Himalayan water towers: snow, dams, and disasters in Asia's next decade

Asia's great rivers are usually discussed as flows: how much water passes a gauge, and who may take it. The more useful frame is storage. The Hindu Kush Himalaya spans eight countries, contains more than 60,000 glaciers holding nearly 6,000 cubic kilometres of ice, and is the source of more than ten major river systems that sustain nearly two billion people (Down To Earth 2026). Snow and ice act as a seasonal reservoir that releases water when rain does not, and the whole system of irrigation, hydropower, and urban supply downstream was built on the assumption that this reservoir refills each winter.

Three forces are now altering that assumption at once. The reservoir itself is thinning and arriving at different times. Upstream states are adding their own storage, and with it the ability to hold or release water. And the mountains are generating hazards that destroy the infrastructure meant to manage water. Each force is documented separately, but their combination is what shapes the next decade, because a downstream state facing less snow, a more assertive upstream neighbour, and a flood that removes its hydropower faces all three together.

Dependence is uneven. ICIMOD, the intergovernmental centre for the mountain region, estimates that seasonal snowmelt supplies about a quarter of total annual runoff from these rivers on average, but about 40 percent for the Indus, 74 percent for the Amu Darya, and 77 percent for the Helmand (Mongabay India 2025). The west, where the monsoon is weak, depends most heavily on snow. The east depends on rain but not on dams it does not control. The risk therefore takes different forms across basins, and a single regional narrative misses the specifics.

The latest ICIMOD assessment finds that snow persistence, the fraction of time snow stays on the ground after snowfall, was 27 percent below the long-term average across the region in winter 2025–26, the fourth consecutive year of below-average snow and worse than the previous record deficit of 23.6 percent. Only two of the 12 major basins analysed, the Ganga and the Irrawaddy, recorded above-average levels. The steepest declines were in the Mekong at 59.5 percent, the Tibetan Plateau at 47.4 percent, and the Salween at 41.8 percent (Down To Earth 2026). ICIMOD warns of reduced runoff, increased groundwater extraction, and higher drought risk, though the final severity depends on spring and early summer conditions.

The longer trajectory is also adverse. ICIMOD's 2023 assessment found that glaciers in the region could lose up to 80 percent of their current volume by the end of the century on high-emissions pathways, that the rate of ice loss has doubled since 2000, and that the changes are largely irreversible (ICIMOD 2023). A glacier that melts faster first raises river flows and then, as it shrinks, reduces them, so many basins face a period of apparent abundance followed by decline, and the timing of the turn varies by basin and elevation.

The immediate effect is on timing and reliability rather than on average volume. A season with little snow reduces early-season melt, which is when irrigated agriculture and run-of-river hydropower depend on it. Downstream states respond with groundwater, which is a one-way draw on a slower-recharging store. The practical consequence is less buffer in dry years, and it falls first on the western basins where snow supplies the most runoff and institutions are weakest.

On 26 August 2026 at about 8:37 in the morning, a mass of ice and rock roughly 600 metres across broke from near Langtang Lirung and fell about 5,600 metres into the Lende Khola, a tributary of the Bhote Koshi. Early reports suspected a glacial lake outburst, but seismic analysis and later accounts describe an ice-and-rock avalanche that cascaded into debris flows and temporary river dams, not the failure of a single lake (Nepal News 2026). The flood reached communities in minutes. By mid-September Nepal's disaster authority reported about 1,399 deaths and more than 5,000 people missing, with a further 43 deaths and 519 missing on the Tibetan side and 261 foreign nationals on China's missing list. Missing-person figures differ between police and disaster-authority lists.

The damage fell on the infrastructure that manages water and energy. A rapid assessment puts physical damage and economic losses at about 408 billion Nepali rupees, around 2.7 billion dollars, and full recovery at about 723 billion rupees, roughly 4.7 to 4.8 billion dollars. Thirteen hydropower projects totalling about 759 megawatts were damaged, including Upper Trishuli 3A and 3B and Chilime, with direct damage of about 889 million dollars and an annual generation loss of about 2,520 gigawatt-hours. Thirty-seven motorable and 68 suspension bridges were damaged, and the Rasuwagadhi crossing to Tibet, a key trade and pilgrimage route, has now been disrupted twice in just over a year (Nepal News 2026).

The warning system failed in the way a cascade defeats it. Flood-forecasting officials received information about 30 minutes after the collapse and sent text messages to six districts within about 40 minutes, reaching more than 600,000 phones, but monitoring stations at Syaphru Besi and Betrawati stopped transmitting almost immediately, and communities in upper Rasuwa had almost no time to react. A new lake forming on the China border continues to rise, and Nepal's disaster authority has warned that it could burst (Al Jazeera 2026). The event shows that the hazard is not a single flood but a chain, and that hydropower built for water management sits in the path.

On the Yarlung Tsangpo, which becomes the Brahmaputra, construction of the Medog hydropower project began on 19 July 2025 under a new central state-owned company, China Yajiang Group, with a planned capacity of about 60 gigawatts, a reported investment of roughly 1 to 1.2 trillion yuan, and commercial operation planned for 2033 (Diplomat 2026). China says the dam will not adversely affect downstream regions. India's response is the Siang Upper Multipurpose Project, planned at about 11,000 megawatts and presented by Arunachal Pradesh's government as a cushion against diversion or sudden release upstream, though local opposition has stalled pre-feasibility surveys (The Citizen 2026).

