Enhanced glacial lake activity threatens numerous communities and infrastructure in the Third Pole

The Third Pole (Tibetan Plateau and surrounding Himalayas, Hindu Kush, and Tianshan Mountains) functions as a critical global water tower but is undergoing accelerated glacier mass loss and retreat due to climate warming. These changes have driven the formation and rapid expansion of glacial lakes, increasing the potential for catastrophic glacial lake outburst floods (GLOFs) that can devastate downstream communities and infrastructure. Despite growing concern, existing regional inventories and GLOF records are inconsistent, with variable definitions, incomplete datasets, and largely qualitative or first-order risk assessments that limit accurate trend analysis and local planning. This study addresses these gaps by integrating consistent glacial lake inventories, compiling an updated GLOF database, and implementing quantitative susceptibility assessment and high-resolution hydraulic modeling to evaluate hazards, exposure, and risk across the Third Pole.

Prior inventories of glacial lakes in the Third Pole vary widely (10,000–30,000 lakes for 2015–2020) due to differing definitions and thresholds, complicating change detection and hazard assessment. Reported GLOF frequencies have been inconsistent, with some studies suggesting decreased or stable trends; however, more recent global and regional compilations indicate increasing reporting and activity, underscoring the need for updated, harmonized datasets. Earlier risk assessments often relied on qualitative or semi-quantitative schemes and simplified flow-path models with coarse exposure data, sufficient for first-order prioritization but inadequate for detailed, site-specific risk estimation. Methodological advances include multi-criteria susceptibility frameworks, empirical relations for lake volume and breach hydrographs, and improved DEM resources. However, supraglacial and ice-dammed lake processes are distinct and often require tailored approaches, and earthquake-triggered events are rare regionally. The literature points to growing proglacial lake expansion, regional clustering of hazards, and the importance of transboundary cooperation and early warning systems.

Glacial lake inventory and change: The authors mapped glacial lakes ≥0.02 km^2 primarily fed by contemporary glacier melt within 10 km of glacier outlines using 878 Sentinel-2A/B images (10 m) from 2018, 2020, and 2022, prioritizing July–November with <10% cloud. Lakes were manually delineated, reviewed multiple times, and classified by topological position relative to glaciers into proglacial, periglacial, extraglacial, supraglacial, and ice-dammed types. Lake volumes were estimated via empirical area–volume relations, and uncertainties for area and change rates were computed using perimeter–resolution-based formulas. Long-term changes were derived by integrating prior inventories (1990, 2000, 2008, 2010, 2015) and additional Landsat scenes for underrepresented regions (Tianshan, Altai). GLOF characteristics: A moraine-dammed GLOF inventory since 1900 was compiled by integrating multiple datasets and satellite analysis, with reconstruction of drainage volumes for 65 events using pre/post area from Landsat and area–volume relations, and establishing a nonlinear peak discharge–drainage volume relation. Susceptibility assessment: For 5,535 moraine-/landslide-dammed lakes (excluding supraglacial and ice-dammed), a quantitative multi-criteria framework was applied using normalized indicators and AHP weighting. Key indicators included mean slope of the parent glacier, potential for mass movement to strike the lake, mean slope of the moraine dam, watershed area, lake perimeter, and the horizontal distance from glacier terminus to lake (to account for ice avalanche potential). Indicators were derived from AW3D30 (30 m) and other datasets. Composite susceptibility was classified into five classes (very low to very high) with natural Jenks; validation used 72 historically outburst lakes. Hydraulic modeling and hazard mapping: For 1,499 lakes with high or very high susceptibility, 2D HEC-RAS simulations were run using a dam-breach hydrograph constructed from estimated drainage volumes (upper-limit relation Va = 2.01·V0^0.65, capped at 20×10^6 m^3; small lakes assumed complete drainage) and an unsteady flow assumption with linear rise/fall in discharge. The modeled domain length was set using an empirical relation between impact distance and drainage volume. DEM inputs prioritized 8 m HMA DEM, supplemented/mosaicked with 12.5 m ALOS PALSAR DEM; depressions were filled and surfaces smoothed. Model parameters included 30 m minimum cell size, Manning’s n = 0.06, and 5 s time step. Outputs included inundation extent, maximum depth, velocity, and arrival time. A per-lake hazard index Hi = S·D·V combined susceptibility (S), modeled depth (D), and velocity (V); regional hazards were obtained by overlaying individual maps. A GLOF probability metric (PGLOF = Ση/N) quantified the relative frequency potential at specific locations. Exposure and risk assessment: Exposure indicators included buildings, hydropower projects (existing/planned), farmland, roads, and bridges. OSM features were audited and comprehensively supplemented by manual digitization from high-resolution imagery within a 200 m buffer of modeled inundation to address low OSM completeness outside Nepal. Exposure indices were weighted via AHP to reflect human impact, and risk was computed as normalized hazard × exposure, then classified (very low to very high) via natural Jenks. A potential disaster intensity (PDI = E/Ai) metric described exposure per unit inundated area. Uncertainty and robustness: The approach focuses on maximum potential scenarios for lakes with high outburst potential, acknowledging uncertainties from DEM quality/coverage (HMA vs PALSAR), Manning’s parameter sensitivity, drainage-volume cap, and exclusion of supraglacial/ice-dammed lakes. Validation indicated 93% accuracy in identifying previously outburst lakes as high/very high hazard.

