The 2015 landslide and tsunami in Taan Fiord, Alaska (2024)

Climate change is driving worldwide glacial retreat and thinning¹ that can expose unstable hillslopes. The removal of glacial ice supporting steep slopes combined with the thawing of permafrost in alpine regions² increases the likelihood of landslides^3,4,5,6. Glaciers undercut slopes, priming them for failure by deepening and widening valley bottoms, and by producing steeper valley walls⁷. Additionally, ice loading produces stress fractures in the underlying bedrock, further preparing slopes for failure⁸. As climate warms and glaciers shrink and retreat, they can no longer support rock slopes, and fractures expand as stresses are released. This slope conditioning leads to rock falls, deep-seated gravitational slope deformation, and occasionally catastrophic rock avalanches^4,9,10.

A further effect of glacial retreat is the creation or extension of bodies of deep water, fresh or marine^11,12, where tsunamis can be generated efficiently (Table1). Along the glacially sculpted coastlines of Alaska, Patagonia, Norway, and Greenland, communities, tourism, and infrastructure are becoming increasingly exposed to such landslides and the tsunamis they may generate. Tsunamis in lakes can create flood risk downstream by flowing into inhabited downstream valleys (e.g.^13,14,15).

Tsunamis triggered by landslide impact can have an order of magnitude shorter periods and higher runups than those driven by tectonics that have dominated tsunami hazard research in recent years¹⁶. While tectonic tsunamis typically have periods in the tens of minutes and peak runups extending up to around 30 m, the best studied landslide tsunami, which occurred in 1958 in Alaska’s Lituya Bay, had a period of about 76 seconds and peak runup of 524 m¹⁷. The geologic traces of the Lituya Bay landslide and tsunami have not been documented, providing no analogue for identification of ancient short-period, large-runup tsunamis in the geologic record, be they caused by landslides, volcanoes, or meteor impacts. The only field data available to constrain these reconstructions are the deposits of the 2000 AD landslide-triggered tsunami in Vaigat Strait, West Greenland¹⁸, and surficial descriptions of deposits from the tsunami in Grewingk Lake in 1967¹⁹. The event we describe here in Taan Fiord, Alaska provides the best example to date of a well-documented subaerial landslide that generated a tsunami, and of its impacts on a fjord, coupled with detailed examination of its deposits (see Supplementary Fig. online). This study provides crucial insight into landslide-triggered tsunami processes and the various traces of such events.

The 2015 Taan Fiord landslide

On 17 October 2015, a massive landslide and tsunami occurred at the head of Taan Fiord, an arm of Icy Bay within Wrangell-St. Elias National Park & Preserve in Alaska (Fig.1). The slope failure was primed by rapid ice loss from a tidewater glacier in a tectonically active setting. Tyndall Glacier filled Taan Fiord as recently as 1961²⁰. Rapid warming over the past half century led to 17 km of terminus retreat and over 400 m of ice thinning between 1961 and 1991. Since 1991, the terminus of Tyndall Glacier has stabilized at a shallow bedrock constriction at the head of the fjord^20,21 (Fig.2). The slope that failed was above the calving front and slid directly into the fjord along the terminus, partially covering the toe of the glacier. Destruction of vegetation and other tsunami traces clearly delineate runup throughout the fjord. Directly across from the landslide, runup reached 193 m, (as compared to 240 m in an initial model estimate²²). Runup exceeded 100 m for 1.5 km, overrunning over 1 km² of area. Further down-fjord, Runup varied dramatically, but generally declined to about 15 m at the mouth of the 17 km long fjord (Fig.2).

Ongoing tectonic deformation likely contributed to the Taan Fiord landslide. The present-day glacier terminus lies along the east-west oriented Chaix Hills Fault, one of many structures that accommodate rapid (4–5 mm a⁻¹) tectonic uplift of poorly lithified Miocene-Holocene rocks to high elevations in the St. Elias orogen^21,23. Uplift of weak and faulted rock likely intensified glacial erosion, leading to rapid valley excavation. Subsequent glacial retreat debuttressed the oversteepened fjord wall, initiating progressive failure of the slope that eventually culminated in catastrophic collapse and a tsunami.

