Dear Dr. Wernli,

We graciously thank the reviewers for their feedback and ideas on this manuscript. We have taken many into account and believe that our manuscript is now both more clear and robust. 

Reviewer comments are shown in bold.

Author comments are in plain text.

Sincerely, 

A Waling and Coauthors

Multiple comments from our reviewers wondered about the -30,000 kg m⁻² s⁻¹ rad⁻² IVT threshold that we used during atmospheric river tracking, thus we choose to address our reasoning here. The -50k threshold of Patricola et al. (2020) and Rhoades et al. (2020) resulted in too few land-falling storms in Greenland. We therefore used a more lenient threshold of -30k that produces an order of magnitude more land-falling ARs, and is still a larger threshold than used in other TempestExtremes AR studies (Ullrich et al. 2020).

Some outlier source points occur in regions which are inconsistent with conventional definitions of ARs. As noted in the text, our tracker parameters are not as selective as used in other studies, and we have made that choice explicitly given the trade-offs between sample size and the strictness of tracker parameters (Line 123).

Patricola, C.M., O’Brien, J.P., Risser, M.D., Rhoades, A.M., O’Brien, T.A., Ullrich, P.A., Stone, D.A., & Collins, W.D. (2020). Maximizing ENSO as a source of western US hydroclimate predictability. Climate Dynamics, 54, 351-372.

Rhoades, A.M., Jones, A.D., O’Brien, T.A., O’Brien, J.P., Ullrich, P.A., & Zarzycki, C.M. (2020). Influences of North Pacific Ocean Domain Extent on the Western U.S. Winter Hydroclimatology in Variable-Resolution CESM. Journal of Geophysical Research: Atmospheres, 125.

For the reviewers: The attached figure shows the same field but using a more strict threshold for the Laplacian of IVT (-50,000 kg m-1 s-2 rad-2). The AR inception points are shifted west and are in closer proximity to the Atlantic Ocean, consistent with the higher IVT threshold for identifying ARs. However, the number of storms dropped significantly, along with the number of storms intersecting Greenland (not shown). Please see attached Figure X: Same as Figure 5 in the manuscript, but using stricter TempestExtremes tracking parameters. 

Reviewer 1: 

This study evaluates atmospheric rivers (ARs) in simulations with varying resolutions. In its current form, I find the paper to be very technical and difficult to follow, so a significant revision of the text is required in my opinion. One of the main results is also not surprising; the authors state that “We suggest that the higher and steeper topography resolved in VRs and the reanalyses prevent ARS from penetrating as far inland as the LL and QU grids”. This result is well known, as higher topography acts as a barrier to water vapor transport (IVT) and hence the inland penetration of AR features.

Per the comment regarding our manuscript’s technical nature, we have revised much of the Model Simulations methods section (Section 2.1). We have left out some of the more technical descriptions of CESM2.2 and have removed most abbreviations used in this section, aside from those deemed absolutely necessary. We have also described the usage of CLM5 and CAM6 within CESM2.2 more explicitly. 

Line 72: “Herrington et al. (2022) ran CESM2.2 simulations using six different grid configurations (Table 1, Figure 1), from 1 January 1979 to 31 December 1998. These include two latitude-longitude (LL) grids, two quasi-uniform unstructured grids (QU), and two variable-resolution (VR) grids. LL grid configurations use the finite-volume (FV) dynamical core and the QU grids use spectral-element (SE) dynamical cores. SE dynamics are solved with high-degree piecewise polynomials, yielding improved numerical accuracy in the horizontal compared with the FV dynamical cores. SE dynamical cores are ideal for high resolution modeling due to their improved computational efficiency on massively parallel systems and the inclusion of condensates that can greatly influence the dynamics of a system at high resolution (Bacmeister et al., 2012; Lauritzen et al., 2018). With their high computational efficiency, SE dynamical cores also support VR grids, including the two presented in our study (Table 1). 

CESM2.2 used the Community Atmosphere Model 6.3 (CAM6) (Craig et al., 2021) for its physics and atmosphere component and the Community Land Model 5.0 (CLM5) (Lawrence et al., 2019) for its land component. The variables used from CAM6 were convective precipitation rate and large-scale stable precipitation rate, which were summed to reach the total atmospheric precipitation. All CAM6 data used in this study was recorded at six-hourly intervals. The ERA5 precipitation variable is also total precipitation and MERRA2 is the bias corrected total precipitation. The IVT fields from the CAM6 simulations were used in AR detection (uIVT, vIVT).

