Dear Authors,

I read your manuscript and I appreciated the idea and motivation, but I was left with many questions.

I do not have a problem with the use of WRF model data to predict hourly rainfall but the prediction needs to be validated with some field data. I understand no hourly data are available in the catchment under study, but are there any data from adjacent areas to verify a degree of correlation? This verification should be done not only for the specific event but in general (e.g., over a whole year) to prove that your approach can be extended and used as a prediction tool. Consider, for example, if your WRF predictions are systematically an overestimation. You still would capture the debris flow events you were seeking, but you would also launch many false alarms.

Further, I understand you validated the approach using a rainfall event during which a disaster was actually triggered. This is ok but it is only half of the validation, namely a true positive identification in space and time. What about another event with similar characteristics that did not trigger debris flows in that catchment? Or the same event but in an adjacent catchment where no debris flows occurred? To be usable as a warning system, your approach should also be able to identify true negatives in space and time.

Further, you had to make assumptions due to lack of field data (e.g., on the pre-rainfall moisture) but you did not discuss how reasonable your choice was or how sensitive the result is to a change in the chosen value. In other words, where does the 5% moisture comes from? Is it supported by field or lab experiments? What changes if you use a moisture of 0% or 20%?

Finally, the DEM resolution. 30 m does not really seem great at your scale, with a catchment of few km. I agree that resampling cannot improve the result (because a smooth DEM remains smooth after resampling), but what about an actual high resolution DEM that more closely follows the roughness of the morphology? If not available in this catchment, couldn't the authors study this sensitivity in another location with better data, to assure the reader that the result remains reasonable?

General comments

Thank you for the opportunity to provide a peer review for this manuscript titled “Numerical model derived intensity-duration thresholds for early warning of rainfall-induced debris flows in the Himalayas” (nhess-2022-297). This study uses the Weather Research and Forecasting (WRF) model to estimate hourly rainfall time series at four meteorological station locations near the Kedarnath catchment, Uttarakhand, India, which record daily rainfall totals, during a debris flow event that occurred in June, 2013. A previous study mapped 120 debris flows resulting from this event in the catchment. This study estimates the volume of debris flows during the 2013 event from this inventory with an empirical relationship originally developed for Taiwan. The authors then use this estimated volume, along with an averaged precipitation time series across the four stations, to calibrate a numerical debris flow initiation and runout model for this event. With the calibrated model, the authors simulated cumulative debris flow volume with time as a function of average rainfall intensity for a range of intensity scenarios. The authors plot the time to initiate a debris flow against the average rainfall intensity in these scenarios to define an I-D threshold for the Kedarnath catchment.

While the idea to use a weather forecasting model coupled with a numerical debris flow model for landslide early warning in regions without available hourly rainfall data is intriguing and within the scientific scope of NHESS, this study will need serious and substantial modifications to both the analysis and manuscript before it can be considered for publication in NHESS or any other journal. At this stage, this manuscript does not meet the standards for publication in NHESS.

I summarize my main comments on the manuscript and the analysis here, and then provide more specific comments in the next sections.

Manuscript:

• The abstract makes various statements that are not supported by references or analysis in the main text and does not report the key results. The abstract suggests that the estimated I-D threshold will be used in a LEWS, but this is not validated or sufficiently discussed in the main text.
• The introduction makes numerous incorrect statements, lacks sufficient supporting literature, and does not clearly define a research question or objective.
• The methods section does not provide sufficient information to reproduce the analysis or to evaluate the validity of the results. It does not meet basic quality standards, such as defining all parameters. Some key parameters for the debris flow model are reportedly set by “calibration and back analysis,” but the details of said calibration are missing altogether.
• The results section, which is only 12 lines long, includes methods descriptions and does not describe the key results of the study.
• The discussion section does not discuss the results of the study or their implications. While it points to some limitations of the analysis, it importantly fails to evaluate the usefulness of the identified threshold for early warning, as suggested in the title.
• The conclusions section repeats introductory and methods material, but does not reach substantial conclusions based on the study’s results.

