On the morning of 7 February 2021 a mass of rock and glacier ice broke off a steep face below Ronti Peak in Chamoli district, Uttarakhand, India, fell roughly 1,800 meters, and turned into a fast, dirty flow that ran down the Rishiganga and Dhauliganga rivers. It destroyed one operating hydropower plant and wrecked another under construction, and more than two hundred people were killed or never found, most of them construction workers. Twelve men came out of a tunnel alive on the first day, and everything after that was recovery, for reasons worth studying carefully.

What came off the mountain

The failure started high, on a steep north-facing wall in the Ronti Gad catchment on the fringe of the Nanda Devi massif, at an elevation reported in the range of about 5,500 meters. A wedge of bedrock with glacier ice attached to and above it detached and fell. The peer-reviewed study published in the journal Science in 2021, led by Dan Shugar with a large international author group, put the detached mass at roughly 27 million cubic meters, about four fifths rock and one fifth ice, and described a vertical drop on the order of 1,800 meters. Other published estimates of the volume differ because the methods differ, and anyone who needs a number for engineering purposes should read the primary literature rather than take mine.

The ice is what made this lethal far downstream. A rock avalanche of that size dissipates enormous energy on impact, and much of the ice in the mass was converted to water during the fall and at the point of impact. That water mixed with pulverized rock and with sediment already sitting in the valley floor, and the result stayed mobile for tens of kilometers instead of stopping at the base of the slope. Published reconstructions using regional seismic records place the detachment at around 10:20 in the morning local time, and the flow reached the project sites downstream within tens of minutes. Accounts differ on exact arrival times at each point, and the Science paper is where to go for the reconstructed timeline.

From the detachment zone the flow ran down Ronti Gad into the Rishiganga, past the village of Raini where the Rishiganga joins the Dhauliganga, and on to the Tapovan area, a path on the order of twenty-five to thirty kilometers. It attenuated as it went, and by the time the water reached the larger channels of the Alaknanda the surge was a river rise rather than a debris flow. The destruction was concentrated in the first thirty kilometers, which is where the two hydropower sites happened to be.

The “glacier burst” that was not a glacier burst

Within hours, national and international coverage described a glacier bursting, and many outlets called it a glacial lake outburst flood, or GLOF. That framing was wrong, and I want to label it as a myth in the same sentence in which I repeat it. The published scientific work found no lake at the source, with no moraine-dammed pond that failed and no supraglacial pond that drained, and the water came instead from ice inside the falling mass. A rock and ice avalanche that supplies its own water leaves a different footprint from a lake failure, and it calls for an entirely different monitoring strategy.

The confusion was understandable on day one. Uttarakhand has a real GLOF problem, state and national disaster authorities have published GLOF guidance, and a February flood in a glaciated headwater with no rain and no snowmelt to explain it looks like a lake failure from a distance. What settled the question was imagery and seismic data, which arrived within days rather than hours and showed a fresh scar on a rock face rather than an emptied basin.

The cost of the wrong label is not academic. If you believe you are dealing with a lake, you send people to survey lakes and you build your watch list from lake inventories. Rock and ice avalanches from steep, ice-loaded bedrock walls do not appear on a lake inventory at all, and no inventory of such faces existed at the resolution needed to have flagged this one in advance.

Two power projects and the people underground

The first structure in the path was the Rishiganga small hydropower project near Raini, a run-of-river plant with a nameplate capacity commonly reported at 13.2 megawatts, which was operating at the time. It was essentially erased, along with workers who were on site. The flow then continued down the Dhauliganga to the Tapovan Vishnugad project of NTPC, a 520 megawatt scheme under construction, where it filled the barrage area, the desilting works, and the tunnel works with rock, sand, and slurry.

The casualty count comes down to where people were standing at 10:20 on a Sunday morning. A construction site of that size concentrates a large workforce in the valley bottom, at the river’s edge, and inside tunnels, which is exactly where the flow went. That is a statement about exposure rather than an argument that the projects caused the avalanche, and I want to keep those two ideas separate because they get blended constantly in coverage of this event. The mountain would have failed with or without the dams, and the number of people in the way was decided years earlier during siting and construction planning.

Residents at Raini had objected to the Rishiganga project in the years before, and there had been earlier flood damage at that site, all of which appears in press accounts and in Indian court filings. After the 2013 Kedarnath disaster, a court-appointed expert body examined the role of hydropower development in the Alaknanda and Bhagirathi basins and recommended against a number of projects. I am summarizing rather than quoting, and readers who want the actual findings should go to those proceedings and reports rather than rely on a paragraph from me.