 CLM5 was coupled to CAM6 and provided daily averaged precipitation…”

Additionally, we have added more information to the Abstract and Introduction that describe the importance of our work to the broader modeling community. For example:

Line 3: “In contrast with traditional modeling studies that rely on coarse, uniform-resolution grids, this project investigates the effectiveness of variable-resolution (VR) grids in modeling ARs and their subsequent precipitation around the GrIS using a study period of 1 January 1979 to 31 December 1998.”

Line 48: “This work will help the atmospheric community determine when the more computationally expensive VR grids are most useful, especially important given the limited in-situ observations available for quantifying the effects of atmospheric rivers over Greenland on precipitation and surface mass balance. The paper also details a replicable method for tracking ARs in the Atlantic Arctic region over a multi-decadal simulation, providing insight and guidance into the objective detection of ARs from model data.”

Finally, we have done a careful, in-depth revision that aims to clarify any technical elements that we chose to still include in our manuscript. For example, we found our two categories of precipitation discussed throughout the text (area-averaged cumulative precipitation and area-integrated cumulative precipitation) to be wordy and confusing at times. Thus, we shortened these two to precipitation rate and area-integrated precipitation, respectively, and use this terminology for clarity and easier reading .

Line 258: “Figure 9a shows the cumulative AR precipitation (hereafter, precipitation rate)...” and

Line 271: “Figure 9c compares the average area-integrated cumulative precipitation (hereafter, area-integrated precipitation...”

Per the comment regarding our main findings not being surprising, we provide further explanation of our argument here. Our study focuses on how well the models are capturing physical behaviors such as precipitation associated with orographic uplift. While there is extensive observation and modeling of ARs over the Pacific and California coast, the focus on ARs reaching Greenland is relatively new. To our knowledge, there have been no studies assessing the influence of topography on precipitation in atmospheric rivers reaching Greenland when considering different grid configurations. Though one would assume orographic uplift would be the main factor influencing this AR derived precipitation in Greenland, our study shows how conventional grid configurations frequently used in the modeling community perform compared to Variable Resolution (VR) grids. We hope to persuade future modeling studies to consider using VR grids as they have been shown in our study to better resolve the well-known dynamics that you described. In addition to this, the IPCC currently relies on coarse resolution Earth System Models to project ice sheet evolution and sea level rise. By understanding in what ways Earth System Models are currently under-performing, future models can improve sea level projections.

Line 20: “While there is extensive observation and modeling of ARs over the Pacific and California coast, the focus on ARs reaching Greenland is relatively new (Mattingly et al., 2018, 2020; Box et al. 2023; Kirbus et al., 2023; Mattingly et al., 2023).”

Line 94: “​​The question becomes how significant these topographic resolutions are to the modeled evolution of ARs and the associated precipitation.”

I did not read the Editor’s comments until after my review. I agree strongly with the Editor of the need for more physical interpretation of the results found, and possibly of case studies, especially of IVT and precipitation fields, to help the reader to follow your work.

We have moved the presentation of specific cases of individual atmospheric rivers from model simulations (Figure 9) as well as an analysis of how well the algorithm captures the areas of precipitation associated with ARs (Figures 11) from the Results to the Discussion section.  We have added text to better link the analysis of these figures to explanations of the discrepancies between VR grid simulations and reanalysis datasets as well as to recommend future considerations when choosing model grids and interpreting their results.

Line 308: “From the above results, VR grids most closely match reanalysis datasets in terms of number (Figure 7a), areal extent of ARs intersecting GriS (Figure 6 and Table 2), and area-averaged precipitation (Figure 9a), noting they still overestimate these metrics. VR grids show less agreement with reanalysis in the area of overlap with the GrIS (Figure 7d) and area-integrated cumulative precipitation (Figure 9c). To provide insight and further investigate the dynamics of modeled ARs, Figure 10 shows individual 95th percentile atmospheric rivers from the simulations.

Figure 10 shows snapshots from the models of the 95 percentile ARs near the time of their maximum overlap with Greenland, and the outline of the detected feature provided in blue. The detected feature represents the moist core of the AR, which relative to the larger synoptic system does not overlap with a large portion of land at any point throughout its lifecycle (Figure 7d). The snapshots indicate the warm front situated out ahead of the AR core contributes a substantial amount of the storm’s precipitation. Additionally large regions of precipitation occur just outside the detected core within the cold front. This begs the question: How much precipitation associated with ARs are not directly under the storm as defined by our detection algorithm? We investigated this dynamical element of the AR in Figure 11.”

We have also added phrases throughout to alert the reader of the dynamics associated with atmospheric rivers.

Line 19: “ARs originate in the low- to mid-latitudes from synoptic scale systems and subsequently travel poleward.”

