We used the glacial lake inventory by Rinzin et al. (2021), which has been developed specifically for Bhutan, and offers greater robustness and accuracy for Bhutan compared to other datasets available at the Pan-HMA scale (Zhang et al., 2023b; Zheng et al., 2021b). Rinzin et al. (2021) mapped 2574 glacial lakes in 2020, located within 10 km of glacier termini and with a minimum lake area threshold of 0.003 km². This dataset includes 85 transboundary glacial lakes, located in the Indian and Chinese territories of the Himalaya, whose drainage flows into the inland regions of Bhutan. Previous records indicate that GLOF originating from a relatively small lake (as small as 0.001 km²) can cause substantial damage in downstream communities, although such cases are rare (Sattar et al., 2025a). For instance, across the HMA region, the median area of glacial lakes with known pre-outburst extents is approximately 0.189 km² (Shrestha et al., 2023). In Bhutan specifically, the smallest glacial lake with a documented outburst history has a present-day area of 0.0506 km² (Rinzin et al., 2021; Komori et al., 2012). Including all glacial lakes for detailed hydraulic modelling would substantially increase the computational demands, whereas resorting to simplified GIS-based approaches (Allen et al., 2019; Zheng et al., 2021b) to cover all lakes would significantly compromise the robustness and accuracy of the resulting flood maps. Therefore, based on the trade-off between model complexity and result reliability, we focused on glacial lakes that (i) are at least 0.05 km² in area and (ii) are located within 1 km of glacier termini. This approach ensures a balance between computational feasibility and the production of reliable flood maps, while still capturing a substantial number of high hazard lakes. Based on these criteria, we identified 278 glacial lakes in Bhutan for flood inundation mapping using hydrodynamic modelling (Fig. S1 in the Supplement). 1V=42.95×A1.408, where V is the volume in 10⁶ m³, and A is the area in km².

Accurate glacial lake volume data is crucial as one of the key determinants of modelled GLOF hydrodynamic characteristics such as flow depth and velocity. However, field-based bathymetric measurement for multiple lakes is costly and not currently possible for some glacial lakes due to their remote location. In the absence of in-situ bathymetric data to determine volume, we calculated the volume of each glacial lake using the area-volume scaling relationship (Eq. 1) proposed by Zhang et al. (2023a) based on the area of each glacial lake as mapped in 2020 (Rinzin et al., 2021) (see Table S1 in the Supplement). This area-volume scaling relationship (maintaining an error margin of ±15 % in the eastern Himalaya), based on recent bathymetric data from the Greater Himalayan region, including 13 representative lakes from Bhutan, is well-suited to approximate Bhutanese glacial lake volumes (Zhang et al., 2023a). However, the lake area is an essential parameter in our study, as we used it as a proxy for calculating volume and peak discharge as follows. Therefore, we examined the sensitivity of model output (depth-velocity) by varying lake size between 0.01 and 5 km² (see the Supplement for details).

It is important to note that not all lakes drain entirely during a GLOF (Maurer et al., 2020; Nie et al., 2018; Zhang et al., 2024). The amount of water drained during a GLOF event depends on numerous factors, such as dam geometry and composition, lake bathymetry, and GLOF trigger – for example, seepage-induced versus landslide-triggered wave overtopping. While consideration of all these factors is crucial for a detailed impact assessment of a particular glacial lake, constraining the drained volume based on these detailed attributes is highly challenging for a study involving numerous glacial lakes. Previous data indicate that smaller lakes are more likely to drain completely during a GLOF event than larger lakes. Here, we used data from Zhang et al. (2023b) documenting the drainage volumes of 64 lakes in the HMA regions. Among these 64 lakes, the median percentage of drainage volume was 98 % for lakes with an area < 0.1 km², 62 % for lakes with an area of 0.1 to 1 km² and 33 % for lakes with an area > 1 km². We used these observed drainage percentages as the basis to calculate the flood volume generated by each lake. For simplicity and recognising that these median drainage values lie within the uncertainty bounds of established area-volume scaling relationships, we adopted the following assumptions: 100 % drainage for lakes < 0.1 km², 60 % drainage for lakes between 0.1 and 1 km², and 30 % drainage for lakes > 1 km². Subsequently, we used Evans's empirical Eq. (2) for moraine-dammed lakes to calculate the possible peak discharge of each lake (Evans, 1986). See Fig. S1 for the distribution of volume and peak flow calculated for each glacial lake in Bhutan (see Table S1). 2Qmax⁡=0.72V0.53 Qmax⁡ is peak discharge, and V is the total volume of the lake calculated using Eq. (1).

