the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 15 Oct 2025
Leveraging hillslope connectivity for improved large-scale assessments of landslide risk
Abstract. Landslides hinder sustainable development in mountain regions, threatening livelihoods and impacting linear and water infrastructure. Susceptibility maps are a common tool for estimating and managing landslide hazards, exposure, and risks. Yet, susceptibility maps omit hillslope connectivity, a critical shortcoming for mapping the magnitude of landslide hazards, including cascading hazards from slope failure and downslope runout. Herein we propose the COHESION (COnnected HillslopE SusceptibIlity for slOpefailure and ruNout) approach to couple susceptibility mapping with an assessment of hillslope connectivity to identify downslope-connected landslide objects (LSOs) and associated runout pathways. As we demonstrate for the Kaligandaki basin in Nepal, analyzing LSOs enables estimating the magnitude of slope failures in terms of mobilized sediment volume and to quantify additional impacts from landslide runout. After calibration using a remotely sensed landslide inventory, we find that 16 % of the basin’s slopes are susceptible to failure, while an additional 9 % of the basin area is impacted by runout. Around 33 % of buildings and 65 % of roads in the basin are on susceptible slopes, while more than 27 % of buildings and 69 % of roads are in landslide runout pathways. Omitting runout from landslide assessments would thus result in a major underestimation of risk. Our results emphasize the importance of connectivity for slope stability modeling on landscape scales, leading to improved assessments of slope hazards and management of river basin sediments.
- Preprint
(8049 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (extended)
-
CC1: 'Comment on egusphere-2025-3733', Max Sutton, 25 Oct 2025
reply
-
AC1: 'Reply on CC1', Rafael Schmitt, 31 Oct 2025
reply
Dear Dr. Sutton, thank you for taking the time to comment on our manuscript. We appreciate your insights and we are glad that you found this manuscript interesting.
- Have you looked into the USGS Grfin tools as a model for connectivity? (Reid et al 2025 https://doi.org/10.3133/tm14A3). It is meant for highly mobile mass movements such as debris flows. If it applies in your case, it can automatically estimate connected areas, runout zones, etc and may be of interest to you.
Thanks for pointing us to the Grfin tool, after careful reading of the manual, this seems a very complimentary approach. Grfin can use a similar H/L ratio approach to what we use in here, in addition it implements more complex algorithms for different types of mass movements, as you say, highly mobile lahars, debris flows etc. Instead Grfin does not account for slope failure, so basically the source areas and size/volumes of those would need to be put in from a separate input. We will certainly discuss this in our manuscript.
- Another way of looking at connectivity is to use slope units (eg Woodard et al. 2024, https://doi.org/10.5194/nhess-24-1-2024) instead of a pixel/grid based approach. Using slope units has been shown to be effective at estimating susceptible area, partially because it inherently accounts for runout zones. It also provides a convenient division for aggregating other parameters such as slope over a vulnerable area. In the context of this paper, you probably would not use slope units for connectivity because it would change your volume and thus runout assessments, but I'm curious if it would be a good way to aggregate/interpolate soil or hydrology parameters
This is an interesting thought. From our reading it seems indeed as slope-units could be interesting to handle certain subcomponents, such as hydrology. However, in general, slope units seem to imply a lumping into larger units based mostly on terrain parameters / topography, which would mask more localized hillslope connectivity
- At what scales is COHESION effective? (Ie, local/basin level, regional level, global level? Similarly, especially given the coarseness of some of your inputs, have you tested if there is a minimum pixel size for which COHESION is reasonable?
We think that COHESION is most applicable on a watershed / regional scale. I think the question of scale really depends on the size of landslides that create most hazards in a given region. I would think that the herein deployed resolution (30m) is towards the upper limit of resolution, also because lower resolutions imply lower extremal values for slope.
