We used multi-source optical remote sensing datasets (Table 2) to track the area of glacier and glacial lakes from 1976 to 2024. Glacier and glacial lake outlines were manually delineated from the geometrically corrected false colour composite images (NIR/red/green) of satellite data by a single expert (Wang et al., 2020). The error associated with the delineation of individual lake boundaries was calculated as the product of half the image resolution and the lake's perimeter. This error calculation is predicted on the assumption that, on average, the lake margin intersects the centres of the pixels along its perimeter (Salerno et al., 2012).

During field investigation (November–December 2024), we engaged with local residents of Thame Village, stakeholders and rural municipality to gather insights on the timing and impacts of this event through interview and informal discussions. These consultations provided qualitative information on the timing and progression of the flood, observed impacts, and community experiences before, during and immediately after the GLOF. We explored the potential causes of this GLOF and employed a Global Navigation Satellite System (GNSS) to measure the lowering of the lake levels after their outbursts. We utilized a laser range finder to assess the width and depth of the breach in the moraine dam of the glacial lakes (Fig. 2), as well as changes in the river morphology and geomorphology in Thame Village. Additionally, the breach parameters for upper lake were also obtained and verified from the field report of International Centre for Integrated Mountain Development (ICIMOD) (Maharjan et al., 2025).

A high-resolution (1 m) DEM was acquired via uncrewed aerial vehicle (UAV) surveys on 24 May 2025. The drainage volume of the upper and lower lakes was reconstructed using the lake outlines and the drone-derived DEM. The post-GLOF lake water surface elevation was estimated by sampling elevations along the lake perimeter from the DEM and assigning the median elevation to all cells within the post-GLOF lake outline, thereby creating a flat-water surface. The pre-GLOF lake surface was reconstructed using the same approach. The reconstructed pre- and post-GLOF water surfaces were differenced, and the resulting elevation differences were multiplied by the grid-cell area and summed to estimate the drained lake volume. For this, lake outline delineated before (15 August 2024) and after (18 August 2024) the event was utilized. The water volumes of upper and lower lakes were empirically estimated using average of nine area-volume relationships (see Tables 3 and S1 in the Supplement). Further, the obtained drained volumes from drone-derived DEM were also compared with empirically estimated drained volume (refer Text S1 and Table S1 in the Supplement).

The peak discharge during the GLOF event and the timing of the dam breach formation were also derived empirically based on the released volume and lake lowering height. The released volume was calculated as described above based on drone-derived DEM, while the lake lowering height and breach dimensions were determined through field measurements (Fig. 2). Various empirical equations commonly found in the literature were utilized for estimation purposes, and the overall average of these calculations was taken as the final estimate for both peak discharge and the timing of breach formation (Tables 3 and S2). The total volume released, peak discharge, and timing of the dam breach for the cascading event were analysed by combining the released volumes from both the upper and lower lakes (Tables S1 and S2).

The different possible triggering factors of the GLOF were analysed based on field observations, published news/reports and through examination of fine-resolution remote sensing images (see Sect. 4.2). Icebergs were manually delineated from PlanetScope (3 m) and Gaofen-7 images (0.65 m) of 15/16 and 18 August 2024, respectively. The icebergs, both pre- and post-GLOF, as well as those that are grounded and drifting, were identified through a comparative analysis of the images (Fig. S2). The volume of icebergs was determined based on the area-volume relationship as shown in Eq. (1). This equation was derived by Watson et al. (2020) through detailed study of calving and iceberg formation of Thulagi glacial lake in Nepal. 1Vi=4.3017A1.19 where, Vi is volume of icebergs in m³ and A is surface area of icebergs in m².

