ADVAKFD IUMFIZICAL MODFLS f OIZ

SIMULATlllCI TSUI1AMI WAVFS A I D RUllUP

This page intentionally left blank
 ADVANCES IN
COASTAL AND OCEAN ENGINEERING
ADVAllCFD llUMFRICAL MODFLS f OR
SIMULATIllB TSUllAMI WAVFS AllD RUllUP
VOLUME 1 0

Editors

Philip 1.-F. liu

Cornell University, USR

Harry Yeh
Oregon State University, USR

Costas Synolakis
University of Southern California, USR

r pWorld Scientific
N E W JERSEY LONDON SINGAPORE - BElJlNG SHANGHAI - HONG KONG - TAIPEI - CHENNAI
A-PDF Merger DEMO : Purchase from www.A-PDF.com to remove the watermark

Published by
World Scientific Publishing Co. Pte. Ltd.
5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

ADVANCED NUMERICAL MODELS FOR SIMULATING

TSUNAMI WAVES AND RUNUP
Copyright © 2008 by World Scientific Publishing Co. Pte. Ltd.
All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.

ISBN-13 978-981-270-012-4
ISBN-10 981-270-012-9

Printed in Singapore.

HueyLing - Adv Numerical Models.pmd 1 8/26/2008, 5:07 PM

August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Preface

Preface

This review volume grew out of the third international workshop

on long-wave runup models that took place June 11–12, 2004 at the
Wrigley Marine Science Center of the University of Southern California on
Catalina Island, California. This workshop and the two previous workshops
were sponsored by the Geo-Hazard program of the National Science Foun-
dation of the United States. Dr. Clifford J. Astill was the director of the
program.
The first long-wave runup workshop was also held in the Wrigley Ma-
rine Science Center, on August 15–17, 1990. Tsunami runup/inundation
and the hazard mitigation were the main topics discussed in the first work-
shop, which brought together researchers from Japan, Russia, England,
and the United States. The proceedings of the workshop were reported by
Liu, et al. (1991). The workshop revitalized research efforts on long-wave
runup, spawning several major research programs on long-wave runup in
the United States as well as overseas. The workshop has also facilitated
further international research collaborations.
From 1992 to 1994 several major tsunamis occurred, including the
Nicaraguan tsunami in 1992; the Flores Island (Indonesia) tsunami in 1992;
the Okushiri (Japan) tsunami in 1993. All three tsunamis caused devas-
tating property damages and many deaths. Moreover, in 1994 alone four
additional tsunamis, including the East Java (Indonesia), the Shikotan Is-
land (Russia/Japan), the Mindoro (Philippines), and the Skagway (Alaska,
USA) tsunamis, occurred around the world. The runup heights along af-
fected coastlines were surveyed and documented by various research teams.
Different research groups have performed numerical simulations of all these
events with different numerical models. The community felt the strong
need to gather researchers together to discuss similarities and dissimilarities
among these models and to discuss the issues in modeling coastal effects of
tsunami.
Consequently, the Second International Workshop on Long-Wave Runup
Models was held at Friday Harbor, San Juan Island, Washington, on
September 12–16, 1995. The participants in the workshop were from Japan,
Russia, England, Brazil, Australia, Canada, Indonesia, and the United

v
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Preface

vi Preface

States. The format of the second workshop was designed to focus more on
discussions than on formal presentations. Four benchmark problems were
selected before the workshop so that numerical models can be compared,
evaluated and discussed among the participants. During the workshop,
seven discussion themes were organized as follows: laboratory, analytical,
finite-difference, finite-element, vertical-plane models, boundary-integral-
element models, and marker-and-cell models. The presentations and dis-
cussions were edited and published in a book entitled “Long-Wave Runup
Models” (Yeh, et al. 1996).
There was no doubt that the benchmark-problem exercises used in the
second workshop proved extremely useful in identifying absolute and com-
parative modeling capabilities. Overall, in terms of tsunami runup mod-
eling, significant advances had been made between two workshops, due
mainly to the advancement of computational capabilities, and because of
the generation of a large 2-D and 3-D laboratory data set and the fortuitous
field measurements in 1992–1995, all of which have contributed to model
calibrations. The tsunami modeling efforts had become more directed to-
wards their implementation for real tsunami predictions and hind-castings
than ever before. At the time of the 1990 Catalina workshop, large differ-
ences between computed runup results and field measurements might have
been attributed to both errors in the seismic estimates of the source mo-
tion and to the hydrodynamic calculations. Much advances had been made
by the 1995 Friday Harbor workshop; researchers became much more con-
fident in the hydrodynamic calculations, at least for non-breaking waves.
It was equally clear that reduction and even elimination of numerical dis-
persion and numerical dissipation effects would — if not already — soon
become reality. At the same time, the workshop participants recognized
additional and important problems arising from modeling improvements,
such as determination of highly accurate initial wave conditions, modeling
the three-dimensional flow effects, and turbulence. Actual tsunami runup
motions are turbulent; the runup flow patterns, impacts, scouring effects,
and sediment transport are all affected by turbulence in the runup motions.
From 1995 to the summer of 2004, there have been at least six additional
large tsunamis resulting in catastrophic loss of life and property. They are
the Irian Jaya tsunami (Indonesia) in 1996, the Peru tsunamis in 1996 and in
2001, the Papua New Guinea tsunami in 1998, the Turkey tsunami in 2000,
and the Stromboli (Italy) tsunami in 2002. Among these six tsunamis, the
Turkey tsunami was definitely caused by land subsidence and slides associ-
ated with earthquakes, and the Stromboli tsunami was caused by landslides
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Preface

Preface vii

caused by volcanic eruption. On the other hand, the source of the Papua
New Guinea tsunami, which killed more than 2000 people and destroyed
completely three villages, remains controversial and has been postulated as
due either to co-seismic seafloor dislocation or sediment slump. Because of
the occurrence of these tsunamis, research interest and efforts on the mod-
eling of landslide generated tsunamis have been intensified in recent years.
The landslide-generated tsunamis have very different characteristics from
the tsunamis generated by earthquakes. The traditional depth-integrated
shallow water equations are not always adequate for modeling the landslide
generated tsunamis. Other noteworthy advances in recent years are the de-
velopment of several computational fluid dynamics models calculating the
nearshore waves and their interactions with structures with the considera-
tion of frequency dispersion and turbulence.
The primary objectives of the third workshop were to provide a
platform for discussing both old and new numerical models and their ap-
plications to various critical issues concerning tsunami runup and wave-
structure interactions. To accomplish this goal, four benchmark problems
were selected and posted on the workshop website before the workshop.
http://www.cee.cornell.edu/longwave/. The workshop participants were
given the solutions, in the form of laboratory data or analytical solutions,
a week before the workshop. The workshop participants discussed their nu-
merical model and the comparisons between their numerical results and
solutions for the benchmark problems. Their presentations are also posted
on the workshop website. The four benchmark problems are as follows:

1. Calculations of the moving shoreline.

2. Tsunami runup onto a complex three-dimensional topography.
3. Landslide generated tsunami.
4. Tsunami forces on a nearshore structure.

This review volume is divided into two parts. The first part includes
five review papers on various numerical models. Pedersen provided a brief
but thorough review on the theoretical background for depth-integrated
wave equations, which are employed to simulate tsunami runup. LeVeque
and George describe high-resolution finite volume methods for solving the
nonlinear shallow water equations. They have focused their discussion on
the applications of these methods to tsunami runup. In recent years, sev-
eral advanced 3D numerical models have been introduced to the field of
Coastal Engineering to calculate breaking waves and wave-structure inter-
actions. These models are still being developed and are at different stage of
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Preface

viii Preface

maturity. Roger and Dalrymple discussed the Smooth Particles Hydrody-

namics (SPH) method, which is a meshless method. Wu and Liu presented
their Large Eddy Simulation (LES) model for simulating the landslide gen-
erated waves. Frandsen introduced the Lattice Boltzmann method with the
consideration of a free surface.
The second part of the review volume contains the descriptions of the
benchmark problems and twelve extended abstracts submitted by the work-
shop participants. All these papers compared their numerical results with
benchmark solutions.
Professor Synolakis and his staff of the University of Southern Califor-
nia were responsible for the local arrangements. Their efforts are deeply
appreciated. The workshop was sponsored by the US National Foundation.

Philip L.-F. Liu, Harry Yeh and Costas Synolakis

References
Liu, P. L.-F., Synolakis, C. and Yeh, H., A report on the international
workshop on long wave runup, J. Fluid Mech. 229 (1991), 678–88.
Yeh, H., Liu, P. L.-F. and Synolakis, C. (ed.), Long-Wave Runup Models
(World Scientific, 1996), 403p.
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

A Dedication to Dr. Cliff J. Astill

This workshop was the third of a series of international workshops, which

discussed the state of arts of analytical and numerical models for tsunami
propagation and runup. The first workshop was held in 1990 on Catalina
Island — the same venue as the present workshop — followed by the second
workshop in 1995 at Friday Harbor, Washington. All three workshops have
been supported by the Geo-Hazard program of the Civil and Mechanical
Systems Division at the US National Science Foundation, which was under
the directorship of the late Dr. Cliff J. Astill, who passed away on 23rd of
December 2004 after a long and courageous fight against cancer. During
his tenure as the Program Director, Dr. Astill was a proactive supporter of
tsunami research, which has not gained as much attention as other earth-
quake research areas do. Coincidentally two days after Dr. Astill’s death,
the Great Indian Ocean Tsunami tragically struck the entire Indian Ocean,
sending tsunami awareness around the globe and urging more accelerated
research in tsunamis. This book is dedicated to Dr. Astill with our deep
appreciation to his effort in promoting and guiding tsunami research.
Dr. Astill was born on March 4, 1936 in Brisbane, Australia. He received
his Bachelor of Engineering degree in Mechanical Engineering from the
University of Queensland in 1957. After the graduation, he worked in both
weapons research establishments and aircraft design industry for almost 11
years. In 1965 Dr. Astill immigrated to the United States after working in
England and Canada. He received his Master of Engineering and Doctoral of
Science (ScD) degrees in Engineering Mechanics from Columbia University
in 1968 and 1971, respectively.
Dr. Astill started working at NSF in June 1971 as an associate program
director in the Engineering Mechanics Program, which was then under the
Research Directorate. He became the Program Director for Solid Mechanics
Program in the Engineering Division in the late seventies and then the
Program Director of Sitting & Geotechnical Systems, Earthquake Hazard
Mitigation in the Division of Civil and Mechanical Systems in the early
eighties. The program has gone through several name changes in the eighties
and nineties. Although NSF started funding tsunami research after the 1960
Chilean tsunami, tsunami research had not received consistent and steady

ix
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

x A Dedication to Dr. Cliff J. Astill

funding until Dr. Astill took over the program. Dr. Astill’s enthusiasm and
steady support in tsunami research over the last 20 years is the catalyst
for all the achievement made to date. He has indeed been a mentor for the
US tsunami research community until his retirement from NSF in January
2004.
Dr. Astill had a vision that tsunami research ought to be interdisci-
plinary. On one hand, he supported the fundamental research in develop-
ing mathematical models and performing basic laboratory experiments and
numerical simulations for various aspects of physical processes related to
tsunami generation, propagation and runup. On the other hand, he rec-
ognized the importance of the hazard mitigation practice by supporting
research in the areas of social sciences, policy-making and planning, and
data management. Fifteen years ago, he initiated such multi-disciplinary
activities by strongly encouraging researchers to submit cooperative group
proposals, which were unheard of in those days. Dr. Astill was always proac-
tive in promoting promising young researchers. In many occasions, using
himself as a source of information, he introduced young researchers to senior
researchers and encouraged them to form a possible collaborative team.
Dr. Astill also firmly believed that an effective means to develop multi-
discipline and cross-disciplinary research programs is through informal
small group meetings and formal workshops. Over the years he has sup-
ported a numerous such meetings and workshops (see the list below).
Through his encouragement, the tsunami community has expanded its
membership to include not only hydrodynamists and seismologists but
also geologists, social scientists, oceanographers, mathematicians, and data
managers. When we look back, Dr. Astill led us to the research mode of “col-
laboratory” more than 15 years ago, which is the fundamental philosophy
of NSF NEES project (Network for Earthquake Engineering Simulations:
http://www.nees.org/index.php). Dr. Astill was not afraid of offering his
opinions and making suggestions to workshop organizers. For example, the
procedure of “Blind Tests” in validating numerical models was first recom-
mended by Dr. Astill and implemented in the 1995 Second International
Workshop on Long-Wave Runup Models at Friday Harbor. Prior to the
workshop — April 1994 — Dr. Astill wrote:

“I don’t know the details, so the following remarks may not

be relevant. However in a recent effort in the area of using
blind predictions (Class A experiments) to validate the proce-
dures used to determine the liquefaction potential of soils, there
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

A Dedication to Dr. Cliff J. Astill xi

were some surprises. (Cliff was referring to VELACS project

(Verification of Liquefaction Analysis by Centrifuge Studies):
http://www.cee.princeton.edu/ radu/soil/velacs/) First, I should
explain these Classes of predictions. Class A: A “single-blind” ex-
periment — using the parlance from the world of medical research
— where the predictions are made before the experiments are run.
Class B: Predictions made AFTER the experiment is run, but with
the experimental results not known to the predictor. Class C: The
predictor knows the results of the experiments to be “predicted”.
There is no class, as far as I know, corresponding to the “double-
blind” experiment of the medical researcher. There is a role for each
of these classes. In the exercise referred to, after the experiment was
run, the predictors were given the actual configuration of the ex-
periment (the prediction was based on the nominal configuration),
but these results were then Class B predictions: the experiment had
been run, but the predictors were still unaware of the results from
the experiment. Finally, there was still a lot of work to do after the
experimental results were made public, and these were then classi-
fied as Class C predictions. Well, so much for naive expectations. In
practice, there were accusations that some predictors had used the
results from prior similar experiments; even that some, who were
also experimentalists, had actually run the experiment themselves
according to the nominal configuration on which they were to make
their predictions! As I said, maybe none of this is germane to your
proposed endeavors. But if it is, do you have specific plans to try
to guarantee that blind predictions are truly blind?
This reminds me of an incident in my early days at NSF, which
overlapped with the last year or two of the mini-skirt fashions. A
blind flower seller used to sit just outside our building on G Street,
selling flowers. One day, a young lady in an extremely short dress
was observed to walk by this blind flower seller, whose head tracked
her passing like it was locked on by a radar beam. Subsequently, I
discovered that there are different degrees of being officially blind.
To be classified as officially blind does not imply the total absence
of sight. So there is the warning.”

