General Comments:

The paper is excellent in terms of organization, writing, and content. The results and figures are convincing, and I am particularly excited about how the authors utilize changes in model structure to address sediment connectivity, which is timely and quite important in terms of advancing watershed sediment simulations. Overall, I have very minor comments regarding some clarifications in a few instances and I believe that a few statements made by the authors should be relaxed a bit. Additionally, I would suggest including a brief paragraph at the end of the discussion regarding limitations of the WaTEM/SEDEM approach, and how we can further move to improve the spatial and temporal resolution of sediment connectivity simulations.

Specific Comments:

• L 32: Perhaps you can also mention that the paper is, for the first time to my knowledge, advancing tools to assess connectivity by quantifying structural uncertainty within the sediment simulations (not referring to structural connectivity here, by the way, just how there are inherent uncertainties within how the model is configured to predict fluxes/loads).
• L 159-167: It would be helpful within the text to tell readers the temporal resolution of the model. It seems like it's yearly according to the RUSLE equations but could be clarified.
• L 180: Is this of the individual pixel or along the slope length?
• L 188: Does this include bank erosion?
• L 188-190: If this is a yearly model then perhaps this statement can be slightly relaxed... For example if the system is known to not aggrade or degrade over longer-term (decadal) timescales then instream erosion and deposition are approximately in equilibrium and so I would not be as concerned with the instream.
• L 205: How did you decide 1200? Is this enough? Sometimes people will utilize 100,000 monte carlo simulations. I’m not saying that you need to run the model for more realizations, just a bit more justification please.
• L 205-208: Right, the typical approach is to calibrate the model and along the way assess sensitivity/uncertainty such that sensitivity/uncertainty of the model is addressed within solution spaces that are plausibly behavioral. I’m not rejecting your approach by any means, but perhaps some additional acknowledgement of the traditional approach and how you are slightly deviating here could be helpful to readers. Some readers might question why you present realizations that will not adequately describe the sediment load/flux in the system.
• L 208: model assumptions – I would clarify you are making assumptions about the structure of the model, so quantifying structural uncertainty.
• L 238: can you clarify why a value for Pcon wouldn’t be applied everywhere in the catchment, but instead for just the forest and buffer strips? What if there is disconnectivity from microtopography in the roadside ditches, for example? Again – I’m not asking for additional analyses, just a sentence or two for clarification and that you might parameterize this other places in the watershed if you had overt reason to.
• L 255-259: can you please add a sentence that details the difference between scenario two and three? The way I understand it is that in scenario three sediment deposition does not occur on the road or in swales/ditches along side the road, but deposition can still occur downstream, for example in between the road and the stream network. In scenario two sediments are automatically connected to the stream, correct?
• L 270: Again I might suggest using the word structural uncertainty of the model.
• L 275-276: This was a bit confusing to me.
• L 306: What is the mean squared error in relation to? The yearly predicted sediment load and the yearly average sediment load from the rating curves? Please clarify.
• L 323: Fig. 6g – what about interrill erosion?
• L 323: Fig 6b,c; L 333: Is it worth showing land use for all the details here?
• L 374: Perhaps you can say in the caption that the short-cut generally overlaps the IQR better than the other 2 scenarios... this could help readers quickly interpret the figure.
• L 385: Out of bound percentage – is this a fraction or a percentage what is presented in the table?
• L 388: It would be great if we could see at this same time scale how SEDEM was performing… but I think this is just a limitation of the model since it runs at a yearly scale, correct?
• L 424: Perhaps also the rating curve is underestimating the load, as you previously mentioned? Which would improve the performance of simulations with respect to the short cutting, correct?
• L 451: I believe Mahoney et al., 2018 talks about importance of road networks a bit in the USA.
• L 463-465: I think this last sentence should be relaxed a bit... quantifying all of the sources of uncertainty due to observation data, model input data, model output data, parameter uncertainty, etc. etc. is quite the undertaking. In fact, in my opinion, it might be an impossible task. Does this invalidate the use of models, however? In my opinion, no, it does not. We can still discern important information from models even though we don't account for 100% of uncertainties. It ultimately will depend on what questions we are trying to answer with the model and what the model is attempting to do, which can be equally as important as quantifying certain uncertainties in my opinion.
• L 465-466: It would be nice if a paragraph on limitations of the modeling approach and future opportunities could be included. For example, while RUSLE is relatively easily implemented and approachable, it would be nice if the RUSLE approach was a bit more physically based. Additionally the RUSLE approach limits the temporal resolution of the model, so seeing event- and seasonal-scale connectivity seems a bit limited. Furthermore, the advanced geospatial data that facilitates this novel connectivity modeling is wonderful, and can help to elucidate hotspots of connectivity. Additionally there is recent sentiment to move towards high-temporal resolution models to quantify hot-moments of connectivity. The yearly timescale inherent to the RUSLE approach perhaps is underserving this sentiment.  
• L 466: I'd suggest perhaps emphasizing that exploring structural uncertainties in the model framework - and not just parameter uncertainties, as is the traditional method - allowed for advanced understanding of connectivity processes. This type of approach in my opinion is quite underserved in modeling work and should be considered in the future where high-resolution geospatial data is available.