We used ERA-5 Land data, the global reanalysis dataset which provides long-term data on climate variables since 1950 (Muñoz-Sabater et al., 2021) and has good performance in the Nepal Himalaya (Chen et al., 2021). ERA5-Land dataset aggregated at daily timestep was accessed using the Google Earth Engine (GEE) platform. Daily aggregates in GEE are pre-calculated daily from hourly values. We extracted the daily aggregated 2 m air temperature (24 h average) and precipitation (24 h total) for the lake area from 1990 to 2024. The days with extreme temperature and precipitation exceeding 90th and 95th percentiles were calculated with respect to long-term time series (1990–2024) data. Additionally, data from the Phortse Automatic Weather Station, located 11 km east of Thame Village at the same elevation (3850 m), was utilized to obtain observed values of temperature and precipitation for the area.

Cascading GLOF processes were simulated using r.avaflow (v. 4.0G), an open-source multi-phase computational tool for complex mass flow dynamics in mountain regions (Mergili and Pudasaini, 2014–2015). Unlike traditional single-phase models, r.avaflow employs a coupled multi-phase approach to accurately characterize phase interactions and material entrainment, enhancing the simulation of cascading hazards (Mergili et al., 2017). Its core algorithm, developed by Pudasaini and Mergili (2019), propagates mass flow across a DEM using depth-averaged continuity and momentum equations (Mergili et al., 2018b). The model employs a second-order TVD-NOC (Total Variation Diminishing Non-Oscillatory Central Differencing) scheme on a staggered grid for spatial discretization, ensuring both accuracy and stability (Tai et al., 2002). Operating with GIS raster cells, it facilitates mass-flow exchanges between adjacent cells at each time step (Mergili et al., 2025). Integrated within GRASS GIS 7, r.avaflow combines Python/C code with R-based statistical tools for efficient simulation and analysis, as detailed by Mergili et al. (2017).

We modelled the cascading GLOF process chain based on two scenarios that encapsulate the extremes of potential mechanisms within the event. In Scenario A, we assumed a rapid inflow generated by the upstream lake breach entered the downstream lake and produced displacement wave that overtopped the dam crest and progressively eroded the moraine dam. However, in Scenario B, the inflow from the upstream lake was assumed to gradually increase the water level (and storage volume) of the downstream lake until it reached the dam crest, subsequently resulting in dam failure.

For GLOF simulation, the volume of the lake, peak discharge, and breach timing were empirically estimated (Sects. 3.2 and 4.2), while other dam and breach parameters were sourced from field measurements (Fig. 2). For both scenarios, lower lake volume modelling was performed using r.lakefill function, a GRASS GIS script used within the r.avaflow and HMA DEM. Following previous moraine dam GLOF studies, the value of the lake dam entrainment coefficient was taken as 10^−5.7 (Zhang et al., 2025). In Scenario A, an idealized triangular hydrograph (Huggel et al., 2002; Froehlich, 2025) was developed based on upper lake estimates of drained volume and peak discharge for input into the lower lake (see Fig. S3). For Scenario B, a GIS-based analysis was conducted, revealing that the drained upstream lake volume resulted in 11.8 m rise in water level prior that exceeded the dam crest leading to the dam's failure.

Other essential input parameters for the simulations in both scenarios are summarized in Table 4 which were mainly derived from existing studies (Mergili et al., 2020b; Rinzin et al., 2025; Zhang et al., 2025; Chen et al., 2026) and obtained through optimization at the early stage of simulation. The selected parameter set yielded good agreement with the observed flood extent and flow routing, resulting in satisfactory accuracy assessment scores (see Sect. 4.4). The HMA-DEM (Table 2) was utilized to represent the terrain owing to its greater resolution and all simulations were executed with a cell size of 10 m. The total simulation duration was set to 3600 s (1 h). The analysis was confined to Thame Village, which is approximately 9 km downstream from the upper lake, as this area experienced the most significant damage. This limitation was necessitated by the unavailability of HMA-DEM data further downstream, as well as to reduce uncertainties associated with the complex topography of the Himalayan region.