Most GLOFs from moraine-dammed lakes start from dam breaching, which is frequently triggered by large mass movement(s) entering the lake from the surrounding hillslopes (Shrestha et al., 2023). However, conducting a dam breach simulation for each lake is challenging due to complexities and uncertainties in constraining the appropriate value for a large range of input parameters (U.S. Army Corps of Engineers et al., 2021). To simplify this, we conducted a flood simulation for each lake by using an input hydrograph as an upstream boundary condition. For each lake, we generated an input hydrograph by fitting the peak flow of each lake to a log-normal distribution curve with a standard deviation (σ) value of 0.75 and a mean of 0, adapting the approach used by earlier studies (Carr et al., 2024; Kropáček et al., 2015). For example, for lake1, the peak flow was 1110 m³ s⁻¹; thus, we constructed a log-normally distributed hydrograph with a peak flow of 1110 m³ s⁻¹ and gradually decreasing flow after this peak. With this assumption, we generated the hydrograph so that the flow rises to its peak rapidly and progressively decreases after attaining the peak, which is consistent with the hydrograph of many previous GLOF events (Maurer et al., 2020; Nie et al., 2020; Zheng et al., 2021a) (see Fig. S2 for a representative hydrograph). The flow duration of the hydrograph of each lake was subsequently adjusted to account for the complete drainage of the estimated drainage volume calculated for each lake. For example, for lake1, the required drainage volume was calculated at 1.036 × 10⁶ m³, and so, the flow duration for this lake was adjusted so that the cumulative flow through the GLOF event was equal to this volume (Table S1).

We used the ALOS Global Digital Surface Model (AW3D30) with ∼ 30 m ground resolution as a source of terrain information for the model setup (Japan Aerospace Exploration Agency, 2021). The DSM was hydrologically corrected by removing spurious depressions and burning in artificial flow paths in locations where deep gorges were incorrectly represented as floodplains (Rinzin et al., 2023). We chose AW3D30 because previous studies (Rinzin et al., 2025) have indicated that it has higher vertical and horizontal accuracy compared to other freely available DEMs for our study area with similar spatial resolution, such as SRTM GL1 (for example, Liu et al., 2019). We assigned an n value of 0.06, which has been used in GLOF modelling in Bhutan previously (Maurer et al., 2020). However, n is one of the most sensitive input parameters in hydraulic modelling. Therefore, we conducted sensitivity analysis by varying n between 0.040 and 0.070, the value range suggested by Chow (1959) for river channel beds with large boulders and cobbles, which characterise river channels in the Bhutan Himalaya (see the Supplement for details).

We created one HEC-RAS project for each major river basin so that a total of seven project files correspond to the seven glaciated basins in Bhutan. For each project, the model domain was established by creating a 1000 m buffer on either side of the centre line of the river originating from each lake. Within this model domain, a computational mesh with a grid resolution equal to the native resolution of AW3D30 (30 × 30 m) was generated. An upstream boundary condition for each lake was defined at the frontal terminus of each lake. We used the same downstream boundary condition for all lakes in the basin, which was defined at the furthest end at the international border between Bhutan and India (for example, Fig. S2). Unique flow data was created for each lake, where we imposed flow hydrographs as the upstream boundary condition for the respective lake and downstream boundary conditions defined by normal depth with an energy slope of 0.01 (U.S. Army Corps of Engineers et al., 2021). Finally, one unsteady flow analysis plan for each lake with corresponding unsteady flow data and boundary conditions was developed. For example, in the Phochu basin, which contains 67 glacial lakes considered for this study, one project file was established. This project file included a single model domain, a downstream boundary condition, 67 upstream boundary conditions and 67 flow datasets (Fig. S2).