Signs of prior hillslope deformation at the location of the 2015 landslide might have provided forewarning. Slumping along the fjord wall at the site was first identified in 1996²¹ and grabens are visible in Landsat images as early as 1995. A comparison of Digital Elevation Models (DEMs) and optical satellite imagery show downslope motion throughout much of the ensuing two decades until the catastrophic failure in October 2015 (Fig.3). While the 2015 Taan Fiord landslide and tsunami did not result in fatalities, actively deforming slopes in more populated places (e.g. Tidal Inlet, Glacier Bay National Park, Alaska²⁴) may be harbingers of more deadly landslide-generated tsunamis in the future. Monitoring gradual downslope motion in mountain ranges around the world, while a technical challenge, would provide a step forward in our ability to mitigate risk.

The final trigger for the landslide is unclear. Seismic waves from a M_W 4.1 earthquake about 500 km away arrived about 2 minutes before the failure, producing ground motion that would not be uncommon multiple times a year in this area²⁵, but might have contributed to the final failure. Similarly, 2015 rains at the nearest gage 110 km away in Yakutat were about 10% more than usual in Sept. and Oct (as-usual the rainiest months of the year). Such deviations above average are common in the years preceding the landslide, but elevated water tables may have contributed at least to the seasonal timing.

Landslide detection and extent

We first identified the 2015 landslide by seismic inversion, using the method of Ekstrom and Stark^26,27. The seismic waves of the Taan Fiord landslide, equivalent to a M 4.9 earthquake, were observed globally. We used an automated landslide detector to identify the seismic signal within hours of the event, which included an abundance of 20 to 100 second period energy, as is typical of large landslides²⁵. Long-period waveforms from the Alaska Regional Network were used to determine the forces associated with the landslide and to refine the estimated location to within 5 km. Seismic inversion suggested an eastward-moving (bearing 96°) landslide that generated peak forces of about 2 × 10¹¹ N and lasting 90 seconds. The landslide source location inferred from seismology was near the calving front of Tyndall Glacier (Fig.2). Based on earlier mapping of fjord geometry¹⁸, the seismogenic motion of the landslide was assigned a length of 1.5 km. These findings, combined with the seismologically determined force history, further suggested a slide mass of 1–1.5 × 10¹¹ kg. Thus, the Taan Fiord landslide was one of the largest non-volcanic landslides in decades^26,28.

These initial estimates were revised within the next year by satellite and aerial imagery, lidar, and ground surveys. The landslide above the terminus of Tyndall Glacier unleashed 7.6 × 10⁷ m³, or 1.8 × 10¹¹ kg of debris. The estimated volume and mass is based on the difference between 2012 and 2016 DEMs, and on an estimate of the slide material remaining in the slide scar. Extending the failure plane beneath onland deposits shows that about 33% of the evacuated volume is still onshore; the rest entered the fjord. Presuming that initial motion was downslope, the landslide moved in a direction similar to that inferred by seismic inversion. The majority of the slide followed the fjord bottom, curving right in an approximately 90° arc blanketing the fjord bottom to its limit 6 km from the source²⁸ (Fig.2). Additional slide material travelled directly eastward through the fjord (Fig.2) and up onto the far shore, depositing hummocks of semi-coherent slide material that blanket the fjord bottom and crest ~15 m above sea level²⁹. This material likely traversed across the bottom of the 90 m-deep fjord and then traveled upwards 105 m to reach its final resting place.

If we assume that the hummocks represent the leading edge of the landslide, the slide velocity must have been at least 45 m s⁻¹ (162 km h⁻¹) for the Taan landslide, similar to values reported for other rock avalanches of comparable dimensions (1903 Frank slide, Alberta, Canada: 3 × 10⁷ m³, 49 m s⁻¹; 1912 Mageik, Alaska: 5.4 × 10⁷ m³, 24 m s⁻¹; 1925 Gros Ventre slide, US-Wyoming: 3.8 × 10⁷ m³, 59 m s⁻¹)³⁰. This estimate is based on the simple conversion of kinetic to potential energy v = (2gh)^0.5 often used in landslide studies to estimate flow velocity from runup height (h)³¹, assuming no potential energy transfers from the body of the slide to the leading edge. These assumptions can overstate maximum velocities in some cases³¹, but also fail to account for friction or the transfer of momentum to water. Alternatively, the hummocks may represent a later phase of the landslide that travelled over earlier deposits that had partly filled in the fjord. In this case, the hummocks would have traversed water as shallow as 50 m, and the minimum flow velocity for the slide would be closer to 36 m s⁻¹ (130 km h⁻¹).

