At nine o'clock on Wednesday morning, a wall of water came down the Bhote Koshi river into Rasuwa district in northern Nepal and took away roads, bridges, a customs post, a market and an unknown number of people.
The most important sentence in the story is this one, from the district's own chief officer, Narendra Pariyar: there was no rainfall in the district headquarters.
It had not been raining. The water came from somewhere else, and it came without warning.
What the numbers actually are
Be careful with the death toll today. Within hours of the flood, outlets were carrying figures ranging from single digits to dozens, and they will keep moving as rescuers reach cut-off settlements. Any number published now is a floor and a guess at once.
The firmer figure is the missing:
| Travellers unaccounted for | 384 |
| Of whom foreign nationals | 291 |
| Indian nationals | around 105 |
| British nationals | 12 |
And the infrastructure number, which is staggering for a country this size:
Nepal's energy ministry says roughly 430 MW of generation was knocked out — about one-eighth of the country's entire installed capacity.
An eighth of the national grid, in a morning, because of weather that did not happen in Nepal.
Where the water came from
The working explanation, and it is preliminary, comes from Rijan Bhakta Kayastha, a glaciologist at Kathmandu University, via the Kathmandu Post: an ice avalanche blocked the Lhende river on the Tibetan side, forming a temporary lake behind the debris. The dam then failed, and everything behind it came down at once.
Kayastha added a detail worth noting: an earthquake occurred on the Tibetan side at around the same time.
Pariyar, speaking to Xinhua, was blunt about the limits of what anyone knows:
"We cannot say for certain whether a glacial lake burst somewhere upstream or something else happened. The flood appeared suddenly."
Why nobody could have been warned
This is the part that generalises beyond Wednesday.
A glacial lake outburst flood — a GLOF — is water held back by ice, moraine or avalanche debris, released all at once when the barrier fails. It does not need a storm. It does not announce itself. In a valley that has had no rain, under a clear sky, the river simply becomes something else.
The trigger for this one is believed to be inside Chinese territory, in a high valley Nepal does not monitor and cannot monitor. Early-warning systems work by watching a threat approach. There is no useful warning time when the threat forms in another country and travels down a mountain.
The International Centre for Integrated Mountain Development (ICIMOD) had already flagged this exact watershed. In a published study it identified the Poiqu watershed in the Tibet Autonomous Region — the river that becomes the Bhote Koshi and then the Sun Koshi once it crosses into Nepal — as "highly prone to glacial lakes outburst floods."
The warning existed. It named the river.
This is the second time in thirteen months
On July 8, 2025, a flash flood hit the same Bhote Koshi. Nepal's Department of Hydrology and Meteorology attributed that one to a glacial lake outburst.
Same river. Same border. Thirteen months.
And the wider frequency data is the thing that should stop people:
- In the 2000s, a glacial-origin flood was expected in the Hindu Kush Himalaya roughly once every five to ten years.
- In May and June 2025 alone, ICIMOD recorded three — in Nepal (Limi), Afghanistan (Andorab valley) and Pakistan (Chitral and Hunza).
Saswata Sanyal, ICIMOD's Disaster Risk Reduction Lead:
"The acceleration of these types of events is completely unprecedented in the Hindu Kush Himalayan region. We need to delve deeper into the triggers that are resulting in cascading impacts."
An event with a one-in-five-year return period is now happening several times a season across the range, and twice on one river in just over a year.
The other flood, and why the difference matters
Nepal was not the only place under water this week. Several hundred miles east, Typhoon Narra has been sitting over the Beibu Gulf, dumping rain on southern China and northern Vietnam.
The photograph at the top of this story is that flood, not this one. We have used it deliberately, because we do not have an image of Rasuwa — and the reason we do not is the story.
In Guangxi, more than 54,000 people were evacuated, including 8,087 into emergency shelters. In the city of Chongzuo, nearly 700mm of rain fell in three days and river levels passed the record set in 2008 — the largest flood ever recorded in the upper reaches of the Zuojiang river. China activated a Level 4 national disaster relief response, and Guangxi alone deployed 459 firefighters, 114 vehicles and 80 rescue boats.
Now hold those two events side by side, because the contrast is the whole argument.
Guangxi knew. A typhoon is visible for days. It has a name, a track, a forecast cone. Tens of thousands of people were moved before the water arrived, into shelters that had been designated in advance, by agencies working from a national response tier. That is what a warning system is for, and it worked.
Rasuwa had none of that. No storm, no forecast, no rain. A slab of ice let go in a valley on the other side of an international border, dammed a river nobody was watching, and the dam failed. The first anyone in Nepal knew about it was when the Bhote Koshi rose seven metres.
Same week, same continent, same element. One disaster you can see coming for three days; one you cannot see coming at all.
What this means downstream
The Bhote Koshi runs into the Trishuli, and the Trishuli eventually crosses into India as the Gandak. Water that starts as ice in Tibet ends up in the Ganges basin. The people at risk from a high-altitude avalanche in one country live in three.
That is the reason this story is not really about a flood. Nepal's exposure here is not a failure of its own forecasting. It is a monitoring problem that stops at a border, on a range that is losing its ice faster than anybody planned for.
What to watch
Whether the toll settles, and where. Whether satellite analysis — the EU has activated its Copernicus emergency mapping service — confirms the avalanche-dam theory or finds a glacial lake. How long the 430 MW takes to come back. And whether anything at all changes about cross-border monitoring of the Poiqu, which was named in a study before it did this twice.
This is a developing story. Casualty figures, the number of missing and the cause of the flood are all provisional and subject to revision. Reported death tolls vary substantially between outlets.
