Seismic Precursors to a 2017 Nuugaatsiaq, Greenland, Earthquake–Landslide–Tsunami Event

Natural Hazards https://doi.org/10.1007/s11069-019-03582-8

ORIGINAL PAPER

Seismic precursors to a 2017 Nuugaatsiaq, Greenland, earthquake–landslide–tsunami event

Rhett Butler1

Received: 31 July 2018 / Accepted: 30 January 2019 © Springer Nature B.V. 2019

Abstract High-frequency (5–20 Hz) seismic signals precursory to and embedded within the June 17, 2017 ML = 4 earthquake–landslide event are analyzed. This event in western Greenland generated a tsunami in Karrat ford inundating Nuugaatsiaq village 32 km distant. Spec- trogram and wavelet analyses of seismic data from the Greenland Ice Sheet Monitoring Network (GLISN) corroborate observations of seismic precursors at Nuugaatsiaq reported by Poli (Geophys Res Lett 44:8832–8836, 2017) and Caplan-Auerbach (in: AGU fall meeting abstracts, 2017) and reveal additional high-frequency arrivals being generated after the apparent initiation of fault rupture. New observations of seismic precursors 181 km from the Event at Upernavik, Greenland are correlated with those seen at Nuugaatsiaq. Wavelet analysis presents > 100 signifcant energy peaks accelerating up to and into the earthquake– landslide event. The precursor events show a distinct, power law distribution, characterized by b values of ~ 2.4. Results are compared and contrasted with small precursors observed in the studies of a natural chalk clif landslide at Mesnil-Val, Haute Normandie, France. The earthquake–landslide appears to have been initiated by seismic precursors located at the fault scarp, leading to a small seismic foreshock and small landslide initiation, followed by a larger earthquake at the fault scarp, precipitating the primary landslide into the Karrat Fjord, which caused the subsequent tsunami.

Keywords Greenland · Seismic precursors · Wavelet analysis · Earthquake–landslide– tsunami

1 Introduction

An earthquake–landslide (generally referred herein as the “Event”) on June 17, 2017, adjacent to Karrat Fjord of west Greenland generated a local tsunami that devastated the nearby village of Nuugaatsiaq, 32 km distant (e.g., Clinton et al. 2017). The landslide survey (Gauthier et al. 2018) revealed the volume of the mass wasting entering the Fjord to be 45 million m3 from a surface rupture area of ¾ km 2 at a mean elevation of 1 km. To place this

* Rhett Butler [email protected]

1 Hawai‘i Institute of Geophysics and Planetology, University of Hawai‘i at Mānoa, 1680 East‑West Road, POST 602, Honolulu, HI 96822, USA

Vol.:(0123456789)1 3 Natural Hazards in perspective, the volume of this Greenland landslide exceeded the volume of the great Lituya Bay, Alaska, landslide of 1958 (Miller 1960) by nearly a factor of two (Gauthier et al. 2018). The Event was well recorded both locally on the Greenland Ice Sheet Monitoring Net- work (GLISN, Kanao et al. 2012; Clinton et al. 2014) and by the Global Seismographic Network (Butler et al. 2004), where the earthquake magnitude measured between ML = 3.5 (GEUS, Geological Survey of Denmark and Greenland) and Mw = 4.8 (GEOFON, Ger- many). The closest GLISN station was NUUG in the village of Nuugaatsiaq, and nearby sites included UPNV, Upernavik, and ILULI, Ilulissat (see Fig. 1). The NUUG STS-2 seis- mometer was fortunately located at 36 m elevation on the outskirts of the village from the ford, and survived to record the whole event. Not only did NUUG record the landslide and tsunami, but also signifcant seismic precursory signals (Poli 2017; Caplan-Auerbach 2017). Analysis of the NUUG and UPNV data will show the presence of signifcant precursory seismic activity starting at least 10 h before the Event. This pattern of precursory seismic signals is quantitatively compared and contrasted with a smaller landslide in Haute Nor- mandie, France. The apparent location of the Greenland seismic precursors is consistent with the locus of the landslide. Hence, the seismic component of the Event preceded the landslide and may be considered causative. The tsunami itself was generated by kinetic energy and momentum of the triggered landslide impacting the ford.