The accuracy assessment of GLOF reconstruction was conducted by comparing the satellite-derived inundation area with the modelled flood extent. An overlay analysis was performed to calculate true positive, false positive, and false negative areas (see Text S2). From these, precision and recall were computed, followed by the F1 Score to provide a balanced measure of overall inundation accuracy (Text S2). These metrics yield a score between 0 and 1, with 1 indicating a perfect alignment between the two datasets being compared (Rawlins et al., 2026).

The impact assessment of the GLOF was conducted by synthesizing the detailed flood incident report on losses and damages compiled by the local government i.e., KPL RM (https://khumbupasanglhamumun.gov.np/, last access: 18 August 2026). Furthermore, additional impacts were gathered from a field study, and news and institutional reports. Reported damage to trekking trails in the Thame Valley was cross-verified using high-resolution GaoFen-2/7 satellite imagery by comparing the reported length of damaged trails with the length of visible post-event damage observed in the imagery.

The evolution of the glacial lake cluster located in the headwaters of Thame Valley is shown in Fig. 3. The upper Ngole lake (Lake 1) started to form in the beginning of the 21st century with progression of its parent glacier shrinkage (Fig. 4a) and reached a considerable size of 0.018 ± 0.009 km² in 2010 (Figs. 3b and 4b). With rapid expansion in recent years (Fig. S4), it reached an area of 0.11 ± 0.004 km² before its outburst. After its outburst, the area slightly reduced by 0.013 km².

The lower Ngole lake (Lake 2) started to evolve before the 1980s, with an area of 0.04 ± 0.01 km² in 1989. It grew by 20 % in the last 35 years (Fig. 4b). The area reduced from 0.048 ± 0.01 km² to 0.015 km² after its outburst in 16 August 2024. The other lakes (lakes 3 and 4) on the southern sides were present from 1989 and a slight fluctuation of ±0.01 km² is observed between the years (Table S3). Lake 5 is transient in nature and is mostly observed in monsoon season when snow-glacier melt and rainfall are common (Fig. S4). The areas of lakes 3, 4 and 5 are 0.042, 0.042 and 0.01 km², respectively as derived from a Gaofen-7 image of 18 August 2024.

Analysing the various potential triggering and conditioning factors (Table 5), it becomes evident that the likely cause for the outburst of the upper glacial lake is potentially linked to extreme temperature-induced snow and glacier melt, coupled with calving from the parent glacier (debris-covered). These processes would increase the lake's water volume, which potentially initiated a dam failure of the thin moraine dam (∼ 3–4 m thick), which overlaid bedrock (Fig. 5). However, the highly fragmented rock slopes along the eastern margin of the upper lake suggest that rock avalanches may have occurred (see Sect. 5.2), potentially generating displacement waves that overtopped the dam and triggered the GLOF. The displacement wave mechanism, irrespective of the potential trigger, is represented by Scenario A in our modelling.

The total surface areas of the delineated drifting icebergs (> 130 m²) in upper lake before the GLOF were 3117 ± 1413 m² with corresponding volume of 61 905 ± 24 145 m³ (Figs. 5 and S2). We estimate icebergs to have a total height of ∼ 15–25 m, based on composite ice thickness raster data (Table 2) with more than 80 % to be considered submerged underwater. The glacier melting and calving events are linked to the extreme air temperatures (see Sect. 4.3). Our estimate from drone-derived DEM calculations indicates that approximately 4.1 × 10⁵ m³ of water drained from the upper glacial lake. The lake level lowered by approximately 4.6 m and dam breached by approximately 21 m (Fig. 2b). The total surface areas of the drifting ice-bergs (> 20 m²) after its outburst were 7092 ± 1210 m² with a volume of 162 921 ± 19 982 m³ (Figs. 5d and S2).

The subsequent outburst flood from the upper glacial lake triggered the GLOF from lower glacial lake. The floodwaters from the upper lake descended approximately 182 m and travelled about 700 m before plunging into the lower lake (Fig. S1). The steepest section is immediately adjacent to the upper lake, where the bedrock step drops 100 m vertically over a 120 m horizontal distance. The rapid influx of water and sediment substantially increased the volume of lower lake and generated surge waves in lake surface water that ultimately led to the failure of the moraine dam. Our estimate indicates that approximately 3.6 × 10⁵ m³ of water drained from the lower glacial lake. Subsequently, the lake level has lowered by approximately 12 m and dam has incised by approximately 30 m (Fig. 2).