- Your model is quite complex. While I appreciate the authors’ consideration of the many factors that impact landslide susceptibility, are all of these inputs relevant at this scale? Have you considered a factor analysis to identify the most important inputs and if you were to make the model simpler, how performance would change? I wonder if performance would actually increase, given how difficult it is to actually measure many of your input parameters? Your sensitivity analysis begins to answer these questions, but a factor analysis or similar technique could be a rigorous way to shed light on relative importance of parameters without having to do separate analyses for individual parameters
This is a good point. Indeed, we have included many parameters mostly to shed a light on potential applications. We did not sample the full parameter space, e.g., through a sobol analysis . We agree that this is an interesting area for more research, and exactly what we aimed to hint on with our parameter sampling. However, we think such an approach would me most yielding for a specific application. In this publication, we demonstrate the full complexity of the model and its modularity - including the stochastic handling of failure probability, the analyses of landuse change, and xposure. Likely, many applications would focus on more simple implementations (e.g., using a pre-derived susceptibility map), or using static rainfall thresholds. For that, it would then be quite useful to explore parameter sensitivity in more detail.
- Line 357: How sensitive are your conclusions to your choice of gamma? Larson 2010 found that small changes in gamma led to large differences in estimated landslide volumes, and that gamma changes with hillslope material type.
We did not yet include any analysis of gamma.
Many thanks again for those very pertinent comments, which we will aim to integrate in the MS.
Citation: https://doi.org/10.5194/egusphere-2025-3733-AC1
-
AC1: 'Reply on CC1', Rafael Schmitt, 31 Oct 2025
reply
-
CC2: 'Comment on egusphere-2025-3733', Max Sutton, 25 Oct 2025
reply
To the authors, thanks for a thought provoking paper! I have a few questions and comments after reading.
- Have you looked into the USGS Grfin tools as a model for connectivity? (Reid et al 2025 https://doi.org/10.3133/tm14A3). It is meant for highly mobile mass movements such as debris flows. If it applies in your case, it can automatically estimate connected areas, runout zones, etc and may be of interest to you.
- Another way of looking at connectivity is to use slope units (eg Woodard et al. 2024, https://doi.org/10.5194/nhess-24-1-2024) instead of a pixel/grid based approach. Using slope units has been shown to be effective at estimating susceptible area, partially because it inherently accounts for runout zones. It also provides a convenient division for aggregating other parameters such as slope over a vulnerable area. In the context of this paper, you probably would not use slope units for connectivity because it would change your volume and thus runout assessments, but I'm curious if it would be a good way to aggregate/interpolate soil or hydrology parameters?
- At what scales is COHESION effective? (Ie, local/basin level, regional level, global level? Similarly, especially given the coarseness of some of your inputs, have you tested if there is a minimum pixel size for which COHESION is reasonable?
- Your model is quite complex. While I appreciate the authors’ consideration of the many factors that impact landslide susceptibility, are all of these inputs relevant at this scale? Have you considered a factor analysis to identify the most important inputs and if you were to make the model simpler, how performance would change? I wonder if performance would actually increase, given how difficult it is to actually measure many of your input parameters? Your sensitivity analysis begins to answer these questions, but a factor analysis or similar technique could be a rigorous way to shed light on relative importance of parameters without having to do separate analyses for individual parameters.
- Line 357: How sensitive are your conclusions to your choice of gamma? Larson 2010 found that small changes in gamma led to large differences in estimated landslide volumes, and that gamma changes with hillslope material type.
Overall, I enjoyed the paper and anticipate sending it to some of my colleagues. Landslide runout is an important consideration in hazard assessment, especially since infrastructure tends to be on flatter runout zones versus steep slopes, and I believe this is an important direction for landslide hazard assessment to head in.
Citation: https://doi.org/10.5194/egusphere-2025-3733-CC2 -
RC1: 'Comment on egusphere-2025-3733', Anonymous Referee #1, 27 Nov 2025
reply
See the attached file.
-
RC2: 'Comment on egusphere-2025-3733', Anatoly Tsyplenkov, 19 Dec 2025
reply
In this manuscript, Schmitt and colleagues present COHESION, a connectivity-aware framework that augments landslide susceptibility mapping with empirical runout modelling. The article addresses an important gap by coupling susceptibility with runout to quantify cascading downslope hazard and exposure at landscape scale. In general, the workflow is clearly staged with helpful schematic figures and an explicit data-needs tables. Overall, the proposed approach is novel but yet reminded me of the framework currently used in New Zealand by Earth Science Institute (ex-GNS) for estimating landslide runouts and further risks (see Brideau et al. (2021) and de Vilder & Massey (2024)).
