the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Jul 2025
Redefining dangerous glacial lakes in Bhutan by integrating hydrodynamic flood mapping and downstream exposure data
Abstract. Dangerous glacial lakes in Bhutan have primarily been identified considering the likelihood of producing a GLOF, which in turn has been assessed only based on upstream lake area/volume and their surrounding topographic conditions. However, this approach is incomplete as it ignores the at-risk downstream exposure and vulnerability thus the actual impacts. Here we redefined dangerous glacial lakes by considering the impact of the simulated most likely scenario GLOF on downstream exposed elements at risk. Our study shows that a total of approximately 22399 people, 2613 buildings, 270 km of road, 402 bridges and 20 km2 of farmland are exposed to potential GLOF inundation in Bhutan. We classified lake130 (Thorthormi Tsho) as a very high danger glacial lake in Bhutan, five lakes as high danger and 21 other lakes as moderate danger. Among these high danger glacial lakes, three of them: lake93 (Phudung Tsho), lake251, and lake278 (Wonney Tsho) were not recognized as dangerous in previous studies. Our assessment further revealed five downstream local government administrative units (LGUs) are associated with very high GLOF danger while nine others are associated with high GLOF danger. Six of these LGUs had not been previously documented as being at risk from GLOF including: Chhoekhor and Bumthang town in Bumthang, Paro town and Lamgong in Paro, Nubi in Trongsa and Khoma in Lhuentse districts. Our study underscores the significance of integrating potential inundation mapping and downstream e data to define dangerous glacial lakes. We recommend strengthening and expanding the existing GLOF disaster preparedness and risk mitigation efforts in Bhutan to reduce future damage and loss in high GLOF danger LGUs identified in this study.
- Preprint
(3100 KB) - Metadata XML
-
Supplement
(1837 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-3206', Anonymous Referee #1, 29 Aug 2025
The study by Rinzin et al. analyses the downstream exposure and vulnerability of infrastructure, buildings and people to glacial lake outburst floods in Bhutan. Their analysis relies on a dataset >200 glacial lakes, a globally available digital elevation model and OSM data. Using HEC-RAS, the authors simulate a scenario for each lake and compare flooding extents, depths and velocity to the locations of the elements at risk.
In general, this is a well-conceived study that leverages hydrodynamic modelling to address some of the weaknesses of previous studies that made quite simplifying assumptions about flood wave propagation and the extent of their impact. However, there are still a few issues with the study which I will outline below. All in all, I recommend major revisions before the manuscript should be published in NHESS.
Major comments:
The parameter choice relies on previously published data (flood volume). However, the choice of parameters does not consider the variability of this data, but rather takes point estimates. For example, the choice of using the median of reported percentages of drainage volume is considered the "most likely flood volume" (L 212). However, if you have a bimodal distribution of partial drainage volumes, then the median is not the most likely flood volume. Thus, it may be useful to not pick out the median scenario, but one that is at the upper end of the distribution, thus giving more weight to extreme scenarios. Same is true for the volume-area relation that may only represents an average of the breadth of possible scenarios. Schwanghart et al. (2016) showed that results of GLOF modelling are not sensitive to uncertainties in the V-A relation for large lakes, but that these uncertainties matter for smaller lakes. I acknowledge that the study already comprises many simulations with quite a heavy computational load. However, it should be at least discussed that the current approach lacks a consideration of the large variability of possible outburst scenarios and that average scenarios may not capture the worst-case scenarios.
A simulation of one or few past GLOFs and comparison of actual with simulated peak discharges would help gaining confidence into the model and its ability to realistically model GLOF dynamics. How can readers evaluate how well your model actually works? This would also enable to tune parameters and eventually study how sensitive the results are to uncertainties in the parameter values.
There are numerous instances were ambiguous or imprecise terms are used. Generally, I think that the terms threat and danger(ous) should be avoided, and that rather terms like hazard (probability of a potentially adverse event happening), exposure (how much are people or infrastructure within reach of a hazardous event), vulnerability (how susceptible are the exposed people or elements) and risk (the combination of the previous, quantitative metrics) should be used as they have a precise and measurable meaning. Threat and danger in turn have a qualitative and subjective meaning. Your work mainly aims to address the exposure of various elements at risk, and you quantify and aggregate the exposure so that it becomes an attribute of each lake. So, to this end, you quantify a lake-specific exposure index.
Minor comments:
41: You state the number of 6907 fatalities, and backpedal later that this number is 80% attributed to a compound event involving the Chorabari outburst. The number of fatalities that can be clearly attributed to the Kedarnath event is probably very uncertain and much lower than those 80%. I would try to tone this more carefully, avoiding reporting numbers with high precision, that actually have a high uncertainty.