Fig. 1 Nuugaatsiaq map with inset map of GLISN seismic stations (green ‘balloons’) discussed in text. Red ‘balloon’ indicates 2017 landslide location. The landslide surface is located in red, showing the approximate strike (105°) and lateral length. NUUG is located at an azimuth of 250° from the source region within the village of Nuugaatsiaq at 36 m elevation. Dotted yellow arrows indicate azimuths of next-nearest seismic stations discussed in text. Radiating red circles simply approximate the tsunami propagation, but neglect the internal refections and seiching within the ford. The white star 22 km WNW of NUUG is the location of a ML = 4.5 comparison earthquake in 2008, observed at both SUMG and SFJD (inset map). Map adapted from GoogleEarth 1 3 Natural Hazards

The tsunami tragedy in Greenland stands as a warning to regions where high-topography is immediately adjacent to a body of water. This concern is high in Hawai‘i where the author has been directly involved with the State of Hawai‘i Emergency Management Agency in assessing tsunami hazards, not only from earthquakes (e.g., Butler et al. 2017), but also landslides. Analysis of the Greenland Event by Taiwan refects the same local concerns (Chao et al. 2018). Toward this end, the analyses herein focus on the advance warning (seismic precursors) leading up to and into the Event to understand and constrain seismological aspects of the source mechanism and its temporal development. In doing so, the seismological precursors identify the initiating event as an earthquake source, triggering the landslide, which set into motion the tsunami.

2 Seismic dataset

All of the seismic data used in this study arise from GLISN, with primary focus on NUUG and UPNV (Fig. 1, inset). Figure 2 shows the 2017 seismic event at NUUG and UPNV, rotated into radial and tangential components with respect to the source. How- ever, while visiting the NUUG station after the Event, it was discovered that the STS-2 had been inadvertently rotated 79° clockwise in 2015 (J. Clinton, personal communica- tion, 2017)—this additional rotation correction has been implemented herein. Whereas the

10-5 1 RAD 32.1 km TAN NUUG 0.5 Z

Meters/Second -0.5 subevent-1 subevent-2 -1 0210 03040506070 10-5 1.5 RAD 181.0 km 1 TAN UPNV Z 0.5 0 -0.5 Meters/Second -1 subevent-1 -1.5 0420 06080 100 120 Time (seconds) from origin time 2017/06/17 23:39:12

Fig. 2 Seismic 3-component data from NUUG and UPNV, respectively, are plotted. Radial (red), vertical Z (black), and tangential (green) data are plotted together on the same axes. Data are band-pass-fltered [1/100 20] Hz. Distances are noted. At NUUG, the initial seismic arrival appears at 9 s, noted by the blue arrow as subevent-1. A second subevent-2 (violet arrow) about 50 s after origin time is evident. Both events refect a change in the character of precursory signals, and embedded wavelet arrivals. Note the initial P wave emergent motion at NUUG is vertical-radial. At UPNV, the P wave is not clearly observable, and frst vis- ible arrival is the S wave. Between the two events, the close correspondence of Z and radial components are indicative of seismic body waves, and suggestive of an initial landslide associated with subevent-1

1 3 Natural Hazards

STS-2 seismometers are essentially fat to velocity from 100 s to 20 Hz, the data were band-pass-fltered at [1/100 20] Hz and corrected only for instrument sensitivity. The 17 June 2017 origin time of the Event varies from 23:39:12 (U.S. Geological Survey, USGS) to 23:39:17.4 (Geological Survey of Denmark and Greenland, GEUS) to 23:39:18.5 (Euro- Mediterranean Seismological Centre, EMSC). I have adopted the USGS solution as earli- est. Figure 3 compares the Event as observed at SFJD and SUMG with a 2008 earthquake located close to the Event in Fig. 1. To view the spectral content as a function of time, spectrograms were generated for NUUG, UPNV, and ILULI using the Thomson (1982) multitaper method applying orthogonal, discrete prolate spheroidal (Slepian, 1962) sequences for measuring the power in each time–frequency window, as implemented by Matlab® pmtm.m. The initial result of this analysis is seen in Fig. 4 for NUUG and UPNV. Similar analysis was conducted for ILULI; however, the noise from the nearby Jakobshavn glacier both dominates and precludes use- ful comparison with NUUG and UPNV.