While I found the topic to be of high scientific and practical interest, I was disappointed by the quality of the paper. I have a feeling that the authors submitted the manuscript in a rush, omitting the proofreading step. This resulted in typos, a lack of measurement units in equations (line 176), inconsistency in abbreviations (for example, both FS and FoS were used for Factor of Safety), incorrectly rendered bibliographic links (see lines 84 and 204), and most importantly, the absence of a Conclusion section. Before applying for the next round of the pipeline, all these issues should be fixed.
That being said, I enjoyed reading this manuscript and found it very thought-provoking. I believe that the manuscript could be published in NHESS after major revisions and a second review round. See my comments below.
Major issues
- While the authors explicitly specified that LSO is the term borrowed from image classification (see line 116), I found it difficult to understand how the LSO corresponds to the landslide scar (or landslide initiation area). That is, as far as I understood it, the LSO is generally bigger than the landslide itself, as it tends to merge several scars together due to the coarse spatial resolution (30 m). It would be beneficial to the reader to illustrate the difference between the digitised landslide and the LSO on a figure (as early as possible in the text, but figure 4 suits this purpose nicely).
- The study area is located in the Nepal Himalayas, a region notorious for its landsliding processes. Indeed, the authors provide extensive information about the hydrometeorological setting (lines 129–136), but almost no data about the landslide processes. The reported quantitative information about the “larger” landslides (see line 143) is too vague from my perspective. It would be nice to read more about the nature of the landsliding processes, what the main drivers, sizes, and volumes are. Such a section in the first half of the manuscript may benefit the reader in terms of better understanding of the (quite) complex model (see my following comments). Yet, I understand that some of the questions I am asking are answered further in section 4.7, which is acceptable, but too late, in my view.
- My next question evolves from the previous one. Overall, I have an impression that the authors are proposing a framework which, by their design, is agnostic to the landslide-triggering event and landslide type. With a larger dataset and a slightly different approach, I would not mention that as an issue. However, since some of the modelling steps involve empirical equations by Rickenmann (2005) for runout length and Larsen, Montgomery, and Korup (2010) for landslide volume estimation, these were developed for a specific type of landslide. Rickenmann (2005) was super explicit that his empirical equation fits only debris flows. Larsen, Montgomery, and Korup (2010) developed equations for bedrock and soil landslides, and from the current manuscript, it is unclear which one the authors used and why. Moreover, the recent findings by Jones et al. (2025) indicate that empirical equations from the global and event datasets, the one the authors used, tend to critically underestimate landslide volume. I would like to see a discussion on this matter in the article.
- The Factor of Safety estimation (section 4.1) and hillslope hydrology (section 4.2) require a lot of very detailed data. However, Tables 1 and 2 show readers that the data sources usually have a pretty coarse resolution, up to 1 km. While it is totally acceptable to use global datasets in data-scarce regions for regional studies (I mostly work in ungauged basins myself, no judgement at all), it is a must to discuss the potential implications of input data resolution on landslide susceptibility and connectivity. This is especially vital for the assessment of threshold subsurface flow, water balance, and so on, which I assume are the most influential factors for landslide modelling.
- Validation is limited to comparing slope distributions via the Kolmogorov-Smirnov statistic, which is generally fine. Nevertheless, I have never seen that in susceptibility-related literature. Quantitative skill assessment against the inventory (e.g., AUROC, F1 measure, and so on) is more common. I would recommend the authors either use the generally accepted quantitative metrics (see Petschko et al. (2014)) or apply a bootstrap analysis of the KS statistic to make your quality assurance process more robust and defendable. In that case, it would be beneficial to the paper to mention why you chose not to use them.
- Last but not least, I found the discussion to be very limited, and the authors did not provide any comparison with existing models and tools for estimating runout distances, susceptibility, and connectivity. It would be nice to see how their approach fits within the existing frequentist frameworks developed in Europe (Steger et al. 2021; Steger et al. 2022) and New Zealand (Spiekermann et al. 2022) for joint landslide connectivity-susceptibility predictions. On the runout side, tools such as Flow-Py (D’Amboise et al. 2022), GPP (Wichmann 2017), Grfin Tools, and others should be discussed, and how COHESION complements or differs from these.