We computed all the simulations using the full momentum shallow water equations since they better represent GLOF rheology (although with a clear water assumption) than the diffusion wave equation (Sattar et al., 2023, 2021). Considering that this study is mainly aimed at providing a GLOF risk overview in Bhutan, all other computational parameters were maintained at the default setting. At a mesh size of 30 m, each model was run stably with a computational time step of 3 s with a Courant number well below 2 (U.S. Army Corps of Engineers et al., 2021). The simulations were executed simultaneously across 15 computers at the Geospatial Laboratory in Newcastle University. We maintained 10 h of simulation time for each model setup, which took 2 to 4 h depending on the lake's size. The utput for each project plan was examined and any models exhibiting instability (e.g., a Courant number above 2 or failed before complete execution) were re-executed by adjusting the position of upstream boundary condition, changing the timesteps, or adding additional features like refinement regions within the 2D model domains to ensure stable model run and reliable results (U.S. Army Corps of Engineers et al., 2021).

We collated the GLOF inundation boundary for each lake generated through the HEC-RAS modelling and calculated the area and length of each inundation. We mapped all buildings, roads, bridges, farmland, and hydropower plants within the GLOF inundation area to identify downstream elements at risk. OpenStreetMap (updated as of 30 April 2025) was used to map buildings, roads and bridges, which we manually verified using Google Earth high-resolution imagery. This resulted in updates of 41 km of missing roads, 152 buildings and 20 bridges using the Google Earth imagery within the flood inundation plain. The local government administrative unit (LGU) boundary map was then used to map these elements at risk in each LGU. As there is no population data with sufficient granularity to map exposed people directly, we therefore used exposed buildings as a proxy for exposed people, assuming that the distribution of people usually corresponds to the location of buildings. Specifically, we divided the total population of each LGU by its total number of buildings, using the 2017 Bhutan population and census data (NSB, 2018) and OpenStreetMap building data. The number of people exposed to GLOF in each LGU was then calculated by multiplying the population per building by the total number of exposed buildings. The ICIMOD's Landsat-based land-use and land cover map of 2023 was used to map farmland (Uddin, 2021) since it is of better quality at the HKH scale than other open-access land cover data, such as Esri Sentinel-2 land cover data (Karra et al., 2021). We considered buildings the most important downstream exposed element because they are the primary space where people live. Thus, we used exposed buildings to calculate the GLOF damage index (Fig. 2), and our risk assessment is biased towards loss of life based on building counts over, for example, asset monetary value where hydropower losses may dominate.

In this study, we defined GLOF hazard based on the downstream damage (calculated here as damage index (Di)) resulting from each GLOF event (Fig. 2). The hazard in this study refers to damage/destruction to buildings that could result from future potential GLOF. We assume that any of our study glacial lakes has the potential to generate a GLOF in the future, and the resulting damage will determine the level of hazard that the glacial lake would pose to those downstream communities. The Di for each element (grid cell) resulting from any GLOF event was calculated as a function of the value of the exposed element (Vi) and the level of damage (Li) following the approach proposed by Petrucci (2012) (Eq. 2) (Fig. 2). Qualitative data such as construction type, occupancy type, and value of the content of the building inside the house are essential to obtain the appropriate value of structure and content. However, such qualitative attributes are incomplete in the existing OpenStreetMap and introduce substantial uncertainties when estimated using other open-access data. In this study, our focus is on providing a relative quantitative comparison of GLOF impacts across different communities instead of determining exact damage values resulting from each GLOF event. Therefore, we considered each building as one unit of Vi uniformly applied across the study domain (Fig. 2).

GLOFs with higher flow velocity can cause more damage to exposed elements than low flow velocity (Federal Emergency Management Agency, 2004). Therefore, we calculated the Li associated with each GLOF event as a function of both velocity and depth, which was accomplished by calculating the depth × velocity (DV) from the HEC-RAS output layer. The level of damage a building suffers also depends on its structural integrity, which in turn is a function of the construction type of the building. The 2017 Bhutan Population and Housing Census reported that the construction type of most Bhutanese buildings is masonry (NSB, 2018). Clausen and Clark (1990) have categorised three damage levels to masonry buildings based on DV as follows: inundation (DV ≤ 2 m² s⁻¹), partial damage (2 m² s⁻¹ < DV ≤ 7 m² s⁻¹) and complete damage (DV > 7 m² s⁻¹). Accordingly, we used these DV value ranges to classify three levels of damage and assigned Li following Petrucci (2012) as follows: Level 1 (Li = 0.25), Level 2 (Li = 0.5), and Level 3 (Li = 1) (Fig. 2).