3 Seismic precursors

A series of precursory arrivals are evident in the spectrogram in Fig. 4 with dominant frequencies (i.e., greater than twice the background noise level) in the 5–20 Hz band at both NUUG and UPNV, extending nearly an hour before the Event. Poli (2017) approached

10-6 10-6 4 4 RAD RAD 520.7 km 3 TAN 529.9 km 3 TAN Z Z 2 2 1 1 0 0 -1 -1 SFJD 2008 SFJD 2017

-2 Meters/Second -2 Meters/Secon d -3 -3 0 50 100 150 200 250 300 0 50 100 150 200 250 300

-5 10-5 1.5 10 1 RAD 532.0 km RAD 484.6 km 1 TAN TAN Z 0.5 Z 0.5 0 0 -0.5 -0.5 Meters/Second -1 Meters/Second SUMG 2008 SUMG 2017 -1.5 0 50 100 150 200 250 300 0 50 100 150 200 250 300 Time (seconds) from origin time Time (seconds) from origin time 2008/06/16 13:28:46 2017/06/17 23:39:12

Fig. 3 The 2017 Event is compared with a ML = 4.5 earthquake (location in Fig. 1) at SUMG, Summit and SFJD, Søndre Strømford, Greenland. These are the only two sites clearly recording both sources. Data are band-pass-fltered [1/100 20] Hz. Plot details are same as for Fig. 2. Time scales are in seconds following source origin time. Relative to the origin time, surface waves from the 2017 Event arrive about 50 s later than the 2008 earthquake over nearly identical paths. This indicates that the primary seismic energy release occurred at subevent-2, about 50 s after subevent-1

1 3 Natural Hazards

Fig. 4 Spectrograms of rotated, horizontal velocity components of NUUG and UPNV show the signal power as a function of frequency and time. Only data with signal-to-noise SNR > 2 are shown against the blue background. The frequency band 5–20 Hz has the best SNR. Distances from the source are noted. Precursors are indicated by the vertical yellow-green stripes, whose occurrences accelerate approaching the Event (black arrow) The initial earthquake (subevent-1) in yellow is followed by a larger, subevent-2 in orange-red (see also Figs. 2, 3). The apparent time ofset (indicated by red arrows) at UPNV, with respect to NUUG, fts the relative propagation times from the landslide location for both the precursors and the Event. Furthermore, the timing of subevent-1 at UPNV is consistent with NUUG. Note that the power in the precursors is larger at NUUG than UPNV, consistent with the greater propagation distance to UPNV. The noise midway into the UPNV spectrogram may refect glacier noise from the Upernavik Glacier located 65 km east of UPNV the measurement of the precursory signals at NUUG using cross-correlation methods. Whereas the spectrogram method illuminates these precursors, the observations depend upon the length of the time windows employed; e.g., there may be more arrivals embedded within the same window, such as near the Event where the precursors merge together into a continuum in Fig. 4. The correlation between NUUG and UPNV is consistent with a source at the locus of the earthquake–landslide Event. For an assumed shear wave velocity of 3.7 km/s, the travel time from the Event to UPNV is about 40 s later than at NUUG—consistent with the observed ofset. Additionally, Poli (2017) has confrmed that individual precursors observed at NUUG have apparent S-P times consistent the source at the site of the landslide. In order to better highlight and defne the character of these precursors, I employed the discrete wavelet transform method, as implemented in Matlab ® routine “wavelet Ana- lyzer”, using Daubechies’ (1988, 1992) orthogonal wavelet D10 for its seismic character, and a level 3 decomposition tree. The wavelet methodology was applied to the each of the three components of NUUG, and plotted together in Figs. 5, 6 and 7, over time windows selected to enhance understanding of Event source features observed in Fig. 2. Precursory signals start at least 600 min prior to the Event. There may be earlier, smaller precursors obscured in the background noise. It is also evident that the rate of the precursors had been accelerating for hours prior to spectrogram data in Fig. 5. In order to decompress the image, Fig. 5 is composed from plots at four, decreasing time scales of minutes, leading up to the timing of the initial seismic arrivals observed in Fig. 2, which I have termed “subevent-1.” The analysis then continues in Figs. 6 and 7 through subevent-1 into the Event itself. where the time scales are in seconds, and the amplitude scale increases from ~ 3 (m 2/s2 × 10−14) in Fig. 6 to ~ 50 and ~ 4500 (m 2/s2 × 10−14) in Figs. 6 and 7, respectively. The precursory signals observed at NUUG (and corroborated at UPNV) extend over a 10-h period leading up to the main Event (Figs. 5, 6, 7). The timing interval of events initially is over 50 min, accelerating to 10’s of minutes, minutes, 10’s of seconds, and then