Line 206: “Despite these outliers occurring at high latitudes, the majority of identified source regions are consistent with atmospheric rivers developing along mid-latitude storm tracks in relation to the baroclinic instability of extratropical cyclones.”

We have also clarified why a climatological approach is preferred over comparative case studies.

Line 53: “This study takes advantage of model output from the multi-decadal simulations…”

Line 181: “It is important to emphasize that CESM2.2 simulations are free-running, constrained by monthly sea-surface temperature and sea-ice extent but not by meteorological observations or reanalysis. We therefore present climatological comparisons among model configurations rather than case studies."

Please find some comments below which I hope are useful to you.

Line 20: “to the poles”. I would suggest rephrasing to either “polewards” or “across the mid-latitudes”.

We appreciate this feedback and have implemented it. 

Lines 20-25: There may be further information for the Introduction in the European State of the Climate Report (https://climate.copernicus.eu/esotc/2022/greenland-heatwaves).

Thank you for this reference, we have included information about the September 2022 atmospheric river in our introduction to help support the claim of increased frequency of extreme events affecting Greenland.

Line 33: “The GrIS experienced multiple major melt events in recent years, including in August 2021 which caused rainfall at Summit Station (Box et al. 2022)  and in September 2022 when at least 23% of the GrIS experienced surface melt (Copernicus Climate Change Service, 2023)” 

Introduction last 2 paragraphs: These do not set up the paper as well as could be expected. These paragraphs could be combined and revised to state clearly the aims or questions addressed in the study. I felt the first line of penultimate paragraph was past research, rather than this study.

We have consolidated and edited these two paragraphs to make them more concise and clear. 

Line 53: “This study takes advantage of model output from the multi-decadal simulations and compares AR characteristics and precipitation produced by six grid configurations using the Community Earth System Model version 2.2 (CESM2.2) (Danabasoglu et al., 2020; Herrington et al., 2022): two latitude-longitude grids, two quasi-uniform unstructured grids, and two VR grids (Zarzycki and Jablonowski, 2015; Zarzycki et al., 2015). The VR grids used in CESM2.2 employ static mesh refinement to yield enhanced resolution around our region of interest, Greenland. We hypothesize that the VR grids will simulate ARs more accurately than the coarser resolution grids through better resolution of finer-scale physical processes and topography, as has been seen in other studies investigating moisture intrusions in the Arctic (Bresson et al., 2022). The model output is compared to ARs detected by ERA5 and MERRA2, two observation-based meteorological reanalysis datasets, as in other studies involving simulated ARs (Bresson et al., 2022; Viceto et al., 2022; Zhou et al., 2022; Mattingly et al., 2023). Section 2 describes the model grids, remapping workflow, AR detection method, precipitation counting method, and the validation datasets used in this study. Section 3 contains the main results and analyses performed in this project. Section 4 discusses the implications of these results. Section 5 summarizes main conclusions from our work and provides direction for future research.”

Methods: CESM2.2 model output used. Then you say “atmosphere simulations used the Community Atmosphere Model 6.3 (CAM6) (Craig et al., 2021)”. Please clarify. The methods are not clear on the models. Please revise.

We have revised this portion of text to make it clear that CAM6 is the atmosphere modeling component of the CESM2.2 Earth System Model. We are running CESM2.2, which used CAM6 physics and runs the CLM5 land model. Please see the text in response to one of your main comments above. 

Lines 92-95: Which simulations? CESM2.2 or CAM6? Are they four ensemble members for each configuration in Figure 2? What are the acronyms for the simulations? Are these bilinear, conservative, or some other remapping technique?

In step with the previous comment, we have clarified that CAM6 and CLM5 are, respectively, the atmospheric model and land surface model components of CESM2.2. The text in Figure 2 has been adjusted to ensure that readers understand that the topography shown is native to each grid and therefore does not undergo any remapping to get these visualizations.

Text in Figure 2: “Native topography of each grid configuration in CLM5…”

Remapping techniques are both conservative; text has been clarified to make sure that readers understand.

Line 100: “The two remapping methods were ESMF (Team et al., 2021) and TempestRemap (Ullrich and Taylor, 2015), both of which are conservative techniques.”

Line 133: CLM. Is this CLM5?

Yes, CLM5. This has been specified.

Figure 4: The whiskers would be better as a percentile (e.g. 1 and 99 percentiles, or 5 and 95 percentiles). The reader will find it difficult to interpret 1.5 x interquartile range.

