Globally Scattered 2011 Tohoku Tsunami Waves From a Seafloor Sensor Array in the Northeast Pacific Ocean

Globally Scattered 2011 Tohoku Tsunami Waves From a

Sea

oor Sensor Array in the Northeast Paci

c Ocean

Monica D. Kohler

, Daniel C. Bowden

2,3

, Jean

‐

Paul Ampuero

2,4

, and Jian Shi

Department of Mechanical and Civil Engineering, California Institute of Technology, Pasadena, CA, USA,

Seismological

Laboratory, California Institute of Technology, Pasadena, CA, USA,

Now at Institute of Geophysics, ETH Zürich, Zürich,

Switzerland,

Now at Université Côte d'

’

Azur, IRD, CNRS, Observatoire de la Côte d'Azur, Géoazur, France

Abstract

Modeling of tsunami wave propagation for forecasting focuses on the arrival time and

amplitude of the earliest tsunami waves reaching coastlines. The complex later tsunami wave

eld, in

which scattering is predominant, poses additional hazards due to possible constructive interference of

coherent packets of wave energy. However, almost no data sets exist to characterize the geographical sources

and temporal evolution of the scattered waves. Here we show how recordings of the 2011 Tohoku tsunami by

an array of pressure gauge sensors in the northeastern Paci

c Ocean reveal coherent waves that are

produced by scattering from distant coastlines including South America and Antarctica, as well as multiple

sea

oor fracture zones, ridges, and island chains. Multiple signal classi

cation analysis and backward

propagation ray tracing provide tight constraints on the origin of each scattered phase and resolve

simultaneous wave arrivals from different scatterers. Incoming waves from constant back azimuths occur

over time durations of several hours, revealing the time persistence of speci

c geographical scatterers. The

results can advance numerical predictions of tsunami wave impact because they provide direct evidence

for the necessity of incorporating both local and distant bathymetry over a range of length scales and for long

time durations, to account for the azimuthal dependence of scatterer strength.

Plain Language Summary

An array of sensors on the sea

oor off the coast of California

recorded the global tsunami produced by the 2011 magnitude 9.0 Tohoku, Japan, earthquake. The

tsunami data recorded across the array show multiple, coherent waves arriving between 1 hr and more than

40 hr after the initial tsunami wave. Such long

‐

duration tsunami waves are of concern because they arrive

at distant coastlines long after the

rst, direct wave arrival. They can become ampli

ed as they are

uenced by the geometries of ports and harbors, sometimes causing damaging runup and inundation. This

study involved a technique that takes advantage of similarities in data among the sensors to investigate

which sea

oor structures in the Paci

c Ocean caused the long

‐

lasting, later

‐

arriving tsunami waves. By

projecting the waves backward in space and time, speci

c structures were identi

ed by pinpointing the

geographical directions from which each wave arrived. The structures include distant coastlines, including

those along South America and Antarctica, as well as many sea

oor fracture zones, island and seamount

chains, and mid

‐

ocean ridges. Unless later

‐

arriving wave amplitudes are adequately modeled over the

deep, open ocean regions through which a tsunami wave propagates, their effects on distant coastlines may

be severely underestimated.

1. Introduction

Accurate identi

cation of large

‐

amplitude scattered tsunami wave arrivals has implications for forecasting

because it enables effective, time

‐

evolving tsunami warnings for speci

c stretches of distant coastline.

Tsunami modeling in operational warning systems is necessarily based on computationally ef

cient but sim-

pli

ed procedures that focus on the main

rst arrival. However, later tsunami arrivals have been damaging

in a number of cases, especially if the population is unaware of the persistence time of the hazard

(Barberopoulou et al., 2014; Contreras et al., 2016; Koshimura et al., 2008; Kowalik et al., 2008;

Pattiaratchi & Wijeratne, 2009). Similar simplifying assumptions are pervasive in the way tsunami data

are currently used to infer earthquake source properties: Usually, the

rst tsunami arrivals are typically con-

sidered in earthquake source inversions, though recent studies focus on how later tsunami arrivals can also

be used to investigate the earthquake source process (Gusman et al., 2017; Kubota et al., 2018; Suppasri

et al., 2017). While the governing equations of trans

‐

oceanic tsunami propagation are well established,

RESEARCH ARTICLE

10.1029/2020JB020221

Key Points:

•

Coherent, long

‐

duration 2011

Tohoku tsunami coda waves were

recorded by a dense pressure gauge

array in open ocean offshore

California

•

The sea

oor sources of tsunami

wave scattering were identi

ed by

application of a high

‐

resolution

beamforming and backprojection

method

•

Sources of scattering include South

America and Antarctica coastlines,

multiple sea

oor fracture zones,

ridges, and island chains

Supporting Information:

•

Supporting Information S1

•

Supporting Information S2

Correspondence to:

M. D. Kohler,

kohler@caltech.edu

Citation:

Kohler, M. D., Bowden, D. C.,

Ampuero, J.

‐

P., & Shi, J. (2020).

Globally scattered 2011 Tohoku

tsunami waves from a sea

oor sensor

array in the northeast Paci

c Ocean.