1 3 Natural Hazards

Radial 3 Z Tangential Pick

1 4

-1 0 0 -600 -500 -400 -300 -200 -100 0

x 10 3 2 ) 2 1

0 -100 -90 -80 -70 -60 -50 -40 -30-20 -10 0

(Meters/sec 2 1

0 -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0

2 1

0 -1.0 -0.9 -0.8 -0.7 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 T ime (minutes) before event1 [2017-6-17 23:39:21]

Fig. 5 Wavelet analysis of NUUG precursors starts 10 h prior to subevent-1 (see Fig. 2). Radial (red), Z vertical (blue), and tangential (green) data are plotted together as velocity-squared. The dominance of radial-vertical arrivals guides peak selection (black circles) larger than background noise. Note the time scale in minutes expands for the successive plots, showing the temporal acceleration of the phenomena— the purple lines highlight the expanding ranges of the plots. Each plot ends at the time of subevent-1 at NUUG, about 9 s after the Origin time. The vertical scale is the same for each plot, indicating that the precursors did not grow larger with time

seconds prior to the Event. In choosing the peaks (circled), a review of the three components showed that the radial component is dominant, and hence, the selections were domi- nated by vertical and radial peaks (seismic energy is coupled between vertical and radial wave propagation). The peaks were selected with signal levels greater than the prior background noise. Peak amplitudes are in velocity-squared for straightforward conversion to

1 3 Natural Hazards

Radial Z 40 Tangential Picks between event1 and event2 Event1, Event2 -14

30 x 10 2 ) 20

10 (Meters/sec

0 9110 520253035404550

Time (seconds) after origin time [2017-6-17 23:39:12]

Fig. 6 Wavelet analysis of NUUG signals embedded within the Event between subevent-1 and subevent-2. Radial (red), Z vertical (blue), and tangential (green) data are plotted together as velocity-squared. Time is seconds following the Origin time. Star-arrow shows subevent-1 and subevent-2 times. Note that after subevent-1, the energy in the signals grows by an order of magnitude, increasing toward subevent-2. Compared with Fig. 2, where the apparent body wave arrivals between subevent-1 and subevent-2 suggest the beginning of seismic arrivals from the landslide

Fig. 7 Wavelet analysis of NUUG signals embedded within the Event for 20 s after subev- 4,000 Radial 4 Z ent-2. Radial (red), Z vertical Tangential -1 (blue), and tangential (green) Picks after event2 data are plotted together as veloc- 3,000 Event2

ity-squared. Star-arrow indicates x 10 the subevent-2 time. Note that 2 after initiation of subevent-2, 2,000 the energy in the signals grows by two orders of magnitude. The primary contribution to the landslide occurs following subev- 1,000 ent-2 with major seismic energy (Meters/sec) release 0 50 55 60 65 70 Time (seconds) after origin time [2017-6-17 23:39:12]

energy. In Fig. 5, the 148 signals extend through the origin time to subevent-1 at NUUG. For comparison, Poli (2017) recovered 95 peaks in the NUUG data over a comparable time period. Further analysis was conducted into the earthquake–landslide seismic energy observed at NUUG following subevent-1. In Fig. 2, the character of the NUUG data changes after

