🇪🇸 Leer este reportaje en español
Investigative report
Where could a disaster like Nepal–Tibet happen again?
On 26 August a mountain came down with no earthquake and no rain. This report searched the whole planet for inhabited valleys where the same ingredients line up, and measured them on the ground so that the people who live there can weigh the risk with data.
The facts: a signal that looked like an earthquake
At 08:37 local time on 26 August 2026, according to the first reports, seismic stations recorded what was read as a magnitude 4.4 earthquake on the border between Nepal and Tibet. Minutes later, the Lende Khola river was coming down as a wall of mud, ice and boulders. The flow entered the Trishuli, swept away the Rasuwagadhi border crossing and kept going downstream for almost 100 kilometres. In places the Trishuli rose nine metres in half an hour.
The United States Geological Survey later revised that signal: it was not an earthquake but the collapse of a slope with glacier involvement, releasing energy equivalent to a magnitude 5.2. About three hours later a second event was recorded, equivalent to a 4.2. Nepal’s meteorological service ruled out heavy rain in the area. Early analyses of satellite images suggest that part of the lower reach of a glacier on the north face of Langtang Lirung, at around 5,200 metres, broke away, dragging rock and sediment with it, and fell more than a kilometre to the valley floor.
That order of events is worth dwelling on, because it changes everything. The shaking the seismographs picked up was not the cause of the collapse: it was its consequence. Millions of tonnes of ice and rock hitting a valley floor make the ground tremble just as a moderate earthquake would, which is why the first reading was what it was. Once the review ruled out an earthquake as the origin, and the meteorological service ruled out rain, what remained was a mountain that fell on its own.
The exact cause is still under investigation and the death toll remained open as this edition closed. But the question that matters here is not how much ice fell. It is why it travelled so far, and where else it could happen.
Why a weather channel is getting involved
As long as the explanation was «an earthquake», this was a geology story and we had nothing to add. A mass of ice and rock that fails with no prior shaking and no rain is something else: it is a problem of temperature, of permafrost that stops holding the wall together, of meltwater working its way into the cracks. In other words, the same variables we follow every day on this channel, only five thousand metres up.
And we have two things that are genuinely useful here. The first is a way of thinking: chaining causes to effects, comparing cases with one another, and respecting a rule we impose on everything we broadcast: better an acknowledged gap than a false number. The second is the means. The terrain models, the satellites and the tools we use to make our maps are public, and they work exactly the same for measuring a Himalayan valley as for drawing tomorrow’s rain over Spain.
This is what we can do. And we are doing it.
What this report is looking for
The aim is concrete: to locate, anywhere on the planet, the places where the ingredients of Nepal–Tibet coincide and, on top of that, where people live within the area a flow like that would reach. The search is not for the biggest glacier or the one melting fastest. It is for the combination of an unstable mass, a path that carries it down, and a population at the end of that path.
The logic followed tries to be as objective as possible: the same rules for every system, the same terrain model to measure them, and three events that have already happened as the yardstick. And all of it is published, limits included, so that it can be checked and corrected.
The intention is neither alarm nor prediction. It is to help those who live in these valleys, and those who make decisions on their behalf, to weigh the risk with data that until now was scattered across dozens of studies. If this screening leads to a single one of these areas being looked at sooner and better, it will have done its job.
The ingredients of a cascade
Nepal–Tibet was neither simply melting nor simply a flood. It was a cascade: an unstable mass high up that falls, gains speed with the drop, enters a channel, picks up water and sediment, turns into a debris flow and finds a narrow canyon that lets it keep its mobility for tens of kilometres until it runs into people.
Chamoli, in India, showed the same chain in February 2021. Between 25 and 27 million cubic metres of rock and ice broke away near the 5,600-metre mark on Ronti Peak. On entering the Ronti Gad the flow picked up channel material, water and sediment, destroyed two hydroelectric plants and killed more than 200 people after a run of over 17 kilometres. Blatten, in the Swiss Alps, repeated the slope–glacier–village configuration in 2025, outside the Himalaya.
