This was not a monsoon flood and, strictly, not a glacial lake outburst either. It was a five-step cascade that began with rock that used to be frozen and ended 76 km downstream. Each step is understood. The reason it keeps happening is understood too.
The failure happened outside Nepal, above a tributary Nepal does not monitor, and arrived at the border in under forty minutes.
Schematic, with strong vertical exaggeration. Settlement elevations are as published; intermediate river elevations are interpolated and hydropower positions along the river are approximate. The two events are plotted on one axis for comparison of distance and height, but they were not on the same watercourse: the 2025 supraglacial lake sat above the Bhote Koshi headwaters, while the 2026 avalanche fell into the Lhende Khola, a tributary joining below it. Sources: ICIMOD, USGS, Nepal Electricity Authority, Kathmandu Post.
A section of glacier and the rock behind it detaches high on a slope in Tibet. Reporting describes a mass roughly 600 m across. This is the part that climate change is making more likely: the ice that insulated and buttressed that face is thinner than it was, and the permafrost cementing the rock is warmer.
The mass drops into the valley of the Lhende Khola, roughly 20 km north-east of the border. The impact pulverises and partially melts the ice, converting a rockfall into a fast, saturated, extremely mobile debris flow. At Chamoli the equivalent flow carried boulders over 20 m across and ran up the valley walls to 220 m.
The debris blocks the Lhende Khola. Water ponds behind it. This dam is unengineered, saturated and temporary, and nothing in Nepal or China is watching it, because it did not exist twenty minutes earlier.
The impoundment overtops or piped through, and releases in one pulse rather than a rising limb. This is why the flood behaved nothing like a monsoon flood: there was no warning shoulder, no gradual rise, no hour to move. Seismic timing puts the surge at Gyirong Port around seven minutes after the collapse.
The surge enters Nepal at Rasuwagadhi. Nepal's Flood Forecasting Division has no notification from the Chinese side and no telemetry on a tributary in another country. The Chief District Officer of Rasuwa describes it striking without warning at around 09:15.
The mechanism has a name and a literature. Permafrost, ground that has stayed frozen year-round, acts as cement in high rock faces, and glacier ice acts as a buttress against them. Warming removes both at once.
More than 70% of documented rock-ice avalanches in High Mountain Asia begin in zones of probable permafrost, and 65% start where ground temperatures already exceed −1.5 °C. That is not a marginal statistic; it means the failures are concentrating precisely where the freezing point is being crossed.
As a glacier retreats, the rock wall it was pressed against loses its support and is exposed to air temperature, rain and repeated freeze-thaw cycling for the first time in centuries. The wall does not fail immediately. It fails at some unpredictable later point, which is what makes this hazard so difficult: the cause is decades of warming, and the trigger is a Tuesday morning.
The long tail is the part that gets least attention. Even if warming is held to 1.5 °C, permafrost at depth continues to thaw for centuries. The supply of destabilised faces above Himalayan valleys is going to keep increasing across the working lifetime of every hydropower plant currently being financed in Nepal.
Honest caveat: no attribution study has tied this specific avalanche to climate change, and researchers investigating it have been careful to say so, while noting unusually warm satellite observations and reduced snow cover beforehand. The mechanism is well established. This one event is not yet proof of it.
47
Potentially dangerous glacial lakes
Across the Koshi, Gandaki and Karnali basins: 25 in the Tibet Autonomous Region of China, 21 in Nepal, 1 in India. 42 of the 47 sit in the Koshi basin alone. Identified and ranked in a joint ICIMOD–UNDP inventory published on 7 September 2020, six years before this flood.
70%+
Rock-ice avalanches starting in permafrost
More than 70% of documented rock-ice avalanches in High Mountain Asia begin in zones of probable permafrost, and 65% start where ground temperatures exceed −1.5 °C. As permafrost warms, the rock faces it was holding together stop being held together.
| Classic GLOF (July 2025) | Debris-dam burst (Aug 2026) | |
|---|---|---|
| What holds the water | A moraine or ice dam, in place for years | A pile of avalanche debris, in place for minutes |
| How long it exists | Months to decades. This one grew Dec 2024 – Jun 2025 and was watched by ICIMOD. | Roughly the time it takes to read this row |
| Can it be monitored | Yes. Satellite and telemetry both work. | Not directly. Only the slope above can be watched. |
| Realistic warning | Days to weeks, with instrumentation | Minutes, and only via the seismic signal |
| What warning requires | A gauge on the lake | A cross-border seismic trigger and a siren |
The 2025 lake sat at 5,150 m and shrank from 0.75 to 0.61 km² when it burst, which is how the outburst was confirmed from satellite. Nothing equivalent exists for a debris dam: by the time it is visible in imagery, the flood has already happened. That is why the only viable warning for the 2026 mechanism is a seismic trigger shared across the border within minutes, and why the failure to build one after 2025 was consequential.
Three events, one mechanism family, and a consistent lesson about what happens when a cascade path has infrastructure in it.
“The acceleration of these types of events is completely unprecedented in the Hindu Kush Himalayan region.”
“Such surprises could be intensified in the coming days due to global warming.”
A section of glacier and rock at roughly 5,200 m in Tibet collapsed and fell about 1,200 m into the Lhende Khola, a small tributary of the Bhote Koshi around 20 km north-east of the Nepal-China border. The mass dammed the stream with debris; when that dam failed the stored water and debris released downstream into Rasuwa. ICIMOD provided the technical assessment.
A landslide. The GFZ Helmholtz Centre initially recorded the signal as a magnitude 4.4 earthquake and Nepal's foreign minister described an earthquake triggering an avalanche, but USGS analysis found seismic waves consistent with a landslide of Ms 5.2 rather than a tectonic event. The mass movement generated the signal; it was not caused by one.
Not in the classic sense, though it is often described as one. A textbook GLOF is the failure of a moraine or ice dam holding a standing glacial lake, which is what happened on this same river in July 2025. The 2026 event was an ice-rock avalanche that created a temporary debris dam and then burst it. The distinction matters for warning: a lake can be watched for months, while a debris dam exists for minutes.
The direct attribution for this specific event has not been established, and researchers noted unusually warm satellite observations and reduced snow cover beforehand. But the mechanism is well understood: more than 70% of documented rock-ice avalanches in High Mountain Asia begin in probable permafrost zones, and 65% start where ground temperatures exceed minus 1.5 degrees Celsius. As glaciers retreat, rock faces they insulated and buttressed become exposed to warming, rain and freeze-thaw cycling. Warming permafrost stops holding rock together.
Yes, at Chamoli in Uttarakhand, India, on 7 February 2021. Around 27 million cubic metres of rock and ice fell from Ronti Peak, the debris flow scoured valley walls up to 220 m above the floor, destroyed the 13.2 MW Rishiganga project and then struck the 520 MW Tapovan-Vishnugad project 10 km downstream. Roughly 200 people were killed or went missing. It was analysed in Science later that year.
The physical drivers are getting stronger, not weaker. Permafrost at depth continues to warm for centuries even under a 1.5 degree scenario, so the supply of destabilised rock faces increases regardless of near-term emissions. Nepal has 21 identified potentially dangerous glacial lakes and 25 more sit upstream in Tibet. The question is not whether but where, and whether anything downstream is watching.
Amarnepal is not a geoscience publication. This page summarises the published assessments of ICIMOD, USGS and the peer-reviewed literature on High Mountain Asia; where those sources are provisional or in disagreement, we say so rather than resolving it ourselves.