GEOPHYSICAL RESEARCH LETTERS, VOL. 26, NO. 19, PAGES 3009-3012, OCTOBER 1, 1999

The landslide-generated tsunami of November 3, 1994 in $kagway Harbor, Alaska: A case study Alexander B. Rabinovich, 1'2 Richard E. Thomson, 2 Evgueni A. Kulikov, Brian D. Bornhold i and Isaac V Fine 1 9 '

Abstract. We examine the origin and behavior of the catastrophic tsunami that impacted Skagway Har- bor at the head of Taiya Inlet, Alaska, on November 3, 1994. Geomorphologic and tide gauge data, com- bined with numerical simulation of the event, reveal that the tsunami was generated by an underwater land- slide formed during the collapse of a cruise-ship dock undergoing construction. Use of a fine-grid model for Skagway Harbor and a coarse-grid model for Taiya Inlet enables us to explain many of the eyewitness accounts and to reproduce the dominant oscillations in the tide gauge record, including the persistent (,- 1 h) 3-min oscillation in Skagway Harbor. The occurrence of the landslide is linked to critical overloading of the slope materials at a time of extreme low tide.

Introduction

At 1912 h Alaska Standard Time (AST) on Novem- ber 3, 1994, a 250 m section of the Pacific and Arctic Railway and Navigation Company (PARN) Dock un- der construction on the eastern side of Skagway Harbor (Fig. 1) slid rapidly into the water. The event occurred about 25 min after an extreme low tide of-1.3 m (rela- tive to mean lower low water), the lowest tide since the dock construction began [Lander, 1996; Cornforth and Lowell, 1996], and was accompanied by a series of large- amplitude tsunami waves estimated by eyewitnesses to be 5-6 m high in the inlet and 9-11 m at the shore- line [Lander, 1996; Kulikov et al., 1996; Raichlen et al., 1996]. A NOAA analog tide gauge (Fig. 1) record- ed mainly 3-min waves with maximum trough-to-crest wave heights of 2 m. The landslide and accompanying tsunami claimed one life and caused an estimated $21 million damage [Lander, 1996; Raichlen et al., 1996].

It has been shown [Lander, 1996; Kulikov et al., 1996; Cornforth and Lowell, 1996; Raichlen et al., 1996] that the November 1994 tsunami likely originated from the failure of the PARN Dock which, in turn, was linked to critical overloading of the slope materials at a time of extreme low tide. An alternative explanation pro- vided by Mader [1997] and Kowalik [1997] is that the waves were generated by a massive submarine slide in

International Tsunami Research, Inc., Sidney, British Columbia, Canada.

2Fisheries and Oceans Canada, Institute of Ocean Sci- ences, Sidney, British Columbia, Canada.

Copyright 1999 by the American Geophysical Union.

Paper number 1999GL002334. 0094-8276/99/1999GL002334505.00

Taiya Inlet (Fig. 1), and that the dock failure was a secondary effect of this "offshore" slide. The main ar- guments for the alternative generation mechanism are that the waves appear to be too large to have been pro- duced by the dock collapse and associated submarine slide and that the spatial scales of the collapsed dock were too small to account for the observed 3-min period and 1-h duration of the observed oscillations.

The purpose of this paper is to examine the 1994 Skagway Harbor event in the context of landslide-gener- ated tsunamis in steep coastal fjords and to show that all aspects of the tsunami waves are readily explained by the dock failure and subsequent submarine slide.

Morphology and Flow Slide Processes Skagway Harbor is a drowned hanging valley branch-

ing off Taiya Inlet (Fig. 1). The results of a detailed geomorphologic study of the harbor (Terra Surveys, pers. comm., 1998) following the dock failure are in good agreement with results of an independent geomor- phologic examination by Cornforth and Lowell [1996]. The upper slope that failed beneath the PARN Dock

Tide Gauge

PARN T a i y a f Dock / Inlet

%% i09m f I 0.5 km

/

135o21 , 135o20 ,

Slide Area

59* 27'30"

59 27'00"

59* 26'30"

135019 '

Figure 1. Northern part of Taiya Inlet, showing Skag- way Harbor and dock slide area. Dashed line denotes the limit of the harbor model domain. Inset shows the location of the inlet model domain and the hypotheti- cal slide proposed by Mader [1997] and Kowalik [1997] (numbers give assumed uplifts and subsidences (m))

3009

3010 RABINOVICH ET AL.- LANDSLIDE-GENERATED TSUNAMI

had gradients of 300 to 350 . The sediment section consisted of grey, soft-to-medium stiff, sandy, slightly clayey silt interstratified with minor variations of sand and clay. These marine sediments were underlain by very coarse colluvium in the nearshore area and by steeply dipping bedrock farther down the slope [Corn- forth and Lowell, 1996].