Finally, the sum of Di for all damaged buildings located within the GLOF inundation boundary of each lake was summed to derive a hazard value (Hg) associated with each lake (Eq. 3). The Hg was then normalised between 0 and 1, and lakes were ranked based on this metric. Additionally, for the classification purpose, using normalised Hg, we categorised glacial lakes into four hazard levels: very high, high, moderate, and low hazard using the Natural Jenks classification system in ArcGIS. 3Hg=∑i∈gDi=∑i∈gViLi,4Rg=Hg×Vlg, where Di is the damage index for exposed element i, Vi is the value of each downstream element, and Li is the damage level for each element. Hg is the aggregated hazard index for geographical unit g (LGU or inundation boundary of the respective lake). Rg is the risk index for geographical unit g (LGU). Vl_g is the vulnerability of the geographical unit g.

We conducted GLOF risk assessment (Rg) for downstream settlements at various geographical scales: 20 districts and 274 local government administrative units (LGUs) (including 205 gewogs [sub-district blocks] and 69 towns). We aggregated the Di of each damage grid located within the respective LGU boundary to calculate Hg for each LGU using Eq. (3). In cases where downstream elements were affected by GLOFs originating from multiple lakes, the combined Hg from each contributing lake was considered in the analysis to account for their exposure to multiple possible GLOFs (Fig. 2).

As well as the magnitude of GLOF and the presence of downstream elements along the flow path, downstream GLOF risk is also determined by the community's capacity to prepare, respond and recover from a GLOF event (Cutter et al., 2008; Zhou et al., 2009). Lack of this capacity, referred to as vulnerability, is influenced by wide-ranging socio-economic factors, including but not limited to the standard of living and the age and gender composition of the community (Cutter and Finch, 2008). Across the world, developed countries were found to be more disaster resilient than developing countries, while disaster-related death and damage have largely spiked in low-income countries (Rahmani et al., 2022). However, identifying specific socio-economic variables that are most relevant to GLOF damage remains a significant challenge, particularly because social data from past events are either unknown or, at times, overlooked. Past studies have used variables such as gross domestic product, population density (Carrivick and Tweed, 2016), human development index, corruption index and social vulnerability index at the national scale (Taylor et al., 2023a). While such data represent a broad overview of the country's socio-economic condition and thus vulnerability to disaster, they do not represent the regional and community-level disparity within the country that influences their ability to respond and recover from the disaster. To address this, drawing upon our local understanding and following earlier studies (Allen et al., 2016; Rinzin et al., 2023), we calculated the relative vulnerability index (Vl_g) using a total of 18 socio-economic indicators from the 2017 Bhutan population and housing census (NSB, 2018) (Table 1). This census data, which is updated every 10 years, represents the most comprehensive and detailed dataset currently available, offering spatial granularity at the individual LGU level. These indicators are essential for evaluating a community's preparedness and response capacity to disasters from hazards like GLOF (Cutter and Finch, 2008). For example, in Bhutan's traditionally gendered societal structure, men often assume more prominent roles in disaster response efforts. The Vl_i for each LGU was calculated as the normalised value across these 18 socio-economic indicators (Fig. 2). The definitions and approaches used for calculating each indicator are summarised in Table 1. Assuming that the LGUs with higher Vl_i are less able to respond to and recover from a future GLOF, Hg for each LGU was multiplied by Vl_g to calculated associated GLOF risk (Rg) (Eq. 4).

Building damage from a GLOF is a function of hydrodynamic factors such as depth and velocity, and the structural integrity of the buildings. On the other hand, human casualties and injuries also depend on the warning/response time, making it essential to consider flood arrival time in GLOF risk assessment. Thus, GLOF arrival time needs to be considered separately from the Di we computed earlier. Accordingly, we determined the flow arrival time of the earliest arriving GLOF for each LGU to determine the amount of response time the people living in the particular LGU will get in case of a future GLOF.