A total of approximately 7.7 × 10⁵ m³ of water was released from both lakes during the event. The drone-derived DEM and empirical methods yielded broadly comparable estimates of total drained volume, differing by 28.3 % (Text S1). The lower lake estimates were in close agreement (2.9 % difference), whereas the upper lake showed a larger discrepancy (64 %).

Interviews with local residents indicated that the days preceding the GLOF were perceived as unusually warm. Although the reported timings varied among respondents, residents consistently stated that the flood occurred during the afternoon of 16 August 2024. Records from local authorities and other stakeholders indicate that the flood reached Thame Village at approximately 13:30 LT. A Planet satellite image acquired at around 10:00 LT on the same day showed no evidence of an outburst at the source area (Fig. 5). Together with the reconstructed flood travel time of approximately 20–25 min from the source lakes to Thame Village (Sect. 4.4), these observations suggest that the GLOF was initiated at around 13:00 LT.

In the month leading up to the occurrence of the GLOF, several days experienced temperatures and precipitation levels that exceeded the 90th and 95th percentiles thresholds derived from long-term (1990–2024) daily climatology (Fig. 6). During summer months (JJAS), climatic conditions are generally warmer and wetter than during preceding spring season (MAM) in Nepal. Specifically, the temperatures on 12–15 August, immediately prior to the GLOF event on August 16, surpassed both the 90th and 95th percentile thresholds. Likewise, the precipitation on 14 and 15 August also exceeded both 90th and 95th percentiles (Fig. 6a and b). Similarly, the exceedances in temperatures and precipitation were observed during July. Compared with the same 30 d period during 1990–2024, the pre-event period (17 July–15 August 2024) was the warmest on record, with temperature exceeding the both 90th and 95th percentile thresholds (Fig. S6a).

The average annual temperature and the maximum monthly temperature during the summer seasons in the Ngole lake region have been observed to be increasing at rates of 0.017 and 0.003 °C yr⁻¹, respectively, between 1990 and 2024 (Fig. 6d). On the day preceding the outburst, the ERA5-Land data recorded minimum and maximum temperatures of 3.7 and 9.7 °C, respectively with consecutive extreme wet days exceeding 10 d (Fig. 6c). Concurrently, data collected from the Phortse automatic weather station, located 11 km east at an elevation of 3850 m, reported minimum and maximum temperatures of 7.45 and 14.13 °C, respectively, along with a total rainfall of 124 mm over a consecutive 14 d wet period prior to the GLOF event (Fig. S6b). This implies that the data of ERA5-Land, though being reanalysis, captures the general climatic conditions and trends of the region (Chen et al., 2021). These climatic conditions are conducive to increased glacier and snow melt, which might have been contributed glacier melt and calving, and initiation of the GLOF.

The reconstructed GLOF flow heights, arrival times, hydrographs and velocity for both scenarios A and B are shown in Figs. 7, 8, 9 and S7. Accuracy assessment of reconstructed GLOF inundation areas show that Scenario A achieved precision, recall, and F1 scores of 0.73, 0.85, and 0.78, respectively, while Scenario B yielded 0.71, 0.83, and 0.77. Overall, these metrics indicate acceptable model performance in capturing the actual flood extent, with Scenario A being marginally more accurate than Scenario B.