Tsunami generation, propagation, and runup

When landslides enter water, the direct hazard they pose (e.g.^32,33) can be extended by the resultant tsunami (e.g.^17,18,34). In Taan Fiord the landslide directly affected about 2 km² of land onshore, while over 20 km² were inundated by the tsunami. We derive the initial tsunami geometry, constrained by landslide volume and aspect ratio, velocity, and duration³⁵. Using a coupled set of solid and fluid mechanics models³⁵, we estimate that the measured landslide dimensions and material properties generated a leading wave near the head of the fjord with crest elevation of 100 m and period of 90 seconds. In the 100 m water depth near the source area, the front of this wave would have started to break at this crest height, approaching the sloping fan on the far side of the fjord as a plunging or surging breaker. To reach its peak elevation of 193 m (Figs2, 4), the tsunami required enough initial kinetic and potential energy to not only climb the slope, but also overcome energy lost to turbulent dissipation and sediment interaction.

The tsunami traveled south away from the landslide source area, down the fjord. Shallow water wave theory (aka long water wave theory³⁶) computes a propagation speed of (gh)^0.5, or about 30 m s⁻¹ in Taan Fjord’s 100 m deep water. The tsunami proceeded to strip alder forest to elevations exceeding 50 m along the upper 7 km of the fjord, before encountering a range of hills (Fig.2). These hills caused complex wave interference, expressed as runup elevations that rise and fall by tens of meters across distances of a few hundred meters (Fig.4). Variability in the flow is further evidenced by uneven stripping of soil, and by the diverse orientations of still-rooted but flattened trees. Farther down the fjord, runup was diminished, reaching between 10 and 30 m elevation. Even with these relatively low runup elevations, the tsunami energy during overland flow was strong enough to leave only soil and debris where young forest with a 10 m canopy previously stood. The leading crest of the tsunami exited the fjord within 12 minutes, based on numerical modeling of the tsunami (see Methods). At distances greater than 5 km from the mouth of Taan Fiord, tsunami runup was below the high tide shoreline and no longer directly measureable during our first field survey six months after the event.

Geologic traces of the tsunami

The tsunami left thick distinctive deposits that were unlike those documented from other modern tsunamis¹⁶ as it overran and resurfaced several alluvial fans along Taan Fiord (see Supplementary Fig.). On the hardest-hit fan (Hoof Hill Fan) the change in surface elevation between DEMs from before and after the event showed the deposit exceeded 5 m thick in places. Even at the most distal fan studied, where the tsunami runup had diminished to 16 m, the deposit was still 40 cm thick. These deposits included many fragments of supple wood, sometimes overlaid pre-tsunami soil, and occasionally included uphill flow-direction indicators. Deposits characterized from numerous recent tectonic tsunamis were typically sandy, less than 10 cm thick, and often normally graded¹⁴. Some of the Taan Fiord tsunami deposits were similarly normally graded as well, however in most ways they were very different. They included abundant coarse sediment ranging up to boulders, and are composed of three distinct units that we could find no analog for in the literature describing tsunami recent historic tsunami deposits.

The three units were most distinct where the tsunami was largest, at Hoof Hill Fan. The lower unit (A) is composed of sand to boulders, while the upper unit (B) is typically well-sorted and composed of cobbles or boulders. A third unit (C), composed of normally graded sand, was found where it infiltrated unit B.

Similar three-part deposits also partially blanketed fans farther down-fjord, although in many cases unit B was thin or absent, and in a few places the deposit was capped by complex layered sediment that we left uncategorized. Unit A might resemble debris flows from upland sources but can be distinguished by evidence of scour and of uphill flow found at the base. Unit B is similar to, but more tabular and widespread than, sieve deposits found on alluvial fans³⁷. DEM differencing shows that these deposits are widespread and commonly meters thick at Hoof Hill Fan (Fig.5), and thus likely to be preserved for millennia. Deposits in more sediment-poor settings are thin and patchy, but include transported boulders up to 5 m in diameter.