Inside the head race tunnel

Two underground rescues happened at Tapovan, and only one of them worked. On the first day, personnel of the Indo-Tibetan Border Police reached a shorter tunnel where a group of workers was trapped behind debris, dug through, and brought twelve men out alive. Those men were close to the portal behind material that could actually be excavated, and the operation was finished in hours. That is the shape of a survivable tunnel entrapment.

The main head race tunnel was a different problem. It is a long, large-diameter bore that curves away from its mouth, and the flow drove a plug of saturated silt, sand, and gravel deep into it. Reports at the time gave varying figures for how far the material extended, and the figures changed as digging progressed, which is normal and is also a reason not to quote any single one as settled. The fine material behaved like wet cement that never sets, so crews would clear a length of tunnel and the slurry would slump back in and refill it, sometimes overnight. Water kept seeping through, and the tunnel’s own drainage and pumping had been destroyed with everything else at the portal.

What followed was weeks of mucking rather than rescue. Crews worked around the clock with heavy equipment inside a confined bore, which carries its own hazards, including engine exhaust in an unventilated space, equipment becoming stuck and blocking the only exit, and a second surge arriving from upstream. A vertical drilling effort was attempted to intersect the tunnel from above in the hope of reaching a void where survivors might have found air, and it did not produce survivors. Nobody came out of the head race tunnel alive after the first day, and the reported number of people believed to have been inside varied at the time, with figures generally given in the range of roughly twenty-five to thirty-five, which is itself a finding worth sitting with.

The limits of high altitude response

Joshimath and Tapovan sit at moderate elevation, around 1,900 meters, so this was not an extreme-altitude rescue in the physiological sense at the work sites. The altitude problem showed up in access and in aviation. The road up the Dhauliganga toward the Niti valley is a single-thread mountain road, and the flow destroyed the bridge at Raini, which cut off villages upstream and left them dependent on helicopters and a temporary footbridge for supply. Helicopter lift capacity falls off with altitude and temperature, February daylight in that valley is short, and mountain weather closes flying windows without asking.

The response strength at Chamoli was proximity. The Indo-Tibetan Border Police maintain a presence in the district because it is a border district, and their people were already acclimatized and close enough to be working at the tunnel the same day, alongside the National Disaster Response Force, the state disaster response force, the Army, the Air Force, state police, and the project owner’s own staff. In remote high terrain, the first effective responder is whoever is already stationed and already acclimatized, and no surge from the lowlands changes that in the first twelve hours.

The rescue also had an unmonitored hazard above it. Debris blocked the Rishiganga valley upstream and impounded water, and there was legitimate concern that the blockage could fail and send a second pulse onto the very site where hundreds of people were working in a hole in the ground. Teams surveyed the impoundment by helicopter, and operations at the tunnel were halted at least once when water levels rose. That is a hard incident command problem, because your only rescue objective sits in the inundation path of a hazard you cannot instrument or drain, and the available control amounts to a spotter, a siren, and a rehearsed evacuation of the work face.

The toll, and where accounts differ

The published figures do not agree, and the reason matters. The Science study led by Shugar states that more than two hundred people were killed or are missing. Figures compiled by Uttarakhand state authorities during 2021 settled at around 204 dead and missing, and various press accounts give 205 or 206. Rather than competing counts of bodies, these are counts of the dead plus the missing, assembled from employer rosters, village reports, and family claims, and the number of bodies actually recovered was substantially lower than the total, with many of the missing never found. If you need a defensible single figure, say that just over two hundred people died or remain missing, and name where the figure comes from.

Roster reconstruction is the underappreciated part of this. When a construction site with contract labor, subcontracted crews, and workers from other states is destroyed in fifteen minutes along with its offices, the list of who was present that morning has to be rebuilt afterward from payroll records, phone records, and relatives arriving to ask. That process takes weeks, it produces revisions, and the revisions get reported as confusion or cover-up when they are usually neither. Plan for it, and tell the public early that the number will move.

Damage figures for the two projects and the destroyed infrastructure have been published in a range of amounts by different sources and in different currencies, and I am not going to pick one. The physical record is clear enough without a rupee figure, with one operating plant destroyed, one large project under construction badly damaged and its tunnel works filled, a road bridge gone, villages isolated, and footbridges and irrigation works along the affected reach wrecked.