We appreciate this feedback but feel that the IQR is an appropriate way to display this data. We find that by using the IQR we get a more accurate view of the 25th to 75th quartile AR seasonalities and are less greatly affected by outliers.

Line 185: It does not look like ERA5 produces the fewest ARs in Fall or possibly Spring. Do you mean in the average number which is not given in the boxplots?

In making this statement we were taking into account the bottom of the box as well as the bottom-most whisker. We see your point in that during the spring and summer ERA5 is at least tied with another configuration for the fewest ARs and have changed the text to reflect this.

Line 196: “ERA5 produces the least, or ties for the least, ARs in all seasons except for winter.”

Figure 5: How are these source points calculated? This needs further explanation, especially on what the physical processes would be to have AR sources in these regions. Are there references for these processes (e.g., Neff et al 2014 is given later) or regions? There are certain simulations with points in Mexico – is this unrealistic? Does the difference between the reanalyses and the other simulations suggest a bias in the models which is not realistic? Please provide further explanation.

Source points are calculated by finding the grid point with the largest IVT contained within the tracked feature, at the first time-sample the feature is detected. Please see comment addressing all reviewers located at the top of this response for information regarding the outlier source points. 

 ERA5 in the northernmost part of Greenland. Is it realistic for the other simulations to have ARs there?

A recent study from Mattingly et al. 2023 investigated extreme melt events in northeast Greenland and linked them to ARs and foehn winds. As such,we have reason to believe that the simulations producing ARs in northeast Greenland is a realistic result. https://doi.org/10.1038/s41467-023-37434-8

Line 379: “Recent studies investigating ARs impacting the northern GrIS support the fact that ARs do occur at such high latitudes in this region (Mattingly et al., 2023).

Lines 211 and 220: I think this is wrong. Figure 7b shows the AR area and not the occurrence. The text surrounding Figure 7 needs to be revised based on this.

Thank you for noticing this. We have revised this section and ensured that the text describing Figure 7 accurately reflects the revised figures. As an example, we found the following line referencing the incorrect figure and corrected it:

Line 223: ”Figure 7a describes the number of ARs that eventually intersect the GrIS based on days relative to time of maximum overlap, and Figure 7c shows the occurrence of these intersecting the GrIS relative to the time of maximum overlap.”

Lines 218-219: This is not necessarily so. It may be that the winds are dropping which will result in smaller IVT and not that precipitation occurred.

This is a good point and we have altered this text to be less proclamatory and more suggestive in nature. For example, we changed “indicating that a large amount of moisture is being transferred” to “indicating that a large amount of moisture may be transferred”. As we did not investigate winds greatly in our experiment we do not have the capacity to analyze the possible impacts of winds in the scope of this  paper.

Line 240: I am missing the link between this sentence and Figure 8. Figure 8 does not show orography, it shows precipitation rate. Please revise this paragraph.

While we see your point, through showing the precipitation rate the orography of the GrIS is shown by indicating portions of the GrIS that are not touched by ARs. When looking at the day of maximum overlap, the southeastern portion of the GrIS has the highest precipitation rates. Following through to T + 24 hr, higher rates can be seen in the eastern portion of the GrIS. This infers the movement of an AR and also describes areas of the ARs cannot penetrate due to topography. To help clarify this, we have revised the first sentence of this paragraph to not include a reference to a figure so that other readers are not similarly confused. 

Line 254: “Many ARs affecting Greenland make landfall on the west coast and travel eastward until they reach the steepest portion of the GrIS.”

Line 316: “Figure 11 shows that precipitation from ARs likely occur within 500 km of the AR detected by our methods.” What is the physical process behind this? If the AR detection point is the core of the AR, then this result may make sense because of possible AR conditions 500 km on either side of the AR core and maximum IVT.

Thank you for your comment. We have included additional text in the manuscript that gives a more complete description of Figure 9; please see below for further explanation of the 500 km contributing area and its physical processes.

Line 322: “Figure 10 shows snapshots from the models of the 95 percentile ARs near the time of their maximum overlap with Greenland, and the outline of the detected feature provided in blue. The detected feature represents the moist core of the AR, which relative to the larger synoptic system does not overlap with a large portion of land at any point throughout its lifecycle (Figure 7d). The snapshots indicate the warm front situated out ahead of the AR core contributes a substantial amount of the storm’s precipitation. Additionally large regions of precipitation occur just outside the detected core within the cold front…”

Figure 11 quantifies the impact of including regions outside the core of the AR in compositing precipitation due to that AR.”