On the Indus, India placed the 1960 treaty in abeyance after the April 2025 Pahalgam attack, and it remains so. Flood information that once passed through the treaty's commission now travels through diplomatic channels: in August 2025 India alerted Pakistan to a high flood on the Tawi, a Chenab tributary, on what an Indian official called humanitarian grounds, and Pakistan objected that the warning should have come through the Indus Waters Commission (The Wire 2025). Indian authorities announced reservoir silt flushing at Salal on the Chenab for 21 to 30 May 2026, and Pakistan's Punjab authority warned that water levels could rise abruptly by two to three metres (Arab News 2026). Given that the Indus draws about 40 percent of its runoff from snowmelt, a snow drought raises the stakes of every dispute about timing.

Data is the quieter lever. China has shared mainstream Mekong flow data with downstream states year-round since 2020, but does not routinely share sediment data (RFA 2020). It supplies flood-season data on the Brahmaputra and Sutlej to India under memoranda renewed in 2018 and 2015, with an earlier interruption after the 2017 Doklam standoff (IPCS n.d.). I could not confirm the current status of those memoranda. In Central Asia, Afghanistan's Qosh Tepa canal, about 290 kilometres long and intended to irrigate around 550,000 hectares, could take up to roughly a third of the Amu Darya's flow by 2028, with no water-sharing agreement with Tajikistan, Uzbekistan, or Turkmenistan (Carnegie 2025; Asia Plus 2026).

The Nepal disaster exposed the weakness that connects the other threats: limited data. Monitoring stations that went silent during the event left officials without real-time information, forecasting relied on telephone and text alerts that arrived too late for the upper valley, and agencies compiled casualty and missing lists from different sources without reconciling them. Police data showed about 4,077 missing, the disaster authority 5,132, and a Prime Minister's Office portal 12,641, a spread that illustrates how hard it is to manage a response when the baseline is uncertain (Nepal News 2026).

The same gap appears across borders. A downstream state cannot plan on an upstream state's reservoir operations if it receives data only seasonally, if the data-sharing agreement lapses with a political dispute, or if the exchange is conducted through diplomatic notes in place of engineers. Hazards ignore borders: the Nepal flood originated near the Tibetan frontier and destroyed the Chinese-funded Syaphrubesi–Rasuwagadhi road and the Gyirong crossing, and Nepal has asked China to help rebuild, to which Beijing responded positively in principle without a signed agreement (Nepal News 2026).

Closing the gap is cheap relative to the losses. Monitoring that survives a cascade needs redundancy: stations placed beyond the likely flow path, satellite and radar observation of unstable slopes and lakes, and automatic sirens in the valleys where minutes decide outcomes. Real-time data exchange across borders needs standing arrangements with engineers on both sides, funded and protected from political disputes, rather than seasonal memoranda that lapse when relations sour. The 4.7 to 4.8 billion dollar recovery bill for a single event is a benchmark against which the cost of such systems can be judged.

The strongest objection is that the alarm is overstated. Snowmelt supplies a quarter of runoff on average, and the monsoon provides most of the rest. One year of deficit is not a trend, as ICIMOD itself notes that severity depends on spring conditions, and the Ganga basin was above average. Upstream dams can also buffer: Medog and Siang could in principle regulate flows and reduce flood peaks as well as divert them, and China points to the Mekong data it has shared. Treaties have survived wars before, and the Indus arrangement lasted more than 60 years. Mountain hazards are rare events whose frequency is hard to separate from natural variability. These points are valid, but they describe the average. Water security is decided in dry years and flood years, and in those years thin data and weak institutions matter most.

Three paths follow. In managed interdependence, upstream states share data in real time, a successor arrangement restores a role for treaty institutions, and regional early-warning systems cover high-risk valleys. In contested infrastructure, Medog, Siang, Qosh Tepa, and the Chenab projects proceed without binding limits, and each state treats the next dry season as a bargaining moment. In shock-driven adaptation, repeated hazard cascades, drought years, and a loss of hydropower force reconstruction and investment under duress. The signposts are observable: the next ICIMOD snow assessments, the status of the Indus treaty and arbitration, Medog construction milestones and any data-sharing agreement, the new lake on the Nepal–China border, the Mekong and Brahmaputra data arrangements, and whether Qosh Tepa begins diverting water by 2028.

The comparative record supports a selective conclusion. Exposure is highest where a basin depends heavily on snowmelt, sits downstream of a larger state's dams, lacks a binding agreement or real-time data, and has hydropower or roads in hazard paths. The Indus, the Amu Darya, and the eastern Himalayan valleys of Nepal fit that profile. Where rain provides most of the flow or where treaties and data exchange are functioning, the risk is lower and more manageable.

For governments, the priorities are investment in monitoring and warning systems that survive a hazard, data-sharing arrangements that do not lapse with politics, and siting rules that keep hydropower and bridges out of known hazard paths. For developers and lenders, hydropower in glaciated headwaters should be underwritten against cascading hazards and reduced dry-season flows, not historical averages. For downstream communities and firms, the prudent assumption is less buffer and less notice, which favours storage, efficiency, and diversification away from single sources.

The mountains are still the water towers of Asia. What has changed is that the towers are thinning, the valves are being installed by others, and the infrastructure below them can be swept away. Water security in the next decade will depend less on how much falls in the mountains than on who can see it, manage it, and withstand what comes down.