At 8:37 on the morning of 26 August, a slab of glacier sheared off a mountainside near the Nepal–China border and fell into the valley below.
It landed in a small river called the Lhende Khola. It dammed it. Water gathered behind the debris for a while. Then the debris gave way.
The ground shook hard enough that the US Geological Survey logged an earthquake. It has since corrected the record. There was no earthquake, the USGS says. The falling mountain itself shook that hard.
As of 27 August, more than 160 people had been killed. Around 750 were still missing, and the figures were still moving.
Now notice what is missing from that sequence.
No rain. No river rising through the night. No gauge reading climbing steadily until somebody issued a bulletin. The people in that valley had no warning because there was nothing to warn them with.
Nepal knows which of its glacial lakes are dangerous. ICIMOD has classified 47 of them.
Watching all of them properly, with sensors, drainage works and alarms downstream, is costed in Nepal's own National Adaptation Plan at around $1 billion by 2030.
The largest commitment so far is a Green Climate Fund project announced in 2025, worth $36 million. It covers four lakes.
Four out of 47. That is the entire story of the 26th of August, expressed as a ratio. Not a failure of Nepali science, which identified these lakes years ago. A failure to fund what the science found.
This is where it stops being someone else's problem.
In October 2021, the Teesta flooded. Bangladesh shares river data with India under the Joint Rivers Commission, so our Flood Forecasting and Warning Centre should have seen it coming.
It did not. The data-sharing arrangement had closed for the season.
Google's flood system, running on satellite data, sent alerts to roughly 3,000 Bangladeshis that week. Our own forecasters had nothing to send. The account is set out in a 2026 study in Water Resources Research.
Both governments, that year, were willing to give their hydrological data to a private company. Neither was giving it to the other.
Closer than most people assume.
The Flood Forecasting and Warning Centre is good on the Ganges, the Brahmaputra and the Meghna. Genuinely good. The World Meteorological Organization has described its basin model as giving lead times of three to five days on the trunk rivers.
The trouble is that flash floods do not arrive down trunk rivers.
Take the Gomati. The same study records that our arrangement with India covers a gauge at Amarpur, while two more stations at Sonamura, less than two kilometres from our border, sit outside it. Those two are the last measuring points before the river becomes ours.
Or take the haor. Researchers report fewer than twelve working gauges across a basin that floods every year. During the 2017 flash flood, three of the nearest upstream gauges were not working at all. That finding is from a working paper and has not yet been peer reviewed, but it matches what practitioners in the region have long described.
None of this is expensive. A gauge is not a barrage. A data-sharing protocol costs nothing to build and only willingness to sign.
Because a warning that works produces nothing anyone can photograph.
A relief camp can be opened, visited, reported. A minister can stand in it. A gauge that prevents a flood produces a disaster that did not happen, and a disaster that did not happen appears in no ledger and wins no credit.
So the money goes where the credit is. Our own National Adaptation Plan, as costed by the government and reported by the Global Center on Adaptation, puts the full response at $230 billion by 2050, about $8.5 billion a year, most of it expected from outside. Against that, monitoring is a rounding error. It is still the line that does not get funded.
The bill arrives anyway, in a different column. Earlier this year, flash floods and upstream inflows drowned tens of thousands of hectares of boro paddy in the haor during harvest, tightening rice supply in a season the country could not afford to lose.
We paid for the missing gauges in rice.
1. Add the border gauges, and stop closing the arrangement in October. The two stations at Sonamura are the clearest example in the country of a fixable blind spot. Adding them is a signature, not a budget.
2. Gauge the haor properly. Twelve working stations for a basin that floods annually, with three dead when it mattered most, is not a resource problem. It is a priority nobody has set.
3. Fund monitoring as infrastructure. Gauges installed under a five-year project stop working in year six. Observation networks need recurring budget lines, the way roads need maintenance.
4. Get Bangladesh into the mountain conversation. We are downstream of Himalayan hazards we do not observe and are not consulted about. A seat at that table costs a fraction of one flood.
No gauge in Bangladesh would have seen that glacier fall. Nothing we build here would have saved a single life in Rasuwa.
But the water in our rivers is decided in places we do not watch, on tributaries nobody thought worth measuring, under an agreement that lapses in the autumn.
Nepal costed the job of looking upstream at $1 billion. Its largest funded project so far covers four lakes out of 47. Our own bill will be larger.
We can decide to pay it. Or we can keep paying in rice.
Sources: US Geological Survey; ICIMOD; UN University; Water Resources Research (2026); arXiv (2026); Global Center on Adaptation; UNDP; The Business Standard. Reporting as of 27 August 2026.
Author: Afifa Mahjabeen, an Associate in the Business Development and Communications team at Innovision Consulting
.png)