Analysis:

• This study uses the WRF model to estimate hourly rainfall across the catchment during the 2013 debris flow event at 1.8 km resolution. However, although daily meteorological station observations are available at four locations near the catchment, there is no validation or analysis of how well the simulated hourly precipitation totals match the daily totals at each station. Such a validation is required, particularly because this study proposes using the WRF model as an approach for areas without hourly data.
• Despite running a spatially explicit weather model and a spatially explicit debris flow model, the authors have chosen to drive the debris flow model of the 2013 event using an averaged hourly precipitation time series at four stations with an elevation difference of ~5000 m. I strongly question this choice, as such an elevation difference likely leads to substantial variations in rainfall intensity across the catchment (Destro et al., 2017; Iadanza et al., 2016). I recommend taking advantage of the available spatially explicit WRF rainfall estimates to drive the debris flow model.
• The debris flow model was calibrated using an empirical estimate for debris flow volume during the 2013 event. The baseline volume estimate was made using an empirical equation originally developed for Taiwan, which is of questionable validity in this different setting. Potentially resulting from this or other sparsely documented modeling choices, the debris flow model substantially overpredicts debris flow areas compared to the mapped inventory, but this is not discussed. Meanwhile, an analysis of how well the model could predict debris flow timing during the 2013 event is missing. Such an analysis is crucial for evaluating this model’s usefulness for early warning.
• The scenario analysis conducted to determine points for the I-D threshold relies on constant precipitation intensities, which is unrealistic for any rainstorm. The shape of the hyetograph is important for determining whether or not landslides are triggered (D’Odorico et al., 2005), and I therefore question whether a constant precipitation intensity can sufficiently represent triggering rainfall for estimating warning thresholds. A potential alternative approach could be to use a precipitation generator to run a suite of scenarios and use these to investigate triggering thresholds (Thomas et al., 2018).
• No uncertainty of the identified I-D threshold is estimated or discussed.
• Although this threshold is apparently intended for use in a territorial LEWS, there is no validation of this threshold’s performance for early warning.

Specific comments

Title

The title is clear, but promises early warning applications, which are not analyzed and barely discussed in the text. Himalayas suggests a broad region, please specify (e.g. “a Himalayan catchment”).

Abstract

• Line 2: Many early warning systems at the territorial scale do not use I-D thresholds (Guzzetti et al., 2020; Scheevel et al., 2017; Peruccacci et al., 2017).

• Line 3-4: Introduction does not provide evidence for this claim.

• Line 5: Specify what the numerical model does. Does it only apply to extreme rainfall? If so, how do you define “extreme”? Not supported in the text.

• Line 7: Which input boundary condition? This is not described in the methods.

• Line 8: Specify which model.

• Line 9: Glossary not mentioned in methods.

• Line 11: Use of this threshold in a LEWS is not evaluated or sufficiently discussed in the main text.

Introduction

• Line 14: Although the frequency and magnitude of extreme rainfall may be increasing, there is to my knowledge so far no empirical evidence that shows that disastrous debris flows have become more frequent. These citations do not show it. Please adjust wording or include the relevant literature.

• Line 17: Debris flow impacts. Non-structural measures do not mitigate debris flows.

• Line 19: Adapt to what? Please specify.

• Line 20: These cover some regions, but few cover entire nations. Please reword.

• Line 21: This statement is incorrect and needs citations. I-D thresholds are rarely estimated using forecasts, but are usually determined using observed rainfall.

• Line 24: Needs citations. Consider (Intrieri et al., 2013; Segoni et al., 2018; Stähli et al., 2015) and references therein.

• Line 30. This statement appears to be incorrect. Figure 6 of (Mathew et al., 2014) presents an I-D threshold with points with <24 hour durations.

• Line 32. This statement needs references.