Minor issues
table 1 — as far as I know, SoilGrids has a resolution of 250 m, not 1 km.
figure 3 — it is not related to the figure per se, but I would like to read some discussion on this topic somewhere in the text. Why did you use reanalysis data, such as ERA-5L, for example? The northern part of the basin is represented with a very limited number of meteostations located mostly in the valley bottom. The quality of interpolation in the north-eastern part of the basin is questionable due to the clustering of stations.
figure 9 — the d facet has never been discussed in the text, and it is, in my opinion, the most controversial. Why does the cumulative percentage of roads for the “reforestation” scenario start from 20 something? Why not from zero? In the current form, I can interpret the results as that for the cumulative percentage of roads of approximately 37%, the failure probability equals roughly 0.6 for both reforestation and deforestation scenarios.
line 293 — specify the kriging approach you used as well as the quality assessment. In other words, how confident are you that your interpolation was good?
References
de Vilder SJ, Massey CI. 2024. Guidelines for natural hazard risk analysis on public conservation lands and waters – Part 4: a commentary on analysing landslide risk to point and linear sites. Lower Hutt (NZ): GNS Science. 81 p. Consultancy Report 2024/38.
Brideau, Marc-André, Saskia de Vilder, Chris Massey, Andrew Mitchell, Scott McDougall, and Jordan Aaron. 2021. “Empirical Relationships to Estimate the Probability of Runout Exceedance for Various Landslide Types.” In Understanding and Reducing Landslide Disaster Risk: Volume 2 From Mapping to Hazard and Risk Zonation, edited by Fausto Guzzetti, Snježana Mihalić Arbanas, Paola Reichenbach, Kyoji Sassa, Peter T. Bobrowsky, and Kaoru Takara, 321–27. ICL Contribution to Landslide Disaster Risk Reduction. Cham: Springer International Publishing. https://doi.org/10.1007/978-3-030-60227-7_36.
D’Amboise, Christopher J. L., Michael Neuhauser, Michaela Teich, Andreas Huber, Andreas Kofler, Frank Perzl, Reinhard Fromm, Karl Kleemayr, and Jan-Thomas Fischer. 2022. “Flow-Py v1.0: a customizable, open-source simulation tool to estimate runout and intensity of gravitational mass flows.” Geoscientific Model Development 15 (6): 2423–39. https://doi.org/10.5194/gmd-15-2423-2022.
Jones, Katie E., Jamie D. Howarth, Chris I. Massey, Biljana Luković, Pascal Sirguey, Corinne Singeisen, Caleb Gasston, Regine Morgenstern, and William Ries. 2025. “An Alternative to Landslide Volume-Area Scaling Relationships: An Ensemble Approach Adopting a Difference Model to Estimate the Total Volume of Landsliding Triggered by the 2016 Kaikōura Earthquake, New Zealand.” Landslides 22 (7): 2219–36. https://doi.org/10.1007/s10346-025-02479-x.
Larsen, Isaac J., David R. Montgomery, and Oliver Korup. 2010. “Landslide Erosion Controlled by Hillslope Material.” Nature Geoscience 3 (4): 247–51. https://doi.org/10.1038/ngeo776.
Petschko, H., A. Brenning, R. Bell, J. Goetz, and T. Glade. 2014. “Assessing the quality of landslide susceptibility maps – case study Lower Austria.” Natural Hazards and Earth System Sciences 14 (1): 95–118. https://doi.org/10.5194/nhess-14-95-2014.
Rickenmann, Dieter. 2005. “Runout Prediction Methods.” In Debris-Flow Hazards and Related Phenomena, edited by Matthias Jakob and Oldrich Hungr, 305–24. Springer Praxis Books. Berlin, Heidelberg: Springer. https://doi.org/10.1007/3-540-27129-5_13.