Reviewer 2:

In the manuscript “Using variable-resolution grids to model precipitation from atmospheric rivers around the Greenland ice sheet” Waling et al. analyze the impact of different horizontal grid configurations and remapping methods on the simulation of atmospheric rivers that make landfall over Greenland. They find that regionally refined grids produce atmospheric rivers that agree best with state-of-the-art reanalyzes products. The paper is interesting, and the visualization of the data is of high quality. The writing is, in parts, hard to follow and could benefit from careful revisions (I am providing a few more concrete examples below). I listed additional general comments and specific comments below.

Thank you for your thoughts. As described in the responses to Reviewer 1, we have revised much of the manuscript to provide clarity and reduce its technical nature. We took many of your comments into account (see below) and believe that our manuscript is improved because of them. 

General Comments

1. I understand that there is a lack of in-situ observations over Greenland but a discussion about the reliability of reanalysis products in this region would be valuable. Specifically, is using ERA5 and MERRA2 precipitation as reference reliable? Are there alternative precipitation estimates available that are more observational-based that could be incorporated into your analysis?

We chose ERA5 and MERRA2 as our observation-based data as these are the standard datasets used for validation in other studies  (Bresson et al. 2022, Collow et al. 2022, Viceto et al. 2022)(Line 175) and an overall lack of observations on the Greenland Ice Sheet. Additionally, our requirements for observational datasets are constrained by the need to have contemporaneous precipitation rates and IVT in order to detect and track AR features; the reanalysis products that we chose conveniently provide these data.

1. It seems like the phase of precipitation is a significant factor in the mass balance of the ice sheet as you indicated in your introduction. How is the precipitation phase simulated and are there any reliable observations that you could use to validate the model except for reanalysis data that might have its own biases?

The precipitation phasing is determined by the land model (CLM5), as a function of the near surface temperature. For temperatures colder than -2 C all precipitation is snow, and for temperatures above 0 C it's all rain, whereas intermediate temperatures produce mixed phase precipitation (linearly extrapolated based on the two end member temperatures for snow/rain).

Note that in CLM, rain can saturate the snow pack and refreeze, thereby adding ice mass to the ice sheet. Total precipitation can be viewed as providing additional information on the mass source since only some fraction of the rain actually runs off the ice sheet.

We are not aware of any rainfall datasets for the Greenland Ice sheet that coincide with our simulation period (1979-1998), though newer on-ice data in southeast Greenland are becoming available for more recent years (Box et al. 2023).

1. I suggest using more descriptive acronyms for your simulations. E.g., approximate dx and QU, VR, … For instance, instead of using ARCTICGRIS you could use VR_0.125, which would be easier to follow for readers that are not familiar with your grid configurations.

We appreciate this feedback and agree that choosing acronyms that will be clear to the majority of readers is essential. Up to this point, we have undergone multiple acronym changes and have decided to stick with the ones that we currently have. Especially for the ARCTIC and ARCTICGRIS grids, we think that these describe the two VR grids well as they have finer resolution in the Arctic and around the GRIS, respectively. We worry that by including numerical portions in the acronyms that it will lead to confusion to those who aren’t modelers.

1. While I agree with your general conclusion that the representation of topography seems to be the dominant factor in your assessment, the current simulations do not allow you to differentiate those from dynamically better-resolving ARs and their associated processes during landfall. Simulations with high horizontal resolution that have smoothed topography that mimics those in a coarser resolution run would have been insightful to differentiate these two effects. I understand that performing such a simulation is significant work. If you do not have the resources to work on this at the moment, I suggest at least discussing this option for future research in the discussion of your manuscript.

Thank you for this recommendation. We agree that the most meaningful way forward would be to run our high resolution simulations using the smoothed topography mimicking the coarser resolution runs. Regrettably, we do not have the resources to perform additional simulations as the lead author has since graduated and taken a new position not funded in this area of research. Funding for the lead student author came primarily through the University of New Hampshire as a departmental Teaching Assistant. Thus, we take your suggestion to include this in our discussion as future research:

Line 361: “The role of topography smoothing could be further verified through running the VR grid with the topography smoothing used by the coarser grids, although we did not perform this experiment.”

We also note that Pollard et al. (2000) and van Kampenhout et al. (2019) have found that simulations using coarse grids precipitate more than observation-based data, and have attributed it to the same phenomena that we have in our study.

Pollard, D., & PMIP Participating Groups. (2000). Comparisons of ice-sheet surface mass budgets from Paleoclimate Modeling Intercomparison Project PMIP simulations. Global and Planetary Change, 24, 79-106.