Three independent cases, three complete chains. That is what turns a high-mountain phenomenon into a land-planning problem.
What is being sought is not the biggest glacier or the one melting fastest: it is the place where an unstable mass high up has a gravitational highway to a population.
The search: one rule and one filter
With those three events as reference, a worldwide screening was defined. The rule is simple and deliberately restrictive: without a permanent population inside the plausible corridor, the system is left off the list. Alaska, Greenland or Patagonia hold enormous and potentially unstable masses of ice, but if the flow has no physical route to a settlement, they are not part of this report.
For each system six questions were reviewed, one per link: whether a significant mass can fail, how much gravitational energy is available, whether the fall path enters a drainage network, whether there is water, a lake or sediment to amplify the flow, whether the valley preserves mobility for kilometres and, finally, whether there is a permanent population in the way. With that criterion, the search concentrates on the Himalaya and the Karakoram, Peru’s Cordillera Blanca, Bhutan and a few Alpine valleys.
Five corridors that deserve immediate analysis
None of these systems is a warning. They are the places where the published evidence best supports every link in the chain, and therefore the first where it makes sense to invest in topography, satellite monitoring and modelling.
Sedongpu–Gyala Peri
Tibet, China. Yarlung Tsangpo corridor.
Here the chain has worked several times. On 16 October 2018 some 130 million cubic metres of debris-covered glacier broke away; the Yarlung Tsangpo was blocked for around 60 hours, the water level rose about 75 metres and at least 6,000 people were evacuated. In March 2021 another ice-and-rock collapse, of around 50 million cubic metres, again produced a mobile flow and a temporary blockage. An early-warning system was installed after these episodes.
The changes visible in Sentinel-1 and Sentinel-2 since 2021, and the exact location of settlements in each scenario. The drop and the valley confinement are measured below.
Badrinath–Mana / Alaknanda
Uttarakhand, India. Population in Badrinath, Mana and Hanuman Chatti.
An inventory published in 2026 identifies 219 hanging glaciers in the basin, with a mean surface slope of almost 34 degrees, in the same mountain system where Chamoli happened. Simulations show flow heights above 50 metres in the Badrinath–Mana reach, and the study foresees a sharp increase in built exposure and population over the coming years.
Working out glacier by glacier which ones discharge directly into inhabited channels, and reconstructing the full profile down to each of the three settlements.
Hualcán – Lake 513 – Carhuaz
Cordillera Blanca, Peru. Chucchún valley.
The chain has already happened here, which is why it is the best-documented case. In April 2010 a rock-and-ice avalanche of some 450,000 cubic metres fell from below the hanging glaciers of Hualcán into Lake 513, generated a 24-metre wave and an overflow that damaged bridges, homes, roads and land along some 20 kilometres down to Carhuaz. The lake had been artificially lowered years earlier; without those works the outcome would have been worse.
Updating the current geometry of the glaciers and slopes above the lake, and assessing how the lowering works change today’s scenario.
Palcacocha – Quillcay – Huaraz
Cordillera Blanca, Peru. City of Huaraz.
Lake Palcacocha holds around 17 million cubic metres of water, and the Quillcay basin has several potential sources of rock-and-ice avalanches. Models chain avalanche, wave, overflow and propagation down to a city of more than 120,000 inhabitants. The precedent is December 1941, when an outburst of the lake destroyed about a third of Huaraz and killed more than 1,800 people, in a city of fewer than 20,000 inhabitants.
Separating today’s hazard from historical memory, incorporating the existing mitigation measures and verifying the exposed population with 2026 mapping.
Lower Barun – Arun
Eastern Nepal. Settlements along the Barun and the Arun; Arun-3 hydroelectric plant.