Dr. Astill was a strong supporter for the reconnaissance field surveys for
tsunamis. Since tsunamis are rare and high impact events, lessons learned
from the field surveys are invaluable. Dr. Astill believed that the strong
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

xii A Dedication to Dr. Cliff J. Astill

links between the real-world observations and the laboratory and/or theo-
retical understandings are crucial for natural hazards research. Based on the
field observations, many topics of fundamental research were identified and
several theoretical hypotheses were validated. Dr. Astill’s effort to support
reconnaissance tsunami survey began after the 1992 Nicaragua Tsunami,
which was the first major tsunami since 1983 Nihonkai Chubu Tsunami.
He developed several options, which enabled researchers to quickly go to
the filed without being burden by paper works. He clearly understood the
importance of the survey and its critical timing for capturing perishable
data. The tsunami community made a numerous tsunami surveys during
his tenure (see the list below), which resulted in substantial advances in
tsunami research.
Dr. Astill also proactively encouraged international collaboration by
inviting international experts to serve on the advisory committees for sev-
eral group research projects. He also suggested that we should participate
more in international conferences and meetings: for example, the Interna-
tional Travel Grants to attend the General Assembly of European Geophys-
ical Society in 2001 and 2002 were funded by Dr. Astill.
Dr. Astill’s career at NSF can be best described by his own words on
May 31, 2004.

“Growing up in Australia was a miserable experience, in retrospect.

My life changed unexpectedly once I left Australia and although my
life has been very prosaic, those 44 years have been like a fairy tale
compared to the 24 years spent in Australia, and that especially
includes the 32 years at NSF. I took so much for granted while
I was at NSF, except for that last year, when it became painfully
obvious, based on the hiring in CMS during that year, that I would
not have been hired if I had applied for my own position. At first
that was a real shock, but then I began to realize just how lucky
I had been to be able to spend 32 years there; 14 years running
the Solid Mechanics Program, and 18 years running the geotechni-
cal earthquake program. While I very much enjoyed those 14 years
with the Solid Mechanics program, and especially the rigor of the
continuum mechanic expertise that the researchers possessed, and
was sorry to leave it, I nonetheless found the earthquake researchers
to be a much nicer group of people to interact with. I very much
missed the expertise in continuum mechanics in geotechnical en-
gineering, and was most grateful to the tsunami community for
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

A Dedication to Dr. Cliff J. Astill xiii

having that expertise. How come that civil engineers specializing

in soil mechanics receive such an abysmal training in continuum
mechanics, while those specializing in fluid mechanics receive it?
......... I’ve been very surprised at the way my thoughts on NSF ba-
sically evaporated the day I retired. After spending 32 years there,
this seems almost unreal, but it has happened. I’ve given a lot of
thought to it, and the fact is that we all sooner or later lose touch
and are forgotten. ...... All in all, the outlook looks very gloomy
except for NEES.”
Dr. Astill was a visionary in tsunami research. He had guided and sup-
ported the tsunami research community in the last twenty years. We were
very fortunate to have the privilege to know him professionally and person-
ally. His open-minded way in dealing with professional issues and his calm
spirits in fighting cancer will forever stay with us. He will be terribly missed
by all of us.
Harry Yeh, Philip Liu and Costas Synolakis

List of Workshops Supported by Cliff:

Hydrodynamic Model Developments:
o International Workshop on Long-Wave Run-up, Catalina Island,
CA, August 1990
o Second International Workshop on Long-Wave Runup Models, Fri-
day Harbor, San Juan Island, WA, September 1995
o The Third International Workshop on Long-Wave Runup Models,
Catalina Island, CA, June 2004
Development of Hazards Mitigation Tools:
o Workshop on the Development of a Tsunami Scenario Simulation
Program, Seattle, WA, September 2002
o Workshop for an Integrated Tsunami Scenario Simulation, Cor-
vallis, OR, August 2003 & San Francisco, CA, October 2004
Tsunami Field Survey and Data Management:
o Tsunami Measurement Workshop, Estes Park, CO, June 1995
o International Workshop on Bathymetry and Coastal Topography
Data Management, Seattle, WA, March 1998 & Birmingham, UK,
July 1999
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

xiv A Dedication to Dr. Cliff J. Astill

Tsunami Sources:
o Workshop on Seafloor Deformation Models, Santa Monica, CA,
May 1997
o Workshop on the Prediction of Underwater Landslide and Slump
Occurrence and Tsunami Hazards, Los Angeles, CA, March 2000
Development of NEES:
o Workshop for Tsunami Research Facilities, Baltimore, May 1998
o Ad-Hoc Planning Working Group for the NSF Major Research
Equipment Program: National Earthquake Engineering Simula-
tions, 1997–1999
o Workshop on Research with NEES Tsunami Facility, Corvallis,
OR, March 2001

List of Tsunami Surveys Supported by Cliff:

Nicaragua Earthquake and Tsunami, September 1992
Flores-Island (Indonesia) Earthquake and Tsunami, December 1992
Okushiri-Island (Japan) Tsunami, July 1993
The East Java Tsunami (Indonesia), June 1994
South-Kuril-Island (Russia/Japan) Earthquake and Tsunami, October
1994
Mindoro (Philippines) Tsunami, November 1994
Skagway (Alaska) Landslide Tsunami, November 1994
Chimbite (Peru) Tsunami, February 1996
Irian Jaya (Indonesia) Tsunami, February 1996
Aitape (Papua New Guinea) Tsunami, July 1998
Vanuatu Tsunami, November 1999
Southern Peru Tsunami, June 2001

List of International Travel Grants Supported by Cliff:

HAZARD-2000, Tokushima, Japan, May 2000
HAZARD-2002, Antalya, Turkey, September 2002
Tsunami Workshop in Petropavlovsk-Kamchatskiy, Kamchatka, Russia,
in August 1996
European Geophysical Society — XXVII General Assembly, Nice,
France, April 2001 & March 2002
January 10, 2008 19:59 WSPC/Trim Size: 9in x 6in for Review Volume Cliff

A Dedication to Dr. Cliff J. Astill xv

Clifford John Astill

March 4, 1936 – December 23, 2004
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Contents

Contents

Preface v

A Dedication to Dr. Cliff J. Astill ix

Part 1: Review Papers

Chapter 1
Modeling Runup with Depth Integrated Equation Models . . . . . . . . . 3
G. Pedersen

Chapter 2
High-Resolution Finite Volume Methods for the Shallow Water
Equations with Bathymetry and Dry States . . . . . . . . . . . . . . . . . . . . . . . 43
R. J. LeVeque and D. L. George

Chapter 3
SPH Modeling of Tsunami Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
B. D. Rogers and R. A. Dalrymple

Chapter 4
A Large Eddy Simulation Model for Tsunami and Runup
Generated by Landslides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
T.-R. Wu and P. L.-F. Liu

Chapter 5
Free-Surface Lattice Boltzmann Modeling in Single Phase Flows . . . 163
J. B. Frandsen

Part 2: Extended Abstracts

Chapter 6
Benchmark Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
P. L.-F. Liu, H. Yeh and C. E. Synolakis

xvii
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Contents

xviii Contents

Chapter 7
Tsunami Runup onto a Plane Beach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
Z. Kowalik, J. Horrillo and E. Kornkven

Chapter 8
Nonlinear Evolution of Long Waves over a Sloping Beach . . . . . . . . . . 237
U. Kânoǧlu

Chapter 9
Amplitude Evolution and Runup of Long Waves;
Comparison of Experimental and Numerical Data
on a 3D Complex Topography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
A. C. Yalciner, F. Imamura and C. E. Synolakis

Chapter 10
Numerical Simulations of Tsunami Runup onto a
Three-Dimensional Beach with Shallow Water Equations . . . . . . . . . . 249
X. Wang, P. L.-F. Liu and A. Orfila

Chapter 11
3D Numerical Simulation of Tsunami Runup onto a
Complex Beach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
T. Kakinuma

Chapter 12
Evaluating Wave Propagation and Inundation Characteristics
of the Most Tsunami Model over a Complex 3D Beach . . . . . . . . . . . . 261
A. Chawla, J. Borrero and V. Titov

Chapter 13
Tsunami Generation and Runup Due to a 2D Landslide . . . . . . . . . . . 269
Z. Kowalik, J. Horrillo and E. Kornkven

Chapter 14
Boussinesq Modeling of Landslide-Generated Waves and Tsunami
Runup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
O. Nwogu
August 26, 2008 16:56 WSPC/Trim Size: 9in x 6in for Review Volume Contents

Contents xix

Chapter 15
Numerical Simulation of Tsunami Runup onto a Complex Beach
with a Boundary-Fitting Cell System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
H. Yasuda

Chapter 16
A 1-D Lattice Boltzmann Model Applied to Tsunami Runup onto a
Plane Beach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
J. B. Frandsen

Chapter 17
A Lagrangian Model Applied to Runup Problems . . . . . . . . . . . . . . . . . 311
G. Pedersen

Appendix
Phase-Averaged Towed PIV Measurements for Regular Head Waves
in a Model Ship Towing Tank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
J. Longo, J. Shao, M. Irvine, L. Gui and F. Stern
April 3, 2008 3:12 WSPC/Trim Size: 9in x 6in for Review Volume divider

Part 1
Review Papers
This page intentionally left blank
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

CHAPTER 1

MODELING RUNUP WITH DEPTH INTEGRATED

EQUATION MODELS

G. Pedersen
Mechanics Division, Department of Mathematics, University of Oslo
E-mail: [email protected]

This survey starts with a sketch of long wave theory, reflecting some of
its diversity. Next, general aspects of runup on sloping beaches are briefly
discussed in the scope of long waves, including important analytical solu-
tions, breaking criteria, significance of non-hydrostatic effects and exper-
iments. Then the literature on numerical modeling of runup is reviewed.
The models are loosely organized into classes and then described chrono-
logically within each class. Primarily, we refer models that are published
in international journals, while matter from the vast number of pro-
ceedings and internal reports are mentioned occasionally. Moreover, we
highlight the verification of the numerical methods by comparison to an-
alytical solutions or experiments. Applications to real tsunami or storm
surge events are, with a few exceptions, not included.

