This manuscript provides a numerical approach explicitly to connect offshore sedimentary provenance records in glacial sediments with their source regions onshore, accounting for variation in sediment productivity and transport. The sediment transport problem is significant as it underpins our knowledge of cryosphere in past climate and therefore guides our ability to interpret past sea level and predict future sea level.

This problem has been approached with a range of techniques from educated guesswork to proximity studies and probabilistic assessments, and spatial modelling of individual parts of the transport system (e.g. subglacial erosion and sediment transport or ocean transport) but has not been comprehensively tackled from source to sink as is done here.  As such the approach presents a unique addition to the ability to model such systems in their totality.

I have several key comments

1 - The paper is not written in the best way for the journal. I would advise a general rewrite with a stronger focus on the new approach, and less on the case-at-hand in which the authors at times get bogged down in details of the case-study and lose sight of the main goal for GMD (and also for uptake of the approach) which is to focus on the approach, its capacity and its veracity.

2 - The degree of case-specific choices in the model is higher than I expected and I am concerned that this might limit the broader application that could make this a truly useful tool (see detailed comments). For example it is not clear if this model could, or could not be applied easily to Greenland or a model from the Pliocene. For GMD I think a more generic standpoint is needed. A simple synthetic model test case might add a lot if the authors can do so.

3 - I have some concerns about the deterministic nature of the approach and the large number of choices that are necessary for it to function. Several variables and assumptions are tested for impact and others are tuned to fit the data but the inter-relationships of parameters is not defined. In between all these moving parts there is overall a low chance that an optimal solution is found - indeed the Nd data is fitted somewhat better than a proximity-based approach but this does not indicate a minimum was found. For the paper I think a clear comment on the potential for unchecked errors to propagate through the model will suffice, but I would encourage the authors to pursue some potential ways to optimise fit to data in a more formal way. 

4 - In my view, the true power of this model, which in each of its parts is relatively basic compared to contemporary approaches, is that it holds the whole system in one model. I would be very interested to know more in the paper about potential for modularity - for example if I wish to do detailed ocean transport but need a glacial input; or conversely if I am modelling the sediment transport in detail but need to model ocean transport to a site. TASP might be the ideal tool for this if I can "plug and play", but if it is a closed process I can't take advantage of it.

Detailed Comments

Introduction

line 32 - in place of qualitative perhaps 'not constrained by a quantitative analysis'

line 34-36 - I think it would be good to express the source-transport-deposition mixture problem formally. You could use Equation 1 of Aitken and Urosevic (2021) or some equivalent

line 38 - I would note that there is no clear basis for changes in provenance to be interpreted to represent retreat and advance events unless other factors are able to be excluded (see introduction to Aitken and Urosevic (2021) and their eq 1 makes this clear). This emphasises the need for a model like TASP to define the system and reduce the potential for misinterpretation.

line 52  - here and elsewhere 'erosion rate' should be replaced with 'erosion potential' as the true rate is never known in TASP

line 69 - I don't think the comparison to Aitken and Urosevic (2021) is particularly relevant - theirs is a probabilistic assessment of sediment production tendencies avoiding the need to model transport. There is no competition (in fact the outputs of their approach could be inputs to this approach)

line 80 - An important simplification applied here is that there is no basal sediment layer. This is conceptually unappealing and also it is included in PSU ice sheet models since Pollard and DeConto, 2003 (https://doi.org/10.1016/S0031-0182(03)00394-8) and sediment transport is included in Pollard and DeConto, 2019. This layer is important as even a few metres of sediment protects the bed from erosion and spatially varying sediment cover would control strongly the provenance derived. It also can store sediments. If this truly cannot be included in TASP, then it must be made clear that the assumption is that sedimentary coverage is relatively uniform over the area.

line 83 - It is important to note also that subglacial fluvial transport is ignored, this too would strongly alter provenance as it can reach hundreds of kilometres into the ice sheet on short periods, and also is not necessarily aligned with ice flow.

