The authors evaluate the performance of a variable-resolution (VR) configuration of the Community Earth System version 2 over the High Mountain Asia (HMA) region, focusing on the cryospheric-hydrological variables. A new VR mesh is produced for this study with its grid spacing refined to ~ 7 km over the HMA region from ~ 1° in the coarse-resolution domain. A new glacier-cover input data is also produced for the VR grid.

The performance of VR configuration is compared to a globally uniform ~1° grid (NE30) through 20-year long simulations (1979-1998) and also evaluated against a variety of observational dataset. While the regionally refined mesh improves some aspects of the simulation quality, such as the circulation patterns forced by the topography, other aspects are degraded from the NE30 simulation. One reason for the degradation is model sensitivity to spatial resolution and time step length, i.e., less optimal tuning. Another VR simulation with several tuning to alleviate such sensitivities showed improved performance, but model biases still remain. They also suggest several future directions to reduce the VR model bias and improve physical representations in both the atmosphere and land models, as well as their coupling method.

The study addresses questions relevant to the scope of TC. The target region (HMA) is an important natural resource for the population in the broad Asia region. The model they evaluate (CESM) is a widely used community model. The grid resolution within the regional refinement is higher than a previous study focusing on the same region using CESM, and the new glacier input data is a original product from this study. The text is well written, and figures are overall high quality, altough I have several minor suggetions. All together, I think this study can be an important contribution to the community.

However, I have one major concern about their atmospheric model configuration and one suggestion of additional experiment that could strengthen the scientific quality and impact of the manuscript.  Please consider the following major comments before publication.

Major comments

1) I believe the spectral element dynamical core used in this study is the hydrostatic version, which is not expected to be appropriate for the 7-km grid spacing. Several previous studies found that hydrostatic and non-hydrostatic schems produce Significantly different solutions in sub-10km grid spacings, or even larger gridcell sizes over steep topography like HMA (Wedi and Smolarkiewicz 2009, Prein et al., 2015, Yang et al., 2017). Errors would appear in vertical acceleration or unphysical propagations of gravitiy waves, which certainly affect moist physics behavior over HMA. No assessment of those aspects were provided in the manuscript.

 

Another related concern is the vertical resolution. The small horizontal grid spacing is probably not balanced by the rather coarse 32 vertical levels (Lindzen and Fox-Rabinovitz 1989, Skamarock et al., 2019). Any testing was done with diffrent vertical levels?

Because I believe that the model is used with the resolution outside of its intended use, I strongly suggest the authors providing justifications and caveats to the readers about their results. The atmospheric model configuration used here should not be recommended as the standard for future modeling studies on the 7km VR grid.

2) First, I appreciate the authors for not only identifying model biases but also suggesting certain model components/characteristics that contribute to the biases, e.g., tropospheric warming due to stronger vertical motion and additional heating from CLUBB at higher resolution and with a shorter time step. But some of bias attribution remain speculative/qualitative or too general (section 3.7), which I feel is limiting the impact of the manuscript. 

Noting that enough materials are presented in the current draft, I'd still like to suggest conducting off-line CLM5 simulations on the same two grids, NE30 and HMA_VR, forced by observed meteorology. Off-line land simulations are much cheaper than the coupled atmosphere-land simulation and would help partition model biases in the surface/near-surface variables, espesically SEB and SMB, into those from the land model alone or those from errornous forcing from the atmosphere (or from problems arising from coupling or feedback). It is also worth asking if finer grid resolution and/or the new glacier data improve the off-line CLM5 performance. Having more accurate knowledge of CLM5 performance will help strengthen the logical basis for the discussion in section 3.7. 

References

Lindzen, R. S., & Fox-Rabinovitz, M. (1989). Consistent vertical and horizontal resolution. Monthly Weather Review, 117, 2575–2583.