• Inventory and changes: In 2022, 5,894 glacial lakes were mapped across the Third Pole (total area 748.79 ± 41.16 km^2; volume 20.13 ± 17.12 km^3). Composition: 869 proglacial (207.33 ± 8.2 km^2; 10.37 ± 9.41 km^3), 2,735 periglacial (222.24 ± 15.74 km^2; 4.47 ± 4.25 km^3), 1,929 extraglacial (290.22 ± 14.74 km^2; 4.47 ± 3.123 km^3), 113 supraglacial (6.92 ± 0.71 km^2; 0.07 ± 0.05 km^3), and 248 ice-dammed (22.08 ± 1.6 km^2; 0.38 ± 0.27 km^3). Small lakes (≤0.1 km^2) comprise 74% of the total. Two elevation clusters at ~3400–3900 m and 4700–5800 m.
• Recent expansion (2018–2022): Net area increase 22.68 ± 8.84 km^2 (3.03%); 83% of increase from proglacial lakes. Mean expansion rate of individual proglacial lakes 0.022 ± 0.002 km^2 per 5 years, an order of magnitude higher than other types. While total area expansion rate was higher in 1990–2018 (31.59 ± 4.03 km^2 per 5 years), proglacial expansion accelerated from 13.82 ± 1.33 to 18.89 ± 2.12 km^2 per 5 years in 2018–2022. Rapidly expanding lakes concentrate in the Eastern Himalayas, Southeastern Tibet, and Hengduan Shan.
• GLOF inventory and trends: 145 GLOFs from 122 lakes since 1900; 93 events after 1980. Mean annual GLOFs rose from 1.5 (1981–1990) to 2.7 (2011–2020), with intensification in Southeastern Tibet and the China–Nepal border. Triggers (n=68 identified, excluding Tianshan): 63% ice avalanches, 9% landslides, 10% melting of buried ice in moraine dams, 15% high temperature and/or heavy precipitation, 3% upstream floods. Tianshan GLOFs are typically short-lived lake drainages via ice tunnels with lower average drainage volumes (0.28×10^6 m^3) than Eastern Himalayas (6.42×10^6 m^3) and Southeastern Tibet (4.37×10^6 m^3). Regional GLOF counts correlate strongly with lake abundance (Spearman r = 0.87) and area change rate (r = 0.86).
• Susceptibility and hazard: Among 5,535 assessed lakes, 379 are very high hazard and 1,120 high hazard. Validation showed 93% (67/72) of historically drained lakes were classified as high/very high before outburst. Modeled potential inundation totals ~6,353 km^2, largest in Inner Tibet (~1,513 km^2), followed by Eastern Himalayas (~850 km^2) and Western Himalayas (~638 km^2). China’s national potential inundation area is ~4,080 km^2.
• GLOF probability and transboundary threats: 28 valleys have PGLOF > 0.24, indicating susceptibility to multiple GLOF sources; many in Eastern Himalayas and Southeastern Tibet. Notable transboundary basins include Poiqu (max PGLOF 0.71), Pengqu (0.62), and Gyigongzangbu (0.71). Identified 112 potential transboundary GLOFs (55 China–Nepal; 28 China–Bhutan).
• Risk and exposure: Of 1,499 high-outburst-potential lakes, 85 are very high risk and 113 high risk; 1,228 pose downstream risk. Estimated exposures: ~55,808 buildings, 105 hydropower projects, 194 km^2 farmland, 5,005 km roads, and 4,038 bridges at risk; approximately 190,000 people directly exposed within modeled paths. Highest exposure in Eastern Himalayas, Southeastern Tibet, and West Tianshan (>10,000 buildings each). China accounts for 46% of exposed buildings, 12% of hydropower projects, 64% of farmland, 59% of roads, and 53% of bridges, exceeding other nations in the region by factors of ~2.9–9.5; Almaty, Kazakhstan, has >6,000 buildings threatened by southern Ile Range GLOFs.
• Proximity and warning time: Median distances from dangerous lakes to nearest exposed settlements are shortest in Southeastern Tibet (7 km) and Eastern Himalayas (9.5 km), yielding basic median warning times of 0.6 h and 0.7 h, compared to 1–1.6 h elsewhere.
• Potential disaster intensity: Highest regional averages in Hindu Kush (0.516), Hissar Alay (0.479), and Southeastern Tibet (0.383), indicating likelihood of smaller but highly destructive events. Results suggest an east-to-west extension of hazard and risk under continued climate-driven lake development.

