
- EndNote: 1
Viewed (geographical distribution)
Since the preprint corresponding to this journal article was posted outside of Copernicus Publications, the preprint-related metrics are limited to HTML views.
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 101 | 56 |
| Germany | 2 | 36 | 20 |
| China | 3 | 4 | 2 |
| United Kingdom | 4 | 4 | 2 |
| India | 5 | 4 | 2 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 101
Kai Zhang
Christopher J. Vogl
Carol S. Woodward
Richard C. Easter
Philip J. Rasch
Hailong Wang
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.