Lu, J., Chen, G., Leung, L. R., Burrows, D. A., Yang, Q., Sakaguchi, K., & Hagos, S. (2015). Toward the Dynamical Convergence on the Jet Stream in Aquaplanet AGCMs. Journal of Climate, 28(17), 6763–6782. https://doi.org/10.1175/JCLI-D-14-00761.1

Pope, V. D., & Stratton, R. A. (2002). The processes governing horizontal resolution sensitivity in a climate model. Climate Dynamics, 19(3–4), 211–236. https://doi.org/10.1007/s00382-001-0222-8

Prein, A. F., Langhans, W., Fosser, G., Ferrone, A., Ban, N., Goergen, K., et al. (2015). A review on regional convection-permitting climate modeling: Demonstrations, prospects, and challenges. Reviews of Geophysics, 53, 1–39. https://doi.org/10.1002/2014RG000475.Received

Rauscher, S. a., & Ringler, T. D. (2014). Impact of variable-resolution meshes on midlatitude baroclinic eddies using CAM-MPAS-A. Monthly Weather Review, 142(11), 4256–4268. https://doi.org/10.1175/MWR-D-13-00366.1

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. https://doi.org/10.1175/mwr-d-19-0043.1

Wedi, N. P., & Smolarkiewicz, P. K. (2009). A framework for testing global non-hydrostatic models. Quarterly Journal of the Royal Meteorological Society, 135, 469–484. https://doi.org/10.1002/qj.377

Yang, Q., Leung, L. R., Lu, J., Lin, Y. L., Hagos, S., Sakaguchi, K., & Gao, Y. (2017). Exploring the effects of a nonhydrostatic dynamical core in high‐resolution aquaplanet simulations. Journal of Geophysical Research: Atmospheres, 122(6), 3245–3265. https://doi.org/10.1002/2016JD025287

Minor comments

L98

"In the western USA and the Chilean Andes, it has been used with regional refinements up to 7 km to simulate regional climate and snowpack (Huang et al., 2016; Rhoades et al., 2016; Bambach et al., 2021; Xu et al., 2021).Also, over the Tibetan Plateau, South Asia, and East Asia, it has been applied to study the regional climate and snow characteristics (Rahimi et al., 2019; Xu et al., 2021). The application of VR-CESM to simulate glacier SMB has been limited, this far, to the Greenland Ice Sheet (van Kampenhout et al., 2019; Herrington et al., 2022)."

I do not find any of the cited studies carrying out 7-km resolution refinement with the CESM-SE or a comparable model; the finest grid spacing seems to be ~ 12km. 

L127~

Deep convection parameterization is not mentioned. Was it turned off for the simulations in this study?  

 

L227 

"including a new glacier region over HMA with a 36-EC scheme in CLM. The new glacier region makes it possible to simulate SMB in multiple (including virtual) ECs in HMA, while retaining the computationally cheaper default behavior of one EC per grid cell in other mountain glacier regions. "

To clarify, in the default CESM/CLM, can a user set different number of ECs for each grid column? Or is this region-dependent EC numbers is a special configuration prepared for this study?

 

L231~

Why the difference in spinup procedure for each VR run? I suppose spin-up procedure affect the model bias against observations, especially those of cryospheric-hydrological variables.  So signals from the spin-up difference are likely to be mixed with those from the configuration/resolution differences. The authors can look at model biases of the HMA_VR7b spinup run and compare them to those after spinup to get some ideas of the spinup impact on the analyzed variables.

L344~

"The monsoonal circulation in the NE30 run has two centers, a broad region of ascent in the southern HMA region, primarily over the Indian Ocean, and a narrower region of ascent over the front range of the Himalayas. 

Please specify which sub-figures or rows are being discussed (e.g., "Figure 6, second row") to help readers follow the text. Maybe it's useful to add indices (a,b,c,...) to each row.

L353-356

"While the warming and drying patterns are largely the result of greater vertical velocities due to the enhanced spatial resolution in the HMA VR runs, the shorter physics timestep also contributes to this warming and drying (not shown), which is a common response to reducing the physics timestep (Williamson, 2008; Herrington et al., 2022)."