• Line 34. References.

• Line 36. This statement is incorrect. Using statistical correlations of past events and monitoring data does not inherently generate inaccuracies. Please rephrase or specify.

• Line 39. This is indeed the case, but needs references.

• Lines 37-43. Needs references throughout.

• Line 45. The Weather Research and Forecast model is incorrectly cited. Please cite the original authors of the model.

• Lines 44-55. Here, a clear research objective or question should be specified. While this section states that this study presents a framework for an early warning system, no such system is described in the methods or results. How such a system would have performed during the 2013 floods is not analyzed, as suggested in lines 46-47.

• Figure 1. Specify what “Location 1-4” mean (meteorological stations). Please use the same elevation color scheme for inside the catchment and outside the catchment. The yellow landslide color is difficult to see, please consider a different color.

Study area and characteristics of the disaster

• Line 58: What is a fragile landscape? Perhaps prone to slope failures?

• Line 59-60: Show these faults on Figure 2.

• Line 61. The major rock types listed here do not match Figure 2. Please revise.

• Line 63. Please define “extreme rainfall” in this case. This suggests that over 6000 landslides occurred, but many fewer are shown in Figures 1 and 2. Why?

• Line 65. Reference for number of casualties and economic impacts needed.

• Figure 2. It is difficult to distinguish the red and black debris flows / slides in 2b.

• Figure 3. Please label Chorabari glacier lake on Figures 1 and 2.

Data and methods

• Data and methods general comment: this section does not provide sufficient detail to reproduce or evaluate the results, and is somewhat difficult to follow. Particularly, not all parameters are defined in the text, models are mis-cited, datasets are not cited, and key modeling choices and approaches are not described. This section must be more thorough.

• Line 75. Please explain briefly what this model is, what it does, and what it is used for in this study. Model needs a citation.

• Line 77. Figure reference wrong, please double check and correct throughout the manuscript.

• Line 79. I infer that Locations 1-4 are meteorological stations that record daily rainfall, but this needs to be specified in the text.

• Figure 5. I am not an expert in weather models, but I suppose that the information presented in this figure would not be sufficient to reproduce the results. I recommend creating supplementary tables that specify the inputs used for all models. All datasets require citations.

• Figure 6. From this figure, I would like to be able to evaluate whether the simulated hourly rainfall time series at each of the stations matches the daily records. Please rescale Fig 6a such that this is possible, or better yet, perform such a validation.

• Line 80. Does the WRF model only output one possible time series? Or did you somehow select this time series from a range of options? How sensitive are these results to inputs and modeling choices? Please document any modeling choices or selections.

• Line 81. This states that the authors have averaged the hourly precipitation time series at the four station locations and used this to drive the debris flow model. I do not understand this choice. From Figure 1, I infer that between Locations 1-2 and 3-4, there are 5000 m of elevation difference. I would expect this to introduce substantial variability in rainfall intensities (Destro et al., 2017; Iadanza et al., 2016) and therefore do not expect an average to appropriately capture this event. I do not understand why, when a 1.8 km resolution time series over the catchment is available, this information was not used to drive the debris flow model. I would recommend taking advantage of this available information, but at the very least, a sound justification of averaging is needed.

• Line 84. Here, please also briefly describe what this model is, what it does, and what it is used for in this study. (e.g. “We use a numerical debris flow initiation and runout model…”). Siva Subramian et al., 2021 is a pre-print; this is not a sufficient citation. More detail is needed in this manuscript describing this model.

• Line 91. Depth of soil or regolith is a very important parameter. How was this determined? Although Figure 8c plots “Soil Depth,” this is not described anywhere in the text. The field work photos from Figure 3 do not suggest much soil development on these slopes, so is this actually regolith depth?

• Line 94. “based in part” – which part? Again, the modeling strategy needs a more thorough description in this text.

• Line 96. All parameter symbols in Table 1 need to be defined.