1 3 Natural Hazards about 50 s into the Event, indicated by the violet arrow as “subevent-2.” Comparing data in Fig. 3 for SUMG and SFJG for both the Event and a nearby ML = 4.5 comparison earthquake in 2008 (Fig. 1) show that the surface wave energy—as measured from apparent origin time—from the Event arrives about 50 s later than for the 2008 earthquake. The initial subevent-1 may correspond with the small initial magnitudes (3.5–3.9) reported by GEUS and EMSC, whereas subevent-2 may correspond with the larger magnitudes (4.2–4.8) based upon long-period, surface wave energy reported by USGS and GEO-FOrschung- sNetz (GEOFON), respectively. For the wavelet analysis of NUUG data from subevent-1 through and into subevent-2 (Figs. 6 and 7), the signal levels were set to be greater than the prior window maxima, i.e., from Figs. 5 and 6, respectively. Figure 6 shows the interval between subevent-1 and subevent-2, expanding the number of observed peaks by 66. Figure 7 shows 20 s of arrivals after the beginning of subevent-2, and 37 additional peaks that dwarf the energies in the precursory signals. The acceleration of precursory signals leading up to subevent-1 fts within the fracture mechanics narrative discussed by Poli (2017). However, the high-frequency, wavelet signals continue and increase in amplitude after subevent-1, suggesting that the fault slip observed seismically as subevent-1 did not end the “precursors.” Rather, the signals continue unabated for another ~ 40 s up to (and through) subevent-2. Contemporaneously, the landslide phase of the Event may have initiated with subevent-1, as refected by the succes- sion of body waves evident in Fig. 2 at NUUG between subevent-1 and subevent-2. Hence, some of the arrivals in Fig. 6 likely correspond with the landslide. Given the surface area of the causative fault, ~ 3/4 km2 (Gauthier et al. 2018), and pattern of subevents, one possible interpretation of the sequence of occurrence follows. Fracture on the fault initiates slowly and intermittently, and accelerates. At subevent-1, a signifcant section of the fault undergoes failure, indicated by the initial ML = 3.5 GEUS magnitude. Using the larger magnitudes (Ms 4.2, Mw 4.8) based upon later surface waves as representative of the total event size, the incipient faulting associated with subevent-1 was associated with < 10% of the eventual fault area. Given the late arrivals of surface waves, the predominant fault rupture likely initiated with subevent-2. However, following subevent-2, the signals in Fig. 7 likely indicate both the fault rupturing (irregularly) and the full mass of the landslide (~ 1011 kg) descending ~ 1 km downslope into the ford.

