General Comments:

This study conducted a sensitivity analysis of major parameters of the GLOF numerical simulation using the open-source software r.avaflow model to determine which parameters significantly affect outputs closely related to GLOF hazard and risk assessment, such as peak discharge, total discharge, flow arrival time, and reach distance. The manuscript is well-organized, and the references are appropriate, demonstrating the research’s significance and originality. All figures are of good quality, and the information is clearly presented.

I have commented on some points requiring minor revision. Please refer to the specific comments below for detailed suggestions for revision.



Specific Comments:

#1    l.86-89: Since the previous sentence already states that r.avaflow is open source, it would be better to delete “is open source and” from this sentence and instead write, “r.avaflow allows modification of all input parameters, ...”.

#2    l.159: The text refers to “Figure 1B and 1E,” but the subfigure labels are written in lowercase. Please make the labeling consistent throughout the manuscript. The same correction applies to the following text , figures, and supplement.

#3    l.165: If you use scientific notation, it might be more appropriate to write 300×10⁶ m³ as 3×10⁸ m³.

#4    l.171: Insert “shows” or another appropriate verb after “The map (a).”

#5    l.254, Figure 2: There is no legend for CFF in the figure.

#6    l.287: It would be better to write this as 10⁶-10⁷ m³.

#7    l.310: It says, “3.2.1 Digital elevation (1) and mesh size (2),” but some of the subsection titles from 3.2.1 to 3.2.6 include numbering, while others do not, which causes confusion. It might be better to remove the numbering from the subsection titles altogether.

#8    l.325: The unit should be written as kg m^-3. Similar unit errors are seen from this point onwards.

#9    l.329: It would be better to write this as 2.94×10⁸ m³.

#10    l.428-430: Table 3 does not show the results described here.

#11    l.431: It would be better to insert the unit (min) after the number 4.37.

#12    l.434-437, Figure 3: The figure title indicates that the left and right panels respectively show maximum flow heights and hydrographs, but this is the opposite. Additionally, the left panel of each subfigure (a)-(d) does not have a horizontal axis label. Also, it is not clear what the dashed lines in the right panel indicate.

#13    l.440-441: Although this description refers to Fig. 3, it is difficult to interpret the flow velocity characteristics from Fig. 3.

#14    l.456, Table 3: What does the “SL no.” in the leftmost column stand for?

#15    l.470-471: Is it not possible to determine from Fig. 4 how the solid volumetric portion exhibit fluctuates?

#16    l.486, Figure 5: What does the light-colored shaded area in each subfigure represent (e.g., 95% confidence interval)? It would be better to clarify this in the legend or caption. Additionally, the regression equation and the coefficient of determination seem to mismatch in some subfigures. Please verify to ensure there are no errors. These comments also apply to Figs. 6 and 7.

#17    l.493, Figure 6: The label on the vertical axis should be replaced with “total discharge.”

#18    l.533: From this point onwards, the CV value is written in units of %, but isn’t it common to treat CV as a dimensionless value if it is derived by dividing the standard deviation by the mean? If the coefficient of variation is expressed as a percentage, it would be better to clearly state this in the text.

#19    l.538-: The following descriptions of the regression analysis results show some differences between the coefficients of determination shown in Figs. 5-7. Please check and correct as necessary.

#20    l.548: Insert “(Fig. 6)” after “total discharge.”

#21    l.558: It states that the volume range is 4×10⁶ m³ to 10×10⁶ m³, but the figure shows normalized parameter values, making it difficult to correlate with the actual volumes. It would be better to include the normalized values ​​here. The same comment applies to the description of C_E in l.560-562.

#22    l.585-586: The description “ranging from HMA-DEM (8 m) to SRTM GL3 (90 m)” would be better described as “with resolutions ranging from 8 m (HMA-DEM) to 90 m (SRTM GL3).”

#23    l.686: This is the first time mention of δ in the discussion section, so it would be better to use the notation “basal friction (δ)” as with the other parameters.