• Line 97. Why do you choose 0.05 m3/m3 as an initial moisture content across the entire catchment? Is this reasonable? From Figure 6, it appears that it had been raining in the days prior to the event, so dry hillslopes may not be an appropriate assumption. Why not spin up the hydrological model with time series from before the event, as this is likely available from the WRF model?

• Line 99. How is the hourly rainfall data used with a time step in seconds? Please specify.

• Line 102. What stream ordering system is referred to here? It would help to label these on Figure 1 or create another figure.

• Table 1. Please describe all symbols used in the text. The “calibration and back analysis” for d50, delta_e, and delta_d is not described anywhere. This is a major issue, as the values of these parameters may have a strong impact on the results. Are these values justified considering your experience in the field? Judging from the photos in Figure 3, I’m not convinced that a d50 of 2.0 mm is appropriate, for example. In any case, some sensitivity analysis should be reported and discussed.

• Line 104. Please use spellcheck throughout. See comments on I-D thresholds for LEWS in intro.

• Line 108. ‘Berti…’ - this should be moved to the discussion.

• Line 110. The choice of “inter-event-time” varies widely between studies. Jiang et al., 2021 will have made one choice, but there are many others in the literature (Segoni et al., 2018). Please describe and justify your choice here.

• Line 112. Please just describe your modeling approach here. The relationship between physical processes and statistical thresholds is material for the discussion.

• Line 116. There is no methodological description of how the model was calibrated “above”. This must be added.

• Line 117. I would make it very clear that you are now moving away from the 2013 event and into scenario analysis. This was hard for me to follow.

• Line 119. Please specify what confluence is referred to here.

• Line 119-120. This method needs much more explanation. There are many statistical methods in use to establish I-D thresholds (Segoni et al., 2018, 2014; Staley et al., 2013; Brunetti et al., 2010; Scheevel et al., 2017). How do you select the threshold here?

• Line 120. I was confused at this point that the text moves back to the 2013 event. I would recommend restructuring to separate the analysis of the 2013 event from the scenario analysis for the I-D threshold.

• Line 126. What values were used for I, D, and C_r in this equation? How did you define the rainfall event? I do not necessarily expect an empirical equation originally developed for Taiwan to be a reasonable approximation of debris flow volume in the Himalaya. Is such an equation transferable? Why? This needs to be discussed.

• Figure 9. It is not clear if this is a schematic figure or if it is results. If it’s a schematic, please note that there is rarely such a clean separation of non-landslides and landslides, such that there are often rainfall events that exceed the threshold but do not trigger landslides.

• Line 128. Geological Index based on lithology needs explanation and documentation of values used.

Results

• General comment on the results section: This section is much too short, and does not describe the key results of the study. These should be presented for the reader.

• Lines 130-136. As I mentioned previously, this calibration needs to be documented in the methods section. The similar volume estimates are by design, as the numerical model was tuned to achieve this. However, judging from Figure 10, the model substantially overestimates the spatial area of debris flow deposits. I question whether such a model can be “considered calibrated.” At the very least, this overestimation must be discussed. I would also like to see evidence that the numerical model can sufficiently reproduce debris flow timing during the 2013 event, not just volume. This is key if such a model is to be used for warning.

• Line 137. The 10 mm/hr plot should be shown in Figure 11. I am not convinced of the choice of intensities. First, using a constant intensity over the course of the scenario is unrealistic, even if I-D thresholds are often based on average intensities. As we can observe from Figure 10, intensities over the course of the 2013 event varied, and the average intensity was certainly less than 20 mm/hr, perhaps less than 15 mm/hr. The peak intensity at any location during this event was less than 40 mm/hr (Figure 6), but the scenarios continue up to 90 mm/hr. Since the shape of the hyetograph influences landsliding (D’Odorico et al., 2005), I question whether the choice of a constant intensity can sufficiently capture triggering rainfall here for use in a warning threshold. An alternative approach could be to use a rainfall generator to run many scenarios and use those to investigate thresholds. See, for example, (Thomas et al., 2018).