4 Seismic precursors to landslides

Reviewing literature for prior seismic precursors to landslides, the studies (Amitrano et al. 2005; Senfaute et al. 2009) of a natural chalk clif at Mesnil-Val, Haute Normandie, France, are similar to Greenland’s situation. Although the scale of the two phenomena difer—the Greenland Event is much larger—nonetheless, there are interesting similarities in the styles of the seismic precursors. Though not discussed herein, Caplan-Auerbach and Huggel (2007) report precursory seismic signals prior to ice avalanches at Iliamna volcano, Alaska. Amitrano et al. (2005) analyzed a 1000–2000 m3 collapse preceded by 2 h of precursory signals, recorded by a geophone and accelerometer sampled at 40 kHz. Similar to Green- land, the incidence of precursory signals accelerated prior to collapse. However, in contrast, the volume of Greenland landslide entering the ford was 45,000,000 m3 (Gauthier et al. 2018), or about 4 orders of magnitude larger. Estimating the relative energy in the precursors, Amitrano et al. (2005) ft the data with −b a Gutenberg and Richter (1954) power law, e.g., N(> E) ∼ E where N is the number of 1 3 Natural Hazards events of energy larger than E, and b refects the distribution of sizes (the b value). For the Mesnil-Val data, a b value of 0.5 was determined, which is substantially diferent from b ~ 1 found in many seismological studies of earthquakes (e.g., Kanamori and Brodsky 2004). Sen- faute et al. (2009) continued the Mesnil-Val study, extending the precursory window to 15 h (224 signals), and measured within 10’s of meters of the collapse. In contrast to Greenland, where the observed precursors had energy between 5 and 20 Hz (at 32–181 km distance), the Senfaute et al. (2009) data had frequencies between 100 and 1000 Hz at ~ 10 m distance. In comparison with the Mesnil-Val studies, I have plotted the cumulative energy vs. time of the observed precursory peaks in Fig. 8. In contrast to the Mesnil-Val which showed about 3 orders of magnitude in energy variation up to collapse, the Nuugaatsiaq Event extended over 6 orders of magnitude in energy. Nonetheless, the shape and character of the cumulative energy plots shown in Fig. 8 herein and in Figure 14 of Senfaute et al. (2009) are very similar. Additionally, Senfaute et al. (2009) noted a signifcant change in frequency content (toward dominantly low frequency, with diminished high frequencies) leading immediately up to fnal failure, and hypothesized these changes were related to the shearing or opening of the existing fractures inducing fnal failure processes. As such, subevent-1 and subevent-2 in Greenland may present a homologous situation. The power law distribution of Greenland precursors is also plotted in Fig. 8. In doing so, the data clearly segregate into three families: precursory signals, signals between subevent-1 and subevent-2, and those following subevent-2. The frst group (green) is distributed with a b value of ~ 2.4 and correlation coefcient R2 ~ 0.94. Large b values have been associated with earthquake swarm activity (e.g., Smith and Sbar 1974), where small events predominate. For the blue data immediately following subevent-1, the trend shallows to a b value of ~ 1.5. The third group (red) representing the initiation of subevent-2 is distributed with a b value of ~ 1.2 and correlation coefcient R2 ~ 0.97. This latter value approaches observed b values ~ 1 for many tectonic earthquakes. Low b values (~ 0.5) seen for Mesnil-Val data have been observed from microfrac- tures during the deformation of rocks in laboratory experiments (Scholz 1968) in a variety of rock types at > 60% fracture stress. In contrast, the Greenland Event started with a relatively high b value~ 2.4, shallowing to b ~ 1.5 following subevent-1, and then decreasing further to b ~ 1.2 after subevent-2, approaching b ~ 1 for common earthquakes. The mean energy levels increase by factors of 6.2 and 48 between the three regimes—the seismic energy level jumps by a factor of ~ 300 in transition from swarm- like activity to tectonic earthquake rates. Plotting the intervals between the NUUG precursory signals (Fig. 9) illustrates the rapidly increasing rate of precursors. Starting at 1000’s of seconds, the time interval between successive signals decreases to ~ 0.1 s prior to subevent-1. Interestingly, the rate slows after subevent-1 to ~ 1 s through subevent-2. Given the orders of magnitude increase in signal energy initiating with subevent-2, details of smaller features are not resolved at this level of analysis.

5 Summary

The Greenland Event is a complex, cascading, interlinked sequence of seismic, subaerial landslide, and tsunami processes. My own interpretation follows. Seismic precursors are observed up to 10 h prior to the tsunami, from NUUG at 32 km (e.g., Poli 2017) and from UPNV at nearly 180 km. The precursory signals continue up to and through

1 3 Natural Hazards 3 10-6 NUUG event 2 NUUG event 1 10-8 EQ Origin

10-10

10-12 -4 -3 -2 -1 0 Cumulative Energy Density, ~ J/m Time (104 sec) with respect to Origin [23:39:12] 2.5 Data NUUG Fit 2 Confidence bounds b = -1.5 1.5 r2 = 0.97