In Scenario A, the simulation results reveal the dynamic characteristics of the event and its downstream impacts. The inflow hydrograph of the upstream lake (Fig. S3) exhibits a sharp peak of approximately 586 m³ s⁻¹ within 699 s. This surge wave with velocities 15 m s⁻¹, released from Input1, first impacted the downstream lake, initiating its breach and generating a compound flood. At the immediate downstream outlet (O1), the discharge hydrograph displays peak exceeding 800 m³ s⁻¹ and lasting for more than 2000 s, reflecting a complex breach evolution process (Fig. 9a). At 800 s after the breach, the liquid-phase discharge at O1 reached a peak of 912 m³ s⁻¹, accompanied by the release of solid-phase material derived from the erosion of the dam body during failure. The combined flood wave then propagated downstream, reaching Thame Village (O2) 1300 s (approximately 22 min, Fig. 8) after the initial breach with maximum discharge of 574 m³ s⁻¹, maximum flow heights up to 4 m and velocity of 10 m s⁻¹ (Figs. 7, 9 and S7). Although significant channel storage effects and energy dissipation reduced the peak discharge at O2 to approximately 548 m³ s⁻¹, the magnitude and duration of the flood remained sufficient with depositions to cause severe damage, highlighting the long-term hazard posed by cascading GLOF events.

In Scenario B, in contrast to the sudden displacement wave in Scenario A, the failure was induced by sustained inflow from the upstream lake, which progressively increased the storage and water level of the downstream lake until overtopping and structural collapse occurred. The hydrographs illustrate the dynamic features of this breach mode. The discharge at O1 (breach output 1) shows a rapid rise to multiple peak flow, a characteristic feature of dam-break floods, followed by gradual recession (Fig. 9b). At 240 s after the breach, the liquid-phase discharge at O1 reached a peak of 1240 m³ s⁻¹, accompanied by the outflow of solid-phase material originating from the dam body. The discharge at O2 (Thame Village) exhibits a delayed and attenuated yet still significant flood wave, arriving 880 s (∼ 15 min) after the breach with maximum flow heights up to 4 m. This scenario highlights a comparatively slower-onset but high-volume failure process primarily driven by cumulative basin filling rather than instantaneous impact, resulting in a distinct hydrograph shape, lower travel time and potentially different downstream impacts compared with displacement-wave-triggered events.

The extent and depth of bed scour observed along the channel between the upper and lower lakes (Fig. 2) provide additional evidence that Scenario A is more plausible than Scenario B. Field observations reveal that a sand plain located immediately downstream of Thengpo Kharka, approximately 500 m long and 150 m wide, temporarily stored ponded water during the event (Fig. S8). Water released during GLOF event was briefly impounded within this sand plain, which likely attenuated the flood wave and delayed its downstream propagation. This phenomenon is accurately captured by the GLOF modelling, where water is shown to have been impounded just downstream of Thengpo Kharka (Fig. 7). This temporary storage provided critical lead time for residents of Thame Village to evacuate safely. Subsequently, the impounded water overtopped and eroded the banks and bed of the ∼ 2 km long channel between the ancient landslide dam and the debris fan near Thame Village, contributing to significant channel incision and sediment transport (Fig. S9).

The GLOF had profound impacts in the downstream areas as it turned into perilous debris flood with Thame Village being the primary deposition zone (Figs. 10 and S9). The roaring sound of the flood and information flow from upstream to downstream areas helped the communities to evacuate during the GLOF event, leaving no casualties. The GLOF had a devastating impact on Thame Village, displacing 135 people and leading to the destruction of several tea houses and buildings, including the primary school, as well as considerable damage to agricultural fields (Fig. 10). The detailed impact assessment shows that GLOF impacted the settlements, trekking trials, land and infrastructures worth total of USD 6.18 million in KPL RM alone (Table 6). The total length of trekking trail damaged in Thame Village was ∼ 500 m cross-verified through images. The impacts of the GLOF were minimal outside the KPL RM; however, it caused substantial erosion and bank cutting up to 50 km downstream and partially damaged a motorable bridge (Fig. 10f).

