A massive rock and ice avalanche caused the 2021 disaster at Chamoli, Indian Himalaya

7 February 2021 hazard cascade

At 4:51 UTC [10:21 Indian Standard Time (IST)], about 26.9 × 10⁶ m³ (95% confidence interval: 26.5 × 10⁶ to 27.3 × 10⁶ m³) of rock and ice (Figs. 1 and 2) detached from the steep north face of Ronti Peak at an elevation of about 5500 m above sea level and impacted the Ronti Gad (“gad” means rivulet) valley floor about 1800 m below. We estimated the onset of this avalanche and its velocity by analyzing seismic data from two distant stations, 160 and 174 km southeast of the source (fig. S6) [(22), section 5.1]. The initial failure happened between 4:51:13 and 4:51:21 UTC, according to a source-sensor wave travel-time correction. We attributed a high-frequency signal 55 to 58 s later to the impact of the avalanche on the valley bottom, indicating a mean speed of the rock and ice avalanche of between 57 and 60 ms⁻¹ (205 to 216 km hour⁻¹) down the ~35° steep mountain face.

Differencing of high-resolution digital elevation models (DEMs) revealed a failure scar that has a vertical difference of up to 180 m and a slope-normal thickness of ~80 m on average, and a slab width of up to ~550 m, including both bedrock and overlying glacier ice (Fig. 2). The lowermost part of the larger eastern glacier is still in place and was not eroded by the rock and ice avalanche moving over it (Fig. 1D), suggesting that the avalanche may have become airborne for a short period during its initial descent. Optical feature tracking detected movement of the failed rock block as early as 2016, with the largest displacement in the summer months of 2017 and 2018 (fig. S4). This movement opened a fracture up to 80 m wide in the glacier and into the underlying bedrock (Fig. 1 and fig. S5). Geodetic analysis and glacier thickness inversions indicate that the collapsed mass comprised ~80% rock and ~20% glacier ice by volume (fig. S10) [(22), section 5.2]. Melt of this ice was essential to the downstream evolution of the flow, because water transformed the rock and ice avalanche into a highly mobile debris flow (23, 24). Media reports (25) suggest that some ice blocks (diameter <1 m) were found in tunnels at the Tapovan Vishnugad hydropower site (hereafter referred to as the Tapovan project), and some videos of the debris flow [(22), section 5.3] show floating blocks that we interpret as ice, indicating that some of the ice survived at considerable distance downstream. In contrast to most previously documented rock avalanches, very little debris is preserved at the base of the failed slope. This is likely due to the large volumes of water [(22), section 5.5] that resulted in a high mobility of the flow.

Geomorphic mapping based on very-high-resolution satellite images (table S2) acquired during and immediately after the event provides evidence of the flow evolution. We detected four components of the catastrophic mass flow, beginning with the main rock and ice avalanche from Ronti Peak described above (component one).

The second component is “splash deposits” (26–28), which are relatively fine-grained, wet sediments that became airborne as the mass flow ran up adjacent slopes. For example, the rock and ice avalanche traveled up a steep slope on the east side of the valley opposite the source zone, and some material became airborne, being deposited at a height of about 120 m above the valley floor. These deposits include boulders of up to ~8 m (a axis length). The bulk of the flow then traveled back to the proximal (west) side of the valley and rode up a ridge ~220 m above the valley floor, before becoming airborne and splashing into a smaller valley to the west (Fig. 2C and figs. S15 and S18). Boulders of up to 13 m (a axis length) were deposited near the top of the ridge. Vegetation remained intact on the lee side of some ridges that were overrun by the splashing mass.

A third component of the mass flow is reflected in airborne dust deposition. A dust cloud is visible in PlanetScope imagery from 5:01 UTC and 5:28 UTC 7 February (10:31 and 10:58 IST). A smooth layer of debris, estimated from satellite imagery to be only a few centimeters in thickness, was deposited higher than the splash deposits, up to ~500 m above the valley floor, although the boundary between the airborne dust deposition and other mass flow deposits is indistinct in places. Signs of the largely airborne splash and dust components can be observed ~3.5 km downstream of the valley impact site. The avalanche also generated a powerful air blast (1) that flattened about 0.2 km² of forest on the west side of the Ronti Gad valley (Fig. 2C).

