Nine metres, thirty minutes

The Lhende is a river most people reading this have never heard of. It comes off high ground north of the Nepal border, runs south, and joins the system that becomes the Trishuli at Rasuwagadhi, where the road crosses into China.

On the morning of 26 August 2026 something failed up there. The current reading is a snow and rock collapse into a glacial system roughly twenty kilometres north of the border crossing. What came down the valley afterward struck settlements along about seventy-two kilometres of river. The Trishuli rose about nine metres in around thirty minutes.

Nine metres in thirty minutes is not a flood you evacuate from. It is a flood you are already inside.

The collapse registered on seismometers. Networks that exist to detect earthquakes picked up the initiating event, logged it first as roughly magnitude 4.4 and later revised it upward toward 5.2, and that signal travelled at the speed of seismic waves through rock, which is to say almost instantly, to institutions on several continents.

The water travelled at the speed of water down a steep valley. Which means that for a window measured in minutes, the fact that something catastrophic had just happened above Rasuwagadhi existed as a data point in seismology networks worldwide, and did not exist in the valley it was about to reach.

This is not an unstudied hazard

That is the part that should be uncomfortable. Glacial lake outburst floods in this range are among the best documented hazards in mountain science. ICIMOD and its collaborators maintain HMAGLOFDB, a version-controlled database of outburst floods across High Mountain Asia. Version 1.0 catalogues 682 events, compiled through 30 June 2022, some of them going back the better part of two centuries.

Six hundred and eighty-two. That is not an emerging risk. That is a pattern with a long, carefully assembled paper trail, maintained by people who have been saying clearly and publicly for years that the pattern is accelerating.

What the record actually shows

Read the events above in order and a shape appears that no single one of them has on its own.

The same corridors keep failing. The Bhote Koshi and Poiqu system appears in 1981, again in 2016, and the 2026 flood came down adjacent terrain. The Khumbu appears in 1985, 1998, and 2024. These are not scattered accidents across an unpredictable range. They are a small number of valleys that have demonstrated repeatedly what they do.

Cascades are becoming the normal case. At Thame in 2024 one lake failed and brought the lake below it down as well. At South Lhonak in 2023 a moraine slab collapsed into the water, raised a wave of around twenty metres, overtopped the dam and sent a flood 385 kilometres down the Teesta into Bangladesh. At Chamoli in 2021 there was no lake failure at all: a rock and ice avalanche behaved like an outburst flood the whole way down. Modelling a single lake and asking whether it will breach is increasingly the wrong question.

The floods do not respect the border, and neither does the risk. 1981, 2016 and 2026 all involved water that formed on one side of the China and Nepal boundary and did its damage on the other. The instruments that see the source and the agencies that must warn the exposed are, in several of these cases, in different countries.

The gap is not measurement

It is worth saying plainly what would and would not have helped on 26 August.

Better satellites would not have helped. The satellites were adequate. Sentinel-1 passes this corridor on a fixed schedule and returns usable radar through the monsoon cloud that makes optical imagery worthless there for months at a time. More research would not have helped either. The research exists, it is good, and it named this hazard class years ago.

What was missing was the last link. A record of which valleys have done this before, held by somebody whose job is that valley. A route from a seismic detection to a district chapter in under an hour. A map of which settlements sit inside the reach, and which health facilities are cut off when the bridges go, held before the event rather than assembled after it.

Every one of those is a delivery problem wearing the costume of a data problem.

What we are doing, and what we are not

Caelus is not on this response. It is worth being blunt about that, because there is a version of this note that implies otherwise by omission and it would be an easy one to write.

In the first week of September we contacted the organisations working Nepal's flood response, offering settlement and health-facility isolation mapping for Rasuwa and upper Nuwakot. Most of those messages bounced or went unanswered, which is what happens when you write to people during the two weeks they are busiest. HOTOSM replied, and the reply was honest: it is late for this activation, and they would welcome support on future ones. They also asked one specific technical question about health facility layers in their datagrid. That question is the useful thing to come out of the whole exchange, and answering it well is worth more than a mapping offer nobody had bandwidth to accept.

So what is actually underway is the analysis. Flood events recorded within ten kilometres of a glacier since 2000, matched against ICIMOD's outburst record. The post-flood corridor read for building destruction, population clustering, and damage to the roads that decide who can be reached. Not because the Trishuli needs another retrospective, but because the next valley does not have one yet, and the point of a historical record is to be finished before the event rather than after it.