Technical Comments:

• L 312: I’m not sure if the different colors are helpful here, maybe consider symbols?
• L 324: typo
• L 364: typo, confusing.

General comments

This paper provides a modelling approach to analyse the effect of linear features on sediment connectivity in a mesoscale catchment. The authors put a lot of work into adapting an existing model such that it is able to account for these effects. I appreciate this effort and I think that this work is important for improving the understanding of sediment connectivity in agricultural catchments. However, I have some major points of criticism which are the following ones:

• The authors state themselves that a comparison between their model results and the measuremed sediment loads in the river should be performed with upmost caution. However, in the results and in the discussion they still make strong conclusions based on exactly such a comparison. The authors should therefore reformulate the discussion such that it reflects this uncertainty better.
• The comparison between the model results for the different scenarios and the measured sediment loads does not make sense for me in the way it is done currently. In their current analysis, the authors ignored important factors influencing the sediment connectivity in the catchment: Firstly, as I understand it from the manuscript, they treated all grassland areas like arable land areas in the model. Since around a third of the catchment area is covered by grassland, this leads to a large overestimation of sediment loads. Secondly, the authors assumed a two-meter grass buffer strip around all agricultural plots. The authors state that they don’t know the real width of buffer strips in the field and that they therefore use a value of two meters for testing the sensitivity of the model. Even though the sensitivity analysis showed that the buffer width had a large influence on the model results, the comparison between modelled and measured sediment loads is only done for the two-meter scenario. Also the conclusions are only drawn based on this scenario. Even though the authors state that the two-meter scenario is more realistic than the scenario without a buffer, this value contradicts the values reported by other studies (e.g. Alder, 2015; Remund, 2021) and the legal requirements. Although this point is addressed in the discussion, I am missing a proper justification why the measurements are only compared to the two-meter scenario. Thirdly, in their “shortcut” scenario, the authors assume that all roads and farm tracks are drained with shortcuts. (At least, this is how I understand it from the manuscript.) I expect that a major part of the roads in the Baldegg catchment is actually not drained by shortcuts. Therefore, it makes sense to use the current “shortcut” scenario in a sensitivity analysis, but not as a realistic scenario. Compared to reality, the current “shortcut” scenario is expected overestimate the real sediment loads.
  
  Consequently, even though the “shortcut” scenario is most similar to the measurements, this is possibly simply caused by a bias in the model input. Even though I agree that roads and shortcuts are in fact important for sediment transport, I think this cannot be concluded from the current analysis. To state that “roads behave as conduits for sediment transport in the catchment”, as it is done in L474f, it is inevitable that the scenarios are revised such that they reflect the reality in the catchment better. (At least for the first point mentioned. Second and third point may also be discussed.)
• The authors state that the catchment is representative for the Swiss plateau. However, they do not further elaborate on this. Other studies, however, rather suggest that the catchment has a low shortcut connectivity compared to other catchments in the Swiss plateau (see comment to L98). The authors should improve on putting their analysis in the right context.

Several specific comments (see below) should be addressed to improve clarity and reproducibility of the study. Additionally, the manuscript should also receive some revisions regarding language and correct spelling.

Specific comments

L36: Talking about a “continuous displacement of small amounts” is wrong here. The discplacement varies strongly between events and years, as you also state below.

L47: Rephrase.

L56: You should also add the most recent publications here, e.g.: https://doi.org/10.1016/j.catena.2021.105290

L66: “assuming they are able to explicitly take connectivity into account”: Difficult to understand. Please write this more clearly.

L73: with a size of few square kilometres

L79: You state above that one major issue of erosion models is the uncertainty of input data. Then you state that you used a high resolution dataset (2x2m DEM). However, for whole Switzerland, a 0.5x0.5m DEM is freely available in the same quality as the 2x2m DEM. You still used the 2x2m DEM. Why did you not use the higher resolution model?