After the rock and ice avalanche impacted the valley floor, most of it moved downvalley in a northwesterly direction. Frictional heating of the ice in the avalanche generated liquid water that allowed the transition in flow characteristics, becoming more fluid downvalley and creating a flow consisting of sediment, water, and blocks of ice. The uppermost part of the valley floor deposits is around 0.75 × 10⁶ m³, with few of the large boulders that typically form the upper surface of rock avalanches [for example, (29, 30)] (Fig. 2G and fig. S16). The mass flow traveled downvalley and superelevated (runup elevation) up to ~130 m above the valley floor around bends (fig. S17). Clear trimlines, at some places at multiple levels, are evident along much of the flow path (Fig. 2, C and D).

At the confluence of the Ronti Gad and Rishiganga River, a ~40-m-thick deposit of debris blocked the Rishiganga valley (Fig. 2, H and I). Deposition in this area probably resulted from deceleration of the mass flow at a sharp turn to the west. During the days after the event, a lake ~700 m long formed behind these deposits in the Rishiganga valley upstream of its confluence with Ronti Gad. The lake was still present 2 months later and had grown since the initial formation. Substantial deposition occurred about 1 km downstream of the confluence, where material up to ~100 m thick was deposited on the valley floor (Fig. 2J). DEM differencing shows that the total deposit volume at the Ronti Gad–Rishiganga River confluence and just downstream was ~8 × 10⁶ m³. These large sediment deposits likely indicate the location where the flow transitioned to a debris flow (31)—the fourth component.

A field reconnaissance by coauthors from the Wadia Institute of Himalayan Geology indicates that the impact of debris flow material (sediment, water, ice, and woody debris) at the confluence of Rishiganga River with Dhauliganga River created a bottleneck and forced some material 150 to 200 m up the Dhauliganga (fig. S15). The release of the water a few minutes later led to the destruction of a temple on the north bank of the Dhauliganga.

A substantial fraction of the fine-grained material involved in the event was transported far downstream. This more dilute flow could be considered a fifth component. Approximately 24 hours after the initial landslide, the sediment plume was visible in PlanetScope and Sentinel-2 imagery in the hydropower project’s reservoir on the Alaknanda River at Srinagar, about 150 km downstream from the source. About 2½ weeks later, increased turbidity was observed at Kanpur on the Ganges River, ~900 km from the source. An official of the Delhi water quality board reported that 8 days after the Chamoli disaster, a chief water source for the city—a canal that draws directly from the Ganga River—had an unprecedented spike in suspended sediment (turbidity) 80 times the permissible level (32). The amount of corresponding sedimentation in hydropower reservoirs and rivers is unknown, but possibly substantial, and may contribute to increased erosion of turbine blades and infilling of reservoirs in the years to come.

Analysis of eyewitness videos permitted estimation of the propagation of the flow front below the Ronti Gad–Rishiganga River confluence (Fig. 3) [(22), section 5.3]. The maximum frontal velocity reconstructed from these videos is ~25 m s⁻¹ near the Rishiganga hydropower project (fig. S11 and table S5), which is about 15 km downstream of the rock and ice avalanche source. Just upstream of the Tapovan project (another ~10 km downriver), the velocity decreased to ~16 m s⁻¹, and just downstream of Tapovan (26 km from source), the velocity was ~12 m s⁻¹. The large reduction in frontal velocity is likely related to impoundment behind the Tapovan project dam. Analysis of PlanetScope images (at 5:01 UTC and 5:28 UTC) suggests that the average frontal velocity between Raini (at Rishiganga hydropower project) and Joshimath (16 km downstream) was ~10 m s⁻¹. We also estimated mean discharge from the videos to be between ~8200 and ~14,200 m³ s⁻¹ at the Rishiganga hydropower project and between ~2900 and ~4900 m³ s⁻¹ downstream of the Tapovan project. Estimates for the debris flow duration are complicated by uncertain volumes, water contents, discharge amounts, and shapes of discharge curves at specific locations. For Rishiganga, for example, we estimate a duration of 10 to 20 min, a number that appears realistic from the information available.