The GLOF-triggered landslides in the Thame area, characterized by active glacio-lacustrine deposits, are ongoing. Deep-seated movement in Thame Village is exacerbated by riverbank erosion and migration (Fig. S10), leading to persistent ground settlement and tension cracking, particularly during wet periods. Elevated water levels during the monsoon further destabilize the slopes, with landslide activity expected to extend into areas farther from the river. Deep-seated rotational landslides are advancing from the east and north sides of the head pond on the right bank, impacting houses and agricultural land on the left bank.

Increase in temperatures in the Himalayan region are driving the retreat and mass loss of glacier (Bhattacharya et al., 2021) and subsequently new lakes are being emerged at higher elevation (> 4600 m a.s.l.) in the Himalaya, including Nepal (Khadka et al., 2018). It is clearly visible in the satellite image that there is no existence of the upper lake before 2000 (Fig. 3a). In the eastern Nepal including Everest region, maximum annual temperature has increased by 0.06 °C yr⁻¹ (Salerno et al., 2015) and particularly, in the study area, the annual average temperature has increased by 0.17 °C per decade during 1994–2024 (Fig. 6d). In response to climatic warming, glaciers have retreated with rates of -6.1 ± 0.2 m yr⁻¹ between 1962 and 2011 (Thakuri et al., 2014). The parent glacier of the upper lake has shrunk by -0.45 km² between 1980 and 2024 (Fig. 4a), consequently the lake started to develop as glacier mass loss accelerated. After the development of the glacial lake, notable expansion occurred leading up to the outburst (Fig. S4).

Although mass movements (e.g., snow/ice avalanches, rockfalls, or landslides) trigger over 70 % of Himalayan GLOFs (Nie et al., 2018; Lützow et al., 2023), a probable snow/ice avalanche trigger was neither reported by government institutions or the Sagarmatha Pollution Control Committee during their immediate post-event field survey (Maharjan et al., 2025), nor was it evident in available remote sensing imagery. Through discounting other potential triggers (Table 5), our analysis indicates that the GLOF from upper lake was likely triggered by a rapid increase in the volume of the upper lake or a displacement wave, driven by extreme temperature-induced snow and glacier melt. The GLOF initiation potentially included calving from the parent glacier, although it is possible this calving occurred post-GLOF caused by the lake level lowering. Our analysis cannot definitively attribute the GLOF to a rockfall, as reported by ICIMOD (Maharjan et al., 2025).

High-resolution remote sensing images (Fig. 5) and geomorphological/geological interpretation provides no definitive evidence of a contemporary rockfall or rockslides (Fig. 11a) impacting the lake. The rock deposits observed along the eastern moraine margins are characterized by a dark, weathered appearance, inconsistent with the fresh, sandy or whitish talus typically associated with recent rock avalanches (Fig. 11). Historical imagery and local accounts confirm these deposits were likely pre-existing (Fig. S5). All historical images (Fig. S5) depict a cluster of boulders that has been present since 2014. It appears that this section of the rockfall is partially situated on debris-covered ice that is melting, rather than directly on the lake bed. Furthermore, fresh small patches of scars observed in the north-eastern zone are an unlikely source for the GLOF (Fig. 11b). The corresponding rockfall volume appears too small to generate a displacement wave capable of significantly impacting the main lake. Even if a wave were generated, the narrow strait would dampen its energy before it could propagate into the main lake. Additionally, the extreme verticality of the terrain limits the reliability of DEM differencing for detecting recent slope failures. Even if a rockfall had occurred as described by Maharjan et al. (2025), its direct impact on the lake is considered unlikely as a substantial already present debris apron (∼ 20–30 m wide) buffers the lake from the lateral moraine slopes, intercepting material and preventing a direct, dynamic hit on the water body.

Therefore, we conclude that the GLOF from upper lake was most likely caused by a hydrological tipping point, a phenomenon that similarly contributed to the 2013 Kedarnath outburst-related disaster (Allen et al., 2016) and the 2009 outburst of Ventisquero Negro in the Patagonian Andes (Worni et al., 2012). Consecutive days of extreme precipitation and anomalously high temperatures prior to the event (Fig. 6) may have contributed to hydro-meteorological preconditioning prior of the lake system. Although direct observations of lake-level changes are unavailable and a causal relationship cannot be conclusively established (Fig. S11), the observed meteorological anomalies support the hypothesis that these conditions increased the lake's susceptibility to failure.