L98: Here you state that the Baldegg catchment is patchy and representative for the Swiss Plateau. Below, you only elaborate on the patchiness, but not at all on the representativeness. Either elaborate on the representativeness below or use another word here. Schönenberger et al. investigated two catchments in proximity or even inside your catchment. Compared to the distribution in the Swiss plateau, these catchments however rather seem to have a low shortcut connectivity. This indicates in my opinion that also your catchment is rather on the lower side with respect to shortcut connectivity.

L103: In Figure 1c, you use the term “infrastructure”, here you use the term settlements. It is unclear how these two terms differ and what the term infrastructure means. Please use consistent terms here. Additionally, how did you treat the areas of roads? Did you include them into the settlement area? Or are they included in the agricultural land/forest area?

L117: Are tile drainage only located in water accumulation zones? (What are water accumulation zones? Are the determined based on topographic index, slope?)

L120: How did you determine these slopes? Which elevation model? The maximal slopes are strongly depending on the model used. You are referring to Figure 1b. However, the slope is not visible in Figure 1b.

L122: “in this case formed by the retreat of the Reuss Glacier in the south to north direction (~18,000 years BP)” -> Not important. Consider removing.

L127: MeteoSwiss -> Please add reference.

L132: What is approximately? Provide the range of numbers of grab samples taken.

L133: What is “opportunistical sampling”?

L136: Why 2020? Above you write that you only sampled till 2019.

L144: What is k? I guess the covariate ID. This should be written explicitly. Additionally, in contrast to Vigiak & Bende-Michl, you are only using the first five covariates, but not the long-term trend covariates (6 and 7). Why?

L149: First column of table: Remove the word “is”. Also the word “water” is not really needed. Second column of the table: This is not the quadratic term of Q_i, but of x_2,i.

L156: You are addressing the variance in sediment concentrations extensively. However, you are not addressing the uncertainty in daily discharge at all. Why?

L165: Shortly explain why you only focused on water erosion.

L193: “usually implemented” -> Rephrase. (It is either implemented or not. Possibly, you could write something like that this version is often used.) Provide references where this version is used.

L199: R version?

L200: SAGA version?

L201: The code does not contain any information on the versions of the packages used. To make sure that the code can still be used in the future, you should at least provide information on the package versions used. Consider also using tools like packrat, checkpoint or docker. To make the code useful, you should also provide examples of input files.

L219: It remains unclear how you derived the land cover map. The reference is not shown in the reference list. Therefore, I don’t understand if you used a vector dataset that you rasterized yourself or if you used a raster dataset provided by Swisstopo. What does the resolution “1:25:000” mean in the latter case? If you used vector datasets: How did you deal with point and line features, e.g. roads or hedges? Did you assume widths for roads and hedges?

L220: The statement that spatially distributed crop statistics are unavailable is wrong. There is a plot-resolution crop dataset from the canton of Lucerne available freely for the whole canton (and accordingly for the whole catchment) for the year 2019. Why was this dataset not used? (CP and K_TC depend on the crop and you reported them to be the most sensitive model parameters.) Lavrieux et al. state that one third of the agricultural area consists of permanent grassland. Therefore, I expect this decision to have a large influence on your results, leading to an overestimation of erosion. You should address this point at least statistically for each of the five subcatchments analysed and for the full catchment. (e.g. look at the fractions of grassland and reduce the estimated amount of sediment load accordingly). Alternatively, you could also do a spatially explicit analysis.

L226: Why did you use the 2x2m resolution DEM? (see comment to L79) How did you process the DEM, e.g. sink filling?

L227: In the K_TC column, consider indicating the land use classes belonging to “high” and “low” (e.g. in brackets). This would make it much easier to read.

L230: Specify that this relates to the maximum CP factor.

L233f: You talk about forests and grass buffer strips in the land cover map. However, the reader is missing were you explain the derivation of forests and grass buffer strips. Consider stating that you are explaining this below.

L246: How much wider? Refer to the article in the corresponding legislation directly, instead of Alder et al..

L249: What were your assumptions on buffer strips along hedges? Did you also use a 2m buffer? Or a buffer corresponding to the legal requirements? How did you treat tree lines?

L250: As mentioned in the comment to L219, I don’t really understand how you derived the road areas.

L252: What is “infrastructure”? Was this also derived from the land cover map?