We conducted numerical simulations with r.avaflow [(22), section 5.4], which indicate that the rock and ice avalanche could not have transitioned to the debris flow seen farther downstream without an accompanying reduction in the debris volume. If such a direct transition had occurred, the modeling suggests that the flow discharge would be approximately one order of magnitude higher than the estimates derived from video recordings [(22), section 5.4]. The deposition patterns we observed in satellite imagery support the hypothesis that the vicinity of the Ronti Gad–Rishiganga River confluence played a key role in flow transition. Our numerical simulations are consistent with the escape of a fluid-rich front from the rock and ice avalanche mass near this confluence (Fig. 4A), reproducing mapped trimlines and estimated flow velocities and discharges down to Tapovan (Fig. 4, B and C). Our simulated discharge estimates at locations P1 to P4 (Fig. 4D) are within the ranges derived from the video analysis [(22), section 5.3], and simulated travel times between P0 and P3 (Fig. 4D) show excellent agreement (<5% difference) with travel times inferred from seismic data, videos, and satellite imagery. We found less agreement between the numerical model results and the reconstructions from videos farther downstream owing to the complex effects of the Tapovan project in slowing the flow, which are at a finer scale than is represented by our model.

Causes and implications

The 7 February rock and ice avalanche was a very large event with an extraordinarily high fall height that resulted in a disaster because of its extreme mobility and the presence of downstream infrastructure. The ~3700-m vertical drop to the Tapovan hydropower project is surpassed clearly by only two known events in the historic record, namely the 1962 and 1970 Huascaran avalanches (11), whereas its mobility (H/L = 0.16 at Tapovan, where H is fall height and L is flow length) is exceeded only by a few recent glacier detachments (10). The location of the failure was due to the extremely steep and high relief of Ronti Peak. The sheared nature of the source rocks and contrasting interbedded rock types likely conditioned the failure [(22), section 1]. The large and expanding fracture (Fig. 1, B and C) at the head scarp may have allowed liquid water to penetrate into the bedrock, increasing pore-water pressures or enhancing freeze-thaw weathering.

Nearly all (190) of the 204 people either killed or missing in the disaster (table S1) [(22), section 2] were workers at the Rishiganga (13.2 MW) and Tapovan (520 MW) project sites (33). Direct economic losses from damage to the two hydropower structures alone are over $223 million USD (34, 35). The high loss of human life and infrastructure damage was due to the debris flow and not the initial rock and ice avalanche. However, not all large, high-mountain rock and ice avalanches transform into highly mobile debris flows that cause destruction far from their source (9).

Our energy balance estimates indicate that most of the ~5 × 10⁶ to 6 × 10⁶ m³ volume of glacier ice first warmed (along with a portion of the rock mass) from approximately –8°C to 0°C and then melted through frictional heating during the avalanche as it descended to the Rishiganga valley, involving a drop of ~3400 m [(22), section 5.5]. Potential other sources of water were considered, including glacier lake outburst floods, catastrophic drainage of water from reservoirs such as surface lakes, ice deposited by earlier avalanches, and enlithic reservoirs. No evidence for such sources was observed in available remote sensing data. A slow-moving storm system moved through the area in the days before 7 February. We estimate that a ~220,000- to 360,000-m³ contribution from precipitation over the Ronti Gad basin was a minor component of the flow, representing only 4 to 7% of the water equivalent contained in the initial glacier ice detachment. Similarly, although water already present in the river, water ejected from groundwater, melting snow, wet sediment, and water released from the run-of-the-river hydroelectric project may have all contributed to the debris flow, even when taken together (with generous error margins), these sum to a small amount compared with the probable range of water volumes in the mass movement. The major effect of ice melt on the mobility of rock and ice avalanches is documented (9, 10), but it appears that the combination of the specific rock/ice fraction (~80/20% by volume) and large fall height of the rock and ice avalanche led to a rare, severe event during which nearly all of the ice melted.