The NCHM, Bhutan, monitors several lakes in Bhutan identified as high hazard (NCHM, 2019). Currently, they have a GLOF early warning system covering Punatsangchu, Mangdechu and Chamkharchu basins, which consists of 13 monitoring stations (Fig. 1). We utilised monitoring station location data from NCHM to evaluate the relationship between Bhutan's existing early warning system, modelled GLOF scenarios originating from these glacial lakes and affected downstream communities. To achieve this, we first overlaid all monitoring stations in Bhutan in a GIS and counted how many of our catalogue of possible GLOFs in the region are covered by the existing early warning system based on their hydrological relationship (NCHM, 2021). We assumed that if a GLOF flow intersects any of the existing EWS monitoring stations, then the event would be monitored by the existing EWS in Bhutan. Similarly, if an LGU is affected by a GLOF that passes through one or more of these stations, the LGU is regarded as being monitored by the existing EWS.

Of the 278 glacial lakes selected for GLOF modelling and downstream risk assessment in this study, the majority (n = 91) of them were in the Punatsangchu basin, followed by the Kurichu basin (n = 64). By contrast, Wangchu (n = 2) and Amochu (n = 5) had the minimum number of lakes meeting our selection criteria (Fig. 3). The volumes of the selected glacial lakes ranged between 0.64 × 10⁶ and 344.1 × 10⁶ m³, with a median volume of 2 × 10⁶ m³. Based on these total volumes, minimum, median and maximum drainage volumes were 0.64 × 10⁶, 1.4 × 10⁶, and 103.2 × 10⁶ m³, respectively. Furthermore, the empirical-based estimation showed that these glacial lakes can produce GLOFs with peak discharges of up to 24 085 m³ s⁻¹, while the median peak discharge is 1552 m³ s⁻¹ (Fig. S3).

Our study revealed that GLOFs from individual glacial lakes can travel as far as 167 km downstream and can inundate a maximum area of 30 km². The modelled GLOFs exhibit a median travel distance of 40 km and an inundation area of 2.9 km² (Fig. S3). Collectively, about 2 % (781 km²) of Bhutan's total land area is exposed to GLOF. The mean flow depth and velocity were 3.3 m and 3.4 m s⁻¹, respectively (Fig. S4). The shortest arrival time to the nearest building was 8 min, and the longest was 10 h (Fig. S3). As a result, a total of 11 322 people, 2613 buildings, 270 km of road, 402 bridges, 19 km² of farmland and 4 hydropower dams are exposed to GLOFs in Bhutan (Table 2). Of the total modelled GLOF events, 71 % (n = 197) affect roads, 42 % (n = 116) affect buildings, and 28 % (n = 77) affect farmland. A GLOF from Thorthormi Tsho could impact 1119 buildings, 72 km of roads, and 4.2 km² of farmland, making it the most consequential event reaching elements at risk. It is followed by lake278, located in the Wangchu basin and Chubdha Tsho in the Chamkharchu basin, both of which are classified as high hazard glacial lakes by this study (Table S1).

Out of 278 glacial lakes selected for flood mapping for this study, 85 (30.6 %) are within catchments that cross the boundaries of India and China and drain into Bhutan inland. GLOFs from these transboundary lakes also affect a substantial number of downstream elements located in Bhutan, including 20 buildings, 0.6 km² of farmland, and 2 km of roads in Bhutan. All these exposed elements are situated within the Kurichu and Drangmechu basins.

The exposed elements are distributed across 17 districts and 88 local government administrative units (LGUs). Bumthang, Paro, Punakha, Wangdue Phodrang and Gasa districts are most affected by the GLOFs. For example, Paro itself has 673 GLOF exposed buildings, 32 km of road and 64 bridges (Table 2). Among the LGUs, the maximum number of exposed buildings is in Bumthang town (n = 283), followed by Punakha town (n = 272) and Paro town (n = 262). The greatest road (roads, footpaths, tracks) inundation occurs in Chhoekhor, followed by Lunana and Saephu, while most farmland is impacted in LGUs such as Bumthang town (2.3 km²), Dzomi Gewog (1.2 km²) and Paro town (1.1 km²) (Table S2).