However, we acknowledge the possibility that a sudden rockfall or rock avalanche might have been occurred, as reported by Maharjan et al. (2025) who visited the site earlier than our team. Given the rapidly changing post-event conditions and the presence of steep, highly fragmented rock slopes along the eastern margin of the upper lake (Fig. 11), a rockfall or rock-avalanche remains plausible. Nevertheless, the available evidence is insufficient to constrain its timing or volume, or to determine whether the observed scour path represents a rockfall trajectory or erosion caused by flowing water into the lake. Notably, the cascading GLOF modelling, through its two scenarios effectively captures the range of possible outburst mechanisms and process (Sects. 3.4 and 4.4), even when the exact triggering factor of the upper lake remains undefined.

The modelling demonstrates that small, previously overlooked glacial lakes can generate hazardous cascading outburst consequences when hydraulically connected. The simulations indicate that downstream impacts were controlled not only by the initial outburst volume but also by progressive lake interactions and sediment entrainment. These findings highlight the importance of considering multi-lake systems in future GLOF hazard assessments.

Comparison of the empirical and drone DEM-derived drained volume estimates (Text S1) shows good agreement for the lower lake, differing by only 2.9 %. In contrast, the empirical approach underestimated the upper lake drained volume by 64 %, resulting in a 28.3 % underestimation of the total drained volume (Text S1). This highlights the advantage of post-GLOF DEM-based reconstruction for estimating drained volume, particularly where lake basin morphology is complex. However, post-GLOF DEM-based drained volume reconstruction is also sensitive to lake outlines. The absence of high-resolution DEMs for the post-GLOF period precluded the incorporation of spatially heterogeneous parameter values and constrained the calibration of model inputs (Zhang et al., 2025). Consequently, the discharge and the simulated results by the model might be underestimated, particularly in the Thame Valley as compared to field – limitations attributable to the lack of fine resolution DEM, post-event topographic data and uncertainties associated with erosion (including large boulders) and deposition rates along the river channel (Fig. S9). This limitation is consistent with previously documented challenges in r.avaflow applications or any hydrodynamic models (Rinzin et al., 2025; Sattar et al., 2025b). Moreover, r.avaflow is sensitive to input parameters and adjusting them can affect mass flow behaviour (Mergili et al., 2018a). Future modelling studies can enhance debris flow dynamics by incorporating high-resolution DEMs along with erosion and deposition rates along the river channel. While alternative hydrodynamic models such as HEC-RAS may offer advantages in accurately reproducing inundation extents (Sattar et al., 2025b), metrics such as travel time and peak discharge are of greater relevance in the context of complex mass flows and cascading processes (Mergili et al., 2020a). In this regard, the performance of r.avaflow remains satisfactory and also the inundation extent produced by it when compared to post-GLOF images with acceptable F1 scores (Fig. 7 vs. Fig. S9). From a model-development perspective, this event highlights both the capability of r.avaflow to simulate complex cascading GLOF processes and the challenges associated with estimating drained lake volumes, and sediment availability and mobilisation. These findings emphasize the need for improved parameter constraints and high spatial resolution DEM for future applications. The GLOF modelling is limited to Thame Village; therefore, future studies could assess GLOF impacts further downstream.

On the impacts side, tangible loss is quantified in this study, however intangible impacts and losses such as cultural loss, mental and physical stress of GLOF is not covered by this study and needs a different evaluation and assessment approach.