Van Kampenhout, L., Rhoades, A.M.,Herrington, A.R., Zarzycki, C.M., Lenaerts, J., Sacks, W.J. & Van Den Broeke, M.R. (2019).Regional grid refinement in an Earth system model: impacts on the simulated Greenland surface mass balance. The Cryosphere, 13, 1547–1564.

1. What about the vertical grid spacing in climate simulations? You increase the horizontal grid spacing by an order of magnitude. Would a higher vertical grid spacing in the horizontally highly resolved simulations be beneficial?

We expect that increasing vertical grid spacing could be beneficial, but increasing horizontal grid spacing is of greater concern.. 

Increasing the vertical resolution would improve the numerical accuracy of the vertical transport and support finer-scale structures such as temperature inversions or cloud macro- and microphysical processes. The CESM3 release will have double the vertical resolution in CAM7. In this co-authors’ experience with CAM7, the increased numerical accuracy leads to a less diffusive solution, with stronger gravity waves and vertical transport (see also Skamarock et al. 2019). And while the overall climatology – the large-scale temperature, humidity, clouds and precipitation rates (incl. mountainous regions) – noticeably changed, these changes are smaller than occur due to increasing horizontal resolution. This is because increasing horizontal resolution can support rougher topography boundary conditions and resolves finer-scale resolved processes in GCMs such as grid-scale updrafts (Herrington et al. 2020).

Herrington, A. R., & Reed, K. A. (2020). On resolution sensitivity in the Community Atmosphere Model. Quarterly Journal of the Royal Meteorological Society, 146(733), 3789-3807.

Skamarock, W. C., Snyder, C., Klemp, J. B., & Park, S. H. (2019). Vertical resolution requirements in atmospheric simulation. Monthly Weather Review, 147(7), 2641-2656.

Specific Comments

Fig. 7b: Does the more rapid decrease in AR area in the reanalysis indicate that the models are not able to extract enough moisture out of the AR once it makes landfall?

We agree with your observation for ERA5, but less so for MERRA2. The water flux to Greenland from ARs can be understood from Figures 10 and 11, which show that the reanalysis transfers less moisture out of the AR’s compared to the coarser grids, which precipitate too much. Although we do not consider nearby ocean points which receive a lot of precipitation as well. 

Fig. 7d: It seems like all simulations largely overestimate the area of overlap. Why is this the case? Are the simulations producing too wide ARs?

This is likely due to the horizontal resolution of all simulations. The VR configurations are able to resolve the most steep topography of the GrIS, thus allowing the LL and QU grids to penetrate further into the ice sheet. In combination with the actual sizes of ARs seen in 7b, we believe that the VR grids are indeed producing larger ARs than the reanalyses but think that the larger impacts are due to the resolution of topography.

While model resolution alleviates most of the discrepancy, the larger overlap areas in the VR runs compared to the reanalysis are evident. While this needs to be investigated further, we suspect that using 6-hourly averages instead of 6-hourly instantaneous output for the reanalysis cases does diffuse the metrics (see also dotted purple line in Figure 11b). All else being equal, the magnitude of the Laplacian of the IVT would be larger for instantaneous output, which may result in larger blob areas, and therefore larger AR overlaps areas. However these potential mask size discrepancies are controlled for in Figure 11, where we include regions outside the AR mask in compositing precipitation. That said, this discrepancy between average output in reanalysis and instantaneous output in the models should have been discussed more clearly in the text, and we have done so at Line 269: "ARCTICGRIS, ARCTIC, and f09 produce higher rates than MERRA2 and ERA5. This could be related to the model outputs being calculated using 6-hourly instantaneous whereas the observation-based data uses 6-hourly averages."

Fig. 8: What is the unit on the color bar? Is this mm/d?

Thank you for pointing this out. Yes, the units are mm/d. We have added this to the caption. 

Fig. 9: The font in this figure is hard to read since it is so small. Additionally, you could consider using a different map projection since the current one results in a lot of white space. Also, it would be helpful to see AR examples from ERA5 and MERRA2 as well.

Thank you for the feedback. We will increase the font sizes and try to improve readability of the Figure in general. We will explore including the reanalyses through either splitting this figure into two, or else as a supplemental.

Fig. 10c,d: The differences in total accumulated AR precipitation are concerning. It seems like these differences are coming from an issue in simulating the AR extent rather than its precipitation rate. The strong grid spacing sensitivity indicates that results have not converged yet and that even higher-resolution runs will continue to change the accumulated precipitation over Greenland. Is this a fair statement?