#24    l.700: This is likely a mistake; it should read, “This is because C_FF controls the mobility of the fluid part.” Also, it would be better to write the “FF” in C_FF as a subscript throughout the text.

#25    l.833: “helpin” should be replaced with “helping.”





That concludes my review.





Takashi Kimura

Assistant Professor, Graduate School of Agriculture, Ehime University

Dear Editors,

Dear Authors,

Thank you for giving me the opportunity to review manuscript egusphere-2024-1819, “Exploring implications of input parameter uncertainties on GLOF modelling results using the state-of-the-art modelling code, r.avaflow” by Sonam Rizin and co-authors. In this study, the authors investigate the sensitivity of key modelling outputs (peak discharge, flood volume, and flood arrival time) in r.avaflow to variations in nine selected input parameters. r.avaflow is a widely used software for simulating catastrophic mass flows, making this analysis highly relevant for researchers and practitioners interested in modelling glacier lake outburst floods (GLOFs). The authors conclude that the volume of landslide material entering the lake has the greatest influence on simulation outcomes, followed by the digital elevation model (DEM) and its resolution, while other parameters show comparatively lower impacts.

This study offers valuable insights for users of r.avaflow, especially those interested in identifying the parameters that exert the strongest control over GLOF modelling results. However, several aspects of the study require clarification and further discussion to strengthen its contributions:

Major Comments

1) Parameter Selection and Exclusions

While focusing on a subset of parameters is practical given the complexity of r.avaflow, the rationale for selecting exactly these nine parameters remains unclear. For instance, the authors treat lake bathymetry and volume as constants, yet these are highly uncertain and challenging to estimate, particularly for remote glacier lakes. Would a shallower lake generate a higher displacement wave, potentially resulting in a larger peak discharge? What about the height of the moraine dam and a potential bedrock sill beneath it? How would this (not uncommon) setting change entrainment and accordingly, peak discharge, once that bedrock sill is hit? Similarly, the study does not examine the effects of varying the velocity or grain size of the landslide entering the lake. These factors may influence wave dynamics and warrant at least a discussion in the context of the available literature.

Moreover, with r.avaflow offering more than 30 tunable parameters, it would be helpful to understand in more detail whether the excluded parameters were found negligible or simply beyond the scope of this study. While a comprehensive sensitivity analysis of all parameters may be impractical, a broader discussion of the omitted parameters' potential roles would add value to the manuscript.

2) Parameter Value Ranges and Physical Plausibility

The ranges of parameter values used in the sensitivity analysis appear to be informed by prior studies, but it is unclear how well these values indeed reflect the physics of GLOFs. For instance, are the chosen ranges realistic for a variety of glacier lake settings? While the authors acknowledge the challenge of equifinality—achieving the "right" results for the "wrong" reasons—this issue is amplified by the absence of validation against real-world cases.

Applying the sensitivity analysis to a documented GLOF event, such as those at Langmale, Salkantay, Elliot Creek, Ranzeria Co, Chongbaxia Co, or Tam Pokhari, could provide a much-needed validation framework. This would allow the authors to test whether the chosen parameter ranges lead to realistic flood scenarios and to assess how uncertainties in key parameters (e.g., landslide volume) translate into variability in flood predictions.

3) Interactions Between Parameters

The current analysis isolates each parameter's effect by varying one parameter at a time. While this approach is useful for identifying individual sensitivities, it does not account for potential interactions among parameters. For example, do certain combinations of parameter changes amplify or mitigate the overall effects on model outputs? Exploring such interactions is crucial for providing a comprehensive understanding of the system dynamics and would greatly enhance the applicability of the study’s findings.

4) Practical Recommendations for Model Users

The manuscript would benefit from more confident and actionable recommendations for users of r.avaflow. Currently, the authors caution against overconfidence in interpreting parameter values but stop short of providing concrete guidance. A “starting point” for parameter selection or a framework for iterative refinement would be invaluable for new users. Additionally, the discussion could include recommendations for scenarios where multiple parameters are varied simultaneously, which more closely reflects real-world uncertainty.