We defined the GLOF hazard based on the total damage associated with each lake. Of the total lakes studied here (278), 164 had zero hazard index as the GLOF from these lakes does not impact any buildings. Among other lakes, with the highest Dg, Thorthormi Tsho in the Punatsangchu basin emerged as the very high PDGL in Bhutan (Fig. 3). Five other lakes are identified as high hazard, which are distributed across the Wangchu (2), Chamkharchu (2), and Punatsangchu (1) basins. Twenty-two of the glacial lakes were in the moderate hazard category: seven in Punatsangchu, five in Mangechu basin, four in Drangmechu basin, three in Chamkharchu, two in Kurichu and one in Wangchu basins (Fig. 3). The remaining (250) were classified as low hazard. None of the high or very high hazard glacial lakes were located within the Chinese and Indian sides of the basins, which drain into Bhutan (Fig. 3).

In this section, we present the GLOF risk ranking and level associated with downstream communities, encompassing 20 districts and 274 LGUs. Punakha is identified as the district that would suffer from the highest GLOF damage in the future, followed by Bumthang and Wangdue Phodrang districts.

Five of the LGUs were classified as having very high GLOF risk, including: Chhoekhor, Punakha town, Lunana, Thedtsho, and Bumthang town (Fig. 4). Further, eight others were associated with high GLOF risk. Likewise, 18 LGUs were associated with moderate GLOF risk, while the rest were identified as low risk LGUs (Fig. 4).

We also ranked the LGUs based on the flow arrival time of the GLOF scenario that would impact on the first buildings in the LGU. Results showed that, for the seven gewogs including Soe, Khoma, Chhoekhor, Lunana, Laya, Saephu, and Kurtoed, the fastest GLOF can impact some of their buildings within 30 min. Some buildings in Khatoed, Tsento, Toedwang and Nubi could be affected within 1 h. Nine gewogs can be affected within 2 to 4 h, another nine within 4–6 h, while the fastest GLOF could take more than 6 h to affect buildings in other LGUs (Fig. 5).

In the LGUs such as Soe and Laya, the first buildings affected by the GLOF are typically isolated and located very close to glacial lakes. Despite their proximity, the GLOF risk ranking for these LGUs remains relatively low due to the limited number of exposed buildings in these LGUs (Fig. 5). Therefore, we compared the Di and the arrival time of the fastest-arriving GLOF within each LGU. This analysis identified Lunana and Chhoekhor as LGUs with very high GLOF risk levels, and the fastest GLOF can arrive in as little as 15 min. On the other hand, the fastest GLOF impacting buildings takes up to 3 h for other LGUs with very high GLOF risk, such as Punakha town and Bumthang town (Fig. 5).

We analysed the distribution of the existing GLOF early warning system in Bhutan with respect to high hazard lakes and downstream communities associated with GLOF risk. Currently, Bhutan has a GLOF early warning system in three basins: Punatsangchu, Mangdechu, and Chamkharchu. Across these basins, the system is equipped with 13 monitoring stations placed at various locations (Fig. 1). Assuming that GLOFs from a lake may be monitored if an EWS monitoring station is located downstream of the glacial lake, our study shows that the existing EWS currently tracks 51 out of the 278 glacial lakes we investigated here. Among these monitored lakes, the network includes the very high hazard glacial lake, Thorthormi Tsho, as well as two of the five high hazard glacial lakes and five of the six moderate hazard glacial lakes. The remaining monitored lakes are classified as low or very low hazard. Notably, the high hazard glacial lakes newly identified here, including lake251, lake262 and lake278, are not monitored by existing early warning systems.

We further examined how many residents within GLOF exposed buildings could receive early warnings based on their hydrological relationship to the existing EWS monitoring stations. Assuming that the buildings located downstream of the EWS monitoring stations can receive an early warning, our study revealed that the existing GLOF monitoring stations can provide early warning to the people living in 1549 buildings, of which about 75 % are in the Punatsangchu basin. Of these, residents in 268 buildings are estimated to have less than 30 m to evacuate after receiving a warning from the EWS monitoring stations located in their respective communities (Fig. 6).

