Castleman et al 2021

This paper present ice sheet model simulations of Thwaites Glacier, Antarctica, and specifically investigates the potential impact of bedrock elevation uncertainty in future projections of this ice stream's future evolution. The manuscript is well written, clear, and will be of interest to the glaciological community. The model employed is state-of-the-art, which lends credence to the findings. The authors conclude from their experiments that ice sheet topography needs to be constrained to no greater than 2 km in the horizontal, and 8 m in the vertical. This presents a target (and a challenge) for future geophysical campaigns. 

l42 - perhaps use '0.59' to avoid it being misread (same applies elsewhere, e.g. l146, l257)

l61-2 - 'nearly impossible...using simplified parameterizations' - I think by virtue of them being 'simplified' that implicitly means that not all oceanic processes are being represented, right? I would just say, 'simplified parameterizations does not capture oceanic processes well', or something like that. 

l62-3 - might be worth mentioning ISMIP6 results here perhaps? Seroussi et al., 2020, Edwards et al., 2021. They show considerable spread, some of which arises from the basal melt rate parameterization. See also IPCC AR6 Chapter 9 (Fox-Kemper et al., 2021). 

l80 (& 82, 83, 86 etc) - 'errors', or 'uncertainties'? Isn't the point that we don't *know* whether the interpolations are accurate or not? Do we know for sure they are 'wrong'? I realise that 'statistical error' has a specific meaning, but from the perspective of clarity, might be worth considering alternative wording for at least a few instances of this, where appropriate.

l303 - Just a question - is there anything in the surface DEM that could help constrain where 'unknown' bedrock features might be present? Usually we see a surface expression of bedrock rises.

l655 - 'highly critical' - surely just 'critical' would do, or just, 'important'?

Fig 2 - GL positions are very hard to distinguish, make thicker perhaps, or put a thicker white line underneath each of the coloured lines?

This study investigates the impact of bed topography error on uncertainty in ice sheet model projections in the Thwaites Glacier basin in West Antarctica. The authors tests a number of different aspects of bed error (e.g. horizontal and vertical resolution), and their impact on grounding line retreat and sea level contribution under an extreme ocean-driven melt rate forcing. This is achieved through a series of experiments using the ISSM ice sheet model and a key contribution of this paper is that these experiments help indicate data resolution requirements needed to keep the uncertainty in SLR projections (due to bed topography error) below a certain threshold (e.g. +/- 2 cm over 200 years).

I don’t have any major comments and I congratulate the authors on a rigorous, well-written study.

Minor comments/suggestions:

L80: plus potentially errors in the interpretation of radiograms -- e.g. misattribution of basal crevasses as bed returns

L115: Do you specify a minimum ice thickness? i.e. in some simulations do you see the ice shelf thins to a point where the ice shelf front has essentially retreats (but is held in position by the imposed minimum ice thickness)?

L153: Personally, I would include the leading zero before the decimal.

L151: Melt rates extrapolate inland, but later on your mention that there is no melt in partially grounded elements. So is this extrapolation inland for newly ungrounded ice only? Perhaps just clarify this.

Section 4.1: Would it be possible to include a conceptual diagram to explain how the wavelet decomposition works or at least what it ultimately means for adding noise to bed topography? Or perhaps some examples in the SI of the various bed realizations (or difference plots between the perturbed and control beds). This is so that readers who don’t follow the details at least can understand what the bed perturbations looks like, aiding their intuitive understanding of the results.

Section 4.2: What is the assumption being made here about bed roughness under Thwaites? Is the minimum resolution of 2 km that you are proposing only required if high frequency variation in topography actually exists?

L341-2: Because you have already described Fig 3b, this sentence about exceeding 2cm around the 2 km mark seems repetitive. I’m not sure what you mean by ‘eventually’…

L344-5: Similarly, this sentence is confusing – what does compensation mean in this context? Does “before 2 km” mean higher or lower resolution?

L246-349: This is an interesting finding. Can you reiterate the context, i.e. this 2 km is therefore the desired minimum horizontal resolution required for bedrock topography data in order to keep uncertainty in SLR projections due to bedrock data to less than 2 cm.

L387-8: Perhaps refer back to Experiment 1 here as it seems relevant.

L404-6: Could you add vertical/horizontal lines on Fig 4 to mark the +/- 2cm SLR and corresponding +/- 8 m vertical res?

L444: “repeat the bedrock sampling experiments, but this time with perturbation of only the ice shelf basal melting rates”. I wasn’t entirely sure what you meant by this – from the Fig 5 caption, I think you mean running the control topography (i.e. bedmachine with no errors) for 300 different melt rate perturbations. Could you make this clearer?

L475 and elsewhere: when you provide the range of the distribution, is this a range between certain percentiles, e.g. 5-95% range (given you are presenting PDFs)?

L496-7: "locations or strong topographical influence on the grounding line, or pinning points" – I would consider a depression which results in no stable GLs / rapid retreat to also be a strong topographical influence on the GL -- could you find an alternative way of phrasing these locations that are particularly conducive to stable grounding lines?

L628-9: Favier et al. 2017 (https://doi.org/10.5194/gmd-12-2255-2019) test a range of melt rate parameterisations and should probably be cited here.