5) Clarity and Presentation

Finally, I suggest a careful revision of the text and figures to improve clarity and polish. For example, the abstract should be revised to better summarize the study's scope and relevance, the study area, and implications. Specific suggestions for improvement are provided below.

Specific comments

L2: The abstract should explain why r.avaflow represents the “state-of-the-art” in GLOF modelling, otherwise please consider removing this phrase.

L9-13: Please try to express one idea per sentence. This opening sentence has at least three, while also including some confusion. As far as I understood, this study does not include direct measurements, which seem to be a core motivation in both forward modelling and back analysis?

L15: How many different GLOFs did you assess in these 78 simulations? What is the key criterion that you evaluated the suitability of the model? Some kind of intersection over union between mapped and simulation runout areas or flow depth?

L16: Please add a motivation why r.avaflow was selected among the many available mass flow models.

L17: You need to introduce which GLOF exactly you model, as certainly not every GLOF is triggered by a mass movement entering the lake. It’s also important to emphasize the dam type: moraine-dammed, bedrock-dammed, or a combination of both?

L17-18: I was surprised not to see lake depth/ volume/ bathymetry in this assessment. Doesn’t the – in many cases unknown – depth of the lake play a fundamental role on the amount of water that can be pushed out of the lake?

L16 and 20: What is ‘GLOF output’? Please rephrase and explain.

L19-21: Somehow repetition of the preceding sentence. You could just add the CV for every variable in the preceding sentence, which could create a bit more space for other findings of your study.

L24: Unclear what you mean with ‘statistically’?

L26-30: Consider updating these numbers with the global glacier lake inventory presented by Zhang et al. (2024), Communications Earth and Environment, as those presented by Shugar et al. 2020 are subject to large errors.

L28: Seems that this study focuses on glacial lakes in HMA, so please state this here.

L32: replace ‘mass inputs’ with ‘steep slopes’?

L33-36: Statements about GLOF frequency seem a bit out of place here, as you are describing the physical process of GLOF triggering before and after?

L46: Remove ‘the’ in front of HMA.

L50: Consider avoiding subjective terms such as ‘unfortunately’.

L58: Consider adding a note that dams not necessarily fail completely?

L63: Including the availability and entrainment of sediment and its grain size distribution?

L72-73: Either ‘most’ or ‘all’

L75: ‘sediment’ entrainment?

L79: Consider adding ‘into Imja Lake, Nepal’

L96: What ‘model outputs’ does r.avaflow provide?

L105: Does the velocity of the mass movement entering the lake also play a role?

L116: ‘data acquisition’?

L119: What kind of ‘data’? Gridded ice thicknesses from ice-flow inversion models?

L120: I would rather deem these ‘arbitrary’ ice thicknesses.

L124: My background is more in statistics, where parameters (coefficients) are usually distinguished from variables (predictors). It would be good if you could define what a ‘parameter’ is in your study because sometimes I have the impression that you are talking about model coefficients or constants that are free to change or optimize in r.avaflow, rather than the input datasets, which I would rather call a variable.

L135: Does ‘employing’ have the same meaning as ‘inferring’ here?

Table S1 is labelled Table S4 in the Supplementary Material. Please correct. In any case, I found the choice of these nine parameters a bit brief and poorly referenced. It seems that you keep all the 29 other parameters in your experiments fixed. So, one might argue that playing around with these 29 parameters could equally reproduce our outcomes if keeping your selected 9 parameters fixed. Please add more motivation why these 9 are so much more important than the others.

Figure S1: Please explain the symbols in the caption or in the x-axis. In addition, I wondered if and why it makes sense to stay within the limits of the parameter values reported previous studies? Do they represent the physically plausible range? Does the local setting match with that represented in your study?

L167: Not sure where you showed the ‘high outburst susceptibility’?