This is the glacier–lake–channel case: a large lake, described as critical for its volume, with steep side walls prone to mass movements. A 2024 study places a potential avalanche zone next to the right shore and expects it to change as the lake grows. 2025 models show the impacts of an outburst reaching far downstream, as far as the Arun-3 hydropower works some 45 kilometres away, with between 700 and 2,000 buildings affected in the highest scenarios.
Explicitly linking the avalanche polygons with the wave and moraine-breach scenarios, and measuring the exposure of permanent population, not only of buildings and infrastructure.
What was measured for this report
Up to this point, the figures were the published ones. From here on they are this report’s own measurements. From each corridor’s list of «still to be measured», this report tackled the part that can be done with open data: rebuilding the eight systems, the five candidates and the three real events, on the same terrain model, Europe’s Copernicus at 30-metre resolution, and measuring them all with the same ruler. It is the first time these eight valleys can be put in a single table.
How to read the figures: the angle of reach
A flow of rock, ice and water is not measured by what falls, but by how far it travels relative to what it dropped. That is the angle of reach: the slope of the imaginary line joining the starting point with the end point. Two flows with the same drop can end up at very different distances; the one with the smaller angle is the one that travelled further. And the further a flow travels, the harder it is for a village downstream to be spared.
The profile of each valley
The profiles show what the map cannot: how much each corridor drops and how it does so. Sedongpu and Lower Barun start from sources above 6,500 metres and fall more than four vertical kilometres. Badrinath–Mana is the shortest: its source is little more than five kilometres from the village as the crow flies. And below, at the same scale, the three real events: Nepal–Tibet fell 5,478 m over 27.0 km of channel, the most brutal drop of the eight.
How far each one reaches
The three real events sit between 12.1° and 21.6°. That is the uncomfortable reading of this report: Palcacocha comes out at 7.0°, below all of them. For a flow from the head of the Quillcay to reach Huaraz it would need a reach greater than Chamoli’s… which is exactly what happened there in 1941. It is not that it is unlikely: it has already happened once, with a city far smaller than today’s.
At the other end, Badrinath–Mana (27.0°) is short and steep. No exceptional reach is needed to get to the bottom, and that means fewer minutes between the collapse and the village. Sedongpu and Hualcán, at 12.7° and 12.3°, sit almost exactly where Chamoli did: inside the range of what has already happened.
And the valley, which decides as much as the mountain
The last column of the table is the width of the valley halfway along the run. A narrow valley does not let the flow spread: it keeps it deep, fast and far-reaching. The two most confined corridors of the eight are Chamoli (387 m) and Palcacocha (503 m). At Chamoli, with an angle of only 12.1°, the corridor measured from the highest source adds up to more than thirty kilometres of channel, and the real 2021 flow ran more than 17; the geometry of the valley explains that distance better than the size of the collapse.
| Corridor | Sourceelevation | Settlementelevation | Dropvertical | Path lengthalong channel | Angle of reachlower = travels further | Valley widthmid-slope |
|---|---|---|---|---|---|---|
| SedongpuGyala | 7,130 m | 2,404 m | 4,726 m | 34.0 km | 12.7° | 844 m |
| Badrinath-ManaMana | 5,556 m | 2,934 m | 2,623 m | 19.0 km | 27.0° | 1,169 m |
| HualcánCarhuaz | 6,056 m | 2,581 m | 3,475 m | 24.5 km | 12.3° | 1,099 m |
| PalcacochaHuaraz | 6,054 m | 2,995 m | 3,058 m | 31.5 km | 7.0° | 503 m |
| Lower BarunArun valley | 6,534 m | 3,001 m | 3,533 m | 17.5 km | 16.4° | 1,013 m |
| Nepal–Tibet 2026Rasuwagadhi · already happened | 7,216 m | 1,738 m | 5,478 m | 27.0 km | 21.6° | 531 m |
| Chamoli 2021Raini · already happened | 6,079 m | 1,732 m | 4,347 m | 35.3 km | 12.1° | 387 m |
| Blatten 2025Blatten · already happened | 3,737 m | 1,460 m | 2,277 m | 11.3 km | 18.5° | 938 m |
Source = the highest point that drains towards the settlement (maximum scenario, not that of the specific event). Path measured along the steepest-descent line on the Copernicus GLO-30 terrain model. Valley width: median at 50 m above the channel in the middle reach.