The study integrates improved glacial lake inventories, an updated GLOF database, quantitative susceptibility assessment, and high-resolution hydraulic modeling to address uncertainties in regional hazard trends and site-specific risks across the Third Pole. Findings demonstrate accelerated proglacial lake expansion and a significant post-1980 increase in GLOF frequency, especially in Southeastern Tibet and along the China–Nepal border, aligning regional lake abundance and change rates with enhanced GLOF activity. The case-by-case hazard, exposure, and risk mapping reveals substantial potential disaster volumes concentrated in economically vulnerable, transboundary mountain regions with short warning times, underscoring the need for coordinated monitoring, early warning, and mitigation. Examples from the Poiqu Basin (engineering drawdown/reinforcement at Jialongco; automated early warning at Cirenmaco) illustrate practical interventions that can reduce magnitude and impacts. The results are directly relevant to land-use planning and the protection of critical infrastructure (e.g., hydropower, transport), and emphasize the urgency of regional cooperation for data sharing, preparedness, and joint response in shared basins.

This work provides a harmonized inventory of glacial lakes and an updated record of GLOFs across the Third Pole, showing rapid proglacial lake growth and increasing GLOF activity since 1980. Through quantitative susceptibility assessment and high-resolution, case-by-case HEC-RAS simulations, the study maps hazards, exposure, and risk, identifying multiple hotspots and transboundary valleys with high GLOF probabilities. It quantifies potential inundation extents and exposed assets, highlighting short warning times and substantial risks to communities and infrastructure, particularly in the Eastern Himalayas and Southeastern Tibet, with growing concern for western regions under future deglaciation. The findings support targeted remediation (e.g., engineered lake lowering, dam reinforcement), expansion of monitoring and early warning networks, and cross-border cooperation. The methodology and datasets provide a transferable framework for repeated assessments elsewhere. Future research should refine triggers and drivers, incorporate improved DEMs and real-time monitoring, address supraglacial and ice-dammed lake hazards with tailored models, and evaluate evolving exposure and adaptive measures under climate change.

• Scope: The susceptibility and simulation focus on moraine-/landslide-dammed lakes; supraglacial and ice-dammed lakes (with distinct mechanisms and periodicity) are excluded from large-scale assessment and require separate frameworks. Earthquake-triggered GLOFs are rare and not explicitly considered.
• Data and modeling: DEM quality/coverage varies; reliance on ALOS PALSAR DEM in some areas necessitated extensive depression filling, potentially underestimating inundation extents, arrival times, and exposure. HMA DEM coverage gaps reduce reliability regionally (e.g., Altai, Southeastern Tibet). Manning’s n and simplified breach hydrographs introduce parameter uncertainty, though sensitivity suggests limited impact on key outputs. A cap of 20×10^6 m^3 on drainage volume constrains large-lake scenarios.
• Exposure data: OSM incompleteness outside Nepal required manual supplementation; while extensive, remaining omissions and boundary uncertainties persist despite 200 m buffers. Exposure represents circa 2016–2020 conditions and may not reflect recent development.
• Risk underestimation: Maximum potential simulations and data limitations likely lead to conservative (underestimated) hazard and exposure in some regions. Climate and seismic indicators were not integrated into susceptibility due to ambiguous/rare influences regionally.