Figure 1b: Loc-4 occurs twice? I cannot read the right black label of the yellow dot, starting with “Lo…”. What does ‘Loc’ actually mean? Please add in the figure caption.

L179: add access date of OSM data.

L185 is this infrastructure within the first 10 km downstream of the lake? Or more?

L189: Check grammar.

L225: What does ‘are scaled with a solid fraction of the flow material’ mean? How do you know the flow material? Please explain.

Figure 2: why do you show here a distinction between H, M, and L scenarios? I guess this stands for High, Medium, and Low, but this appears neither in the figure caption, no in the manuscript, no?

L258: This reference (Mergili and Pudasaini, 2024) is not part of the list of the list of references, but what I understand is that this is a link to a website? https://www.landslidemodels.org/r.avaflow/direct.php

L287: Are these landslide volumes representative for previous landslide-generated lake outburst floods? I still wonder how much the landslide velocity modulates the displacement of water. If the landslide enters slowly (<1 m s-1) the water body, would you expect a much lower wave height?

L292: Why is that range ‘conservative’? What I read from these references is that these values were rather ‘informed by the expertise of the model developers’? In my opinion, ‘conservative’ implies that the selected values rather seek to model the lower bound of potential GLOF discharges to avoid gross overestimates. By contrast, you seem to rather aim for the mean estimate for those parameters by excluding values outside of the interquartile range.

L314-316: I wondered why you decided not to use the ALOS PALSAR DEM (12.5 m resolution) as a compromise between the high-resolution HMA and low resolution SRTM DEMs?

L318: Do these mesh sizes imply that you did not run the models using the original mesh size of a given DEM? If so, why? In addition, which resampling algorithm did you choose to change the grid resolution? The DEM in some of the figures looks a bit ‘edgy’, and I wonder if that the effect of a Nearest Neighbour interpolation.

L322: Add ‘the’ or ‘a’ in front of GLOF.

L323: Remove ‘into’

L329: In my opinion, this is a really important point in the present manuscript: what is the volume of this lake? Table S2 suggests that the empirical equations alone differ by a factor of two (205 to 381 Mil m³), without accounting for uncertainties in the model parameters itself. What the equation yield is only the mean volume for a given lake area; the underlying data in lake-area-volume-relationships, however, may cover one to two orders of magnitude of estimated volumes for a given lake area. So, I wonder how representative these estimates are in order to provide a meaningful estimate of flood volume and discharge. It was also interesting to see that the volume from ice thickness models (last row in Table S2) almost triples the mean of means that you seem to use. In L338, you mention that you somehow adjusted the bathymetry, but how? Did you make the lake shallower? In any case, I would strongly encourage to also include a varying lake bathymetry in the variables that you assessed, as it remains unclear how much uncertainties in that variable propagate in your overall model result.

L340-345: For non-experts, it might be good to show a histogram showing volumes of historic mass movements that generated GLOFs. This might help underline if these ten values cover a physically plausible range or not.

Figure S6: needs a color key that distinguishes the DEMs

L354-355: So these six source locations alone give you 60 (6 locations x 10 volumes from 1 to 10 Mil m³) simulations? Or do you consider that all these sources produce mass movements simultaneously?

L371-379: Again some assumptions where we need more information: why can only the moraine provide material for entrainment? Isn’t the broad floodplain downstream of Thorthormi Glacier full with sediments that can be entrained during a flood? In addition, what is the entrainment height? How did you measure the height of the moraine? How do you know that there isn’t a bedrock sill, at which erosion might come to a halt during a GLOF?

L381-389: How much sense does it make to treat these variables independent of each other? I.e. only varying the internal friction angle, while keeping all other fixed? This is not really physically plausible, isn’t it?

L401: What are these ‘outputs’?

L409: Isn’t that standard deviation strongly dependent on the range of input values that you assessed? A narrower range in the input parameters will give you a smaller range in output values?

Figure 3: what do the dashed lines show? Typo in some panels: ‘reolution’

L445-446: Check grammar.