1. Introduction
Comprehension of wave runup on beaches is essential for prediction of beach
erosion and coastal impact of tsunamis and storm surges. This is one reason
for the lasting attention that runup topics have received in the literature
of hydrodynamics and coastal engineering. Moreover, this complicated, yet
commonplace phenomenon, that can be observed at every beach with in-
cident swells or wind waves, pose mathematical and conceptual challenges
that appeal to many scientists. Wave runup also inherits similarities to other
important phenomena in hydrodynamics, such as “green water” (wave over-
topping on vessels), slamming and interaction between the ocean pycnocline
and bathymetry.
As for many other wave phenomena, long wave theory has been crucial
for our understanding of wave runup on beaches and derivation of closed
form solutions supporting this understanding. Also numerical models for

3
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

4 G. Pedersen

large scale ocean waves in general, such as tides, storm surges and tsunamis,
have traditionally been based on shallow water theory, due to the simplic-
ity and efficiency of this formulation. Certainly, more general models are in
progress, also for runup on beaches23 – 25,43,50,51,70,109 , but do still involve
heavy computations. For instance, most fair sized three-dimensional wave
problems are still dependent on approximate methods, such as long wave
theory. In fact, even an idealized, two-dimensional simulation modeling the
propagation of a tsunami wave or swell from deep water, through shoal-
ing/surf and to runup on a shore will often border on the limits of models
based on the Navier-Stokes equations or full potential theory. Therefore,
for many applications we must rely on long wave models also in the years
to come.
In the context of depth integrated equations, the moving shoreline on a
beach is in some respects similar to a free water surface that is deformed
by waves; the flow field is confined by a time dependent boundary with a
position that is unknown a priori. Furthermore, in both cases the surface
excursions correspond to motion of material particles with gravity as restor-
ing force. However, there are important differences. In runup the medium is
“thinning” toward the shoreline, resulting in a singularity in the governing
equations, though not in the physically relevant solutions. In addition, wave
breaking and topographical effects are often important for the near-shore
flow. If we look beyond the long wave (depth-integrated) description, other
aspects appear, such as the dynamics of a contact point between a free sur-
face and a rigid no-slip boundary, boundary layers, and effects of bottom
porosity. These are presumably most important on laboratory scale and are
anyway outside the scope of the present article.
An early review on long wave runup is found in Meyer and Taylor
(1972)67 . The predecessors to the “The Third International Workshop On
Long-Wave Runup Models”, leading to this volume, also initiated articles
summarizing aspects of runup research53,113. Among the contributions are
overviews of finite difference and element methods for tsunamis, that also
address long wave runup32,105 .
This review is meant to provide a brief and updated status on long wave
runup models, with an inclination toward applicability of the different long
wave approximations. We do not include related themes such as runup on
vertical walls, edge waves, rip currents, and impact problems. Moreover,
we focus on the principal aspects of physics and modeling rather than case
studies of real events of tsunamis and storm surges. Wave breaking is the
subject of Chapter 2 (by LeVeque and George) in this volume48 . Still, for
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 5

completeness, models with breaking, as well as theory and experiments, will

be discussed briefly herein.
We start with a short presentation of relevant parts of long wave theory.
Then the runup phenomenon is discussed with selected analytic solutions
and experiments, before the diversity of models in the literature will be sum-
marized. No complete assessment of the various methods will be attempted,
but documentation of performance through comparison with analytic so-
lutions, experiments, other models and grid refinement will generally be
emphasized. Significance of physical processes like wave dispersion will also
be addressed when appropriate.

2. Long Wave Equations

Following standard conventions we employ a typical depth d and a wave-
length L as vertical and horizontal length scales, respectively. The choice of
L, in particular, is ambiguous and it may also correspond to other lengths
√
than that of a wave. Identifying the time scale tc = L/ gd, where g is the
acceleration of gravity, and the dimensionless amplitude, , we define L/tc ,
d/tc and d, respectively, as scales for horizontal velocity, vertical veloc-
ity and surface elevation. When we instead refer to dimensional quantities
they are marked by a star. Different long wave equations can be obtained
through perturbation expansions in µ ≡ d/L and  and may then be clas-
sified according to which orders of these parameters that are retained in
the equations. Omission of all µ terms yields the nonlinear shallow water
(NLSW) equations, while retaining second-order in µ yields Boussinesq type
equations that are available in a series of varieties61,79,110 . Long wave ap-
proximations prescribe simple vertical structures for the field variables. In
NLSW theory, the horizontal velocity is vertically uniform and the pressure
hydrostatic. Therefore, the vertical coordinate, z, vanishes from the equa-
tions. When O(µ2 ) terms are retained there are vertical variations in the
horizontal velocity. Still, the explicit appearance of z is removed from the
continuity and momentum equations by integration. Hence, we denote them
as “depth integrated equations”. The spatial dimension of the partial dif-
ferential equations is then reduced by one. Furthermore, the nonlinear free
surface appears only through nonlinear coefficients. These features make
long wave formulations well suited for numerical solution. Regardless of the
mathematical reduction of dimension, we will still refer to problems as two-
or three-dimensional according to the physical configuration.
The order of a long wave equation is reflected in the dispersion char-
acteristics. According to full potential theory the dispersion relation of a
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

6 G. Pedersen

linear, sinusoidal wave reads (depth: h = 1)

1 1 2
c2 = tanh(µk) = 1 − (µk)2 + (µk)4 + ..., (1)
kµ 3 15
where c is the phase speed and k is the wave number. Shallow water theory
only reproduces the first term on the right hand, while traditional Boussi-
nesq equations yield the first two. However, as we will see below, different
formulations valid to O(µ2 ) give different dispersion properties. They may
even reproduce the O(µ4 ) term in (1) correctly or display en extended va-
lidity range.

2.1. Boussinesq Theory

During the last fifteen years fully nonlinear Boussinesq type equations with
improved dispersion properties have been put to work in computer models.
Following Hsiao et al. (2002)29 we write a set of fully nonlinear Boussinesq
equations on the form

ηt = −∇ · (h + η)(v + µ2 M) + O(µ4 ),
 
(2)
 
 2 2 1 2
vt + ∇(v ) = −∇η − µ z ∇∇ · vt + zα ∇∇ · (hvt )
2 2 α
+ µ2 ∇(D1 + D2 + 2 D3 ) + O(µ4 ) + N + E, (3)

with
   
1 2 1 2 2 2 1
M = z − (h − hη +  η ) ∇∇ · v + zα + (h − η) ∇∇ · (hv),
2 α 6 2
1 2 1
D1 = η∇ · (hvt ) − zα v · ∇∇v − zα v · ∇∇ · (hv) − (∇ · (hv))2 ,
2 2
1
D2 = η 2 ∇ · vt + ηv∇∇ · (hv) − η∇ · (hv)∇ · v,
2
1 
= η 2 v · ∇∇ · v − (∇ · v)2 ,

D3
2
and where the index t denotes temporal differentiation, h is the equilibrium
depth, η is the surface elevation, v is the horizontal velocity evaluated at
z = zα (x, y) and ∇ is the horizontal component of the gradient operator.
The heuristic terms N and E in the momentum equation (3) represent
bottom drag and artificial diffusion and will be discussed below. This par-
ticular form of the leading dispersion (0 µ2 ) term was discussed and tested
by Nwogu (1993)71 , while additional nonlinearity was added by Wei et al.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 7

(1995)107 . A similar formulation, with the velocity potential as primary un-

known instead of the velocity, is found in Chen and Liu (1995)7 . Applying
transformations and deletion of various higher order terms to the set (2)
and (3), we may reproduce a number of other Boussinesq type equations
from the literature42,71,107 . As pointed out in the references107 weakly non-
linear versions, in the sense that some or all O(µ2 ) terms are omitted, may
sometimes yield nonzero volume flux at the shore. This should in particular
be avoided for runup simulations.
When h is constant and  → 0 Boussinesq equations p on the form (2,3)
yield c2 as a rational function of µ2 k 2 . For zα = ( 1/5−1)h this expression
reproduces the first three terms in the expansion (1), while zα = −0.531h
yields a particularly favorable dispersion relation p over an extended range
of wave numbers. (see Fig. 1). With zα = ( 1/3 − 1)h, the velocity v
differs from the depth averaged velocity (v) by O(µ2 ), only. If we then
delete all O(µ2 ) terms in (2) and (3) we retrieve the traditional Boussinesq
equations for constant depth. These possess a dispersion relation that is
clearly inferior to the optimal version of (2,3) (Fig. 1). In general, the form
of the volume flux in (2) impliesa v = v +µ2 M. Differentiating this relation
with respect to time, inserting the resulting expression for vt in (3), and
invoking v = v + O(µ2 ) in nonlinear and dispersive terms we obtain the
standard Boussinesq equations that inherit errors of order µ2 , µ4

ηt = −∇ · [(h + η)v] , (4)

h2
 
 h
vt + ∇(v2 ) = −∇η + µ2 ∇∇ · (hvt ) − ∇∇ · vt
2 2 6

+ O(µ2 , µ4 ) + N + E. (5)

This set has been used much in numerical simulations and for the first time
by Peregrine (1967)78 for the shoaling of a solitary wave. We will subse-
quently refer to formulations like (4,5) as “standard” Boussinesq equations,
whereas those with more accurate dispersion properties, such as (2,3) with
zα = −0.531h, will be denoted as “improved”.
Further information on, and extensions of, higher order long wave equa-
tions are found in a number of papers in the literature16,58 – 61 . However,
progress has primarily been made with respect to higher order nonlinearity

a To reproduce v from v in non-constant depth we would have to employ a time dependent

zα .
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

8 G. Pedersen

Eq. 1 q
∗
√c zα = ( 13 − 1)h
gd
zα = −0.531h
KdV
LSW

λ/d
Fig. 1. The phase speed as function of wavelength for long wave equations compared
to that of the fully inviscid set. The curve for the Korteweg-deVries (KdV) equation is
included for comparison.

and dispersion during propagation in mildly varying bathymetries. Cor-

responding improvement has not yet been reported for a rapidly varying
bottom. The shoreline, with its mathematical singularity and abrupt change
of wave characteristics, as discussed below Eq. (7), may hardly be regarded
as belonging to a region of mild slope. Hence, while the higher order Boussi-
nesq formulations are of great value in finite depths, it remains to be seen
what may be locally achieved in runup on a sloping beach. In fact, as will
be discussed subsequently, even the significance of leading order dispersion
is debatable in the vicinity of the shoreline.

2.2. Shallow Water Equations

Deletion of all O(µ2 ) terms simplifies (2,3) to the NLSW equations. The
pressure gradient in the momentum equation then becomes ∇η, correspond-
ing to hydrostatic pressure. As a consequence the horizontal velocity be-
comes independent of z. The NLSW equations lack the important effect of
wave dispersion and may lead to erroneous results for long term propaga-
tion, even if the waves are long compared to the depth79 . However, they are
still quite a reasonable option for surf zone dynamics and are, by far, the
most commonly used framework for runup calculations, as well as tsunami
and storm surge models. Also the NLSW equations may be reformulated in
different ways. For modeling shocks (bores), in particular, it is important
to recast the momentum equations into a conservative form. When F and
B are omitted we may write the x-component as
1
(Hu)t + (Hu2 + H 2 )x + (Huv)y = Hhx , (6)
2
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 9

where subscripts indicate partial derivation, H = h+η, and u and v are the
velocity components in the x- and y-directions, respectively. Naturally, a
corresponding equation applies to the y-direction. We observe that the right
hand side is a momentum source due to a sloping bottom. Often the depth
integrated fluxes are used as primary unknowns instead of the velocities.
This may cause difficulties during runup, since velocities are needed for the
propagation of the shoreline and the relation between fluxes and velocities
may be poorly defined when H → 0.
A popular transformation of the NLSW equations is obtained through
introduction of characteristics. For plane waves the characteristic equations
become
p
βt± + (u ± c)βx± = hx , where β ± = u ± 2c, c = h + η. (7)
Characteristics played an important role in the early analytic work on runup
of breaking waves41,87 and are used in some numerical runup models as
well72,101,102 . In these references the technique is extended to two horizontal
directions by operator splitting102 or a partial characteristic description72 .
At the shoreline c becomes zero and both the characteristic equations and
variables in (7) coalesce. Alternative equations or extrapolation must then
be employed in a numeric model. Moreover, as long as ∂(h + η)/∂x is
nonzero at the shoreline, the spatial derivative of c is non-finite, which is a
challenge from a computational point of view. Still, as demonstrated in the
references runup can be treated accurately and efficiently by (7).
If the two-dimensional NLSW equations are linearized, we may eliminate
the velocity to obtain a very simple equation
ηtt − (hηx )x = 0. (8)
It might seem counter-intuitive for such an apparently nonlinear phe-
nomenon, but linear equations work surprisingly well on a number of runup
problems.