N(peaks) ≥ E 1 b = -2.4 10 r2 = 0.95 b = -1.2 log 0.5 r2 = 0.98

0 -11 -10 -9 -8 -7 3 log10 Energy J/m

Fig. 8 [upper panel] The cumulative energy of precursory and embedded signals from Fig. 5 (green), Fig. 6 (blue), and Fig. 7 (red) are plotted with time. Origin, subevent-1, and subevent-2 times are circled. Compare with Figure 14 of Senfaute et al. (2009). [lower panel] The log 10 of number of signals with energy ≥ E are plotted against the log 10 of E, the signal energy density. Colors correspond with upper panel. The signals fall naturally into three groups. The linear models give power law slopes, from which b values are ~ 2.4 and ~ 1.5 for data before and after subevent-1, respectively, which then shallows further to b value~ 1.2 after the initiation of subevent-2. Note the substantial correlation coefcients

a ML = 3.5 “foreshock,” which initiates the landslide. About 50 s following this foreshock, a second, larger subevent-2 (M > 4) occurs. This latter event generates both the surface wave magnitude (and arrival time), as well as the largest part of the landslide. The Greenland seismic precursors extend over a longer duration, and were signifcantly

1 3 Natural Hazards

104

102

100

10-2 -3.5 -3 -2.5 -2 -1.5-1 -0.5 0 0.5 Mid-Point Time ( 104 seconds ) with respect to Origin time [23:39:12] 102

al between signals (seconds) 1 rv 10 te In 100

10-1 DETAIL

10-2 -20 -10 01020430 05060 70 Mid-Point Time (seconds) with respect to Origin time [23:39:12]

Fig. 9 [upper panel] Interval times between successive signals are plotted from Fig. 8 at their midpoint. The increasing acceleration of signal rate is evident. [lower panel] Detail near the Origin time. Times of subevent-1 and subevent-2 are indicated by stars. Note acceleration of signals in the 10 s prior to subevent-1 more energetic than a comparable landslide associated with failures of coastal chalk clifs in Normandie, France (Amitrano et al. 2005; Senfaute et al. 2009). The Greenland seismic precursors, the body waves of foreshock subevent-1, and the surface waves for the main subevent-2 were generated on the fault surface exposed by the landslide. The latter interpretation is supported by the moment tensor earthquake mechanism determined by GEOFON which is consistent with the strike and dip geome- try at the fault. In essence, the earthquake ruptured the fault surface, releasing the hang- ing wall nearly a km above the Fjord (and creating the head scarp), which then became a rock avalanche, accelerating downhill, impacting the Fjord with a mass ~ 1011 kg, dis- placing water in the Fjord, and generating the destructive tsunami. Hence, we may con- sider the Greenland earthquake–landslide to have been a series of seismic precursors, and small earthquakes which precipitated the subaerial landslide, which in turn generated the Nuugaatsiaq tsunami. The key conclusion is that in addition to the physical threat of a subaerial landslide, a steep slope abutting the ocean is a potential candidate for a landslide-generated tsunami. Whether or not seismological precursors are observable, the threat remains cred- ible. The concatenation of events wherein a small earthquake (M < 5) may lead to a dev- astating tsunami via an earthquake-triggered landslide must be considered by coastal communities with steep topography. Higher awareness of very-shallow seismic events in such regions is necessary. 1 3 Natural Hazards

Acknowledgements I thank the operators of the GLISN and the GSN for free and open data access from seismic stations in Greenland. I thank John Clinton at ETH Zurich and both Tine Larsen and Trine Dahl- Jensen at GEUS Denmark for correspondence regarding their observations of the Event. HIGP contribution number 2369. SOEST contribution number 10642.

Data statement All seismic data utilized are available from the IRIS Data Management System. Data from NUUG following the tsunami inundation and consequent power/communications disruption came from John Clinton at ETH Zurich. Earthquake catalog information comes via USGS (https://earthquake.usgs. gov/earthquakes/), GEUS (http://seis.geus.net/quakes/dnk-2017-06.html), and GEOFON (http://geofon.gfz- potsdam.de/eqinfo/event.php?id=gfz2017luxw)