L457: Please explain which volume you used in this simulation from the different sources (Loc-1 to Loc-6).

L465: Peak discharge is measured m³ s-1. Do you mean 180 x 10³ m³ s-1?

L466: 60,000 ³ s-1? I am really unsure about the values and the scale you show in Figure 4. How realistic is a peak discharge of 180000 ³ s-1? This would be one of the largest GLOF magnitudes in human history.

Figure 5: Just to make sure that I understood it correctly: Linearly decreasing, and particularly negative, values mean that this parameter decreases the variance in the output? In other words, if you would still increase the value of this parameter, then your variance in GLOF discharge or volume would almost vanish?

Figure 6 and 7: sometimes you use double brackets )). Does that have a specific reason?

L535: Word(s) missing at the end: GLOF … discharge, volume, arrival time?

L575: remove ‘the’ in front of ‘multiple’

L580: either ‘datum’ or ‘dataset’

L589: The effect of the DEM on GLOF output is indeed really interesting, but I could not follow your argument of river channel conveyance changing this output? What is this effect, could you explain this in more detail? I initially speculated that it’s the lower surface roughness and friction stemming from coarser DEM resolutions that causes higher discharges? What is your take on this?

L594-597: Interesting thought, do you have any evidence for this effect? Specifically why those changes might amplify GLOF magnitude?

L604: rephrase to ‘a co-registration algorithm developed by Shean et al. (2016) or so?

L611: ‘DEMs’

L615-617: A problem with this conclusion is that you inherently infer that the simulations using the HMA-DEM are better/ more realistic or whatsoever, as they produce smaller discharges. However, as you do not provide any validation/ reference dataset, it is difficult to judge if one DEM really outperforms the others. You also have no independent validation dataset to show that the HMA-DEM has fewer errors than the others, nor do you show how the noise in these DEMs propagates in your simulations.

L646: To be fair, you also did not explicitly mention the angles and directions of these mass movements anywhere in your manuscript.

L659: A provocative conclusion would be that you can obtain highly contrasting GLOF discharges for the same lake, just by moving the initiation zone of your mass movement to another location of the slope. How much sense do these worst-case scenarios make then, if there is – in theory -  infinite combinations of source location, volume, velocity, etc. of a landslide entering a lake?

L661-662: I am not sure whether this conclusion is valid. You modelled the effect of increasing landslide size only for the case with the highest consequences (loc-1). Would you expect similar effects on GLOF peak discharge, if you were to model landslide impacts from the other locations, say loc-3, where the wave might be dampened as it is first pushed against the opposite valley wall?

L697: How would one measure δin the field? Could you advise? And what is a statistically substantial sensitivity analysis and how would one do that?

L700: Verb missing.

L717: How should that ‘careful treatment’ be done? By using a range of values and aggregating the results? Many papers claim that the research was done with utmost care – here you have space to explain and give recommendations how to approach these uncertainties in GLOF modelling.

L722: Check grammar.

L729: You often mention that results should be interpreted in caution. I agree, but what exactly is it that make you caution against these results and what is your suggestion to move forward? Users might want to have some guideline how to attain a certain level of confidence in their r.avaflow model results.

758: one ‘arrival and’ used too often?

L770-781: That paragraph offers no new insights and can be deleted entirely in my opinion.

L788-789: Again, I feel a bit stranded with this note of caution. In which settings do you expect the tested parameters to be substantially different from your setting?

L800-801: It is really surprising to me that this study did not assess any parameter interaction, especially as you point at the strong effects of some of the parameters. What would happen with flood volume and discharge, if a user selects a very large DEM resolution, a high landslide volume, and a high entrainment coefficient? Will you just get super high GLOF outputs or do you expect them to cancel out each other? I understand that your sensitivity analysis seeks to model the output by varying one parameter and keeping all others constant; however, I would be delighted to see some kind of recommendation, which parameter values might be suitable for a first try, and in particular, what would be a really bad initial set of parameters in GLOF forward modelling.