The failure took place after part of the mid-section of the original dock had been removed during construction and replaced with four 10-m wide cellular bulkheads (cells). At the time of the event, pilings were still being removed from the empty cells and only minor backfilling had been initiated. Some water was trapped inside the cells above the tidal level. The failure appears to have begun in the vicinity of the cells and to have swept rapidly southward along the original dock, reaching its southern limit a short distance beyond the dock. The flow slide accelerated down the steep slope of the fjord and was diverted southwards through the trough at the base of the slope into Taiya Inlet. Sediment was eroded from the trough and carried along with the flow slide, resulting in the deepening of the trough by more than 20 m locally (Fig. 2). A similar failure had taken place on October 29, 1966 (also at extreme low tide) immediately south of the original dock as a result of fill construction [Cornforth and owett, 1996].

Numerical Simulation

o : 4 6 8(m) 4 2 0 -2 -4 iscc ! sc

IOscc 10scc

50sec

Figure 3. Animation frames from the harbor model simulations at 1, 10, and 50 s after the dock failure. (left panels) Movement of the slide body; and (right panels) propagation of the tsunami waves. The initial tsunami consists of a leading crest (red) followed by a trough (blue).

We have simulated landslide-generated waves in Skag- way Harbor using a three-dimensional numerical model for a viscous landslide with full slide-wave interactions proposed by Jian# and eBlond [1994]. The model was generalized to include the subaerial slide and actual bot- tom topography [Fine et al., 1998]. The model starts from rest using measured properties of the landslide and established equations of motion for both the waves and submarine landslide. A finite-difference method was used to solve the non-linear shallow water equations, with a one-dimensional radiation condition applied at the open boundary.

Figure 2. Depth changes (m) in the area of the dock failure. The thick solid line marks the main slide area used for the extended Taiya Inlet model. The dashed line marks the subaerial part of the slide.

The Skagway Harbor model was used to determine the wave structure immediately following the PARN Dock collapse. Based on data from Terra Surveys Inc. (pers. comm., 1998) and Cornforth and Lowell [1996], we take the slide density to be 2.0 g cm -a, slide viscos- ity to be 0.05 m 2 s -, and assume that the initial slide covered a rectangular area of 330 m x 160 m (Fig. 1) with parabolic cross-sections for both horizontal axes. The mean thickness of the slide body was assumed to be 15 m, in accordance with the estimated slide volume of 0.8 x 106 n a. The subaerial slide accounted for 10% of the total slide volume. The localized Skagway Har- bor model covers a 260 x 160 grid having grid steps Ax = Ay: 5 m and time steps At = 0.0167 s.

Figure 3 shows snapshots of the modeled slide body movement and associated slide-generated surface waves. The dynamics of the computed slide motion correspond well to the results of the geomorphologic study pre- sented in the previous section (Fig. 2). The leading wave, propagating in front of the slide, arrived at the ferry dock and NOAA tide gauge site as a positive wave (crest), in agreement with the tide gauge record (cf. Ku- likov e! al. [1996]; Lander [1996]), and with the results of the laboratory modeling by Raichlen et al. [1996]. In contrast, the hypothetical massive slide in Taiya Inlet proposed by Mader [1997] and Kowalik [1997] causes a trough to arrive first at the harbor site. To account for the observed leading crest in the N OAA tide gauge record for November 3, 1994, Kowalik [1997] has to rely on an analysis by Nottingham [1997] which claims that the leading crest was associated with "... an almost instantaneous atmospheric pressure change ... of two inches of Mercury ... caused by the large crest wave" in Taiya Inlet (two inches of Mercury m 67 mb). We fur- ther note that the surface waves propagate much faster than the slide moves (cf. Fig. 3) and there is no pos-

RABINOVICH ET AL' LANDSLIDE-GENERATED TSUNAMI 3011

sibility of resonance coupling (in contradiction to the assumption by Kowalik [1997]). The lead wave crest was followed by a broad wave trough (Fig. 4a).