We would completely agree the results are concerning at coarse resolution. We discuss this finding in lines 312-319. If one were wanting to forecast AR behavior in Greenland a higher resolution configuration would be extremely advantageous. This opens the interesting question as to what “good enough” is regarding intent of study. The similarity of the ARCTIC and ARCTICGRIS suggest the solutions are converging, and therefore addressing its differences with the reanalysis products are the only thing left to determine if these grids are “good enough.” Thank you for this comment. 

L12: "..., smoothing from coarser resolution latitude-longitude and quasi-uniform grids". It took me a few times to understand what you mean here. I suggest rephrasing this sentence to something like "the coarser resolution latitude-longitude and quasi-uniform grids allow ARS to penetrate further inland due to their smoother topography. “

We agree. Here is the revised line:

Line 12: “In contrast, ARs are allowed to penetrate further inland on the GrIS due to topographic smoothing from coarser resolution latitude-longitude and quasi-uniform grids.”

L13-15: This sentence is hard to understand. I was not able to understand the reason why the VR grid has lower area-integrated cumulative precipitation and why area-average cumulative precipitation is similar from only reading the abstract.

We have revised this sentence to hopefully be more clear. Please see:

Line 13: “Precipitation rates are similar for the VR, latitude-longitude, and quasi-uniform grids, thus leaving the reduced areal extent in VR grids to produce lower area-integrated precipitation.”

L19: Is this true? I thought ARs can also originate in the tropics or sub-tropics.

ARs can originate in both low- to mid-latitudes; we have revised our statement to include the tropics.

L41-41: An issue with lat/lon grids is also the odd shape of grid cells in the polar regions, which are high-resolution in the zonal direction but low-resolution in the meridional direction.

Thank you for this, it is a good point. We have included this sentiment at Line 42: “In addition to this numerical instability, the "stretched" shape of latitude-longitude grids leads to high resolution in the zonal direction but lower in the meridional.”

L85: Something is missing after "...30-second resolution by”

The 30-second resolution is provided by Rastner et al. (2012) (Line 90).

L119: Why did you use a larger gradient than Ullrich et al. (2021)?

We wanted to choose a stricter gradient to predict ARs which will be more likely to cause detriment to the GrIS. Please see the response at the beginning of this document for a more detailed description.

L195: Fig. 5 misses a closing bracket.

This has been fixed, thank you!

L250: I am unsure where to see the 30 mm difference in Fig. 10.

This can be found by comparing the highest and lowest produced cumulative precipitations in Fig 10a. 

Line 260: “After the study period, there is a difference of around 30 mm between the highest and lowest depths produced by mean values from the grid configurations and reanalyses, as can be found by comparing the highest and lowest average precipitation at the end of the study period (t + 1.0 days).”

L308-309:  Ikeda et al. (2010) and Ikeda et al. (2021) would be good additional references here. They show that fairly high-resolution models (down to km-scale) might be needed to resolve complex flow and precipitation interactions with topography.

Great references, thank you! We have included this at Line 194: “Ikeda et al. (2010) and Ikeda et al. (2021) have found similar results describing the high resolution needed to resolve precipitation and flow around steep topography in the western United States.”

Ikeda, K., Rasmussen, R., Liu, C., Newman, A., Chen, F., Barlage, M., Gutmann, E., Dudhia, J., Dai, A., Luce, C. and Musselman, K., 2021. Snowfall and snowpack in the Western US as captured by convection-permitting climate simulations: current climate and pseudo global warming future climate. Climate Dynamics, 57(7-8), pp.2191-2215.

Ikeda, K., Rasmussen, R., Liu, C., Gochis, D., Yates, D., Chen, F., Tewari, M., Barlage, M., Dudhia, J., Miller, K. and Arsenault, K., 2010. Simulation of seasonal snowfall over Colorado. Atmospheric Research, 97(4), pp.462-477.

L322: What would be the reference for such a bias correction?

At this time we are not sure. We put this statement in the manuscript hoping that it would spark the imagination of someone interested in our work and expand upon the benefits and deficiencies of VR grids.  

L327: What do you mean with path tracking here and why would it be beneficial?

We have removed this statement from the manuscript upon further reflection. 

L348-350: How about running high-resolution global versions of CEMS2? This might improve ARs in multiple basins and the moisture export into the Arctic in general. Also, how are other models performing in this region? The HighResMIP simulations could be a good opportunity for analysis in future studies.

Our hopes with this study are that the lower resolution outside of the area of interest are refined enough to adequately resolve moisture transport into the Arctic. With this, it is a good idea for future research, but we do not currently have the resources to run high-resolution global versions of CESM2.2. We have added a sentence to our future work referencing HighResMIP.