Soon after the disaster, media reports and expert opinions started to circulate, postulating links of the event to climate change. Recent attribution studies demonstrated that glacier mass loss on global, regional, and local scales is to a large extent attributable to anthropogenic greenhouse gas forcing (36, 37). High-mountain slope failures in rock and ice, however, pose additional challenges to attribution owing to multiple factors and processes involved in such events. Although long-term trends of increasing slope failure occurrence in some regions could be attributed to climate change (16, 38, 39), attribution of single events such as the Chamoli event remains largely elusive. Nevertheless, certain elements of the Chamoli event have potential links to climate and weather, as described below. Furthermore, the Chamoli event may be seen in the context of a change in geomorphological sensitivity (40) and might therefore be seen as a precursor for an increase in such events as climate warming proceeds.

The stability of glacierized and perennially frozen high-mountain slopes is indeed particularly sensitive to climate change (16). Our analysis suggests regional climate and related cryospheric change could have interacted in a complex way with the geologic and topographic setting to produce this massive slope failure. Air and surface temperatures have been increasing across the Himalayan region, with greater rates of warming during the second half of the 20th century and at higher elevations (41, 42). Most glaciers in the Himalaya are shrinking, and mass loss rates are accelerating across the region [(22), section 5.4, and (43–46)]. Glacier shrinkage uncovers and destabilizes mountain flanks and strongly alters the hydrological and thermal regimes of the underlying rock.

The detachment zone at Ronti Peak is about 1 km higher than the regional lower limit of permafrost at around 4000 to 4500 m above sea level, as indicated by rock glaciers in the region and global permafrost maps (47, 48). Exposed rock on the north face of Ronti Peak likely contains cold permafrost, with rock temperatures several degrees below 0°C. In connection with glaciers, however, ground temperatures can be locally higher. The ice-free south face of Ronti Peak is certainly substantially warmer, with rock temperatures perhaps around or above 0°C, causing strong south-to-north lateral heat fluxes (49). Permafrost temperatures are increasing worldwide, particularly in cold permafrost (16, 50, 51), leading to long-term and deep-seated thermal anomalies and even permafrost degradation (49). Increasing ground temperatures at the failure site of the Chamoli avalanche could have resulted in reduced strength of the frozen rock mass by altering the rock hydrology and the mechanical properties of discontinuities and the failed rock mass (52).

The geology of the failed rocks includes several observed or inferred critical attributes [(22), section 1]: (i) The rocks are cut by multiple directions of planar weaknesses; the failed mass detached along four of these. (ii) The rock mass is close to a major thrust fault, with many local shear fractures, which—along with other discontinuities—would have facilitated aqueous chemical weathering. (iii) The rock types (schist and gneiss), even when nominally unweathered, contain abundant soft, platy, oriented, and geomechanically anisotropic minerals (phyllosilicates and kyanite especially); weathering will further weaken these rocks, and they will be more likely to disintegrate into fine material upon impact, which would influence the rheology and likely enhance the mobility of the mass flow.

The 7 February failure considerably changed the stress regime and thermal conditions in the area of the detachment zone. Only detailed investigations and monitoring will determine whether rock or ice adjacent to the failed block (including a large hanging rock block above the scarp) were destabilized because of these changes and present an ongoing hazard. Similarly, the impoundment at the Ronti Gad–Rishiganga River confluence requires careful monitoring because embedded ice in the dam deposits may melt with warmer temperatures, increasing the risk of an outburst flood by reducing lake freeboard of the dam, and/or reducing structural coherence of the dam.

Videos of the event, including the ones broadcast on social media in real time [(22), section 5.3], showed that the people directly at risk had little to no warning. This leads us to question what could have happened if a warning system had been installed. We estimate that a suitably designed early warning system might have allowed for 6 to 10 min of warning before the arrival of the debris flow at the Tapovan project [perhaps up to 20 min if situated near the landslide source, or if a dense seismic network was leveraged (53)], which may have provided enough time to evacuate at least some workers from the power project. After the event, a new flood warning system was installed near Raini (fig. S15D) [(22), section 2.1]. Studies show that early warning system design and installation is technically feasible, but rapid communication of reliable warnings and appropriate responses by individuals to alerts are complex (54). Previous research indicates that effective early warning requires public education, including drills, which would increase awareness of potential hazards and improve ability to take action when disaster strikes (55, 56). Considering the repeated failures from the same slope in the past two decades [(22), §1], public education and drills in the Chamoli region would be very beneficial.