2.3. Lagrangian Coordinates and ALE Formulations

Problems involving material boundaries that are moving make application
of Lagrangian coordinates tempting. In the limit µ → 0 the horizontal
velocity is uniform in depth and it displays only a weak vertical varia-
tion for small µ. A Lagrangian description may thus be based on material
columns of water for the NLSW equations and averaged columns for Boussi-
nesq equations. A common choice for coordinates is the initial positions of
the columns, but every mapping of these are feasible and do not lead to
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

10 G. Pedersen

more complicated equations. The transformation between Eulerian and La-

grangian coordinates, (a, b), expresses that the latter are constant for a fluid
particle; at + v · ∇a = 0 and bt + v · ∇b = 0. Applying this transformation
to equations of the form (2,3), with a value of zα corresponding to depth
averaged velocities, we arrive at equations for x, y and η according to
∂(x, y)
(h + η) = F (a, b), (9)
∂(a, b)

∂(x, y) xtt ∂(η, y) ∂(x, y) ytt ∂(x, η)

=− + O(µ2 ), =− + O(µ2 ), (10)
∂(a, b)  ∂(a, b) ∂(a, b)  ∂(a, b)
where only the NLSW terms are spelled out and the Jacobi determinants
are defined according to
∂(x, y) ∂x ∂y ∂x ∂y
= − .
∂(a, b) ∂a ∂b ∂b ∂a
We note that time derivatives of positions are of order  such that the
ordering of (10) is consistent. Both a standard form of the Boussinesq
equations115 and a fully nonlinear version for plane waves34,75 are avail-
able. Comparing (9,10) to the Eulerian counterpart we observe that the
convective terms have vanished and that ungracious Jacobi determinants
have appeared. Moreover, the continuity equation has been integrated once
in time to directly express the volume conservation in a material fluid col-
umn, with F representing volume per area in the a, b plane.
For wave propagation in one horizontal direction the choice of marker co-
ordinates is unproblematic, but non-rectangular two-dimensional horizontal
domains have to be resolved with curvilinear grids or finite elements. For
runup the Lagrangian description has the major advantage that the shore-
line is located at temporally fixed values of a and b; the computational
domain is independent of time. As a bonus this sometimes extends also
to piston type wave paddles that may be cumbersome to incorporate in
Eulerian models.
To the knowledge of the author no Lagrangian version of improved
Boussinesq equations akin to (2,3) have been reported. Presumably, there
are no principal difficulties in the transformation, but it will result in very
lengthy expressions.
The fully Lagrangian description, where particles are traced throughout
the fluid, is in fact an over-treatment of the moving boundary problem. We
need only to follow the boundary particles and may thus employ a wider
class of transformations, that may be referred to as Arbitrary Lagrangian
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 11

Eulerian (ALE). There is a large diversity with respect to selection of coor-

dinates and dependent variables. In the simplest case with runup of plane
waves an obvious choice is a time dependent stretching of the spatial coordi-
nate in accordance with the shoreline motion. For simple three-dimensional
geometries, typically having a single, moderately curved shoreline, a sim-
ilar stretching with an along-beach variation may be employed72,84 . With
more complex geometries the transformation to time dependent curvilinear
coordinates may be found by optimizing with respect to local refinement,
variability and skewness of the grid (measured in the physical plane), while
fixed grid-boundaries (in the coordinate plane) are constrained to follow
the shoreline(s)88 .
In three-dimensional runup problems the ALE technique may be formu-
lated more robust with respect to interior grid deformations due to currents
than the purely Lagrangian ones. However, both traditional Lagrangian and
ALE methods are susceptible to grid deformation when large inundation
takes place in complex geometries. In addition bathymetric data must be
interpolated at each time step in numerical simulations.

2.4. The Shoreline

We observe that the linear wave equation (8) has a singularity for h = 0.
Provided that the gradient of h is finite at the shoreline both singular and
regular solutions exist. The simplest example is that of a linear standing
wave with frequency ω on an inclined plane defined by h = αx. The regular
solution, that has physical significance, then reads
√
η = AJ0 (2ω αx) cos(ωt + δ), (11)

where A, δ are constants and J0 is the Bessel function of zero order that
may be expanded in even powers of its argument. For the singular solution
of (8) we have η ∼ ln(x) for small x, while the horizontal velocity has a pole
of order 1. As a consequence there is a volume source at the singularity.
Hence, runup models that do not conserve volume at the shoreline should be
checked extra carefully for accuracy and spurious behavior. The mathemat-
ical structure with regular and singular solutions carries over to nonlinear
equations as well as full potential theory33 and is further discussed in a long
wave context by Meyer65,66 and others.
For nonlinear equations the shoreline is signified by H = h + η =
0. In addition the position of the shoreline must be traced. We confine
the discussion to two dimensions and assume that the shoreline always
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

12 G. Pedersen

consists of the same material particle. The shoreline position, x = ξ, then

moves according to dξ dt = u(ξ, t). In Lagrangian description this is only an
instance of the unknown x(a, t), but in Eulerian models the equation for
ξ must somehow be integrated. This is often achieved implicitly by filling
and emptying of grid cells. For very steep and breaking waves, outside the
long wave regime, the shore may not always correspond the same particle(s).
Moreover, the waterline is affected both by bottom roughness, viscosity and
capillary effects. The latter are generally more significant on the laboratory
scale than for real tsunamis and storm surges, say. One should alway bear
this in mind when models are calibrated against experiments.

2.5. Additional Effects

For waves propagating across oceans the rotation of the Earth must be taken
into account as the Coriolis force. However, the local runup dynamics on
a shore is hardly influenced by such effects and further discussion of this
topic is thus omitted.
Bottom friction is often included in long wave models as a square of the
velocity. A common form is
K
N=− |v|v, (12)
µH
where K is an empiric constant. Unfortunately, much of the systematic
empiric data for bottom friction stem from steady channel flow and may
not be relevant for the thin and rapidly changing fluid layers in runup.
Moreover, there is no consensus concerning the use of drag terms like (12).
Zelt and Raichlen (1991)116 calibrated the formula against experiments94
and then settled for K = 5×10−5. Comparing with the same dataset, Lynett
et al. (2002)57 suggested a value an order of 100 larger. We also observe
that the above formula in principle becomes singular at the shoreline.
Bores may be included in the NLSW description as discontinuities in the
velocity and the surface elevation. In numeric models these may be repro-
duced by special shock capturing schemes, generally applying approximate
solvers of Riemann problems (evolution of step distributions). Alternatives
that give finite width bores and may be used also with Boussinesq type
equations are the roller model62 or inclusion of a diffusion term, E, in (3).
For plane wave this term typically reads

E (x) = H −1 (ν(Hu)x )x . (13)

August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 13

Such terms have often been included to stabilize numerical models, under
cover of being a physical eddy viscosity effect. With a constant ν (13)
is quite inefficient as a tool for controlling wave breaking. Zelt (1991)114
made ν dependent on the velocity gradient in a similar term. ν was zero
until the gradient reached a threshold value and then increased with the
gradient beyond this. The critical value of the velocity gradient was related
to the highest stable solitary wave, which has an amplitude 0.78 times the
depthb . Kennedy et al.42 expressed ν by ηt and pursued the idea further by
including an explicit time relaxation for ν after activation of the diffusion.
This method has been demonstrated to reproduce bores from experiments
very well42,57 . However, as pointed out in a reference56 , this procedure may
not recognize a nearly stationary bore on a current, like the one that often
forms during drawdown at beaches.

2.6. Numerical Modeling of Long Waves

There are numerous papers on numerical solution of NLSW and standard
Boussinesq equations in general. Finite difference (FD) techniques domi-
nate, but quite a number of finite element (FE) methods are reported as
well. In traditional ocean and tsunami models the NLSW equations are
often solved by explicit methods and discretized on staggered grids64 (see
Fig. 2), denoted as C-grid, B-grid etc. Many FD models employ discretiza-
tion in finite volumes with fluxes defined at the interfaces to assure con-
servation of mass and momentum. This is particularly important for shock
propagation models. Under-resolved models inherit a strong artificial dis-
persion that occasionally may mimic true dispersion. However, this cannot
be accounted upon as a reliable feature and should be avoided. Grid sensi-
tivity tests are crucial in this context.
Boussinesq equations must be simulated by implicit methods and lead
to more CPU-time consuming models than the NLSW counterparts. As
compared to standard Boussinesq formulations the additional higher or-
der derivatives of the formulation (2,3) lead to extended computational
molecules and more difficult boundary conditions. This, in addition to the
lengthy expressions, render the implementation of this more general formu-
lation somewhat deterring. However, two FD codes for numerical solution of
this set may be downloaded from the Internet, namely the FUNWAVE44,45

b Recent investigations indicate that cross-wise instability appears at smaller

amplitudes40 .
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

14 G. Pedersen

v u, v u, v
y y
u η u ∆y η ∆y

v u, v u, v

x x

Fig. 2. Left panel: The C-grid that defines a basic configuration for mass conserving
FD methods. It is used also for many Navier-Stokes models with η replaced by pressure.
Right panel: The less used B-grid. This is well suited for representation of deformation,
as needed in Lagrangian models. u and v are the velocity compenents in the x- and
y-directions, respectively.

and COULWAVE56 models. These contain a number of features, including

runup facilities. Both models employ fourth order (five point) and second
order differences for the O(µ0 ) and the O(µ2 ) terms, respectively, combined
with higher order time integration. As a consequence, the leading disper-
sion in the wave celerity, namely the second term in (1), is kept clean of
discretization errors, which is a natural objective for numerical solution of
enhanced Boussinesq equations.
Solution of (2,3) by a finite element method is reported by Woo and
Liu108 , where the problem with higher order spatial derivatives is resolved
by introduction of extra dependent variables.

3. The Runup Phenomenon; Reference Solutions and

Experiments
Runup is generally part of a more complete problem, where some incident
wave is present in finite depth. This then undergoes shoaling and amplifi-
cation, and may turn into a bore, before it reaches the beach and starts to
run up (or down), often as a long and thin tongue that is retarded by grav-
ity. The withdrawal phase is then characterized by the run-down of a thin
layer of fluid, often with a strong bore, oriented onshore, evolving at the
toe. This whole sequence involves many stages and different physical mech-
anisms. There is a rich literature on both shoaling and surf-zone dynamics
that is outside the scope of this paper. Turbulence and sedimentation in the
swash zone12,54,83,85 are directly related to runup, but are also side issues
in the present context.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 15

3.1. The Non-Breaking Regime

The closed form solutions for runup are mostly associated with hydrostatic
(µ → 0) theory. We will discuss some aspects of these solutions with ref-
erence to a wave in a generic wave tank, as depicted in Fig. 3, with an
incident wave characterized by a length λ and amplitude A. The quantity
` is the propagation distance in constant depth, for instance the distance
from a wave maker to the slope. In the limit µ → 0 the equations are in-
dependent of the choice for the scaling factor L, introduced at the start of
Sec. 2. Rescaling with the length of the slope, d cot θ, as L we then realize
that the solution of the problem in Fig. 3 is governed by the parameters
 = A/d, κ = L/λ and `/λ.

z∗ λ

A x∗
θ ` d

Fig. 3. Definition sketch of generic wave tank.

In 1958 Carrier and Greenspan5 presented their much celebrated non-

linear, analytic solutions of the NLSW equations for runup on an inclined
plane. They invoked a transformation that in dimensionless form reads
√ φσ φσ u2
σ = 4 x + η, λ = 2(t − u), u= , η=− − . (14)
σ 4 2
The potential, φ, then solves the linear wave equation

φλλ − (σφσ )σ = 0.

Among their explicit solutions were the nonlinear generalization of (11),

the standing wave, as well as an initial value problem for runup. We will
subsequently refer to these solutions as CG-s5 and CG-i5 , respectively. A
kind of breaking, in the sense that η becomes multivalued as function of x,
is also contained in the standing wave solution. A major problem with the
transformation is the specification of an incident wave. Carrier (1966)4 sug-
gested the coupling of the nonlinear solution with a linear outer solution,
possibly including dispersion and a general bottom profile, through linear
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

16 G. Pedersen

patching conditions. This idea has later been exploited by others, among
which the work of Synolakis (1987)94 has made most impact. Before going
into details we will remark on some general properties of this approach.
Since the transformed equations, as well as far-field and patching are lin-
ear, the solutions in σ and λ equal linear solutions in the physical plane
with σ 2 /16 and λ/2 substituted for x and t, respectively. Nonlinearity en-
ters first through the transformation (14) back to the physical variables.
When u is zero, as at the point of maximum excursion of the shoreline,
the nonlinearity also disappears from this transformation. Hence, the max-
imum runup height for the combined linear/non-linear theory will be as for
a purely linear one. Moreover, ` only gives a time shift when the constant
depth region is governed by the LSW equations. For the wave tank problem
in Fig. 3 the maximum runup height then becomes

R = AF (κ), (15)

where F is a function depending on the actual shape of the incident wave

(an example is given below).
Solitary waves68 are much used in theoretical runup investigations, even
though they are more rarely seen in nature. They have permanent form,
finite width and are easy to create in a laboratory. To leading order in µ
and  the solitary wave solution in dimensional form reads
r  
∗ 2 ∗ ∗ 1 3A p 1A
η = ASec (k(x − ct )), k = , c = gd 1 + . (16)
d 4d 2d
In reality solitary waves are both nonlinear and dispersive, but the ex-
pression (16) may anyhow be used as initial condition in the combined
linear/nonlinear calculations described above. Synolakis94 applied Fourier
transforms to find a simple asymptotic formula for the maximum runup
height of solitary waves. Defining λ as 1/k we make the identification
κ = kd cot θ and write the formula on the form (15)
  41
1 1 A
R = 3.042Aκ 2 = 2.831A(cot θ) 2 for κ → ∞. (17)
d
When A and k in (16) are regarded as independent we observe that the
runup decreases with the wavelength. This is general feature of the present
theory that is found also for periodic waves. There are two main sources of
errors in (17), namely the asymptotic assumption and the lack of dispersion
on the slope. To check the effect of the first we may simply find R by solving
(8) numerically. Application of a five point centered difference scheme, with
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 17