Williamson (2008) does not specifically discuss warming and drying as seen in this work. 

He illustrated model sensitivities to both the timeestep and spatial resolution.

According to Herrington et al. (2022), shorter time step contributes only to the warming of the lower troposphere. 

Pope and Straten (2002) found a similar warming of the mid-latitude troposphere because of eddy flux and its convergence are enhanced with higher resolution. Although large-scale feature, mid-latitude waves simulated at different resolutions converge only at 50 km or finer grid spacing according to Lu et al.(2015). An enhanced mid-latitude eddies inside regional refinement is reported by Rauscher and Ringler (2014) in their VR simulations as well, so the same processes may be occuring in the simulations here.  

Section 3.3 & 3.4

Please clarify what "absolute monthly mean xxxx (e.g., precipitation) differences" means. 

Precipitation, snow cover, and snow depth are positive quantities, so no need to use their absolute values. 

Section 3.6

Why is the NE30 result not included in this section?

L517, typo

"Figure 15 shows the relation between grid-cell-mean SMB and glacier fraction (GCF)"

->"Figure 14 shows..."

Figure 8

The minimum and maximum biases are not good to be used as whiskers for this figure because they

push the y-axis limits so wide that we cannot see the differences in the quantile boxes. 

It's probably better to use 95th percentile, 3 standard deviations, or any other quantities that only moderately widens the y-axis range compared to the quantile range. 

typo in the caption

"for rainfall (mm month-1) (a-b) and snowfall (mm month-1) (c-d)"

should be

"for rainfall (mm month-1) (a,c) and snowfall (mm month-1) (b,d)"

Figure 11, typo in the caption

Obs, blue

-> Obs, black

Figure 12, typo in the caption

"The black box in Figure 14a denote"

-> "The black box in Figure 12a denotes"

Code and Data Availability.

"Data will be available before publication in Zenodo."

What data are you referring to? Will the codes that produced the input data be publicly available?

Supplement material

Section S1 describes the workflow to produce the new glacier data in great detail.  

A critical information missing is the codes and/or applications used. Please consider sharing those information as well. Without, it is difficult for other researchers to reproduce the data or apply the same procedure to other regions. 

Figure S1

Not sure how the redbox in the inset of Figure S1b represents the outline of HMA.

Do the two insets cover the same area?

Summary

Wijngaard et al. in “Exploring the ability of the variable-resolution CESM to simulate cryospheric-hydrological variables in High Mountain Asia” evaluate a first-of-its-kind study on High Mountain Asia glaciers using the variable-resolution capabilities in the Community Earth System Model at 9 km horizontal refinement.  The authors do an admirable job of comprehensibly evaluating all aspects of snow energy and mass balance (SEB and SMB) drivers of HMA mountain glaciers.  Unfortunately, it is identified that over the 1979-1998 period HMA mountain glacier SMB loss in VR-CESM is 10x higher than expected (expected SMB is -20 GT/yr, VR-CESM SMB is, at best, -200 GT/yr), although not due to any fault of the authors.  The authors provide several plausible bias sources and hypotheses to test in future VR-CESM studies.

Overall, I think the paper fits perfectly within the scope of The Cryosphere and could be, given more work, an extremely valuable contribution to the scientific community.  The findings have both scientific and societal impact as mountain glacier modeling, particularly in Earth system models, is a recent model development across the community of models (requiring comprehensive historical evaluation studies such as this one) and glacial melt supplements water supplies and poses significant flood hazards (e.g., glacier lake outburst flood events) to billions of people that reside downstream of them. 

I think there are still several major(ish) revisions that need to happen prior to this paper being accepted.  Most of my suggested revisions are minor, however a few (if feasible) may require some time to address, and I wouldn’t want the authors to be time pressured by a quick turnaround with a suggestion of minor revisions.