Watching a mountain nobody was watching

There was no instrumentation on the face that failed, no seismic station, no camera, and no radar aimed at that wall, and under the prevailing priorities there was no reason for there to have been, because Ronti Gad is uninhabited and the hazard inventories in use focused on glacial lakes. What did exist was satellite imagery, and the published work reports that a developing fracture on that face is visible in archive imagery from years before the collapse. The signal was in the record, and nobody was reading that particular part of the record in real time.

Researchers have discussed warming and permafrost degradation in steep bedrock as a plausible contributing factor in this class of failure, and the published work on Chamoli is careful about how far it goes in attributing this specific collapse to that mechanism. I will be careful too. Saying that high mountain rock walls are losing the ice that holds them together is well supported as a general trend, while saying that a specific wedge failed on a specific February morning because of it is a hypothesis, and the literature treats it as one.

The practical lesson is about screening rather than sensors. India has capable institutions in this space, from the Geological Survey of India to the Wadia Institute of Himalayan Geology, and the constraint is coverage of an enormous area rather than talent. Imagery-based screening of steep ice-loaded faces upstream of populated valleys and major infrastructure is cheap compared with instrumenting individual slopes, and it produces a prioritized short list on which instruments and cameras can then be justified. Anyone building such a program should get current methods and current watch lists from the responsible national and state agencies rather than from an article like this one.

What this means for planners elsewhere

Most readers of this site will never manage a Himalayan rock avalanche, and will at some point manage a confined space entrapment at an industrial site, a water treatment plant, a mine, or a utility tunnel. Chamoli is a hard lesson in the fact that underground rescue outcomes are largely determined before the incident. The twelve men who lived were close to a portal behind debris that could be dug, while the people deep in the head race tunnel were behind a plug of slurry that reflowed faster than it could be removed, and no amount of equipment or courage at the portal changes that arithmetic. Pre-incident engineering, meaning second egress, refuge chambers, a hardened communications path, and a real accountability system at the portal, is what produces survivors.

The second lesson concerns hazards that sit above your rescue. Any operation in a valley bottom, a trench, a tunnel, or a debris field can be inside the footprint of a follow-on event, and someone in the command structure has to own that specific question, with a named observer, a defined trigger, an audible signal, and a rehearsed movement to a defined safe point. Chamoli had that problem in the form of an impounded lake upstream, and the same logic applies to an unstable structure, an untested slope, or a tank that has not been isolated.

The third lesson is about how you talk while you are still ignorant. On day one the mechanism was misnamed nationally, and neither the number of people underground nor the eventual total of the missing was known, with the missing total revised for weeks afterward. All of that was unavoidable, and what is avoidable is stating any of it in the declarative voice. Describe what is known to have happened, say plainly that early figures and early explanations are preliminary and will change, and point people to the agency that will produce the record. That posture costs nothing, and it is the only one that survives the final report.

Takeaways

• The 7 February 2021 Chamoli disaster began as a rock and ice avalanche from a steep face below Ronti Peak, and the study published in Science in 2021 put the detached mass at roughly 27 million cubic meters, about four fifths rock and one fifth ice, with other published estimates differing.
• Widely repeated early descriptions of a glacier bursting or a glacial lake outburst flood were wrong, because the published work found no lake at the source and the water came from ice melted during the fall and impact.
• Twelve workers were rescued alive from a shorter tunnel at the Tapovan site on the first day, and no one was brought out of the long head race tunnel alive, because saturated silt reflowed into the bore faster than crews could clear it.
• The number of people believed trapped in the head race tunnel was reported at the time in a range of roughly twenty-five to thirty-five, which shows that portal accountability was itself a failure point.
• Death and missing totals differ by source, with the Science study saying more than two hundred killed or missing and Uttarakhand state figures compiled in 2021 giving around 204, while some press accounts give 205 or 206, and many of the missing were never recovered.
• Travel time from the failure to the two hydropower sites was on the order of tens of minutes, so near-field exposure could only have been reduced by siting and drills, while downstream districts had enough lead time for phone alerts and dam notifications that did work.
• No instrumentation was watching the face that failed, and a developing fracture visible in archive satellite imagery went unread in real time because the hazard inventories in use were built around glacial lakes rather than steep ice-loaded rock walls.
• Underground rescue outcomes are set by pre-incident engineering, meaning second egress, refuge chambers with stored air, a communications path that works in rock, and an accountability system that does not wash away with the site office.