3◦
5◦
7.18◦

R/A

Eq. (17)
10.54◦ Boussinesq
LSW

A/d
Fig. 4. Shallow water and Boussinesq predictions for maximum runup heights of solitary
waves for different beach inclinations as indicated on the curves. The thick and thin
columns represent breaking limits, (20), during runup and drawdown, respectively.

the closest node half a grid increment from the shore to implement the
condition of zero flux, makes this a trivial task. The results are subsequently
shown in Fig. 4.
The effect of dispersion on solitary wave runup is frequently overlooked
when NLSW models are compared to experiments or dispersive models are
compared with analytic NLSW solutions. This makes the model assessments
imprecise, with a negative bias concerning model performance. We may
employ optics to obtain a simple estimate on the variation of dispersion
with depth. Assuming that the period is unchanged during shoaling, we
1
find that the wavelength is reduced as h− 2 and the relative size of the
leading dispersion term in the wave celerity vanishes proportional to h near
shore. However, this result will not be valid in the proximity of the shore and
nonlinear effects will lead to steepening of the wave front, which in turn
enhance dispersion. Therefore, we resort to employment of a Lagrangian
Boussinesq type model34,76 (Sec. 4.1), that was readily at hand for the
author. Converged results of both the LSW and Boussinesq models are
compared to (17) in Fig. 4, that also contains the breaking criteria as given
in Eq. (20). Shallow water theory over-predicts runup slightly for the larger
A/d, but is surprisingly close to the Boussinesq theory even beyond the
hydrostatic breaking limit. For large θ and small A/d, meaning small κ,
the numeric solutions coincide, while the asymptotic formula (17) deviates.
Hence, we clearly observe the asymptotic nature of the closed form solution
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

18 G. Pedersen

3◦

R/A
Eq. (17)
Boussinesq
10.54◦ 0.5
0.9
NLSW

A/d
Fig. 5. Maximum runup heights for solitary waves. Results from combined Boussi-
nesq/NLSW theory are marked by the transition depth h(ac ) at equilibrium.

and that the agreement is good down to κ = 1, say, that corresponds to

R/A ≈ 3.04 in the figure.
We have also performed some numerical experiments to unravel where
dispersion effects are important. To this end we introduce a variable µ(a),
where a corresponds to the initial distance from the shore (Sec. 2.3). In
deep water µ has its proper value, but over an interval (ac , 1.1 × ac ) it
is reduced to zero. As shown in Fig. 5, omission of dispersion over half
the slope does not influence the maximum runup. Even if the hydrostatic
region is increased to nearly the whole slope, the change in R is very small,
though breaking occurs for a smaller A. The largest error occur when also
the uniform depth region is made non-dispersive (data marked NLSW in the
figure). This deviation depends strongly
p on ` that in the present simulation
is chosen as ` = − ln(0.00025)d/ 3A/d, which corresponds to an initial η ∗
equal to A/1000 at the start of the slope. In Fig. 6 we have compared the
particle acceleration in the Boussinesq model to the contribution from the
hydrostatic part of the pressure gradient, given by ∂η/∂x, for a moderately
steep incident wave. In finite depth there is a mild influence of the non-
hydrostatic pressure (right panel). When the wave have just reached the
shore the surface has an inclination of 36◦ , which is steeper than an extreme
(maximum amplitude) Stokes or solitary wave. Still, the non-hydrostatic
part of the pressure gradient is noticeable only in a small region close to
the shore.
Naturally, here we have merely scratched the surface concerning applica-
bility of long wave theories. Still, it is indicated that for non-breaking waves
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 19

t∗ = 12.1 d/g t∗ = 19.4 d/g

p p

t∗ = 21.8 d/g t∗ = 29.8 d/g

p p

η∗
d

x∗ /d

1 Du∗
g Dt∗
full full
hydrostat. hydrostat.

x∗ /d x∗ /d
Fig. 6. Runup of a solitary wave with A/d = 0.15 and θ = 7.18◦ . Upper panel: Surfaces
for the incident wave, an early stage in runup, an intermediate stage and maximum
runup. Lower panels: Comparison of the full acceleration to the part that stems from
the hydrostatic pressure gradient. Left: Vicinity of beach for the second time (thin solid
line in upper panel). Right: The incident wave (rightmost wave profile in upper panel).

the NLSW equation will often do for the near-shore region and runup. An
equally important lesson is that dispersion should not readily be ignored for
the constant depth propagation in simulations of wave tank experiments.
There are other analytic runup solutions that deserve mentioning. Some
are related to those above, but involve different initial conditions, slides or
modified geometries3,6,39,49,74,77,92,95,97,103 .
Another class of solutions contains the eigenoscillations in basins of
parabolic bottom shapes that are summarized by Thacker100. There exist
closed form solutions of the NLSW equations for the two lowest eigenmodes.
In the first mode the whole fluid body moves back and forth in a uniform
horizontal motion, driven by a uniform pressure gradient associated with a
linear surface. Setting d equal to the maximum depth, identifying the half-
width of the of equilibrium surface with L and choosing d as the amplitude
of the lowest harmonic in η ∗ , we obtain simple formulas for the 2D case
√  √ 1 √
η = cos( 2t)x − cos2 ( 2t), u = − √ sin( 2t). (18)
4 2
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

20 G. Pedersen

It is noteworthy that the eigenfrequency is independent of the amplitude.

The second mode inherits a linear spatial variation of the velocity and
a second order polynomial for the surface. The first eigenmode is much
used for comparison with and validation of numerical methods. However,
this may be a deceiving test due to the extremely simple spatial structure
of the flow that is reproduced exactly by many methods that may give
large discretization errors under other circumstances. It will serve best for
validation of code or performance check of crude shoreline approximations.
The second mode has somewhat better perspectives for assessment of runup
models, but is used less often.

c0

H∗
d
2c0

x∗ /d
Fig. 7. The dambreak solution according to (19). Surfaces are depicted for t ∗ = 0,
∆t∗ .., where ∆t∗ = 2d/c0 .

An important analytic solution of the shallow water equations is that of

dam-break93 , where a mound of water, usually step sized, is released from
rest at an horizontal plane. The initial acceleration distribution and early
development of the flow pattern are also extended to full potential theory63 .
Even though the initial water front is vertical, a smooth thin tongue will
evolve. Assuming an initial condition with fluid height d for x∗ < 0 and a
dry bed for x∗ > 0 we obtain a self-similar dilution wave from the NLSW
equations. For −c0 t∗ < x∗ < 2c0 t∗ the solution reads (see Fig. 7)
2
x∗ x∗
  
d 2
H∗ = −2 , u∗ =
c0 + ∗ , (19)
9 c 0 t∗ 3 t
√
where H ∗ is the total water depth and c0 = gd is the long wave speed
in depth d. Dam-break is a standard test case for Navier-Stokes models,
and is quite common for long wave solvers as well. NLSW models may
strive to reproduce the break at the limit to quiescent water, the huge
accelerations for small times, and perhaps the thin and rapidly moving fluid
tip. Otherwise, the solution (19) share the weakness of the eigenoscillations
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 21

in parabolic basins concerning verification of NLSW models. The spatial

structure of the flow is too simple to put many techniques to a proper test.

3.2. Transition to Breaking

Relevant overviews and background information on wave breaking may
be found in, for instance, Cokelet11 , Peregrine80 and, Synolakis and
Skjelbreia96 .
From the NLSW solutions of the preceding section we also have breaking
criteria for solitary waves during runup and drawdown on a plane beach on
the form
  54
A 10 5 A 3 9
< C(tan θ) 9 or κ 2 < C4. (20)
d d 4
For runup the constant C = 0.82 is reported94 , while C = 0.48 is obtained
for drawdown15. These findings are consistent with the commonly observed
feature that waves break more easily during withdrawal. The breaking lim-
its are depicted in Fig. 4. From the figure we observe that the breaking
criterion for runup is too strict compared to the Boussinesq results. This
is as expected since dispersion slow down the steepening during shoaling.
Grilli et al.24 have investigated breaking of solitary waves on a beach by
a boundary integral method for full potential theory. The reference rec-
ognized two breaking regimes: one with breaking before the shoreline, or
on-shore plunging, and an intermediate regime with a (nearly) vertical front
at, and a little beyond the equilibrium shoreline, which transforms into a
smooth swash without any appreciable plunging or spilling. Applying curve
fitting to the breaking/no-breaking domains in the A, θ plane, approximate
criteria were presented as
A
= B tan2 θ, (21)
d
where B = 16.9 and B = 25.7 correspond to the lower boundaries of the
intermediate and strongly breaking regimes, respectively. The criterion (21)
is generally much more relaxed than (20). However, it must be remarked
that the definition of breaking is ambiguous and that (21) is based on few
data points, in particular for small A/d and θ where (20) is most likely to
apply. Experiments34 have confirmed the existence of the intermediate zone
and even revealed waves that were overhanging at the shore, but still did
not proceed to breaking. Instead, the steepening process was reversed and
a smooth swash emerged from the toe of the wave.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

22 G. Pedersen

Plunging breakers are outside the long wave regime, but fully developed
bores may be crudely presented as discontinuities in shallow water solutions.
This theme is treated in Chapter 2 (by LeVeque and George) in this volume
and we will merely attach a few comments on the runup aspects. The most
remarkable feature for bore runup28,41,67,87 is that the bore collapses rapidly
at the shoreline, giving birth to a long thin runup tongue with a height
that is proportional to the square of the distance from the instantaneous
shoreline. The swash is dominated by gravity and the runup trajectory
becomes a parabola in the x, t plane. This transition from a steep incident
wave to a swash is closely related to dam-break34,67,81 (see Fig. 7). The
runup of steep non-breaking waves and finite width bores are quite similar
to that of the idealized bore solution; a steep front is transformed into a
thin swash zone, where gravity often dominates over pressure effects34,114 .
Moreover, there are large accelerations and velocities in the early phases of
runup, when the steep front vanishes. Small timing errors in experiments or
models may lead to large deviations temporarily, while quantities like runup
height and overall pattern is less influenced. This may cause an under-rating
of model performance. The occurrence of a very thin runup layer that even
may vanish as the square of the distance from the tip, seemingly argue
against the application of a linear onshore extrapolation of the surface that
is employed in some models. Even more so, perhaps, since the rundown
phase display even thinner layers. On the other hand, bottom roughness,
finite width bores and turbulence will modify the swash zone dynamics.
Since, in addition, the feedback of a thin swash to the main body of fluid
may be small, it is quite likely that surface extrapolation may be acceptable
also in such cases.

3.3. Experiments
There is only a limited number of published experimental investigations
that are commonly used for assessment of theoretic models. Hall and Watts
(1953)26 measured runup heights of solitary waves for different ampli-
tudes and beach inclinations. Synolakis (1987)94 published runup heights
of breaking and non-breaking solitary waves on a 1.0 in 19.85 slope and
amplitudes in the range 0.004 < A/d < 0.6. In addition, data from a set
of wave gauges, on and off the beach, were combined to rather complete
surface elevations at different times. These surfaces have been valuable for
validation of numerical models, in particular for the breaking cases. One
such data-set is displayed in Fig. 10.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 23

island base
4.64

7.2

13
equilibrium shoreline

27.43

Wavemaker

Fig. 8. Definition sketch of conical island experiment, with scales in meter. The outer
dashed circle is the base of the island, the fully drawn circle is the equilibrium shoreline
when d = 0.32m.