Line 384: “ In the spirit of accurately studying ARs and their precipitation around Greenland, HighResMIP (Haarsma et al., 2016) could also be compared to VR simulations. Zhao (2022) used HighResMIP to simulate ARs globally and found that they resolved the mean precipitation from these events.”

L364: There is a question mark in this citation (?Kirbus et al., 2023).

We have fixed this, thank you!

L375-376: There is a "that" in the last sentence that should be deleted "We therefore that”

Great point, this has been fixed.

Reviewer 3:

This technical paper evaluates the ability of CESM2 to simulate Atmospheric Rivers (ARs) reaching the Greenland ice sheet as well as the sensitivity of the spatial resolution used to model them. 

The discussion about the impact of the resolution is not original and has already been discussed in Ettema et al. (2009) and Franco et al. (2012) for example. Franco et al. (2012) has exactly the same conclusion than here for example. The ability of CESM2 to simulate ARs is a bit more interesting. 

We appreciate the examples that you have brought forth which further support our findings regarding the impacts of resolution on topography. We have included references to both Ettema et al. (2009) and Franco et al. (2012) in our text (see below). Though that aspect of our study has been discussed already, we believe that our work is of value because we identify the processes responsible for the greater precipitation at coarser resolutions, through looking at individual ARs and their collective behavior, which illustrate greater penetration of storms into the ice sheet interior at coarse resolution.

Line 299: “Additionally, studies from Ettema et al. (2009) and Franco et al. (2012) that used regional models forced by reanalysis data support the idea of improved climate simulation from topographical smoothing due to grid resolution specifically in Greenland.”

In addition to the justified remarks of the 2 other reviewers, I have 2 additional major remarks before a potential acceptation in WCD.

• Before studying the impact on simulated ARs, the ability of CESM2 + spatial resolution to simulate the mean annual precipitation should be evaluated. As CESM2 is not forced by ERA5, ARs simulated by CESM2 do not occur in the same time than ERA5/MERRA2 and have not the same initial intensity or water content. Therefore, we have differences because ARs are initially not the same ones. As we sometimes say, “apples” are here compared with “pears”. How does CESM compare with the mean 1980-1999 annual precipitation? Is there a precip overestimation as for ARs? Do we see the same resolution sensitivity than for ARs? Precipitation from RACMO, MAR or GrSMBMIP should be used as reference for the mean annual precipitation. 

Herrington et al. 2022 evaluated the climatological precipitation and surface mass balance over Greenland in the runs used for this study. There it is shown that mean annual precipitation in the VR grids compare favorably with RACMO products driven by ERAI and ERA5, while the coarser grids precipitated too much over the ice sheet. We will be sure to add this important context to our findings in this study (Line 307). 

Ensembles are the preferred method of comparing CESM, a free running model constrained only by boundary conditions, to reanalysis. However, CESM is so sensitive to resolution that a single realization can provide robust estimates of climatological variation due to grid and/or dynamical core changes (Herrington et al. 2022).

Herrington, A. R., Lauritzen, P. H., Lofverstrom, M., Lipscomb, W. H., Gettelman, A., & Taylor, M. A. (2022). Impact of grids and dynamical cores in CESM2. 2 on the surface mass balance of the Greenland Ice Sheet. Journal of Advances in Modeling Earth Systems, 14(11).

1. About the spatial resolution sensitivity experiments, the low resolution topography should be used in the high resolution simulations to confirm that the differences are well due to the ability to resolve the ice sheet topography (the mountain barrier effect) and not to the ability of CESM to resolve precipitation/cloud processes at different spatial resolutions. 

While we agree that this would make for an interesting experiment, we do not have the time or resources to perform further analyses. Funding for the lead student author for this Master’s thesis work came primarily through the University of New Hampshire as a departmental Teaching Assistant. Thus, as this is a good suggestion, we have added it to the future research section to pose to other interested scientists doing similar work.

Line 361: “The role of topography smoothing could be further verified through running the VR grid with the topography smoothing used by the coarser grids, although we did not perform this experiment.”

Ref:

Ettema, J., M. R. van den Broeke, E. van Meijgaard, W. J. van de Berg, J. L. Bamber, J. E. Box, and R. C. Bales (2009), Higher surface mass balance of the Greenland ice sheet revealed by high-resolution climate modeling, Geophys. Res. Lett., 36, L12501, doi:10.1029/2009GL038110.

Franco, B., Fettweis, X., Lang, C., and Erpicum, M.: Impact of spatial resolution on the modelling of the Greenland ice sheet surface mass balance between 1990–2010, using the regional climate model MAR, The Cryosphere, 6, 695–711, https://doi.org/10.5194/tc-6-695-2012, 2012.