Conversely, people living in 1050 exposed buildings, that is, at least 41 % of exposed buildings, do not have access to early warning coverage. Approximately half of these unserved buildings are clustered in downstream LGUs with high GLOF risk, including Lamgong and Paro Town in Paro districts. Although EWS is in place in the Chamkharchu basin, a cluster of about 82 buildings in Chhoekhor in Bumthang is not covered by EWS. This is because the flood waves from the potential GLOF can arrive at these buildings before activating the monitoring stations at Khagtang and Kurjey (Fig. 6).

Some of the most damaging historical GLOF events in the world have occurred from seemingly inconspicuous glacial lakes (Allen et al., 2015; Petrakov et al., 2020), whilst some large-magnitude GLOF events have caused minimal or no downstream damage (Shrestha et al., 2023; Lützow et al., 2023). This is because the GLOF magnitude alone does not determine downstream damage caused by the GLOF event; instead, it is the interaction between GLOF magnitude and the downstream exposed elements that determines the extent of damage (Taylor et al., 2023a). For example, the greatest number of building damages associated with the 2023 South Lhoknak Lake GLOF event in the Indian state of Sikkim occurred between 200 and 385 km downstream of the glacial lake, with 59 % of these impacted structures constructed within the past decade (Sattar et al., 2025b). This highlights the escalating exposure and so the risk due to infrastructure expansion and settlement growth in GLOF-exposed areas (Nie et al., 2023) and underscores the importance of considering exposure data in GLOF risk assessment. To address this in Bhutan, we improved our understanding of GLOF risk by coupling flood characteristic modelling of individual lakes and downstream exposure data. Accordingly, we have produced flood mapping and GLOF hazard ranking for 278 glacial lakes, along with comprehensive GLOF risk assessments for 274 local government administrative units (LGUs). As a result, we classified lake130 (Thorthormi Tsho) as a very high hazard glacial lake in Bhutan, five lakes (lake93, lake251, lake262, lake276 and lake278) as high hazard and 20 other lakes as moderate hazard. Likewise, five downstream LGUs were associated with very high GLOF risk, while eight others were associated with high GLOF risk.

Our approach departs from many existing practices of identifying high hazard glacial lakes, which are primarily based on the susceptibility of lakes to produce GLOF without regarding the characteristics of settlements located downstream of the lakes (Rinzin et al., 2021; NCHM, 2019), in two ways: (1) incorporation of GLOF Hydrodynamic characteristics: we considered flow velocity and flow depth, which are both primary components of the GLOF flow that determine damage to the downstream elements (Federal Emergency Management Agency, 2004). (2) Interaction of flow depth and velocity with downstream exposed buildings: we mapped potential downstream building damage associated with each GLOF event based on the interaction between the depth-velocity and downstream at risk elements. By focusing on the interaction between flood magnitude and downstream exposed buildings, our method classifies glacial lakes as high hazard only when their potential flood poses a threat to downstream elements, making it a more practical and effective strategy for bespoke GLOF risk reduction activities. For example, lake278 and lake251 are small, and they produce relatively small GLOFs with their estimated peak discharge approximately 2000 m³ s⁻¹. However, both were classified as high hazard as the GLOF from these lakes impacts hundreds of downstream buildings (Fig. 7). Likewise, our approach assigns a higher GLOF risk ranking to communities that are either affected by GLOFs from multiple lakes, impacted by high-magnitude GLOFs, or have multiple buildings located within the GLOF inundation area, whilst also considering the community's vulnerability. For example, Chhoekhor was identified as having the highest GLOF risk in Bhutan because at least 191 buildings were potentially impacted by GLOF from as many as 14 lakes in the basin. On the other hand, gewogs such as Toedwang in Punakha are also classified as having very high GLOF risk, despite having a comparatively low number (60) of potentially impacted buildings, because these buildings could be impacted by very high magnitude GLOFs in terms of depth and velocity (Fig. 7).

We classified only one lake (Thorthormi Tsho) as a very high hazard glacial lake. This is because it is not only the largest glacial lake but is also located in the Punatsangchu basin, which is among Bhutan's most populated basins, resulting in exposure of up to 1119 buildings, again, an order of magnitude more exposure than the next lake associated with the highest exposure. These findings further highlight the importance of strengthening risk mitigation measures for Thorthormi Tsho and the affected downstream settlements.