Suggested Revisions

Line 14 – change “could help…” to “can help through targeted grid refinement”

Line 18 – delete “to evaluate the outcomes”

Line 22 – add “but is still underestimated”

Line 37 – change “differs per region” to “is regionally dependent”

Line 41 – change “as a response to climate change” to “in response to climate change”

Line 52-53 – consider citing - Li, D., Lu, X., Walling, D.E. et al. High Mountain Asia hydropower systems threatened by climate-driven landscape instability. Nat. Geosci. 15, 520–530 (2022). https://doi.org/10.1038/s41561-022-00953-y

Line 60 – change “with or without” to “with”

Line 63 – change “high horizontal…” to “fine horizontal…across many glaciers”

Line 71 – change “giving” to “providing”

Line 79 and Line 99 – consider citing - Rhoades, A. M., Ullrich, P. A., Zarzycki, C. M., Johansen, H., Margulis, S. A., Morrison, H., et al. (2018). Sensitivity of mountain hydroclimate simulations in variable-resolution CESM to microphysics and horizontal resolution. Journal of Advances in Modeling Earth Systems, 10, 1357– 1380. https://doi.org/10.1029/2018MS001326

Line 94 – FWIW Rhoades et al. 2018 provides model throughput timing for a wider range of refinement regions (including a 7km refinement patch)

Line 102 – change “this far” to “thus far”

Line 106 – why was 1979-1998 chosen rather than a period that encapsulates more of the satellite record?

Line 152 – is a constant lapse rate and relative humidity with elevation an appropriate assumption?  At the very least, is there a study to cite here?

Line 156 – cite the topography dataset developer manuscript/data archive.

Equation 1 and 2 – can black/brown carbon deposition influence SEB and SMB in VR-CESM?  If so, is black/brown carbon deposition too high and helping to drive the SMB bias?  Also, can surface melt pools and flow channels develop (influences albedo and, potentially, energy transport through glacier)? 

Line 195 – delete “more”

Line 218-219 – could the authors add the observed max elevation for each HMA region so that readers can contrast with model representation max elevations?

Line 224 – glaciers are assumed constant?  Is this true for these simulations or just the default setting?  If so, how would keeping glaciers constant (I’m guessing areal extent of glacier cover in grid cell?) shape the results?

Line 235 and 250 – could these fixed, solely temperature-based thresholds influence SMB bias?  Would implementing Jennings et al. 2018 temperature-relative humidity-based rain-snow thresholds help (see Equations 3-4 in Jennings et al. 2018) to increase the probability of more snowfall and a more positive SMB (see discussion on hydrometeor energy balance in Jennings et al. 2018 for physical intuition on why accounting for humidity matters for snowfall)?  This might be difficult given computational limitations (although several authors are at NCAR and may have access to additional computational time on Cheyenne), but could the authors run an experiment (5-10 years) with the HMA 7km grid and simply swap in the Jennings et al., 2018 temperature-humidity based rain-snow partition scheme to compare how SMB biases are altered (maybe add to Table 3)?

Jennings, K.S., Winchell, T.S., Livneh, B. et al. Spatial variation of the rain–snow temperature threshold across the Northern Hemisphere. Nat Commun 9, 1148 (2018). https://doi.org/10.1038/s41467-018-03629-7

Line 138, 232 and 247 – the assumption of capping snow depths at 1 m and 5 m w.e. in HMA seems like another culprit for SMB bias.  How often does snow depth hit the cap over the 20-year simulations?  Could the authors provide a cumulative annual “snow loss” estimate due to snow capping and compare contrast with SMB bias (I’m guessing this might influence refreezing portion of SMB in Table 3 values)?