line 110 - Perhaps add a comment here on how it might be interfaced with complementary environment-specific transport modelling such as SUGSET or Parcels

Methods

I find the description of Nd data to be overly long for the paper, and too specific - it seems the model tracks a numerical quantity that can be safely mixed (i.e. it cannot track categorical data such as rock types, or numerical data that cannot be mixed (such as U-Pb zircon ages) ... but it could probably be used to track bulk chemistry, for example. 

line 148 - For the purpose of this work, the choice to use offshore data to constrain onshore distribution introduces a problematic circularity...what would be the result with onshore data alone?

line 152/153 - uncertainty here  should probably be confidence

line 167 - erosion potential as it is not realised as a rate

line 169 - eq 2 - sedimentary armoring of the bed is neglected. This limitation should be recognised as it is a common process to include in subglacial sediment models - Q could this be included?

line 177 - the choice of erosion scaling I think is not very important and neither is model resolution - I don't think this paragraph adds much to the paper

Figure 3 - can we have a zoomed in view of the streamlines?

line 190 - 195  A comment here (or perhaps in discussion) is needed for how TASP might scale up to a more dynamic model, or an ensemble. 8 hours is not too much to ask, but if you wanted even to do 20 or 30 models it would become a problem. Perhaps a representative random sample of points would suffice?    

line 196 to 197 To have unique streamlines for each cell-outlet pair seems excessive (perhaps I misunderstand). A more efficient approach might be to accumulate sedimentary material as it flows (e.g. using D8/Dinf algorithm and a flow accumulator) 

 line 198 -201 this description of mixing could be better expressed with an in line equation I think

line 210 - While I appreciate it is a steady state analysis - if I understand correctly you treat it as instantaneous delivery. I think there needs to be some expression here of the timeframe to transport...at 0.1 to 1 km a year you might be looking at several millennia to transport the sediment to the outlet; in somewhere like the Siple Coast, that is certainly enough time for the flow to reorganise substantially

line 213 to 221 - I don't think you can ignore subglacial fluvial transport even in Antarctica - high pressure channels exist and are at work evacuating sediments from far inland beneath the ice sheet. I think it is sufficient to say that TASP does not currently include this process - You could add a citation to the model codes that do tackle this such as SUGSET and GraphSSeT and if these could be integrated somehow with TASP.

line 225 - Similarly here I think you needn't say it is infeasible, but it is not part of TASP and that is OK, so long as if I did want to do this in detail somewhere I can still use TASP for the rest!

Section 2.3 - I am less familiar with the oceans modelling sphere, but I do know there are a range of codes that can handle this in the specifics such as ROMS (Eulerian) and Parcels (Lagrangian). Similarly, to the above I think TASP has a simple approach relative to the dedicated codes and does not replace them, but gives a useful complement. Some degree of comparison is warranted

Section 2.3.3

This section shows that with detailed observational data, we can get an acceptable representation of modern-day iceberg trajectories -- but how might this perform for, e.g. the Pliocene? Does the accuracy degrade to the point where we might as well say they travel west and not east?

line 327-329 - this ocean-ice harmonisation process was not very clear to me

line 347 - eq 5 - the format of this equation is not very clear. It would be clearer I think to split the melt rate from the transit time d/v. Also the brackets are not necessary

line 416 to 427  - Are these processes Antarctic specific or might the processes be better represented by global data or data from data-rich margins rather than sparse local data?

line 429 - are there not problems from the sharp cutoff? I think this could be better represented as a gradual transition   

line 479 - 483 - Is this the same as the D8 algorithm? and it stops when al adjacent sells are above the central cell?

lines 592 to 560 - can global data or studies from data rich regions support this better?