Spiekermann, Raphael I., Hugh G. Smith, Sam McColl, Lucy Burkitt, and Ian C. Fuller. 2022. “Development of a Morphometric Connectivity Model to Mitigate Sediment Derived from Storm-Driven Shallow Landslides.” Ecological Engineering 180 (July): 106676. https://doi.org/10.1016/j.ecoleng.2022.106676.
Steger, Stefan, Volkmar Mair, Christian Kofler, Massimiliano Pittore, Marc Zebisch, and Stefan Schneiderbauer. 2021. “Correlation Does Not Imply Geomorphic Causation in Data-Driven Landslide Susceptibility Modelling – Benefits of Exploring Landslide Data Collection Effects.” Science of The Total Environment 776 (July): 145935. https://doi.org/10.1016/j.scitotenv.2021.145935.
Steger, Stefan, Vittoria Scorpio, Francesco Comiti, and Marco Cavalli. 2022. “Data-Driven Modelling of Joint Debris Flow Release Susceptibility and Connectivity.” Earth Surface Processes and Landforms 47 (11): 2740–64. https://doi.org/10.1002/esp.5421.
Wichmann, Volker. 2017. “The Gravitational Process Path (GPP) model (v1.0) – a GIS-based simulation framework for gravitational processes.” Geoscientific Model Development 10 (9): 3309–27. https://doi.org/10.5194/gmd-10-3309-2017.Citation: https://doi.org/10.5194/egusphere-2025-3733-RC2 -
RC3: 'Comment on egusphere-2025-3733', Takashi Kimura, 26 Dec 2025
reply
General Comments
This study proposes a GIS-based analytical framework for landslide hazard and risk assessment that incorporates the concept of hillslope connectivity. By analyzing the proximity of unstable slopes and their connectivity to landslide runout pathways, the authors map the probability that mobilized sediment reaches buildings and roads. The work has clear potential to provide valuable insights that could inform landslide mitigation measures and future land-use and river-basin management.However, I have the following concerns regarding the methodology:
Major concern — Potential inconsistency between landslide-volume estimation and slope-stability analysis
In this study, the factor of safety is computed using the soil depth assigned to each grid cell when evaluating landslide probability. This setup implicitly fixes the rupture (slip) surface at a specific depth—namely, the base of the soil layer. If so, the landslide volume associated with each landslide object (LOS) should be calculated in a manner consistent with that assumed failure depth (e.g., by integrating soil thickness over the grid cells contained within each LOS).Instead, the authors estimate landslide volume for each LOS using the global area–volume scaling relationship (Larsen et al., 2010), independent of the soil-thickness distribution used in the infinite-slope stability analysis. This approach could imply a failure depth that differs from (i.e., is shallower or deeper than) the rupture-surface depth assumed in the stability analysis.
If the authors wish to retain Eq. (21) for landslide-volume estimation, they should explicitly discuss and justify its consistency with the assumed rupture-surface depth (i.e., the soil–bedrock interface) used in the slope-stability analysis.
In addition to the major concern outlined above, I identified several minor points that require revision to improve clarity and consistency. Please see the specific comments below.
Specific Comments
Page 1, line 74–75
“(see, e.g., reviews in (Intrieri et al., 2019; Jiang et al., 2022)).” → “(see, e.g., reviews in Intrieri et al. (2019) and Jiang et al. (2022)).”Page 2, line 84
“Montgomery David R. and Dietrich William E., 2010;” → “Montgomery and Dietrich, 2010;”
I noticed similar issues elsewhere; please check the manuscript throughout for consistent citation formatting.Page 3, line 117
“(Ghorbanzadeh et al., 2022)).” → remove the extra parenthesis.
Also, please clarify the intended distinction between volume and mass in this context, as they are not interchangeable.Page 4, line 137–142 (Figure 1)
“and roads (B)” → “and roads (C)”.Page 5, line 161–162
“Marc et al. (2019a), Figure 2).” → “Marc et al. (2019a) (Figure 2).”Page 6, line 176–177
The variable defined as τm, i is written as τmi in the subsequent equation. Please standardize notation throughout.
