Nepal's deluge began as a glacier falling a kilometre into a river it then dammed

The fall is the part with the clearest evidence. Dan Shugar at the University of Calgary describes a glacier that rolls over a cliff at about 5,200 metres and snapped along what he calls a perfect straight line, dropping more than a kilometre to the valley bottom ^[15]. Jakob Steiner, a geoscientist at the University of Graz, read the same imagery as a lower section cut off as if with a huge knife ^[33]. Two candidate triggers are on the table, and they sit at different depths. Shugar's is surface water: warm weather melted snow and ice, satellite images from consecutive days show nearby glaciers losing a fair bit of snow, and meltwater percolating to the bed could have lubricated it ^[16]. Alton Byers at the University of Colorado, Boulder points lower, to permafrost beneath the ice losing structural integrity as it thaws ^[21]. Neither hypothesis can be tested against a monitoring record, because the glacier that failed had not been heavily studied or closely watched ^[20].

Where the water came from matters as much as why the ice let go. Shugar's reading is that the potential energy of millions of tonnes of ice falling a kilometre, released as heat, melted enough of that ice to feed the flood ^[17]. That is the mechanism invoked for the Chamoli disaster in India in February 2021, which was smaller ^[18]. ICIMOD's own preliminary assessment found no anomalies on the Chinese side of the border and describes a rock-ice avalanche inside Nepal blocking the Lhende and driving a cascading flood ^[14].

The instrument that caught this was a seismometer. Kristen Cook at the University of Grenoble says the event appears clearly in regional records, including a Nepali national network station near Kathmandu, beginning with a high-amplitude signal carrying the character of mass wasting rather than an earthquake, then continuing for hours with the long signal typical of a large debris flow ^[19]. That is a physically distinct fingerprint, and it was read after the fact: the event was first listed as a magnitude 4.4 earthquake before analysis of long-period waves reassigned it ^[6]. Whether seismic detection becomes warning depends on how fast that classification runs and whether anyone downstream is on the other end of it.

The margins downstream are thin. The gauge rise at Galchhi works out to roughly 0.3 metres a minute ^[28]. Hatim Sharif, a hydrologist at the University of Texas San Antonio, says running or driving is useless against a flow he compares to liquid concrete, and that the only option is to climb at least 6 metres ^[11]. That is 3 metres below the peak rise recorded at that gauge ^[30], though the two numbers measure different things: one is river height far downstream, the other a rule of thumb about the depth of flow you are trying to get above. His proposed remedy is real-time river gauges buying 20 to 30 minutes ^[12], the same order of time as the rise itself.

The thing this does not tell you is how much of it was warming. Attribution requires simulations comparing today's fossil-fuel-warmed climate against the preindustrial one ^[25], and ICIMOD's Saswat Sanyal says the connection should not be drawn yet ^[26]. Shugar's position is narrower and defensible: steep glaciers break for reasons unrelated to climate, but the rising frequency of these failures points to heating as a contributor ^[27]. The exposure figures are firmer than the attribution. Around 78% of the region's glacier area sits between 4,500 and 6,000 metres and is highly exposed to warming, spread across more than 63,000 glaciers that feed at least ten major Asian river systems ^[24]. The global tally of people at risk from glacial lake floods runs to about 15 million ^[23], and that tally counts lakes that already exist.