Tsunami Spectra and Energy Decay The localized Skagway Harbor model is limited be-

cause simulated waves are allowed to escape without reflection through the open boundary at the entrance to the harbor. In reality, abrupt changes in the domain width and water depth at the harbor entrance will re- flect waves back into the harbor to produce standing oscillations. These abrupt changes presumably account for the protracted "ringing" of the tsunami waves record- ed by the tide gauge. We have examined these effects by increasing the computational domain to include the northern half of Taiya Inlet (Fig. 1, inset). The ex- tended model has grid dimensions 763 x 311, grid steps Ax = Ay = 10 m and time step At = 0.0385 s. Two different initial source domains have been used for this extended Taiya Inlet model: (1) The rectangular slide used in the previous section for the Skagway Harbor domain, and (2) a more realistic slide domain based on data from Terra Surveys (Fig. 2). The slide volumes (m 0.8 x 106 m a) and results for both models are similar, with model (2) yielding wave heights about 15% higher than model(i).

The model simulation of the NOAA tide gauge record (Fig. 4b) is similar to the observed sea level record (Fig. 4a) with two major differences. First, the simu- lated wave heights are about 2.5 times greater than ob- served and second, the simulated records contain signif- icant high-frequency oscillations which are absent in the observed record. These differences appear to be related to the response characteristics of the gauge. Raichlen et al. [1996] present a laboratory calibration of a nitrogen

3

-1

19

Observed

(tide gauge record) i I i i

_(b) Simulated (no adjustment)

i i i i i I i i

_(c) Simulated (adjusted for 3/4 turn)

_

i i i i i I i i 20

November 3, 1994 AST (hr)

Figure 4. Observed and simulated tide gauge records for the NOAA site. (a) Observed tide gauge record af- ter subtraction of the astronomical tides; (b) simulated water level records from the numerical model; (c) as in (b) but corrected for a a/4-turn valve opening.

1.0

104 95%

,o102

,, S'm lat u ed- {no'adjustment) \..,'11 lu ......... Simulated (adjusted for 3/4 turn) I . , . i . , . i . ,.1 i i i i I . , . i . , .

Frequency (cpm) Figure 5. Spectra of observed and simulated tsunami records in Skagway Harbor. The inset shows the ad- mittance function for the tide gauge with 3/4-turn valve opening (from Raichlen et al. [1996]).

bubbler tide gauge of the type used in Skagway. Re- sults show that the dynamic response of the gauge is a function of the needle valve opening located at the end of a gas-filled tube. We used the calibration curve (admittance function) for a valve opening of 3/4 turn (Fig. 5) which, according to NOAA [Raichlen et al., 1996], is the nominal setting for the gauge. The direct and inverted Fourier transforms allowed us to "correct" the simulated sea level records by mimicking the effect of the gauge opening. The corrected simulated record is presented in Fig. 4c and the spectra for the observed, simulated and adjusted records are presented in Fig. 5. Adjustment of the simulated waves effectively suppress- es the high-frequency oscillations except at the begin- ning of the record. For the actual tide gauge record, high-frequency oscillations at the leading edge of the tsunami would have been damped by the non-linear gauge-response effects.

An encouraging feature of our model is the close agreement between the corrected sea level records (Fig. 4c) and the observed sea level record (Fig. 4a). The fundamental observed period of 3.0 min agrees well with the computed period for the dominant oscillations (Fig. 5). Smaller wave heights of the simulated 3-rain oscilla- tions and the presence of high-frequency oscillations in the adjusted record (Fig. 4b) but not in the observed record (Fig. 4a) suggest that the admittance amplitude of the NOAA gauge is steeper than shown in Fig. 5.

The observed residual (nontidal) oscillations and the simulated oscillations adjusted to the 3/4-turn valve opening were used to analyze the heights and periods of individual waves and to estimate the Q-factor for the harbor motions; Q = r/ST, where 5 is the decay coeffi- cient and T is the wave period. The estimated Q values for the dominant oscillations were large for model and observations (Fig. 6). Specifically, Q = 24, 5 m 2.7 h - for the observed waves and Q 21, 5 m 3.0 h - for the simulated waves, suggesting that wave energy leaks slowly from Skagway Harbor into Taiya Inlet. The

3012 RABINOVICH ET AL.: LANDSLIDE-GENERATED TSUNAMI

'a, , 1 = 2.7 hr '1 I

150 r]l,'[3 Q= 24 I X& = 3.0 h r"l - Q = 21 [ t

0 / I I I I I I I I I I I I

2

0 20 November 3, 1 4 AST (hr)

igure 6. Variation in observed and simulated wave heights (top pane]s) and periods (lower pane]s) at the time of the Skagway Harbor tsunami. Estimates are based on successive crests to troughs (triangles) and troughs to crests (squares). The fitted exponential func- tions approximate the wave height decay.

sharp contrast between the harbor depth (m 10 m) and the inlet depth ( 200 m) causes the original tsunami energy to be trapped in the harbor.