Also, this equation (the first equation appearing in the manuscript) should be numbered.Page 6, line 184
The sentences appear broken; “While we do not account for deep percolation,. We use Kent’s (1973) curve number model to estimate surface”
Perhaps revise to: “While we do not account for deep percolation, we use Kent’s (1973) curve number model to estimate surface runoff.”
Please correct as appropriate.Page 7, line 226–227
“fricvtion” → “friction” appears elsewhere. Please standardize.Page 8, line 237
“(Rawls et al., 1992)..” → “(e.g., Rawls et al., 1992).”Page 8, line 256–259
Eq. (13) appears incorrect: as written, mi*×bi×Ti would always equal 1. Please re-check the equation.
Also, “(Vogl et al., 2019a, b; World Bank, 2019)” should be cited immediately after the phrase “a threshold subsurface flow,” before the equation.Page 10, line 300
“roduce” → likely a typo (perhaps “reduce” or “reduces”). Please confirm.Page 10, line 301–305 (Figure 3)
Figure 3 would be easier to interpret if it also showed the spatial distribution of annual maximum precipitation.Page 10, line 314
Remove the unnecessary comma in “unstable, cells”.Page 10, line 316–320
Please confirm whether FSi is maximized at mi=0 and minimized at mi=1. As written, the explanation may be reversed.
Also, please number these two equations.Page 11, line 331
Consider italicizing k (landslide object index) for consistency with standard mathematical notation.Page 13, line 361–363
Eq. (23) has overlapping superscripts/subscripts and is difficult to read. Please rewrite in a clearer format (e.g., by grouping terms in parentheses). The same applies to Eq. (24).Page 13, line 377
Eq. (30) is not in the text. Please check numbering and citations.Page 13, line 382
The variable defined as δHh is written as δh in Figure 5. Please ensure consistency.Page 14, line 385–389 (Figure 5)
The equation shown in the upper-right subfigure differs from Eq. (24) (the V multiplier appears to be missing). Please check which version is correct.
“(γ\gammaγ)” → likely a typo
Also, please clarify how the river-line position and extent were determined.Page 14, line 400
“Marc et al. (2019)” → please specify 2018, 2019a, or 2019b to match the reference list.Page 17, line 435–439 (Table 2)
“Value source” does not need parentheses.
Replace parenthetical citation style with consistent author-year format where appropriate, e.g.:
- (Vanacker et al., 2003) → Vanacker et al. (2003)
- (Dhakal and Sidle, 2003; McGuire et al., 2016; Sidle, 1991) → Dhakal and Sidle (2003); McGuire et al. (2016); Sidle (1991)Also, “Estimated average from Vanacker et al., 2003” → “Estimated average from Vanacker et al. (2003).”
Please add the land-cover coefficient kf to the parameter list.
Page 18, line 454
Is it necessary to restate an equation already defined earlier as Eq. (2) (line 193–194)?
Consider removing redundancy.Page 24, line 566–574 (Figure 9)
Since LOS failure and sediment arrival are more likely at higher probabilities, the plots may be clearer if x-axis is reversed (probability 1.0 → 0.1) and the cumulative curves are shown in descending order.
Also, why does the cumulative curve for the Reforestation scenario in Fig. 9c stop at 0.9?Page 26, line 602–604
This sentence is unclear and contains typos (e.g., “areas area”). Please rewrite for grammatical correctness and clarity.Page 27, line 639
Delete the double period.Page 27, line 663–664
Remove the duplicated wording “such as CASCADE.”Citation: https://doi.org/10.5194/egusphere-2025-3733-RC3
Model code and software
COHESION slope stability model Rafael Schmitt and Shikshita Bhandari https://zenodo.org/records/16595396
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 290 | 144 | 35 | 469 | 22 | 21 |
- HTML: 290
- PDF: 144
- XML: 35
- Total: 469
- BibTeX: 22
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
To the authors, thanks for a thought provoking paper! I have a few questions and comments after reading.
Overall, I enjoyed the paper and anticipate sending it to some of my colleagues. Landslide runout is an important consideration in hazard assessment, especially since infrastructure tends to be on flatter runout zones versus steep slopes, and I believe this is an important direction for landslide hazard assessment to head in.