Briggs et al. (1995)2 investigated runup on a conical island, with slope

1 in 4, situated on plane bed of equilibrium depth, d, 0.32m or 0.42m
(Figs. 8 and 9). This experiment was inspired by observed runup patterns
at a rounded island for the Flores tsunami in 1992112 . Among the incident
waves were plane solitary waves with amplitudes a little short of A/d =
0.05, 0.1, 0.2. Wave heights were measured at different offshore wave gauges
and special gauges at the sloping bottom provided coarse digital series of
inundation and withdrawal around the island. The experiments are also
described in detail by Liu et al. (1995)52 with the first numeric simulations
on this problem. It is noteworthy that the breaking criterion (21) states
that no solitary wave will break during runup on a plane 1 in 4 beach.
Solitary waves of amplitude 0.2, or less, may then come close to breakingc
only during withdrawal or due to three-dimensional effects. The criterion
(20) yields A/d = 0.175, but the slope is too steep and the solitary wave too
high (and short) for shallow water theory to apply. The most remarkable
feature of the measured runup was a pronounced local maximum at the
rear of the island, due to interaction of waves coming from either side.
It is noteworthy that linear theories14,38 reproduce the experiments very
well, save for the lee side and then particularly the local maximum for the
intermediate amplitude A/d = 0.1. The conical island experiment was used
as a benchmark problem for the Second workshop on long wave runup113 .

c TheBoussinesq model34 employed in Sec. 3.1 indicates breaking during retreat for
A/d = 0.2, but not for the smaller amplitudes.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

24 G. Pedersen

90 1 a) 90 1 b) 90 1 c)
120 60 120 60 120 60

150 0.5 30 150 0.5 30 150 0.5 30

180 0 180 0 180 0

210 330 210 330 210 330

240 300 240 300 240 300

270 270 270

Fig. 9. Measured2 (stars) runup on conical island with recent simulations by Lynett
et al.57 (lines). (a): A/d = 0.05, (b): A/d = 0.1, (c): A/d = 0.2. Reproduced from Lynett
et al. by courtesy of the Coastal Engineering.

There are other experimental investigations of runup that, so far, are

less used for verification of long wave models34,49,50,67,69,111 .

4. Runup Models
We may recognize at least three key problems in runup modeling:

(1) The shoreline position is unknown a priori, it must be determined as a

part of the solution procedure. The computational domain must some-
how be adjusted to the shoreline motion.
(2) The singularity at the shoreline means that there do exist singular
solutions of the PDEs that may be exited by numeric errors.
(3) It is difficult to find a sufficient number of accurate boundary condi-
tions, in contrast to, for instance, the symmetry condition at a ver-
tical wall. In finite difference methods one sided representations are
generally required at shoreline. These may be implemented directly or
indirectly by fictitious points combined with extrapolation. Common
consequences are that the order of accuracy of the numerical method
is reduced at the waterline and a production of noise that must be
controlled by filtering or dissipation.

In view of these points the performance of runup models should be carefully

assessed. Results should be compared to the analytic solutions or experi-
ments on runup, or at least compared to computational results obtained
independently. In addition, grid dependence in the solution, which always
can be investigated, should be reported. It is surprising and disappointing
how seldom the latter is found in the literature.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 25

Some of the literature that addresses moving shorelines is focused on

storm surges and tides. Such applications involve smaller gradients and
longer time scales than runup of tsunamis, or swells, and are thus less
challenging. The contributions of this kind often omit both testing and
detailed descriptions of shoreline dynamics. Hence, it is difficult to evaluate
the potential of such published methods concerning runup of steeper waves.
In the subsequent review, priority will be given to articles where vali-
dation in at least one of the above fashions is documented. The methods
are grouped in two: those that remove the time dependence of the compu-
tational domain by a transformation, and those formulated on a fixed grid,
possibly with some auxiliary nodes at the shoreline. Both groups contain
finite difference as well as finite element methods. The reader should always
bear in mind that the referenced articles vary in style and objective. Hence,
when we are enabled to refer discussions on errors and shortcomings from a
particular work this may equally well reflect quality and open-mindedness,
on the authors side, as inferior model performance relative to other
publications.

4.1. Transformed Domain Methods

Application of a Lagrangian, or an arbitrary Eulerian-Lagrangian (ALE),
description generally eliminates the tracing of an a priori unknown shoreline
(point 1 above). Several runup models of this kind have been reported, even
though they are in minority relative to the Eulerian counterparts.
An early discussion of runup and Lagrangian coordinates is given by
Shuto (1967)89 , but for the linear shallow water equations only. Hence, the
potential of the Lagrangian description was not really exploited. Later Goto
(1979)20 and Goto and Shuto (1980, 1983)21,22 compared 2D numeric re-
sults with linear solutions and the nonlinear standing wave solution CG-s5 .
The description of the method was sketchy, but the agreement with the
analytic solution was very good and markedly better than for some early
Eulerian models. In 1983 Pedersen and Gjevik75 applied Lagrangian coor-
dinates to a 2D Boussinesq set with standard linear dispersion properties.
Employing finite differences on a staggered grid the authors reproduced
analytical solutions92 accurately. Computed runup of non-breaking solitary
waves were compared to experiments, with good agreement for moderately
steep slopes. The model has later been revisited and improved by Jensen
et al. (2003)34 , who also discussed the significance of different O(µ2 ) terms
in the runup. A 3D Lagrangian FD model was reported by Johnsgard and
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

26 G. Pedersen

Pedersen (1997)37 and Johnsgard (1999)36 with applications to runup and

wave generation by sub-aerial slides in idealized lake geometries. The model
was based on a B-grid and verified by extensive grid refinement and com-
parison with CG-s5 and oscillations in a parabolic basin100 .
Lynch and Gray (1980)55 described a finite element technique with time
dependent shape functions and a grid that could be adapted to a moving
shoreline. However, their model was not tested on any standard problem.
The first runup model that inherited both three-dimensionality and dis-
persion was published by Zelt and Raichlen in 1990115 . They employed
a finite element technique with a partition in quadrilaterals in the La-
grangian coordinates. One appealing feature of this kind of formulation is
the so-called natural boundary condition for the shoreline. After reproduc-
ing CG-s5 accurately the authors applied the model to the response of an
idealized harbour to incident swells. Their equations were similar to the
standard Boussinesq equations, apart from some extra error terms of order
2 (see Sec. 2) that appeared for curved beach profiles. This minor fault was
eliminated in the succeeding work114 , where wave inundation of a horizon-
tal shelf were modeled with a bottom friction term, giving good agreement
with experiments. In a subsequent paper114 Zelt included wave breaking
by means of a diffusion (Sec. 2.5) dependent on the local wave steepness,
or, rather, the compression rate of the Lagrangian water columns. Com-
putations were compared with non-breaking (A/d = 0.04) and breaking
(A/d = 0.28) solitary wave runup experiments94 . Excellent agreement was
observed for non-breaking runup, while there were discrepancies for break-
ing waves. In particular, a spurious, nearly vertical front persisted during
inland excursion, which presumably was due to the diffusion term. Com-
parison was also made between Boussinesq, as well as NLSW, simulations
and experiments19 of solitary wave runup on a steep (θ ≈ 20◦ ) slope. It is
noteworthy that the Boussinesq results reproduced the experiments much
closer (see discussion in Sec. 5). Later another Lagrangian FE method for
the NLSW equations was reported by Petera and Nassehi (1996)82 . Un-
fortunately, little documentation on the runup was given and there is no
evidence of any improvement over the FE technique described above.
In two dimensions, the fluid domain can be mapped onto a fixed inter-
val through a time dependent, but spatially uniform, coordinate stretching
in accordance with the waterline motion. An example was given for the
NLSW equations by Johns (1982)35 who employed a staggered grid and
linear extrapolation of η to obtain the shoreline position. A numeric solu-
tion showed fairly good agreement with an CG-s5 solution with a very small
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 27

amplitude. A similar technique was reported by Takeda (1984)99 , who used

a semi-characteristic method to reduce the need for one-sided approxima-
tions at the shore. Good agreement with CG-s5 was found for runup, while
a somewhat degraded performance was observed for drawdown. The pa-
per contains a general discussion on employment of asymmetric difference
equations at the shore, but no model test that could support any firm con-
clusion was included. The approach of Johns35 was generalized to three
dimensions by Shi and Sun (1995)88 who invoked an orthogonal transfor-
mation and then a C-grid discretization on a fixed coordinate domain that
was composed of rectangles. The technique was applied to a real storm surge
inundation. The large deformation of the grid imposed by the adaptation
to a real geometry clearly called for grid-refinement studies, but none were
presented.
In Özkan-Haller and Kirby (1997)72 , the NLSW equations were rewrit-
ten in a partial characteristic form and solved on a transformed domain
of rectangular shape by spectral collocation. Chebyshev polynomials and
trigonometric functions were used for the cross- and along-shore directions,
respectively. The CG-s5 solution was reproduced with high accuracy and
the method was then applied to edge-wave instability. An advantage over
fully Lagrangian techniques is that strong shear currents may be included
without grid distortion. Recently, Prasad and Svendsen (2003)84 have em-
ployed a similar transformation with a finite difference technique for the
NLSW equations and obtained close agreement with both CG-s5 and CG-
i5 . The authors also presented a FD technique on a fixed grid, where they
went to some length in describing the shoreline accurately by carefully dis-
cretized boundary conditions. Also this technique displayed good results,
but was markedly more prone to shoreline fluctuations than the transfor-
mation method and displayed a slower convergence according to grid refine-
ment tests. Zhou et al. (2004)117 make the same simple transformation as
Johns35 for the two-dimensional problem. The NLSW equations are then
solved by a mesh-less technique, utilizing smooth shape functions and col-
location. The paper is brief and only dam-break is presented as verification.

4.2. Fixed Grid Methods

In a fixed grid the key problem is to include/exempt points and allot proper
values to field variables at the boundary. For finite difference methods this is
most commonly attempted by more or less ingenious extrapolation schemes.
A number of such are reviewed in Imamura (1996)32 . Balzano (1998)1 tested
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

28 G. Pedersen

a series of NLSW methods for tidal flooding, mainly based on the C-grid, on
a small amplitude second eigenmode in a parabolic basin100 . To a greater
or lesser extent all techniques gave near-shore fluctuations. However, the
degree of nonlinearity was too small to allow any inference on the general
runup performance of the models.
Apart from simulations of flooding based on linear theory86 , the first
proper runup model was reported by Sielecki and Wurtele (1970)90 . They
employed three different finite difference techniques, among them the Lax-
Wendroff method, for the NLSW equations with the Coriolis term. A sim-
ple extrapolation combined with an asymmetric representation was used to
follow the shoreline. Among the test cases were CG-i5 , that was well repro-
duced except for a small discrepancy at shoreline, and a three-dimensional
eigenoscillation in a parabolic basin100 that was simulated perfectly. In 1975
Flather and Heaps13 published a 3D FD technique for the NLSW equations,
based on the C-grid, with a scheme to update wet and dry regions. However,
the model was applied to tidal oscillations in a bay without any verification
of its runup performance. Hibberd and Peregrine (1979)27 employed the
Lax-Wendroff technique to the NLSW equations in conservative form and
obtained a model for non-breaking waves as well as bores. According to the
authors the simple shoreline method of Sielicki and Wurtele90 did not suf-
fice for these problems. After attempting various methods they settled for a
predictor-corrector sequence of linear extrapolations and applications of the
physical equations in the vicinity of the shoreline. Even though some near-
shore deviations were identified, both CG-i5 and CG-s5 were reproduced
well. Very close agreement were reported with analytical runup heights67 ,
while the theoretical values overshot measurements69 . This was explained
by viscous effects that must influence the thin swash zone significantly on
laboratory scale. Later Kobayashi (1987)46 adopted the method and applied
it to runup on rough slopes.
Kowalik and Murty (1993)47 employed linear onshore extrapolation of
velocity and surface elevation with a C-grid discretization of the NLSW
equations. Their method gave reasonably good results for CG-i5 and the
first mode in a parabolic basin, but with significant noise at the shoreline.
Then they used the method to compute the inundation in a real tsunami
event.
With the experiments on solitary wave runup on a conical island Briggs
et al. (1995)2 (Sec. 3.3) provided what still is the only 3D dataset well
suited for verification of models. In a companion paper Liu et al.52 pre-
sented NLSW simulations with a C-grid model with upwind representation
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 29

of the convective terms. The moving shoreline was represented through a

sequence of over-toppings and dry-outs in a step-bathymetry. In spite of
some deviation the maximum runup heights were generally predicted very
well for the smaller amplitudes. Moreover, even though the sharp local max-
imum at the lee side was not fully reproduced for A/d = 0.1, the nonlinear
model fared much better than linear solutions14,38 in this respect. Recently,
the same method was re-applied to this case for A/d = 0.1, utilizing a
hierarchical grid to obtain selective grid-refinement around the island10 .
However, no substantial improvement was obtained. The results of Liu
et al. for the highest amplitude were less good. According to the discus-
sion in Sec. 3.3, this case really requires a dispersive model. At the time the
only well documented Boussinesq type model for three-dimensional runup
was the Lagrangian method of Zelt and Raichlen115 , described in Sec. 4.1.
Regretfully, to the knowledge of the author this finite element model is
never applied to the conical island problem.
Titov and Synolakis (1995)101 used the characteristic form of the 2D
NLSW equations that were solved by a corrected leap-frog method allowing
for non-uniform grids. This enabled the inclusion of an auxiliary shoreline
point in the otherwise fixed grid. The position and velocity for the extra
point were projected from the nearest wet point. Good agreement with
experiments94 and the formula (17) was obtained for non-breaking solitary
wave runup. The method also gave rather good agreement with experi-
ments on a breaking wave (A/d = 0.3), save for some pronounced deviation
in the early phase of runup. However, this may partly be due to the sensi-
tivity of the solutions to small time differences at this stage (Sec. 3.2), but
omission of dispersion did probably contribute as well. Later the method
was generalized to three dimensions by operator splitting102 . Comparison
was then made to the conical island experiment2 with very good agreement
for A/d = 0.1 and somewhat larger errors for A/d = 0.2. The latter may
again be due to the absence of dispersion. Physical tsunami applications
displayed encouraging results indeed and demonstrated that the method is
robust in complex situations and handles features as merging of shorelines.
It is incorporated in the tsunami model known as MOST73 .
Madsen et al.62 reported a 3D Boussinesq technique with improved lin-
ear dispersion characteristics, based on depth integrated volume fluxes, and
a surface roller model for breaking. The beach was made slightly permeable
with a porosity, rapidly decreasing from unity to a small residual value be-
neath the surface. As a consequence the equations could be solved through-
out a region that was extended beyond the shoreline, with no explicit
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