49: Provide a definition of danger, in particular if your aim is to quantify it. Rather, as pointed out above, avoid this term entirely.
81: This should be 31%, not 0.31%.
280: How was the HEC-RAS interfaced with? It would be great if you could add a technical description in a paragraph that details how you interfaced with HEC-RAS. I assume that you used the HEC-RAS controller to automate the tasks.
318f: Is it common to take the product of depth and velocity as damage level? Is it useful that the damage level of a water depth of 1 m and velocity of 5 m/s is the same for a water depth of 5 m and a velocity of 1 m/s?
540: It would be helpful to use a stringent and precise terminology here. What is devastating in comparison to damaging? Was the Missoula flood devastating, but not damaging, because no humans were affected (not sure whether this is true)? In simple terms, the risk of GLOFs is mainly determined by the exposed elements at risk, not by their hazard?
594f: I don't think that your approach challenges traditional susceptibility analyses. Rather, your approach may complement them. In contrast to susceptibility studies, your analysis assumes that the outburst probability is homogeneous, thus neglecting any variations in dam stability and lake exposure to avalanches and landslides.
719-721: Considering a risk framework, this is a somewhat trivial statement.
729: Please avoid the high precision of numbers when their estimates are prone to large uncertainties.
Table S1: As the table spans several pages, it would be great to have the header row of this table on each page.
Table S2: Be consistent in the number of digits that you report. Up to 8 digits behind the decimal point suggest an accuracy that you probably don't have in your measurements. When reporting counts (Bridges), use integers.
Citation: https://doi.org/10.5194/egusphere-2025-3206-RC1 -
RC2: 'Comment on egusphere-2025-3206', Adam Emmer, 12 Sep 2025
This study models and assesses the impacts of potential future GLOFs in Bhutan and takes into account elements at risk. The methodology is innovative and goes beyond susceptibility / hazard assessment. It is important to acknowledge the amount of work behind it and appreciate implications for the development of GLOF EWS in Bhutan. I have read the study with great interest and have number of questions / comments to it:
L1: how do the authors know these are the “most likely” scenarios without looking at triggers and dam properties of individual lakes?
L37: water addition is not the driver; expansion of existing glacial lake basins / formation of new basins is
L47: is there any study actually showing that moraine-dammed lakes are closer to settlements? I argue that the reason explaining this observation is rather linked to specific GLOF mechanisms from moraine-dammed lakes (https://www.nature.com/articles/s44221-024-00254-1)
L73: I suggest to use established terminology - susceptibility, hazard, vulnerability and risk
L91: “hydrodynamics” do not interact, GLOFs do interact
L108: the problem usually is not the susceptibility assessment itself but lake size threshold used
L114: using the 50,000 m2 threshold is arbitrary; there are examples of damaging or even deadly GLOFs from much smaller lakes (e.g. the Independencia/Huaráz GLOF earlier this year coming from approx. 10,000 m2 lake)
L124: please provide a link to this portal
L139: Rizin et al. 2021 also used 50,000 m2 threshold in their assessment, meaning that the danger of GLOFs from small lakes (regardless their proximity to settlements or infrastructure) is systematically overlooked
Fig. 1: elevation is not readable since the country is coloured in green; consider using the same format for all coordinates (90°E vs. 90°30’E)
L177: yes, because GLOFs from smaller lakes and / or farther in the past tend to happen unnoticed
L183-185: this is a common hazard-oriented filtering; considering the points made in the intro, I wonder why a distance from settlements / infrastructure is not considered as a primary filter rather than lake size and distance from glaciers?
Fig. 2: it is not clear to me from the figure how a damage level which is calculated for each cell of the mesh by multiplying max. velocity with max. depth is translated into one damage index for each lake? A normalized sum? It is important to acknowledge that normalisation is sensitive to outliers (very large lake130 is in order of magnitude larger than any other lake in the dataset); DV is not defined correctly - L1 is a subset of L2 (<5 is a subset of <10), 10 does not fall to any category; please check and revise
L212-213: this is a justified approach but it does not give the most likely volume scenario (which would have to be treated lake by lake, considering dam geometry and properties and potential triggers)
L221: most GLOFs do not originate from moraine-dammed lakes (as mentioned earlier in the Intro)
L246: justification of parameters with the use in previous studies should be accompanied by performance evaluation (from the original study or elsewhere), otherwise it is prone to proliferate the use of possibly irrelevant parameter values
L246: again, better justification for the use of this value would be with the values recommended for different Channel types -(https://www.fsl.orst.edu/geowater/FX3/help/8_Hydraulic_Reference/Mannings_n_Tables.htm) I think 2b fits well most of the GLOF streams
L269: same as above - previous use doesn’t guarantee satisfactory performance and suitability
L283-287: population density is suitable for hazards affecting large continuous areas (e.g., earthquakes, heat waves) but GLOF impact areas are very localized and the approach ignores spatial patterns of population distribution; since the authors have detailed data about individual buildings, I was wondering why not to estimate exposed population as a function of built-up area, instead of using population density multiplied by inundation area?