General comments

This paper steps through a series of experiments exploring the impact of uncertainty in bed topography beneath Thwaites Glacier on future sea level rise. Each experiment is clearly explained and there is a logical progression between them, with sensitivity tests included. The authors conclude that there is still significant variation in the future trajectory of Thwaites Glacier given current bedrock uncertainty, and that in order to constrain sea level rise to ±2cm at least a 2 km spatial and 8m vertical resolution is required.

Given that the paper is well written and structured, I would recommend it for publication, although there are a couple of points that I think need addressing in the discussion, and a few minor changes to consider. 

Specific comments

In Experiment 1: Minimum and Maximum bedrock resulting spreads, you make the assumption that the minimum bedrock topography represents the maximum possible retreat and the maximum bedrock topography represents the minimum possible retreat. This is supported by your figures for max bedrock (4.8cm SLR), control (21.1 cm SLR) and min bedrock (26.7 cm SLR). However, beds which contain a mixture of plastic and viscous areas can exhibit behaviour which is outside the range bounded by purely viscous and purely plastic beds. (Koellner et al., 2019, https://doi.org/10.1016/j.epsl.2019.03.026). Do you think that there is a possibility that more retreat could be seen by a scenario between the max bedrock and min bedrock endmembers? The smoothest and roughest beds available within the 3σ bounds from Bedmachine might provide a different set of extremes? Nias et al (2016, https://doi.org/10.1017/jog.2016.40) might be a useful reference here.

Melt rate obviously has an impact on sea level rise, as you discuss in experiment 5. However, it’s not really addressed in this paper whether there is any interplay between the topography sensitivity and the melt rate. If the ocean warmed faster than expected, would a higher melt (ie 2x instead of the max 1.8x you use) mean that a higher topographic resolution is required?

You also don’t mention at any point the effect of variations in basal slipperiness, or basal sliding law, on the future sea level rise. It would be good to see this addressed in the discussion/future work section.

Technical comments

“Ice sheet model”, “Ice sheet instability” and “sea level rise” should technically be written with a hyphen as ice-sheet model, ice-sheet instability and sea-level rise. There are a few instances of this throughout the paper.

L17 Upper case G, Glacier

L50 For basal melting rates, because stochastic evolution is prevalent in ocean circulation, especially beneath ice shelves, melting rates are difficult to accurately observe, model, and predict. >> Basal melting rates are difficult to accurately observe and model, because stochastic evolution is prevalent in ocean circulation, especially under ice shelves.

L71 Data are technically plural, so ‘The data are interpolated’

L75 It’s confusing to say that bedrock topography is directly measurable using remote sensing, because this is not true. We can measure surface elevation using remote sensing, and then use this to interpolate between radar lines to get the bed topography. Instead you could say something along the lines of ‘bed topography is directly measurable using ice-penetrating radar (see Holt et al., 2006, Holschuh et al., 2020 etc)(https://doi.org/10.1029/2005GL025561, https://doi.org/10.1130/G46772.1).

L77 Although there are a variety of different digital elevation models, the modelling community has primarily only used Bedmap (v1 and 2) and Bedmachine Antarctica. So this sentence is maybe misleading, and could be removed without altering the paragraph significantly.

L107 Shelfy-Stream (hyphen and capital missing)

L112 Can you add a sentence clarifying what the modification from Yu et al. (2018) does?

L114 Is the model of Schlegel et al. (2018) from Schegel et al. (2013)? It seems like you don’t need the Schlegel et al (2013) reference here.

L140 Schlegel et al. (2018)

L158 Is there a missing ‘a’ here? “…using a sub-element on a partially floating…”

L166 I don’t think provided is necessary here

L194 I would change yield to is  “The resulting SLR difference between A and B is 21.9cm”

L196 Figure 2 only shows the grounding line position after 200 years, and not the evolution of the grounding line. Can you confirm here somehow that there is a stabilisation of the grounding line, and that it wouldn’t retreat further if the model was allowed to run for longer?

L226 “extreme” not “extremes”

L227 Thwaites Glacier (insert glacier after Thwaites)

L298 ‘produces’ not ‘produce’ (Noise amplification … produces geological landforms)

L325 I think it’s clearer to say ‘See Appendix (Sect 9.2) for further details on…’

L330 ‘Many’ rather than ‘much’ (Many of the perturbations are…)

L425 You’re missing a ‘the’ here (We use the LHS sampling algorithm)

L596 The wording here is confusing because Thwaites Glacier is not an active landmass, it’s a glacier which sits on top of an active landmass. You could just rephrase this as ‘Thwaites glacier is affected by vertical land motion’.

L601/602 Change ‘its’ to ‘their’ (We hope to explore their contribution), because couplings are plural.

L606 Change ‘promotes’ to ‘promote’

L610 ‘…would therefore not be correctly configure for the sensitive experiments we perform’ >> sensitivity experiments?

L655 it is highly critical to better constrain uncertainty in simulated SLR contribution that is sourced in bedrock topography >> it is critical to constrain the uncertainty from bed topography in simulated SLR contribution OR it is critical to constrain how much of the uncertainty in simulated SLR contribution is sourced from bedrock topography? Something about this sentence is a little unclear, it maybe needs rewording.

L668 upper case G, Glacier

Your reference list is also not consistently formatted, and it’s probably faster for you to tidy this up in your files before the typesetting stage