30 G. Pedersen

reference to its position. However, as communicated by the authors, this

technique was not without weaknesses. The dispersion terms was turned off
in the onshore motion, noise had to filtered and there was a mass deficit
in the swash tongue due to the porosity. The latter caused marked de-
viations from the CG-s5 solution, while the overall agreement was still
reasonably good. Subsequently, the porous shore technique were amended
and incorporated in the FUNWAVE model45 for improved Boussinesq
equations (Sec. 2.6), with results presented in a series of papers. In a
paper on rip-currents Chen et al. (1999)8 interpreted the porosity as
slots in the seabed and counterbalanced the volume loss by adjusting the
inter-slot beach level. Wave breaking was represented by an eddy viscos-
ity depending on the temporal surface gradient (Sec. 2.5). This model
was put to test by Kennedy et al. (2000)42 through comparison with
experiments on breaking in finite depth and the analytic CG-s5 solu-
tion. The model reproduced the breaking well, while the agreement with
CG-s5 was substantially improved relative to Madsen et al.62 , even if
some deviation persisted. Chen et al. (2000)9 did the conical island test.
The agreement for the smaller amplitudes, A/d = 0.05, 0.1 was fairly
good, but poorer rather than better relative to the pre-existing NLSW
computations52,102 . For the higher amplitude, A/d = 0.2, where disper-
sion is more important, the results improved, at least compared to Liu
et al. (1995)52 . Unfortunately, simulations based on the porous bed tech-
nique appeared not to have been compared to the detailed experimental
surface profiles for breaking and non-breaking waves94.
The runup facility in the COULWAVE model56 for the improved Boussi-
nesq equations was described and tested in Lynett et al. (2002)57 . Again
wet and dry nodes were recognized by the total water depth, while the
discrete equations at the shoreline were based systematically on fictitious
nodes beyond the shoreline. Linear extrapolation from neighboring points
in the fluid was used to assign values to the onshore nodes. Higher order
differences were then used throughout the computational domain. In prin-
ciple, this is a simple procedure that may seem to depend heavily on the
analytic nature of the solution. The authors presented a good description of
the implications of the procedure for the computational molecules adjacent
to the shore. In addition to the general loss of accuracy and noise (arrested
by a 9 point filter) at the shoreline, the higher order spatial derivatives were
annihilated, which imply omission of dispersion terms. Wave breaking was
included in a similar manner as in Kennedy et al. (2000)42 . Model results
agreed well with CG-s5 and the second eigenmode in a parabolic basin100
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 31

0.6

0.4

0.2

0 a)

−0.2
−20 −15 −10 −5 0 5 10

0.6

0.4

0.2

0 b)

−0.2
−20 −15 −10 −5 0 5 10

0.6

0.4
z/h

0.2

0 c)

−0.2
−20 −15 −10 −5 0 5 10

0.6

0.4

0.2

0 d)

−0.2
−20 −15 −10 −5 0 5 10

Fig. 10. Breaking solitary wave runup. Experiments from Synolakis94 are indicated
by markers, while the fully drawn lines are solutions from Lynett et al.57 . Panel (c)
includes also results from Zelt114 (dots), Titov and Synolakis101 (dash-dot) and the N-S
computations of Lin et al.51 (dashes). Reproduced from Lynett et al. by courtesy of the
Coastal Engineering.

was accurately reproduced. Also the conical island test was made with
good results. For runup of a breaking wave (A/d = 0.3) impressive agree-
ment was obtained with both experiments94 and a turbulent Navier-Stokes
simulation51 (Fig. 10). Even though this is unlikely to be the reason for the
good outcome, it must be noted that the choice of the coefficient in the
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

32 G. Pedersen

bottom-drag was optimized. As pointed out by the authors, there are still
some unresolved shortcomings in the model for features like over-topping
and reentry of a runup tongue into quiescent water56 .
Another line of runup modeling is based on the NLSW equations in con-
servative form, finite volume techniques and approximate Riemann solvers
to include bores as sharp jumps in the solution. This branch will be covered
in depth in Chapter 2 in this volume by LeVeque and George48 . Hence, we
refer briefly only a selection of contributions, with emphasis on runup tests.
The shoreline may be treated by the special Riemann problem related to
penetration of gas into vacuum or dam-break. However, this leads to overes-
timation of the pressure gradient and acceleration at the shoreline. Hence,
modelers often resort to complete or partial exclusion of very shallow cells
from the general computation scheme. The flux and source terms in the
momentum equation (see below Eq. (6)) are generally separated by opera-
tor splitting, even though Watson et al. (1992)106 suggested the inclusion
of the source term in the Riemann problem by a local transformation to an
accelerating frame of reference. Sleigh et al. (1998)91 invoked a discretiza-
tion into triangular volumes which enables flexible local grid refinement.
The method was tested on various dam-break problems and bore-runup.
Good agreement was observed with the analytical solution for the standard
dam-break problem, save from small deviations at the toe. Two-dimensional
simulations were presented by Hu et al. (2000)30 including runup of non-
breaking solitary waves, that agreed closely with an analytic solution 94 ,
and the standard dam-break problem. The latter case demonstrated that
exemption of shallow cells provide a better solution near the fluid tip than
application of a modified Riemann solution. In Brocchini et al. (2001)3 oper-
ator splitting was employed on a rectangular grid and a Riemann technique
was used for the shoreline. Results compare well with the analytic solution
CG-i5 and the standard dam-break problem. However, for the latter there
was noticeable spurious behavior close to the water line. The particular
strength of Hubbard and Dodd (2002)31 is a hierarchical mesh refinement,
which means that finer rectangular grids may be superimposed on coarser
ones in regions that require high resolution, as the shore and swash zone.
Close agreement, save for small irregularities at the shore, was obtained
with CG-s5 . Surprisingly, the reproduction of the first nonlinear eigenmode
in a parabolic basin was less good.
A common objective for this kind of models is to preserve sharp bore
fronts. It would have been interesting to see a detailed comparison with
real, finite width bores, as breaking solitary waves94 .
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 33

The finite element method has a much weaker merit in Eulerian runup
modeling than the finite difference counterpart. Several methods for in-
cluding a moving shoreline in a static FE grid have been proposed105 , but
few have been validated by comparison to reference solutions or system-
atic grid refinement. An interesting attempt on solving a two-dimensional
Boussinesq equation with the differentiable Hermitian cubic shape func-
tions was made by Gopalakrishnan and Tung (1983)18 . The pressure gra-
dient was used directly to compute the shoreline acceleration which then
was integrated to yield the position. The shore element was then modified
at each time step, and split/deleted when appropriate. However, the only
application was solitary wave runup, without any comparison to other so-
lutions nor application of grid refinement tests. Moreover, the presented
solutions clearly displayed noise and spurious features. Later the method
was modified17 to the three-dimensional NLSW equations and applied to
tidal flooding, again with inadequate testing of the runup performance.
Umetsu (1995)104 employed partially dry triangular elements, where the
velocities on dry near-shore nodes are imposed as averages of the values
from the adjacent wet ones. The technique performed well for broken dam
flow, but produced outspoken deviation from the CG-i5 , in particular close
to the shore. In Takagi (1996)98 the technique was applied to the conical
island problem with reasonably good, but not convincing, result.

5. Discussion
The crucial point in runup modeling is the representation of the shoreline
motion, for which the presence of a singularity (Sec. 2.4) calls for caution.
The literature refer many rather crude as well as sophisticated methods for
the moving boundary. Even though noise in the vicinity of the shoreline is
often observed there are few severe problems that have been attributed to
the singularity. It is tempting to assume that this problem is not that grave
after all, or that a logarithmic singularity is too weak to be fatal. On the
other hand, too many authors have treated numerical convergence rather
loosely. More rigid grid refinement studies might reveal that there indeed
are irregularities at the shore, even though they in most applications will
be overshadowed by other error sources. Anyway, the bottom line is that
there is a diversity of working runup models about.
The recent progress in long wave runup modeling may roughly be di-
vided into two directions: one where the runup facility is integrated in gen-
eral purpose wave propagation models with high order inherent dispersion
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

34 G. Pedersen

and another where increasingly robust or accurate shoreline representations

are included in NLSW models, with features like adaptive grids and shock
capturing. Some of the NLSW models are both well tested, efficient, and
flexible enough for, for instance, realistic case studies of tsunami, while the
existing Boussinesq type models still may miss a little on runup/drawdown,
bore presentation and computer efficiency for such applications. On the
other hand, dispersion is important for wave propagation in finite depth,
not only for swells, but also for many tsunamis. This leads to the questions;
when are the different kind of long wave models appropriate and how may
they be combined in applications that require the strengths of both kind of
models?
Our little study of significance of dispersion, reflected in Fig. 5, sug-
gests that the NLSW equations may be very accurate for the near-shore
region and runup for non-breaking waves. Regretfully, it is difficult to at-
tain further insight on this point from the literature. Truly, in some of
the cases of non-breaking solitary wave runup, reviewed in the previous
section, the inclusion of dispersion led to improved agreement with experi-
ments or more general models57,114 . However, the difference may be due to
accumulation of dispersion effects in finite depth propagation and shoaling
(Sec. 3.1) that change the incident wave before the shoreline is reached, and
that are missed out in the NLSW solution. For the case of breaking soli-
tary wave runup94 the improved Boussinesq solution57 is clearly superior
to those of the NLSW101 and standard Boussinesq equations114 (Fig. 10).
Still, again the effect of dispersion on the incident presumably is important.
In addition the bore representation are very different in the respective mod-
els, and even somewhat dubious for the standard Boussinesq formulation
in question. Hence, it is difficult to draw any firm conclusion on the local
near-shore effect of non-hydrostatic pressure and more investigations are
required, where local effects are emphasized.
If, for a given application, the dispersion can be neglected near-shore,
but not elsewhere, nesting of models may be desirable. For instance, a higher
order Boussinesq type description for deep water propagation and shoaling
should be coupled with an efficient NLSW solver with shock handling and
robust runup properties for the surf and swash regions. A one way cou-
pling from Boussinesq to NLSW equations is principally straightforward.
On the other hand, a two way coupling involves conceptual problems. The
NLSW solver will produce and preserve shocks sharp, except at the shore-
line. In finite depths the shocks may vanish only due to dissipation. Hence,
the reflected waves from shallow region may be difficult to convey to a
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 35

Boussinesq type model without severe problems. To the knowledge of the

author no such attempt on a complete two way coupling has been reported
in the literature.
For Lagrangian, and curvilinear, models very good performance is re-
ported for runup of waves in 2D or simple 3D geometries. This makes trans-
formed domain models an excellent tool for academic studies. However, a
widely accepted weakness of such methods is the strong grid deformation
that may occur in demanding applications. No well documented attempt
to apply such models to real tsunami studies has been reported. In fact,
in recent years Lagrangian coordinates, in general, have been most popu-
lar in connection with smooth particle hydrodynamics that is a meshless
method. It would have been interesting to see how well such a method could
be employed with the NLSW equations and free boundaries.

Acknowledgments
The author wishes to thank Professor P. L.-F. Liu and Professor P. Lynett
for their helpful assistance.