References

Amitrano D, Grasso JR, Senfaute G (2005) Seismic precursory patterns before a clif collapse and criti- cal point phenomena. Geophys Res Lett 32(8):1. https://doi.org/10.1029/2004gl022270 Butler R, Creager K, Earl P, Fischer K, Gaherty J, Laske G, Leith W, Park J, Ritzwoller M, Tromp J, Wen L (2004) The Global Seismographic Network surpasses its design goal. EOS Trans AGU 85(23):225–232 Butler R, Walsh D, Richards K (2017) Extreme tsunami inundation in Hawai‘i from Aleutian–Alaska sub- duction zone earthquakes. Nat Hazards 85(3):1591–1619 Caplan-Auerbach J (2017) Precursory seismicity associated with landslides, including the 2017 tsunamigenic landslide in the Karrat Fjord, Greenland. In: AGU fall meeting abstracts Caplan-Auerbach J, Huggel C (2007) Precursory seismicity associated with frequent, large ice avalanches on Iliamna volcano, Alaska, USA. J Glaciol 53(180):128–140 Chao WA, Wu TR, Ma KF, Kuo YT, Wu YM, Zhao L, Chung MJ, Wu H, Tsai YL (2018) The large greenland landslide of 2017: Was a tsunami warning possible? Seismol Res Lett 89(4):1335–1344. https:// doi.org/10.1785/0220170160 Clinton JF, Nettles M, Walter F, Anderson K, Dahl-Jensen T, Giardini D, Govoni A, Hanka W, Lasocki S, Lee WS, McCormack D, Mykkeltveit S, Stutzmann E, Tsuboi S (2014) Seismic network in Greenland monitors earth and ice system. EOS Trans Am Geophys Union 95:13–14 Clinton J, Larsen T, Dahl-Jensen T, Voss P, Nettles M (2017). Seismic observations from Nuugatsiaq slide/ tsunami. https://ds.iris.edu/ds/nodes/dmc/specialevents/2017/06/22/nuugaatsiaq-greenland-landslide- and-tsunami/. Accessed 5 July 2017 Daubechies I (1988) Orthonormal bases of compactly supported wavelets. Commun Pure Appl Math 41(7):909–996 Daubechies I (1992) Ten lectures on wavelets. Soc Ind Appl Math 1992:194 Gauthier D, Anderson SA, Fritz HM, Giachetti T (2018) Karrat Fjord (Greenland) tsunamigenic landslide of 17 June 2017: initial 3D observations. Landslides 2018(15):327–332. https://doi.org/10.1007/s1034 6-017-0926-4 Gutenberg B, Richter CF (1954) Seismicity of the earth and related phenomena. Princeton University Press, Princeton, p 310 Kanamori H, Brodsky EE (2004) The physics of earthquakes. Rep Prog Phys 67(8):1429–1496 Kanao M, Tsuboi S, Butler R, Anderson K, Dahl-Jensen T, Larsen T, Nettles M, Voss P, Childs D, Clinton J, Stutzmann E, Himeno T, Toyokuni G, Tanaka S, Tono Y (2012) Greenland ice sheet dynamics and glacial earthquake activities, Chapter 4. In: Müller J, Koc L (eds) Ice sheets: dynamics, formation and environmental concerns. Nova Science Publishers Inc., Hauppauge, pp 93–120 Miller DJ (1960) The Alaska earthquake of July 10, 1958: giant wave in Lituya Bay. Bull Seismol Soc Am 50(2):253–266 Poli P (2017) Creep and slip: seismic precursors to the Nuugaatsiaq landslide (Greenland). Geophys Res Lett 44:8832–8836. https://doi.org/10.1002/2017gl075039 Scholz CH (1968) The frequency-magnitude relation of microfracturing in rock and its relation to earthquakes. Bull Seismol Soc Am 58(1):399–415 Senfaute G, Duperret A, Lawrence JA (2009) Micro-seismic precursory cracks prior to rock-fall on coastal chalk clifs: a case study at Mesnil-Val, Normandie, NW France. Nat Hazards Earth Syst Sci 9(5):1625–1641 Slepian D (1962) Prolate spheroidal wave functions, Fourier analysis and uncertainty—IV: extensions to many dimensions; generalized prolate spheroidal functions. Bell Syst Tech J 43:3009–3057

1 3 Natural Hazards

Smith RB, Sbar ML (1974) Contemporary tectonics and seismicity of the western United States with emphasis on the Intermountain Seismic Belt. Geol Soc Am Bull 85(8):1205–1218 Thomson DJ (1982) Spectrum estimation and harmonic analysis. Proc IEEE 70:1055–1096

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afliations.

1 3