• Line 143. It is not clear what event is referred to here, is it the 2013 event, or is this threshold valid for any rainfall event. “Material parameters similar…” were these adjusted after calibrating the model or are they the same?

Discussion

• Lines 148 – 150. These require references.

• Line 149. This argument states that previous thresholds are insufficient, but this study has provided no evidence that the estimated threshold would perform better in a warning system. Such evidence is required to support this argument.

• Line 151. This argument states that runoff induced erosion occurs during extreme rainfall lasting only a few hours, but the 2013 event studied here appears to have lasted for days. This is a break in logic.

• Line 152. See comments in intro on LEWS in other countries. Also, ID thresholds are estimated, not forecasted.

• Figure 12. The comparison between the Lakhera et al., 2020 threshold and the threshold estimated in this study is not valid. The Lakhera et al., 2020 threshold plotted here is specified as I_max, whereas the threshold estimated in this study is based on average intensity. Lakhera et al., 2020, specify thresholds for debris flows and debris slides, but the threshold plotted here is for all mass movements.

• Line 154. In the introduction, Mathew et al., 2014 is cited, which provides an I-D threshold that appears to be based on hourly data. This would be an additional point of comparison. Furthermore, as this is a publication for an international journal, a comparison of these results with the international literature is warranted. (Guzzetti et al., 2008; Segoni et al., 2018) are starting points. Importantly, there should be a discussion of why the results found here may be similar or different to other results reported in the literature. (Bogaard and Greco, 2018) may be helpful for this.

• Overall comment on the discussion: Since this study intended to estimate I-D thresholds for early warning, there must be a discussion of performance for early warning, but this is missing altogether. At a bare minimum, would this threshold have successfully warned for the 2013 event? How often would the threshold be exceeded otherwise, resulting in false alarms? Is there any case in which missed alarms would occur? (Staley et al., 2013) could be a starting point for considering this.

• Additional overall comment on the discussion: many limitations are listed, but without discussing how these might impact the identified threshold. Indeed, many modeling choices were made throughout the study, and these may induce uncertainty in the threshold, but that uncertainty is not quantified or discussed. The discussion should address these sources of uncertainty.

Conclusions

• Overall comment on the conclusions: The conclusions section repeats introductory and methods material, but does not reach conclusions based on the results presented in this study. The final statement that the approach presented in this study is promising for establishing Te-LEWS in new geological settings is not supported by the analysis or results presented in the study.

Technical corrections

I refrain from making further technical corrections at this stage, but recommend that the authors consult a native English speaker for proofreading. I also recommend that the authors review the quality standards for submission to NHESS or any other journal and ensure that their manuscript meets these standards prior to submission.

Review references

Bogaard, T. and Greco, R.: Invited perspectives: Hydrological perspectives on precipitation intensity-duration thresholds for landslide initiation: proposing hydro-meteorological thresholds, Natural Hazards and Earth System Sciences, 18, 31–39, https://doi.org/10.5194/nhess-18-31-2018, 2018.

Brunetti, M. T., Peruccacci, S., Rossi, M., Luciani, S., Valigi, D., and Guzzetti, F.: Rainfall thresholds for the possible occurrence of landslides in Italy, Natural Hazards and Earth System Sciences, 10, 447–458, https://doi.org/10.5194/nhess-10-447-2010, 2010.

Destro, E., Marra, F., Nikolopoulos, E. I., Zoccatelli, D., Creutin, J. D., and Borga, M.: Spatial estimation of debris flows-triggering rainfall and its dependence on rainfall return period, Geomorphology, 278, 269–279, https://doi.org/10.1016/j.geomorph.2016.11.019, 2017.

D’Odorico, P., Fagherazzi, S., and Rigon, R.: Potential for landsliding: Dependence on hyetograph characteristics, Journal of Geophysical Research: Earth Surface, 110, https://doi.org/10.1029/2004JF000127, 2005.