Results

In the context of the GMD journal this section is overly focused on the case study -- which is in any case not a good basis for an accuracy test as the true result is not well known. The improved data fit is fairly equivocal due to the influence of a) parameter tuning to fit the data (which I assume was NOT done for the inverse distance) and b) I would say it is (probably) not a statistically significant outcome given the scatter in the data  - although I do not have a good gauge as to expected errors in eNd data, there is a lot of horizontal scatter in Figure 9.  

line 610 - realistic  looking and reasonable results is a weak expression

line 631 - close agreement is a bit of a stretch given the amount of scatter in the data and R-squared of just 0.58

line 648 - MSE of 3.77, while clearly worse, is fairly close to 3.05 given there was not any tuning applied. Unless you can prove statistical significance you should delete 'considerably' and also 'much' on line 649

line 649 - I don't think you can prove outright that the transport modelling was what caused the difference, therefore delete 'therefore'

Discussion

line 674 to 676 - The need for a high resolution observational record here works against the scope of the model for long-term examples...Add a comment here on if/how this process might be represented on long timescales to match the long-term assumptions? This is particularly true of the past

line 705 to 745 - This is an overly detailed accounting for a detail of the specific application and not very relevant to the development of the model. Suggest to delete or shorten considerably

line 726 - why was 200 km chosen?

line 750 Figure 12 -- this figure is fairly poor and seems in part to have been clipped from a previous figure. The coastline and annotations are peculiarly chunky -- suggest to use digital coastlines from IMBIE or measures

Conclusion

line 783 to 786 - the model seems to have confirmed the main features of sediment transport in the ocean...at least today

line 795 to 800 -  I am less convinced by the paleo ice sheet application - it is not clear to me how the surface ocean transport can be modelled to a comparable standard without the observations and the approach has not been demonstrated with degraded data 

line 802 - It might be worthwhile to point out a potential use for predictive targeting of core sites

line 806 - In terms of proxies, the model seems to be restricted to those that can be numerically mixed, which is probably fine for Nd, but problematic for more categorical proxies listed...

Appendix

I do not include detailed comments on the appendix for reasons of length in this commentary, and it is not very relevant to the development of the TASP model, only the application.  My recommendation would be to publish the model here and the application (including this mapping) in another journal.

line 880 - no data regions should just be left as no data I think

line 900 - I think to include offshore data in the definition of onshore data that is then modelled to fit offshore data introduces a problem, however small its effect   

This paper introduces and then tests a computational framework for predicting the provenance of sediment delivered from the Antarctic continent to the Antarctic continental shelf (and slightly beyond) on the basis of interpreting ice sheet model output. This work fills a significant gap in capability in terms of helping understand the pathways that detrital particles take as they are eroded subglacially and then transported to and beyond the ice sheet margin and theoretically allows the tracing to be completed using any ice sheet model output. The model takes into account the movement of particles such that the neodymium isotopic composition of the material is computed – this is beneficial because it can be compared directly to sediment collected offshore in a number of locations. Thus the framework should enable erosion and sediment transport and thus sediment provenance to be computed under different ice sheet regimes which should therefore produce different maps of neodymium compositions – this will allow certain ice sheet models to be ruled in or out based on their fit to measured neodymium compositions. The framework incorporates a set of appropriate transport and concentration processes although it has to make some assumptions as it does so. Processes include glacial erosion, transport subglacially, movement and rain-out from icebergs, ocean bottom currents and downslope sediment transport based on slope and the overall result is a seafloor map of Nd isotopic composition.

The code and dataset for comparison is openly available and well documented - all very clear.

General Comments:

The paper is largely well written and easy to follow and is largely well structured with useful figures. I have some relatively minor comments on the science and writing which would be good to see addressed.

First, in places the paper mixes description of the model with descriptions of the test area where it is being applied. For example at lines 129-135 there is discussion of the area that you examine in West Antarctica. However, for a paper that is introducing a framework/piece of software that should be applicable to anywhere, I think it would be beneficial to fully describe the model framework itself before then showing how it can be applied to a particular region. I think this means reviewing carefully the introductory and methodological sections to make sure that the framework is fully introduced and then you can roll into a description of the particular ice sheet model being tested and the particular region being tested. This would make the model easier to understand and potentially easier to apply to other areas of interest.