Line 265-290 – the authors might consider evaluating SWE model estimates from VR-CESM compared with Liu et al. 2021 (note the period of record does not overlap with the VR-CESM simulations, but a climatological comparison might still be useful).  In addition, ERA5-Land might also be useful too.  This would be especially insightful for the snow capping assumption/issue. “…It can be accessed through https://nsidc.org/data/HMA_SR_D/ (last access: 22 April 2021) or https://doi.org/10.5067/HNAUGJQXSCVU (Liu et al., 2021). The dataset is provided as NetCDF files for each 1∘×1∘ tile shown in Fig. 1, available at 16 arcsec (∼500 m) and daily resolution from WYs 2000 to 2017…”

Liu, Y., Fang, Y., and Margulis, S. A.: Spatiotemporal distribution of seasonal snow water equivalent in High Mountain Asia from an 18-year Landsat–MODIS era snow reanalysis dataset, The Cryosphere, 15, 5261–5280, https://doi.org/10.5194/tc-15-5261-2021, 2021.

https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview

 Line 305 – eddy or geopotential height?

Line 314-321 – very informative!  It might be worth mentioning here that 2m surface temperatures don’t always correspond/agree with free/upper atmosphere warm biases, particularly over HMA (the authors later show that 2m surface temperatures are colder than expected in most HMA regions) to hedge reader expectation/confusion.

Figure 3 – for the temperature anomaly plots, the authors might consider decreasing the number of color bins to 0.5 deg C bins (current precision is hard to interpret).

Figure 6 – could the authors add subpanel labels (a, b, …) and point to them while describing physical interpretation in Lines 343-356 (which is well explained, but pointing to specific plots might help those that aren’t super familiar with monsoonal/precipitation dynamics in this region).  Define 2^nd y-axes on first column of subpanels (or remove).

Line 358-376 – should median biases be stated?  I think the authors should at least highlight that most of the distribution of the box-and-whiskers fall below the 0 deg C bias line (cold bias), save for JJA in the NW- and NE-HMA, and particularly VR-based distributions.

Figure 7 – “absolute temperature” shouldn’t all values then be positive?

Figure 8 – the use of a-b and c-d are incorrect, should be a-c (rainfall) and b-d (snowfall)

Line 368-376, 323-330 and 358-376  – Bambach et al. 2022 and Rhoades et al. 2018 (and others) have shown that a cold bias with elevation is also seen in non-glaciated regions in mid-latitude mountain regions (and without the application of the downscaling/ECs).  Do the authors think that the thinner clouds could be a culprit for the colder surface temperatures (Figure 7)? I’m thinking during the day there could be more shortwave insolation, but in the night (lower cloud fraction/thinner clouds) there could be enhanced radiative cooling with the net daily balance being negative?  Or is this surface temperature cold bias driven more by a lack of longwave feedbacks and/or minimal boundary layer turbulence over snow?  Slater et al. (2001) had a hypothesis for a positive feedback loop that creates stable boundary layers, particularly over winter/snow conditions (see Figure 8 in Slater et al., 2001).  Also, if this cold bias is fixed, it would likely worsen the SMB bias (unless the cold bias is partly due to the melt energy extracted from the atmosphere to the glacier, as the authors hypothesize).

Bambach, N. E., Rhoades, A. M., Hatchett, B. J., Jones, A. D., Ullrich, P. A., & Zarzycki, C. M. (2022). Projecting climate change in South America using variable-resolution Community Earth System Model: An application to Chile. International Journal of Climatology, 42( 4), 2514– 2542. https://doi.org/10.1002/joc.7379

Rhoades, A. M., Ullrich, P. A., Zarzycki, C. M., Johansen, H., Margulis, S. A., Morrison, H., et al. (2018). Sensitivity of mountain hydroclimate simulations in variable-resolution CESM to microphysics and horizontal resolution. Journal of Advances in Modeling Earth Systems, 10, 1357– 1380. https://doi.org/10.1029/2018MS001326