Guzzetti, F., Peruccacci, S., Rossi, M., and Stark, C. P.: The rainfall intensity–duration control of shallow landslides and debris flows: an update, Landslides, 5, 3–17, https://doi.org/10.1007/s10346-007-0112-1, 2008.

Guzzetti, F., Gariano, S. L., Peruccacci, S., Brunetti, M. T., Marchesini, I., Rossi, M., and Melillo, M.: Geographical landslide early warning systems, Earth-Science Reviews, 200, 102973, https://doi.org/10.1016/j.earscirev.2019.102973, 2020.

Iadanza, C., Trigila, A., and Napolitano, F.: Identification and characterization of rainfall events responsible for triggering of debris flows and shallow landslides, Journal of Hydrology, 541, 230–245, https://doi.org/10.1016/j.jhydrol.2016.01.018, 2016.

Intrieri, E., Gigli, G., Casagli, N., and Nadim, F.: Brief communication: Landslide Early Warning System: toolbox and general concepts, Nat. Hazards Earth Syst. Sci., 13, 85–90, https://doi.org/10.5194/nhess-13-85-2013, 2013.

Mathew, J., Babu, D. G., Kundu, S., Kumar, K. V., and Pant, C. C.: Integrating intensity–duration-based rainfall threshold and antecedent rainfall-based probability estimate towards generating early warning for rainfall-induced landslides in parts of the Garhwal Himalaya, India, Landslides, 11, 575–588, https://doi.org/10.1007/s10346-013-0408-2, 2014.

Peruccacci, S., Brunetti, M. T., Gariano, S. L., Melillo, M., Rossi, M., and Guzzetti, F.: Rainfall thresholds for possible landslide occurrence in Italy, Geomorphology, 290, 39–57, https://doi.org/10.1016/j.geomorph.2017.03.031, 2017.

Scheevel, C. R., Baum, R. L., Mirus, B. B., and Smith, J. B.: Precipitation thresholds for landslide occurrence near Seattle, Mukilteo, and Everett, Washington, Precipitation thresholds for landslide occurrence near Seattle, Mukilteo, and Everett, Washington, U.S. Geological Survey, Reston, VA, https://doi.org/10.3133/ofr20171039, 2017.

Segoni, S., Rosi, A., Rossi, G., Catani, F., and Casagli, N.: Analysing the relationship between rainfalls and landslides to define a mosaic of triggering thresholds for regional-scale warning systems, Natural Hazards and Earth System Sciences, 14, 2637–2648, https://doi.org/10.5194/nhess-14-2637-2014, 2014.

Segoni, S., Piciullo, L., and Gariano, S. L.: A review of the recent literature on rainfall thresholds for landslide occurrence, Landslides, 15, 1483–1501, https://doi.org/10.1007/s10346-018-0966-4, 2018.

Stähli, M., Sättele, M., Huggel, C., McArdell, B. W., Lehmann, P., Van Herwijnen, A., Berne, A., Schleiss, M., Ferrari, A., Kos, A., Or, D., and Springman, S. M.: Monitoring and prediction in early warning systems for rapid mass movements, Natural Hazards and Earth System Sciences, 15, 905–917, https://doi.org/10.5194/nhess-15-905-2015, 2015.

Staley, D. M., Kean, J. W., Cannon, S. H., Schmidt, K. M., and Laber, J. L.: Objective definition of rainfall intensity–duration thresholds for the initiation of post-fire debris flows in southern California, Landslides, 10, 547–562, https://doi.org/10.1007/s10346-012-0341-9, 2013.

Thomas, M. A., Mirus, B. B., and Collins, B. D.: Identifying Physics-Based Thresholds for Rainfall-Induced Landsliding, Geophysical Research Letters, 45, 9651–9661, https://doi.org/10.1029/2018GL079662, 2018.