The difference between the deposits in Taan Fiord and those that have been recently described in other tsunamis might be due to differences in sediment source, depositional setting, or wave shape, among other things. The difference in deposit composition may simply reflect a difference in sediment source: In both Taan Fiord and tsunami deposits elsewhere, the bulk composition of the deposit is similar to the source sediment, whether that source is sandy beaches or bouldery alluvial fans. Also, a fan is importantly different from a coastal plain because the retreating wave may have the capacity to rework significant sediment as it runs down the sloping surface. In contrast, coastal plains drain more slowly, and typically sediment is only mobilized in localized constrictions during withdrawal¹⁶. Finally, while tectonic tsunamis have a long period, usually over 10 minutes, the period of the Taan Fiord tsunami was likely similar to the 90 seconds it took the slide to do most of its acceleration and deceleration. This difference in period likely had large impacts on temporal and spatial variability in the tsunami flow as it moved onland, and thus on the erosion and deposition of sediment.

The tsunami deposits in Taan Fiord may be particularly useful in identifying or interpreting deposits of similar events that produce short-period waves, and send those waves over sloping surfaces with diverse sediment available. For example, deposits interpreted to have been generated by landslide tsunamis have been documented in Hawaii³⁸ and the Canary Islands³⁹, pre-Quaternary deposits in the rock-record have been interpreted as records of large landslide or impact-generated tsunamis (summary in 14), and possible impact-generated tsunami traces have been described on Mars^40,41. The deposits in Taan Fiord provide the first well-constrained example that might be used to interpret these and other ancient deposits, in order to better understand the frequency and magnitude of landslide and bolide impact tsunamis. If viewed independently, and out of geomorphic context, neither of the sedimentary units left by the Taan tsunami are necessarily indicative of a landslide-triggered, short-period tsunami. However, taken together and contextualized with the other evidence, the sedimentary deposits may prove sufficiently distinct to aid in the identification of paleo tsunamis. We provide more detailed descriptions of the tsunami sedimentology in the Supplementary Figure.

Implications for hazards assessment

The landslide and tsunami predicated by glacial retreat at Taan Fiord represents a hazard occasioned by climate change. More such landslides are likely to occur as mountain glaciers continue to shrink and alpine permafrost thaws. These landslides can more often be expected to produce tsunamis as water bodies grow and extend landward, closer to steep mountain slopes. Other notable landslides have occurred in recently deglaciated regions (e.g.^3,4,9) and some have produced tsunamis (e.g.^{17,19,34,42,43}, Table1). Their locations, though mostly remote, are attracting tourism and development. For example, incomplete failure and ongoing slow slip have been documented on a slope at Tidal Inlet, a fjord in Glacier Bay National Park, 6 km from a channel visited by dozens of cruise ships during summer months²⁴. On 28 June, 2016, an approximately 1.5 × 10¹¹ kg landslide collapsed onto Lamplugh Glacier, also in Glacier Bay National Park, but luckily did not reach tidewater and so did not trigger a tsunami^44,45. Then on 17 June, 2017, a landslide in Rink Fiord triggered a tsunami that killed 4 people in Nuugaatsiaq, Greenland, 30 km away, highlighting the need for further study of these phenomena.

Comparison of pre- and post-event data at Taan Fiord constrain the dimensions of the landslide and tsunami, and in turn may aid in identifying other such events in the recent geologic past. In order to mitigate the risk associated with landslide-triggered tsunamis, we suggest the following: 1) revisit geologic records of paleotsunamis to better understand frequency and causal mechanisms of past occurrences; 2) assess areas of potential failure given known glacial histories and evidence of precursory motion; and in areas of particular concern, 3) map areas of likely impact using glacier, landslide, and tsunami inundation models in order to reduce impacts should an event occur; and finally 4) monitor for landslides using seismic and remote-sensing techniques.