Discussion and Conclusions

Geomorphologic data, analysis of the tide gauge rec- ord, and numerical simulation of the PARN Dock col- lapse and subsequent wave propagation reveal that the tsunami of November 3, 1994 in Skagway Harbor was generated by the collapse of the dock at a time of ex- treme low water. Kulikov et al. [1998] provide a physi- cal mechanism relating extreme low tide and landslides, and show that this mechanism is applicable only for sub- aerial slides. The strong correlation between extreme low tide and subaerial landslides is an additional argu- ment that the 1994 tsunami was induced by the dock failure and not by an external submarine slide.

The results of our numerical computations for the combined fine- and coarse-grid models are in good agree- ment with observational data. The computed period of the fundamental mode of 3.0 min for Skagway Harbor is nearly identical to the observed period. Estimated and observed Q-factor values suggest a significant tsunami energy retention in the harbor due to the large contrast in depth between the harbor and inlet. Both our numer- ical model and the tide gauge record show that a wave crest arrived first at the tide gauge, contrary to the models proposed by Mader [1997] and Kowalik [1997] which require that a wave trough arrive first. To ex- plain this contradiction, Nottingham [1997] argues that a sudden atmospheric pressure drop of 67 mb occurred seconds before the arrival of the tsunami event. Not

only does this inverted barometer effect not apply at such short time scale (tens of seconds) but also pressure changes of this magnitude can only be associated with major atmospheric phenomena, none of which were ob- served at the time of the tsunami event. Our model of the submarine landslide associated with the dock failure in Skagway Harbor accounts for all aspects of the ob- served wave field without any additional assumptions concerning simultaneous, hypothetical geophysical or hydrometeorological events in the adjoining inlet.

References

Cornforth, D. H., and J. A. Lowell, The 1994 submarine slope failure at Skagway, Alaska, in Landslides, edited by K. Senneset, pp. 527-532, Balkema, Rotterdam, 1996.

Fine, I. V., A. B. Rabinovich, E. A. Kulikov, R. E. Thomson, and B. D. Bornhold, Numerical modelling of landslide- generated tsunamis with application to the Skagway Har- bor tsunami of November 3, 1994, Proc. Tsunarni Syrnp., Paris, 1998.

Jiang, L., and P. H. LeBlond, Three-dimensional modeling of tsunami generation due to a submarine mudslide, J. Phys. Oceanogr., 24{(3), 559-572, 1994.

Kowalik, Z., Landslide-generated tsunami in Skagway, Alas- ka, Sci. Tsunarni Hazards, 15(2), 89-106, 1997.

Kulikov, E. A., A. B. Rabinovich, R. E. Thomson, and B. D. Bornhold, The landslide tsunami of November 3, 1994, Skagway Harbor, Alaska,. J. Geophys. Res., 101(C3), 6609-6615, 1996.

Kulikov, E. A., A. B. Rabinovich, I. V. Fine, B. D. Bornhold, and R. E. Thomson, Tsunami generation by landslides at the Pacific coast of North America and the role of tides, Oceanology, 38(3), 323-328, 1998.

Lander J. F., Tsunamis Affecting Alaska, 1737-1996, 195 pp., Natl. Geophys. Data Center, Natl. Oceanic and At- mos. Admin., Boulder, Colo., 1996.

Mader C. L., Modeling the 1994 Skagway tsunami, Sci. Tsunarni Hazards, 15(1), 41-48, 1997.

Nottingham, D. P. E., The 1994 Skagway tsunami tide gage record, Sci. Tsunarni Hazards, 15(2), 81-88, 1997.

Raichlen, F., J. J. Lee, C. Petroff, and P. Watts, The gener- ation of waves by a landslide: Skagway, Alaska - A case study, Proc. 25th Coastal Eng. Conf., ASCE, Orlando, Florida, 1996.

B. D. Bornhold, I. V. Fine, E. A. Kulikov and A. B. Rabinovich, International Tsunami Research, Inc., 11321 Chalet Road, Sidney, BC, VSL 5M1, Canada. (e-mail: itri@ii.ca)

R. E. Thomson, Fisheries and Oceans Canada, Institute of Ocean Sciences, 9860 West Saanich Road, Sidney, BC, VSL 4B2, Canada.

(Received April 9, 1999; accepted June 24, 1999.)