Slater, A. G., Schlosser, C. A., Desborough, C. E., Pitman, A. J., Henderson-Sellers, A., Robock, A., Vinnikov, K. Y., Entin, J., Mitchell, K., Chen, F., Boone, A., Etchevers, P., Habets, F., Noilhan, J., Braden, H., Cox, P. M., de Rosnay, P., Dickinson, R. E., Yang, Z., Dai, Y., Zeng, Q., Duan, Q., Koren, V., Schaake, S., Gedney, N., Gusev, Y. M., Nasonova, O. N., Kim, J., Kowalczyk, E. A., Shmakin, A. B., Smirnova, T. G., Verseghy, D., Wetzel, P., & Xue, Y. (2001). The Representation of Snow in Land Surface Schemes: Results from PILPS 2(d), Journal of Hydrometeorology, 2(1), 7-25. Retrieved Feb 13, 2023, from https://journals.ametsoc.org/view/journals/hydr/2/1/1525-7541_2001_002_0007_trosil_2_0_co_2.xml

Line 377-391 – could the authors provide cumulative annual/monthly precipitation totals to better contextualize monthly average biases?  I’m not familiar with the HMA region’s cumulative precipitation totals and can’t contextualize the +/- monthly biases stated.

Line 381 – change “worst” to “worse”

Line 386 – I’m curious how accounting for both temperature and humidity within the Jennings et al. 2018 rain-snow partitioning scheme(s) would alter this snowfall bias

Line 400-406 and Figure 9 – I’d recommend comparison with Liu et al. (2021) to better verify snow depth/SWE.  The authors might also consider using ERA5-Land.  ERA5-Land could be useful for other surface variable comparisons (esp. since it’s at a much more comparable resolution, ~9km, to VR simulations and the authors wouldn’t need to coarsen for comparison).

Liu, Y., Fang, Y., and Margulis, S. A.: Spatiotemporal distribution of seasonal snow water equivalent in High Mountain Asia from an 18-year Landsat–MODIS era snow reanalysis dataset, The Cryosphere, 15, 5261–5280, https://doi.org/10.5194/tc-15-5261-2021, 2021.

https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview

Line 411-413 – this finding matches one of my earlier comments/questions about LW feedbacks potentially driving surface temperature cold biases.  Although fixing this bias might further impact SMB bias since snow/ice are nearly blackbodies in LW spectrum…

Line 427 – I may have misinterpreted this, but I thought that there was reduced/thinner cloud cover (see Line 327-330) not greater, as stated here.

Line 436 – the extensive ice melt (and heat extraction from environment) could also be driving the surface temperature bias?

Line 441-457 and Figure 13 – given that SMB doesn’t appear to be water limited (i.e., VR-CESM produces too much precipitation/snowfall in most HMA regions in Figure 8), how much does the atmospheric variable SMB downscaling method influence the SMB bias (particularly temperature/surface energy fluxes)?  I’m wondering how the use of a fixed lapse rate (6 K/km and 32 W/m^2*km) and constant relative humidity with elevation shapes the negative SMB, particularly in mountains where lapse rates can vary quite a lot, even ones much smaller and with less heterogeneity than HMA (see Lute and Abatzoglou, 2020)

Lute, AC, Abatzoglou, JT. Best practices for estimating near-surface air temperature lapse rates. Int J Climatol. 2021; 41 (Suppl. 1): E110– E125. https://doi.org/10.1002/joc.6668

Figure 11 – do the initial condition spin ups of the VR-CESM experiments have anything to do with why SMB starts, even in the first year, with such large losses compared to both observations/WRF?  This is likely a naïve comment (not a glacier expert), but would there be any way to initialize the VR-CESM simulations to start with glacier/snow thickness comparable to observations/WRF instead of relying on the CESM spin up procedure?  Or could the authors use ERA5 to spin up a standalone CLM simulation that could then provide initial conditions to the AMIP VR-CESM simulations? 

Line 455 – delete “than”

Line 486-501 – agreed… major bummer on the 10x SMB issue in HMA (expected SMB is -20 GT/yr, VR-CESM is, at best, -200 GT/yr).  I hope the nudging exercise can help alleviate some of the large-scale biases in temperature.  I hope some of my random ideas above help too.

Line 546 – I’d argue most regions simulated by VR-CESM have cold biases (and across most seasons too)