Lakes, slopes and faults: the mountain frontier

Lakes that grow and burst

A glacial lake can empty in hours when the moraine that holds it fails or a slope falls into it. In October 2023 that happened at South Lhonak Lake in north Sikkim. A multinational study found that up to 14.7 million m³ of frozen moraine collapsed into the lake and raised a wave about 20 m high; the flood breached the Teesta III dam at Chungthang and left 55 people dead and 70 missing (ICIMOD, 4 Feb 2025, summarising the study in Science).

In Nepal and the catchments that cross into it, there are more of these lakes, covering more ground. The Glacial Lake Observatory mapped 4,150 lakes of at least 0.001 km² in Nepal and the catchments that cross into it, using Sentinel-2 images from 2017 to 2024, and found that both their number and their total area grew over those eight years (Rawlins et al., Earth System Science Data, 21 Jul 2026).

This year showed how little warning there can be downstream. On 26 August a flood came down the Bhote Koshi into Rasuwa district in Nepal at about 9 am, while it was not raining at the district headquarters. Satellite images showed heavy rain on the Tibetan side, experts suspected a lake outburst there, and power production stopped at Rasuwagadhi (111 MW), Chilime and Trishuli-3A (The Week, 26 Aug 2026). On 11 October, 46 days later, the Budhi Gandaki flooded in Gorkha district with no reports of heavy rain upstream. An outburst was suspected; the head of Nepal's disaster authority suggested an avalanche had sent snow and debris into the river instead (Annapurna Express, 11 Oct 2026). Both rivers feed the Gandaki system, which crosses into India as the Gandak (Wikipedia, a secondary source).

India already watches part of this. The National Remote Sensing Centre has mapped more than 7,500 of about 28,000 glacial lakes, the Central Water Commission issues a fortnightly bulletin on high-risk lakes from June to October, and NDMA lists 195 high-risk lakes (PIB, 4 Dec 2024). NDMA runs a ₹150 crore programme to reduce outburst risk in four states (NDMA reply to the National Green Tribunal).

What is thin is the off-season and the far side of the border. The bulletin stops in October, cloud hides lakes from optical satellites for much of the monsoon, and some of the lakes above Indian valleys lie in Tibet or Nepal. Sarovara follows lake area in Sentinel-2 images through the year, flags lakes that are growing, raises a lake's level when an earthquake or heavy rain comes near it, and estimates how long a flood wave would take to reach the first town downstream, calibrated on the 2023 Teesta flood. It shows lakes across the border as hazards only. It does not detect or forecast an outburst: it says where to look, and the agencies decide.

Slopes and the only road in

The mountain monsoon kills on the roads as well as in the rivers. Of the 267 deaths that Himachal Pradesh's State Emergency Operations Centre counted from 30 June to 4 September 2026, 159 were in road accidents, and 92 roads were shut on 5 September (Rozana Spokesman, 5 Sep 2026).

Warnings for slopes are thinnest where the frontier is. The Geological Survey of India issues landslide bulletins for 21 districts, but they are operational only in Darjeeling, Kalimpong and the Nilgiris. Ladakh, Jammu and Kashmir and Arunachal Pradesh get none, although 21%, 12% and 15% of their mapped terrain is rated highly susceptible (PIB, 20 Aug 2025).

Marga turns public forecasts into a 72-hour outlook, in six-hour blocks, for eight corridors and 70 points along them: rain that can trigger landslides, snow, ice, fog and storms. Beside each road it lists the official alerts that concern it, exactly as issued. By late October the hazard has moved from rain to fog and ice. In the outlook issued at 14:30 IST on 21 October, four corridors were at warning, all for dense fog, three were at watch for ice or fog, and one was normal.

Source: Marga board, issued 21 Oct 2026, 14:30 IST, from Open-Meteo forecasts. Marga reissues the board with every forecast run, so the linked file may hold a newer outlook than this table. Eleven official alerts concerned these corridors in the week before; none was active (alerts.json). The thresholds have not yet been checked against recorded closures, so the levels are advisory. Public roads only, never convoy timings. For road status follow the Border Roads Organisation, state public works departments, NHIDCL and district administrations.

Faults with no warning

The government told Parliament this year that "no proven system exists in the country to provide early earthquake warning" (PIB, 1 Apr 2026). The Rajya Sabha committee on home affairs has asked for denser seismic networks and for communication facilities to be retrofitted (Report 261, 7 Aug 2026, paragraphs 4.4.12 to 4.4.14).

The ground moves often. From 1 January 2016 to 21 October 2026 the USGS catalogue lists 2,470 earthquakes of magnitude 4.5 or more in the area Kampana studies, from 5° to 38° N and 66° to 98° E, which covers India and its neighbours. 552 reached magnitude 5, about 51 a year; 30 reached 6; two reached 7, near Tingri in Tibet on 7 January 2025 (M7.1) and near Mandalay in Myanmar on 28 March 2025 (M7.7).

Andaman arc148

Hindu Kush108

Tibetan Plateau84

Northeast65

Central Himalaya48

W Himalaya31

Myanmar belt29

Sulaiman, Kirthar22

E Himalaya11

Plains and seas5

Kutch1

In the 90 days to 21 October the USGS catalogue has 138 earthquakes of magnitude 3.5 or more in the same area, 15 of them magnitude 5 or more. For each one, Kampana estimates shaking at a fixed list of public reference towns, and for events near India at mapped public infrastructure too, and raises the event to watch when any estimate reaches intensity V, where very light damage becomes possible. Three reached watch: a magnitude 5.1 on 6 October 31 km northwest of Bageshwar in Uttarakhand, a 4.6 on 5 October 20 km east of Nagaon in Assam, and a 4.3 on 9 August 19 km northwest of Uttarkashi. For the Bageshwar event Kampana estimated intensity IV at the nearest reference town, 31 km away, with a range of III to V, and shaking that would be felt, intensity III or more, at 20 reference towns; the USGS received 50 "Did You Feel It?" reports (USGS event page).

How good are the estimates? For four recent events with a USGS ShakeMap, Kampana's estimates at 119 towns came within one intensity unit of ShakeMap 96% of the time, with a mean difference of −0.1 and a root-mean-square difference of 0.5. That is close enough for triage, deciding which bridges, roads and lakes to inspect first, and not for anything more. Kampana uses the intensity equation of Allen, Wald and Worden (2012) for an average site, so soft ground and steep slopes can shake harder than it says. It covers public infrastructure only; military and armed-police sites are excluded before anything is stored.

What would help

• Lakes watched through the year, including those across the border, so that growth is noticed before the monsoon rather than after a flood. That is Sarovara.
• Road outlooks that change with the season, rain in the monsoon and fog and ice in winter, shown beside the official alert for the same road. That is Marga, with official alerts relayed unchanged by Sandesha.
• Triage within minutes of an earthquake: which towns, bridges, roads and lakes were inside the estimated shaking, ranked for inspection. That is Kampana.

Each of these is advisory. None replaces the Central Water Commission, NDMA, the Geological Survey of India, the National Center for Seismology, the India Meteorological Department, the Border Roads Organisation or a state disaster management authority.