L253: If roads act as sinks, why is this related to field drainages? From the text, I don’t understand where you assume the sediments to be trapped. On the road? In the drainage system? In sludge collectors? The scenario seems to make sense for me, but you should specify your assumptions more clearly.

L256: Did you assume here that all roads are acting as shortcut? Or only a part of the roads?

L262f: How many directions?

L266-268: In my opinion, some (or probably all) of the additional packages (e.g. “doParallel”, “foreach”) are not worth mentioning here, as they are only used to speed up the calculation process, but not important for reproducibility of your work.

L278: What does RFA stand for?

L280: Version?

L299: Do I understand correctly that Mg yr^-1 means tons per year? Probably write tons instead.

L299: For me, the differences between GS and NGS scenarios are not well visible in the plots. Consider making this better visible, e.g. by adding a moving average per category or something similar.

L317-L318: How did you quantify this for the whole catchment? Just by eye? If yes, you should provide the respective plots (e.g. in the appendix). Otherwise, can you provide a quantitative assessment?

L324: In Figures 6b, 6c, 6e, 6f, and 6g, arrows indicating the flow direction would help to understand the plot better.

L358: In my opinion, Table 3 would be understandable easier if you would write 25%, 50%, and 75% instead of Q1, Q2, Q3.

L366f: As I understand from L244ff, you say that you don’t really know what the real widths of buffer strips are in the catchment. It therefore makes sense to me that you use a fixed width of 2m and use it for testing the model sensitivity by running two scenarios – one with and one without the buffer strip. However, in L366 you now state that the 2m scenario is more realistic than the “no buffer” scenario. I agree that a 2m scenario is probably the more realistic scenario along forests. However, I expect the effect of a grass buffer along forests to be small as sediments are trapped by forests anyways. Along roads, I doubt that the 2m scenario is more realistic than the “no buffer” scenario, since the legal requirement is only 0.5m (as you write in L430). (However, I might be wrong with these doubts.) Since I expect the buffer width along roads to be much more important for your model results than the ones along forests, I think you should report both scenarios here, or give a clear explanation on why you think or why you can show that the 2m scenario is more realistic.

L379-382: In the “Ron” stream, the 95% prediction interval seems much narrower than in the other rivers (Figure 8). Therefore, the observed values are mostly outside of this interval and the out of bound percentage is much higher than for the other streams. Can you explain this?

L389: Consider using the same y axis limits for all plots. Like this, it is difficult to see the differences between the streams. At least the zero line should be visible in all plots.

L422: Which physiographical statistics did you analyse? Please provide details. I guess you did not analyse crop types (e.g. fraction of arable land (without grassland) on the total catchment area)? Could this also be a reason for the difference?

L443: Where do you show in your study that model resolution is important for your results? I don’t think that you can conclude this from your study.

L444-445: In L285 you state “Hence, modelled hillslope yields and suspended loads are not fully commensurable, and we did not focus on a rejectionist framework for model testing.” Here, you state that “soil redistribution rates and patterns are intrinsically linked to linear features”. The strength of the latter statement does not really fit to the caution you demand in the first statement. Therefore, you should reformulate this sentence.

L472-475: I very much like this part. In contrast to the discussion, I feel that here the strength of statements fits together with what you did in your work and with the related uncertainties.

L478: the effects -> the potential effects

L484: Would you not rather recommend a proper validation of you model before upscaling it?

Technical corrections

L11: "In particular": Seems to be the wrong transitional phrase. Please rewrite.

L13: “grass-buffer-strips”: Is that the correct term/correct spelling? (Revise in whole manuscript.)

L31: increase -> increases

L44: infra-structure -> infrastructure

L75: weekly -> weakly

L89: Baldegg Lake -> Lake Baldegg (revise in whole manuscript; see for example https://doi.org/10.1039/D0EM00317D)

L97: of the canton of Lucerne. (revise in whole manuscript)

L111: field-blocks -> field blocks (Revise in whole manuscript.)

L114: Consider just writing km^-1

L120: Elevation -> The elevation

L141: water discharge values -> discharges

L194: Make a reference from this. No URL directly in the text. Check if there’s a permanent identifier/URL.

L221: Wrong table referenced.

L235-238: Difficult to read. Rephrase.

L292: This can be easily visualised -> This is shown

L257: Is it a hydrological or hydraulic shortcut?

L306: increased in -> by or to

L372: In Figure 7, you write “short-cut”, while in the whole manuscript you wrote “shortcut”. Please adapt.