Second, I would like to see a clearer discussion of uncertainty or a clearer capability to embrace uncertainty within the framework. This is because at the moment particular parameter choices are made to fit the particular model area but there is no real discussion of the extent to which the result would vary if particular parameters were adjusted. For example, which parameters perturb the result most strongly? Or in other words, are there particular processes that dominate the result? It would be great to see some more discussion of that. In addition, it would be useful to see a table of parameter choices made here as this would serve to show in quick summary what was done, but would also show clearly what parameters are generally changeable within the framework so that people applying it to other areas can then make their own adjustments. Much of that information is in the user's guide but a table in the paper would give that additional help to readers.

Specific Comments:

Beyond these comments, most of my suggestions are minor – they are outlined below by line number:

Equation 2 and associated text – please explain/justify why that particular erosion rule was used. Other erosion rules are available (e.g. Herman et al, erosion under an alpine glacier) – would the results differ significantly? (link to discussion about uncertainty perhaps).

175: here you discuss a negligible change to the match with seafloor sediment Nd values in terms of how you choose k – this feels like a result or a point for discussion as opposed to something which should appear in the methods.

177-180 – you discuss the Aitken and Urosevic approach and mention your approach differs, but you don’t really say why/if your approach is more appropriate or provide evidence for why it might be better etc. – you could link that to my point about equation 2.

188: 'Standard Euler method' could do with more explanation. Also, explain why you think this is an appropriate method and also in terms of velocity, are you using basal velocity or ice surface velocity?

193-195: The computational demand is rather specific to a particular machine (which we have no real info about in terms of its specification). Therefore you could perhaps mention the types of CPUs. I would lose the point about using fewer CPUs and less memory – it seems obvious although I guess one option is to rephrase in such a way as to indicate whether there might be minimum specs that would allow the model to run (or perhaps to compare how long it would take to run on a more standard machine (assuming it would run on such a machine).

205-210: I wonder whether we might benefit from a schematic diagram to illustrate how particles might move through a grid framework (e.g. showing how particles will travel subglacially and then into the marine realm) . It could have some grid cells schematically drawn out with a start and end point for the particle. to show how these values are calculated as it moves through different processes. This schematic could be done just for this component of sediment transport, or another option would be to schematize the entire sediment transport process from source to iceberg to deposition in the ocean - this might help us understand the overarching structure of the processes being accounted for in TASP.

210: steady-state ice flow is assumed. Please discuss this assumption and its validity either here or in the discussion section later.

222-227: This feels a little out of place/tacked on. It could be mentioned earlier where you mention the erosion rule that is implemented, or it could be saved for discussion in the discussion section later.

233: Its not immediately clear why debris distribution in the ice column is in a section on ocean surface currents and iceberg rafting - move as appropriate or flag at the outset why the debris in the ice column is related to the iceberg processes.

Figure 4: Fig. 4 is a result figure, but also it doesn’t really tell us whether the fit is good or not. Thus is it a helpful figure right here? Perhaps Fig. 4b is relevant here, but fig 4a does seem to be a result figure which could be saved for later when results of the tests on the specific location are presented.

272: A value of 4 m debris-rich ice layer thickness. Is this also based on observations of debris layer thicknesses at all?

610: It is not indicated why/how you know this is a ‘realistic-looking’ distribution. Please elaborate.

653: TASP is better in deep waters. Better in comparison to what?

666: You could quantify some of the values (predicted vs. measured) so that we understand the fit. Overall, if quantification at particular core sites is being done, then a proper description of E_nd distribution would be good as part of the example science question you are addressing by applying TASP.

Figure 12: Can core-top sites be shown? I don’t know how much of this map is actually constrained by measurements.

Section 5 (Conclusions): The conclusion could be clearer about the key processes incorporated in TASP before it gets into particular science findings. Thus add a few points about the processes at the start of conclusions to properly say what the framework does. This is important because it’s a GMB paper which needs to therefore show the key points of any model before also showing any location specific results etc.