L314: what is the meaning of these values 8 and 10?
L330: Is there any specific reason why the values were categorized? And any specific reason why into these three categories? Categories blur differences. Why not normalize the values similarly to the SE indicators? It would be good to show at least the histogram of values so it is easier to understand how the values are distributed and how the dataset is split into categories
L331: please check if the first category really includes 0? categories are not defined correctly - 5 falls to both L1 and L2
L336: equation (iii) and (iv) can be merged? the sum function needs definition of i (typically from 1 to n, where n is the number of impacted grid cells)
L344: what is a gewog? please give approximate translation or explanation
L373-374: the selection of SE indicators is not justified; how sensitive is the outcome to changes in inputs? If 18 indicators are available, it does not necessarily mean that the set of 18 indicators is the best to use. How sensitive is it to outliers?
L375: all indicators should be relative values before they are standardized, otherwise large LGUs are getting higher values and so higher normalized values; for example, it is stated that the measure of the indicator “Reliable source of energy for lighting” is defined as “Households with a main source of energy for lighting as electricity”; if so, this is the absolute value which is meaningless to normalize unless it is related to the number of households in individual LGUs, i.e. “Proportion of households with a main source of energy for lighting as electricity”
L380: building damage is not only a function of hydrodynamics factors but construction type and construction quality factors too
L386-389: please consider elaborating this earlier - Intro or study area section; which lakes they found hazardous?
L392: how?
L396: missing verb in the sentence?
Table 2: this shows the downside of using population density and ignoring spatial patterns of settlement location (see my comment above); comparing Bumthang Town and Lunana, the mean population of a mean building ranges from 0.71 person per building (Lunana) to 20.3 persons per building (Bumthang), i.e. in two orders of magnitude
L402: the structure in methods starts with hazard, followed by exposure and vulnerability; it is contraintuitive that Results section starts with Impact and exposure; I suggest to unify the structure in methods and results
L433: and SE vulnerability indicators
L435-440: consider moving to methods
L445-446: can any of the lakes located in Bhutan generate transboundary flood to any of the neighbouring countries?
Fig. 3: why is building count expressed per km2?
Fig. 4: building damage level in shades of blue is not very intuitive
Fig. 5: flow arrival time is shown on x axis, right? What is the difference to flow arrival time indicated by the color of a circle? For example, why is the arrival time to Bumthang (45) 3 hours when reading from the x axis but 10 hours according to the color scale?
L523: how many residents in GLOF exposed buildings?
L549: escalating exposure and so risk
L594: it does not challenge but expands on more common GLOF hazard assessment studies
L597-611: yes, but this is not contradicting or surprising that there are differences, is it? These are different approaches used for different purposes; one says which lakes are more or less likely to produce a GLOF while the other says which lakes are more or less likely to produce a damage in case of a GLOF
L622-624: it is not one or the other; all components (h, e, v) need to be taken into consideration in GLOF risk management and I suggest to reword this sentence
L631: lakes located outside Bhutan’s border; how about transboundary GLOF danger originating in Bhutan?
L647: please consider structuring this sub-section into sub-sub-sections
L647: for the future work, please also highlight the need for developing site-specific GLOF scenarios that consider different triggers, dam types and geometries (which is (understandable) source of uncertainty in this country-wide study
L647: please comment on the sensitivity of the assessment procedure to outliers (see my comment above); please comment on the selection of the SE indicators and categorisation vs. normalisation that are used concurrently in different steps of the procedure
L723: GLOFGLOF
L729: “approximately” should not be followed by very precise (though highly uncertain) numbers
Fig. 9: consider unifying graphical style with other maps in the study (please remove pink background from the legend)
Table S1: the distribution of normalized danger index values is anomalous; removing one outlier of the lake 130 would likely generate way different results in terms of what is shown in Fig. 8a - please discuss this (see my comment above)
Overall, this is a state-of-the-art country-wide GLOF risk assessment study with some innovative methodological aspects. The manuscript would benefit from the use of standard risk terminology. Parts of the methodology require better justification / discussion. I recommend moderate revisions.
Citation: https://doi.org/10.5194/egusphere-2025-3206-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 971 | 128 | 13 | 1,112 | 35 | 17 | 23 |
- HTML: 971
- PDF: 128
- XML: 13
- Total: 1,112
- Supplement: 35
- BibTeX: 17
- EndNote: 23
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1