Our study complements conventional PDGL assessment approaches by redefining which glacial lakes pose the greatest hazard to the downstream settlements. As a result, we identified three new high hazard glacial lakes, including lake93 (Phudung Tsho), lake251, and lake278 (Wonney Tsho), which are not recognised as high hazard or PDGL by any of the previous studies. Also, 53 of the previously identified 64 very highly susceptible to GLOFs lakes (Rinzin et al., 2021) are categorised as low GLOF hazard lakes. Conversely, 12 lakes classified as low or very low GLOF susceptibility emerge as moderate to high hazard in our study (Rinzin et al., 2021). Likewise, nine of the high hazard lakes monitored by NCHM (six in Punatsangchu basin, one each in Mangdechu, Chamkharchu and Kurichu basins) (NCHM, 2019) are categorized as low hazard in our study (Fig. 8). These discrepancies arise because we classified lakes as high hazard only if a potential GLOF would affect a significant number of downstream buildings, whereas earlier studies' definitions are motivated by the likelihood of producing GLOF based on characteristics of lake and surrounding terrains (Rinzin et al., 2021; NCHM, 2019) (Fig. 8). For example, lake278 in the Wangchu headwaters is classified as high hazard in our study because a potential GLOF could impact 636 buildings across seven LGUs in Paro while the earlier studies considered this lake as safe as it does not have geomorphological characteristics and lake condition to qualify as high hazard (Rinzin et al., 2021).

By ranking GLOF risk for all 274 LGUs, we discovered potential new GLOF risk hotspots, such as Paro town and Lamgong gewog in Paro and Chhoekhor gewog in Bumthang. GLOF risk in these places was previously not quantified, and existing GLOF early warning systems in Bhutan currently do not cover these high GLOF risk LGUs (NCHM, 2021). We therefore recommend prioritising monitoring of glacial lakes in Bhutan based on all components of risk (hazard, exposure and vulnerability. Specifically, Bhutan's glacial lake monitoring and downstream risk mitigation efforts should expand beyond Lunana to include other high GLOF hazard lakes and vulnerable downstream settlements such as Paro Town and Chhoekhor gewog, while emphasising that higher granularity studies might be needed to guide bespoke risk reduction efforts in these respective areas.

None of the transboundary lakes were classified as very high or high hazard based on potential GLOF impacts in Bhutan. This is because damage was minimal, mainly inundating uninhabited parts of Bhutan, located in deep, inaccessible gorges. However, we identified GLOF from four lakes in the Drangmechu basin (located in Arunachal Pradesh, India) and 11 in the Kurichu basin (located in the Tibetan Autonomous Region, China), which could potentially impact several buildings in Bhutan. Furthermore, GLOF from 20 lakes located in the Indian and Chinese territories of the Himalaya enter Bhutan, although they do not impact any buildings (Fig. 9). The modelled transboundary GLOFs sourced from lakes located within Bhutan attenuate before they cross the international border between Bhutan and India. However, we acknowledge that some of the GLOF events in the future can impact settlements along transboundary river floodplains in India, especially under the worst-case scenarios and when their flows are amplified by the addition of material along the flow path, such as from a landslide deposit (Cook et al., 2018) and/or hydropower dams (Sattar et al., 2025b). Identifying potential transboundary GLOFs is vital, given their potential destructive power and long run-out distances and challenges stemming from the absence of transboundary GLOF risk mitigation mechanisms. For example, a recent GLOF from South Lhoknak Lake in the Indian Himalaya travelled over 300 km downstream, causing significant damage in Bangladesh (Sattar et al., 2025b). The absence of transboundary cooperation for GLOF risk mitigation (for example, flood magnitude communication, shared hydropower reservoir draw-down management) between Bhutan, China, and India complicates efforts to monitor and manage such risks. Establishing regional cooperation is essential to enhance early warning systems, facilitate data sharing, and implement coordinated risk reduction strategies, such as one proposed by Zhang et al. (2025a), thereby minimising the potential damage from future transboundary GLOFs.