GLOF hazard refers to the likelihood and magnitude of an outburst event from a given lake, considering a broad range of contributing factors (Allen et al., 2022). In this context, both national (Rounce et al., 2017) and basin scale (Khadka et al., 2021) hazard assessments have overlooked the upper and lower lakes, primarily due to minimum size thresholds used in their studies (Table 1). The Thame event suggests that existing GLOF hazard assessment schemes should be revised in two key ways: (1) lake-size thresholds used to screen potentially dangerous lakes should be lowered to include small lakes capable of producing damaging outbursts, and (2) assessments should explicitly evaluate cascading processes, whereby failure of an upstream lake can trigger downstream lake outbursts and amplify flood impacts. To address this gap, a first-order assessment of future GLOF hazard for both lakes was systematically conducted by analysing twenty GLOF hazard factors (GAPHAZ, 2017) that encompass atmospheric, cryospheric, geotechnical and geomorphological, conditioning and triggering variables (Table S4).

A repeated GLOF from the upper lake is possible, since it is still connected to a calving glacier, has potential for upward expansion and remains susceptible to ice/snow avalanches and rockfall (Fig. 11). Following the 2024 event, however, much of the moraine and debris that previously formed the dam was removed, leaving the lake largely confined by bedrock. As a result, the likelihood of a dam breach failure similar to that of 2024 event is reduced. Nevertheless, overtopping induced by displacement waves generated by ice or rock avalanches cannot be ruled out, particularly as a substantial proportion of the upper lake volume remains. In contrast, the lower lake has lost ∼ 80 % of its volume in the recent event (Table S1). It no longer has direct contact with a glacier (Fig. 1) and is not susceptible to direct ice- or rock-fall triggers. Although a renewed outburst remains possible if a GLOF were to originate from the upper lake, the reduced lower lake volume suggests that the magnitude of future cascading floods is likely to be lower than that of the 2024 event in the near term. Continued glacier retreat and future lake evolution may, however, alter the hazard potential over longer timescales. Therefore, regular monitoring of upper glacial lake is recommended using remote sensing technologies and field assessments.

The Thame GLOF event underscores the need for proactive, integrated disaster risk management that addresses compound hazards in the Thame Valley, including glacio-lacustrine landslides, monsoon-driven flooding, channel instability, and potential future GLOFs from small lakes (Sattar et al., 2025a). Risk in the valley arises not only from the probability of extreme outburst floods but also from the interaction of ongoing geomorphic processes with settlements and critical infrastructure along the Thame stream (Fig. S10). Effective risk management must therefore integrate hazard assessment, exposure reduction, and vulnerability mitigation within a unified framework (Emmer, 2024; Niggli et al., 2024; Wang et al., 2026).

Short-term measures should focus on reducing hazard impacts and immediate exposure rather than attempting to eliminate underlying geomorphic drivers. Targeted structural interventions – such as selective, over-toppable flow-diversion berms near the debris-fan apex and localized bank protection within the village – can help limit flood intrusion and channel migration under recurrent monsoon conditions. Concurrently, GLOF exposure and vulnerability must be reduced through land-use restrictions or zoning, evacuation of structurally compromised buildings and systematic ground deformation monitoring. Low-cost monitoring approaches can support adaptive decision-making, including evacuation thresholds and infrastructure relocation, while large-scale river engineering works are not recommended due to high cost, technical uncertainty, and limited long-term effectiveness.

Over the long term, sustainable GLOF risk reduction depends primarily on exposure management through risk-informed land-use planning and the progressive relocation of permanent settlements and critical infrastructure to lower-hazard zones (Khadka et al., 2024a). Areas subject to frequent flooding or landsliding should be reserved for low-exposure areas. Given persistent residual risk, a robust GLOF early warning system is also essential to reduce vulnerability and prevent loss of life. Such a system must combine reliable hydrological monitoring, communication mechanisms with redundancy, clear evacuation planning, and strong institutional support for maintenance and periodic reassessment (Wang et al., 2026). Overall, the Thame case highlights that in high-mountain environments, reducing GLOF exposure and vulnerability is more effective than attempting to fully control dynamic geomorphic hazards and implementing expensive lake mitigation projects such as lake lowering.