Journal of Geophysical Research: Solid

Earth

125

, e2020JB020221. https://doi.

org/10.1029/2020JB020221

Received 19 MAY 2020

Accepted 25 OCT 2020

Accepted article online 30 OCT 2020

KOHLER ET AL.

1of14

some implications of the theory have received attention only after limitations of the available data were over-

come. For instance, it is thanks to recordings of huge tsunamis in the past decade that research on

long

‐

period tsunami dispersion has gained new impetus (Tsai et al., 2013). Munk (1963) was among the

rst

to identify tsunami scattering properties such as reverberations, decay times, and basin

‐

wide energy diffu-

sion using long

‐

duration data from the 1960

9.5 Chile earthquake (Miller et al., 1962). Scattering by

mid

‐

ocean bathymetry features is theoretically expected to affect tsunami waves, modifying their ampli-

tudes, energy packet travel times, duration, and propagation direction (Mofjeld et al., 2000, 2001, 2004;

Saito & Furumura, 2009). However, it is dif

cult to validate the theory of tsunami scattering, primarily

because only a handful of large or regional tsunamis have been recorded on pressure gauge arrays with inter-

station spacing small enough, or for long

‐

enough continuous time durations (Fukao et al., 2008; Gusman

et al., 2016; Kubota et al., 2020; Matsumoto et al., 2017; Mizutani & Yomogida, 2019; Sandanbata et al., 2018;

Sheehan et al., 2019; Thomson et al., 2011; Wang et al., 2019), to resolve the source of scattered waves.

Shorter

‐

period waves (~10

–

30 min) in particular are in

uenced by the geometries of ports, harbors, and mar-

itime facilities; this dependence, in turn, results in later

‐

arriving, hazardous, tsunami

‐

induced effects such as

strong, persistent currents, jets, and large eddies (Borrero et al., 2015). In addition, this period band encom-

passes the range of periods seen in waves causing damaging runup and inundation in different locations

around the world (Okal, Fritz, Raad et al., 2006; Okal, Fritz, Raveloson et al., 2006). This band is complemen-

tary to long

‐

period tsunami observations (Heidarzadeh & Satake, 2014), but it is much more rarely observed

in arrays of multiple sensors because of limited sample rates that are almost always too low and limited

recording time durations where records either stop after a few hours or are further severely downsampled

at the longer times.

Here we present a high

‐

resolution analysis of a unique data set with the aim of advancing our understanding

of tsunami wave scattering. The 11 March 2011

9.0 Tohoku, Japan, earthquake resulted in a major tsu-

nami that propagated across the Paci

c Ocean. A temporary array of ocean bottom seismometers (OBS) and

differential pressure gauges (DPG) was deployed far off the coast of Southern California (Figure 1a), where it

recorded the subsequent tsunami with unprecedented spatial and temporal resolution (Figure 1b) (Kohler &

Science Team, 2010, 2011; Lin et al., 2015). This array comprised the ALBACORE (Asthenospheric and

Lithospheric Broadband Architecture from the California Offshore Region Experiment) deployment and

spanned a region that was 150 km north

‐

south by 400 km east

‐

west. It extended into the deep open

Paci

c Ocean, west of the continental shelf edge. In this array, 22 stations with an average spacing of

75 km were equipped with DPGs that recorded waveform time series data continuously at 50 samples per

second (sps). The spatial and temporal resolution of the tsunami data enabled by the dense ALBACORE

array con

guration makes it possible to investigate features in the tsunami waveforms, especially the later

but still prominent and coherent tsunami waves, which have not previously been observed with this level

of clarity on these spatial and temporal scales.

The ability to identify potential sources of sea

oor and coastline topographic scattering of tsunami waves is

necessary to characterize the physical mechanisms of the long

‐

duration scattering process. Classical beam-

forming has been previously applied to tsunami waveforms recorded by arrays of sensors to determine the

direction

‐

arrival of the incoming wave energy packets (e.g., Hanson & Bowman, 2005; Shi et al., 2017);

however, it has limited ability to discriminate between true signals and noise unrelated to the tsunami

energy, even after narrowband

ltering. Nor does it distinguish between contemporaneous arrivals of energy

from different scattering sources. In this study, we show how the application of a high

‐

resolution array

‐

pro-

cessing technique to the dense ALBACORE tsunami data set leads to the identi

cation of scattered tsunami

wave arrivals and the accurate estimation of their direction

‐

arrival (back azimuth). We apply the Multiple

Signal Classi

cation (MUSIC) beamforming method (Schmidt, 1986) to the Tohoku tsunami data, because

of its increased resolution in determining the direction

‐

arrivals over conventional delay

‐

and

‐

sum and

coherence beamforming methods.

The speci

c sea

oor bathymetric structures that caused scattering are then identi

ed through backprojec-

tion (reverse wave tracking) based on the direction

‐

arrival estimated from narrow band tsunami wave sig-

nals. Backprojection techniques have been applied to many

elds including seismological and

oceanographic phenomena to de

ne spatiotemporal features of source processes (Hayashi et al., 2011;

Heidarzadeh & Satake, 2014; Meng et al., 2011; Zhang et al., 2009). The spatial and temporal evolution of

10.1029/2020JB020221

Journal of Geophysical Research: Solid Earth

KOHLER ET AL.

2of14