How it was done, and what it is not
Everything comes from open data and can be redone. Each corridor’s path is traced by following the line of steepest descent from the source, after filling the depressions in the terrain model; without that step, the trace gets stuck in the first hollow. The source is taken as the highest point that drains towards the settlement: it is the maximum credible scenario, not that of the specific event, so the drops in the table are an upper bound and should be read as such. Each trace was checked in a simple way: if the traced channel did not pass through the village, the corridor was not accepted.
And what this is not: it is not a dynamic simulation. It does not compute velocities, wave heights or arrival times; those need flow models, calibrated parameters and a glaciologist behind them. It is geometry measured on real terrain, and it serves what it serves: ranking corridors and seeing which ones resemble what has already happened. None of these figures is a forecast, and none replaces the warning systems that already exist in several of these valleys.
What is known and what is not
The difference between an honest report and an alarmist headline lies in one detail: which part of the chain has been observed in a real event, which part has been modelled in published studies, and which part is still a hypothesis awaiting verification on the ground. The next figure summarises that reading for the five corridors.
What this work does not do matters too. It does not calculate an annual probability of collapse. It does not replace a dynamic flow simulation. It does not yet quantify the exposed population with a single data layer, so it compares valleys, not people. And it is neither a warning nor a timed prediction: none of these glaciers «is going to fall» within a timeframe anyone can set.
On climate, the cautious formulation is this: warming is changing the stability conditions of high mountains, with ice loss and thinning, permafrost degradation and more meltwater, and it can increase the susceptibility of particular systems. At Chamoli there were thermal anomalies, fractures and progressive deformation observable before the collapse. That does not mean every individual collapse can be attributed to climate change without a specific study.
The next phase is also measurable, and part of it is already done: the eight corridors now share a single terrain model. Still missing are deformation monitoring with satellite radar, a homogeneous population layer and a flow simulation for the three or four best-supported systems. And the most demanding validation test has been passed halfway: applied backwards, the method places Nepal–Tibet, Chamoli and Blatten within a narrow range, and the five candidates fall inside that range or below it. Had they all fallen far above it, the method would be wrong.
The right question
The question is not only how much ice is left in the mountains. It is what happens when a system that is losing stability is connected, through a complete physical chain, to a community. Nepal–Tibet answered that question on 26 August. The aim of this screening is not to have to wait for the next answer before starting to look. And if someone who lives in one of these valleys, or works to protect them, finds here a figure that helps, or one that is wrong, that is exactly the conversation we wanted to open.
Main sources
- U.S. Geological Survey (2026). 2026 Nepal Debris Avalanche and Flash Flood.
- Ahmed et al. (2026). Ice-rock avalanches in a warming Himalaya indicate pathways toward effective preparedness. Communications Earth & Environment.
- Büntgen et al. (2025). The 2025 Blatten disaster in the Swiss Alps followed exceptional warming. Communications Earth & Environment.
- Krishnan et al. (2026). Basin-scale inventory and exposure assessment of hanging glaciers, Central Himalaya. npj Natural Hazards.
- NHESS (2023). Early warning system for ice collapses and river blockages in the Sedongpu Valley. Natural Hazards and Earth System Sciences.
- Landslides (2023). Simulating glacier lake outburst floods, Lake 513 validation case. Landslides.
- Frey et al. (2018). Multi-Source Glacial Lake Outburst Flood Hazard Assessment and Mapping for Huaraz. Frontiers in Earth Science.
- Science of the Total Environment (2024). Evolution of the Lower Barun lake and its exposure to potential mass movement slopes. STOTEN.
- NHESS (2025). Assessing economic impacts of future GLOFs in Nepal’s Everest region. Natural Hazards and Earth System Sciences.