References
1. A. Balzano. Evaluation of methods for numerical simulation of wetting and
drying in shallow water flow models. Coast. Eng. 34, 83–107 (1998).
2. M. J. Briggs, C. E. Synolakis, G. S. Harkins and D. R. Green. Laboratory
experiments of tsunami runup on circular island. Pure and Applied Geo-
physics 144(3/4), 569–593 (1995).
3. M. Brocchini, R. Bernetti, A. Mancinelli and G. Albertini. An efficient solver
for nearshore flows based on the WAF method. Coast. Eng. 43, 105–129
(2001).
4. G. F. Carrier. Gravity waves on water of variable depth. J. Fluid Mech. 24,
641–659 (1966).
5. G. F. Carrier and H. P. Greenspan. Water waves of finite amplitude on a
sloping beach. J. Fluid Mech. 4, 97–109 (1958).
6. G. F. Carrier, T. T. Wu and H. Yeh. Tsunami run-up and draw-down on a
plane beach. J. Fluid Mech. 475, 79–99 (2003).
7. Y. Chen and P. L.-F. Liu. Modified Boussinesq equations and associated
parabolic models for water wave propagation. J. Fluid Mech. 288, 351–381
(1995).
8. Q. Chen, R. A. Dalrymple, J. T. Kirby, A. B. Kennedy and M. C. Haller.
Boussinesq modeling of a rip current system. J. Geophys. Res. 104, 20617–
20637 (1999).
9. Q. Chen, J. T. Kirby, R. A. Dalrymple, A. B. Kennedy and A. Chawla.
Boussinesq modeling of wave transformation, breaking, and run-up. Part
II: 2D. J. Waterw., Port, Coast., Ocean Engrg. 126(1), 48–56 (2000).
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

36 G. Pedersen

10. Y.-S. Cho, K.-Y. Park and T.-H. Lin. Run-up heights of nearshore tsunamis
based on quadtree grid system. Ocean engineering 31, 1093–1109 (2004).
11. E. D. Cokelet. Breaking waves. Nature 267, 769–774 (1977).
12. B. Elfrink and T. Baldock. Hydrodynamics and sediment transport in the
swash zone: a review and perspectives. Coastal Engineering 45, 149–167
(2002).
13. R. A. Flather and N. S. Heaps. Tidal computations for Morecambe Bay.
Geopshys. J.R. Astr. Soc. 42, 489–517 (1975).
14. K. Fujima. Application of linear theory to the computation of runup of
solitary waves on a conical island. In Long-wave runup models, Eds. H. Yeh,
C. E. Synolakis and P. L.-F. Liu (World Scientific Publishing Co., 1996),
p. 221–230.
15. B. Gjevik and G. Pedersen. Run-up of long waves on an inclined plane.
Preprint series in applied mathematics, Dept. of Mathematics, University
of Oslo (1981).
16. M. F. Gobbi, J. T. Kirby and G. Wei. A fully nonlinear Boussinesq model
for surface waves. Part 2. Extension to O(kh)4 . J. Fluid Mech. 405, 181–210
(2000).
17. T. C. Gopalakrishnan. A moving boundary circulation model for regions
with large tidal flats. Int. J. Num. Meth. Eng. 28, 245–260 (1989).
18. T. C. Gopalakrishnan and C. C. Tung. Numerical analysis of a moving
boundary problem in coastal hydrodynamics. Int. J. Num. Meth. Fluids 3,
179–200 (1983).
19. D. G. Goring. Tsunamis: The propagation of long waves onto a shelf.
Ph.D. thesis Rep. KH-R-38, Calif. Inst. Technol., Pasadena, California
(2003).
20. C. Goto. Nonlinear equations of long waves in the Lagrangian description.
Coast. Eng. Japan 22, 1–9 (1979).
21. C. Goto and N. Shuto. Run-up of tsunamis by linear and nonlinear theories.
In Coastal Engineering. Proc. of the seventeenth Coastal Engineering Conf.,
vol. 1 (Sydney, Australia, 1980), p. 695–70.
22. C. Goto and N. Shuto, Numerical simulation of tsunami propagations and
run-up. In Tsunamis–Their Science and Engineering, Eds. K. Lida and
T. Iwasaki (TERRAPUB, 1983), p. 439–451.
23. S. Grilli and I. A. Svendsen. Computation of nonlinear wave kinematics dur-
ing propagation and run-up on a slope. In Water wave kinematics (Kluwer
Academic Publishers, 1990), p. 378–412.
24. S. T. Grilli, I. A. Svendsen and R. Subramanya. Breaking criterion and
characteristics for solitary waves on slopes. J. Waterw., Port, Coast., Ocean
Engrg. (1997).
25. S. Guignard, R. Marcer, V. Rey, K. Kharif and Fraunié. Solitary wave break-
ing on sloping beaches; 2-D two phase flow numerical simulation by SL-VOF
method. Eur. J. Mech. B-Fluids 20, 57–74 (2001).
26. J. V. Hall and J. W. Watts. Laboratory investigation of the vertical rise
of solitary waves on impermeable slopes. Tech. Memo. 33, Beach Erosion
Board, U.S. Army Corps of Engrs. (1953).
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 37

27. S. Hibberd and D. H. Peregrine. Surf and run-up on a beach: a uniform

bore. J. Fluid Mech 95, 323–345 (1979).
28. D. V. Ho, R. E. Meyer and M. C. Chen. Long surf. J. Mar. Res. 21, 219–232
(1963).
29. S.-H. Hsiao, P. L.-F. Liu and Y. Chen. Nonlinear water waves propagating
over a permeable bed. Phil. Trans. R. Soc. Lond. A 458, 1291–1322 (2002).
30. K. Hu, C. G. Mingham and D. M. Causon. Numerical simulation of wave
overtopping of coastal structures using the non-linear shallow water equa-
tions. Coast. Eng. 41, 433–465 (2000).
31. M. E. Hubbard and N. Dodd. A 2D numerical model of wave run-up and
overtopping. Coast. Eng. 47, 1–26 (2002).
32. F. Imamura. Review of tsunami simulation with a finite difference method.
In Long-wave runup models, Eds. H. Yeh, C. E. Synolakis and P. L.-F. Liu
(World Scientific Publishing Co., 1996), p. 25–42.
33. E. Isaacson. Water waves over a sloping bottom. Comm. Pure Appl. Math 3,
11–31 (1950).
34. A. Jensen, G. Pedersen and D. J. Wood. An experimental study of wave
run-up at a steep beach. J. Fluid. Mech. 486, 161–188 (2003).
35. B. Johns. Numerical integration of the shallow water equations over a slop-
ing shelf. Int. J. Numer. Methods Fluids 2, 253–261 (1982).
36. H. Johnsgard. A numerical model for run-up of breaking waves. Int. J. Num.
Meth. Fluids 31, 1321–1331 (1999).
37. H. Johnsgard and G. Pedersen. A numerical model for three-dimensional
run-up. Int. J. Num. Meth. Fluids 24, 913–931 (1997).
38. U. Kânoǧlu and C. E. Synolakis, Analytic solutions of solitary wave runup
on the conical island and on the reverse beach. In Long-wave runup models,
Eds. H. Yeh, C. E. Synolakis and P. L.-F. Liu (World Scientific Publishing
Co., 1996), p. 215–220.
39. U. Kânoǧlu and C. E. Synolakis. Long wave runup on piecewise linear to-
pographies. J. Fluid Mech. 374, 1–28 (1998).
40. T. Kataoka and M. Tsutahara. Transverse instability of surface solitary
waves. J. Fluid Mech. 512, 211–221 (2004).
41. H. B. Keller, A. D. Levine and G. B. Whitham. Motion of a bore over a
sloping beach. J. Fluid Mech. 7, 302–316 (1960).
42. A. B. Kennedy, Q. Chen, J. T. Kirby and R. A. Dalrymple. Boussinesq
modeling of wave transformation, breaking, and run-up. Part I: 1D. J. Wa-
terw., Port, Coast., Ocean Engrg. 126(1), 39–47 (2000).
43. S. K. Kim, P. L.-F. Liu and J. A. Liggett. Boundary integral solutions for
solitary wave generation, propagation and run-up. Coast. Eng. 7, 299–317
(1983).
44. J. T. Kirby. Funwave software download page (1998).
URL: http://chinacat.coastal.udel.edu.
45. J. T. Kirby, G. Wei, , Q. Chen, A. B. Kennedy and R. A. Dalrymple.
Fully nonlinear Boussinesq wave model documentation and user’s manual.
Research Report CACR-98-06, Center for applied Coastal research, De-
partment of Civil Engineering, University of Delaware, Newark, DE 19716,
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

38 G. Pedersen

September 1998.
URL: http://chinacat.coastal.udel.edu.
46. N. Kobayashi, A. K. Otta and I. Roy. Wave reflection and run-up on rough
slopes. J. Waterw., Port, Coast., Ocean Engrg. Div. ASCE. 113, 282–298
(1987).
47. Z. Kowalik and T. S. Murty. Numerical simulation of two-dimensional
tsunami run-up. Marine Geodesy 16, 87–100 (1993).
48. R. J. LeVeque and D. L. George. High-resolution finite volume methods for
the shallow water equations with bathymetry and dry states. In Advanced
Numerical Models for Simulating Tsunami Waves and Runup, eds. P. L.-
F. Liu, H. Yeh and C. Synolakis (World Scientific Publishing Co., 2008),
p. 43–73.
49. Y. Li and F. Raichlen. Solitary wave runup on plane slopes. J. Wa-
terw., Port, Coast., Ocean Engrg. 127(1), 33–44 (2001).
50. Y. Li and F. Raichlen. Non-breaking and breaking solitary wave run-up.
J. Fluid Mech. 456, 295–318 (2002).
51. Pengzhi Lin, Kuang-An Chang and Philip L.-F. Liu. Runup and rundown
of solitary waves on sloping beaches. J. Waterw., Port, Coast., Ocean En-
grg. 125(5), 247–255 (1999).
52. P. L.-F. Liu, Y.-S. Cho, M. J. Briggs, U. Kânoǧlu and C. E. Synolakis.
Runup of solitary waves on a circular island. J. Fluid Mech. 302, 259–285
(1995).
53. P. L.-F. Liu, C. E. Synolakis and H. Yeh. Report on the international work-
shop on long-wave run-up. J. Fluid Mech. 229, 675–688 (1991).
54. S. Longo, M. Petti and I. J. Losada. Turbulence in the swash and surf zones:
a review. Coastal Engineering 45, 129–147 (2002).
55. D. R. Lynch and W. G. Gray. Finite element simulation of flow in deforming
regions. J. Comp. Phys. 36, 135–153 (1980).
56. P. J. Lynett and P. L.-F. Liu. Coulwave model page (2004).
URL: http://ceprofs.tamu.edu/plynett/COULWAVE/default.htm.
57. P. J. Lynett, T.-R. Wu and P. L.-F. Liu. Modeling wave runup with depth-
integrated equations. Coast. Eng. 46, 89–107 (2002).
58. P. A. Madsen, H. B. Bingham and H. Liu. A new Boussinesq method for
fully nonlinear waves from shallow to deep water. J. Fluid Mech. 462, 1–30
(2002).
59. P. A. Madsen, R. Murray and O. R. Sörensen. A new form of the Boussinesq
equations with improved linear dispersion characteristics. Coast. Eng. 15,
371–388 (1991).
60. P. A. Madsen and H. A. Schäffer. Higher-order Boussinesq-type equations
for surface gravity waves: derivation and analysis. Phil. Trans. R. Soc. Lond.
A 356, 3123–3184 (1998).
61. P. A. Madsen and H. A. Schäffer. A review of Boussinesq-type equations for
surface gravity waves. Advances in Coastal and Ocean Engineering, vol. 5
(World Scientific Publishing Co., 1999), p. 1–95.
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 39

62. P. A. Madsen, O. R. Sörensen and H. A. Schäffer. Surf zone dynamics simu-

lated by a Boussinesq-type model: Part I. model description and and cross-
shore motion of regular waves. Coast. Eng. 32, 255–287 (1997).
63. J. C. Martin, W. J. Moyce, W. G. Penney, A. T. Price and C. K. Thornhill.
Some gravity wave problems in the motion of perfect liquids. Philosophical
Transactions of Royal Society of London, Ser. A 244, 231–334 (1952).
64. F. Mesinger and A. Arakawa. Numerical methods used in atmospheric mod-
els. GARP, Publ. Ser. WMO 17, (1976).
65. R. E. Meyer. On the shore singularity of water wave theory. Part 1. The
local model. Physics of Fluids 19, 3142–3163 (1986).
66. R. E. Meyer. On the shore singularity of water wave theory. Part 2. Small
waves do not break on gentle beaches. Physics of Fluids 19, 3164–3171
(1986).
67. R. E. Meyer and A. D. Taylor. Run-up on beaches. In Waves on Beaches
and Resulting Sediment Transport (Academic Press, 1972) p. 357–411.
68. J. W. Miles. Solitary waves. Ann. Rev. Fluid Mech. 12, 11–43 (1980).
69. R. L. Miller. Experimental determination of run-up of undular and fully
developed bores. J. Geophys. Res. 73, 4497–4510 (1968).
70. J. J. Monaghan and A. Kos. Solitary waves on a Cretan beach. J. Wa-
terw., Port, Coast., Ocean Engrg. 125(3), 145–154 (1999).
71. O. Nwogu. Alternative form of Boussinesq equations for nearshore wave
propagation. J. Waterw., Port, Coast., Ocean Engrg. 119(6), 618–638
(1993).
72. H. T. Özkan-Haller and J. T. Kirby. A Fourier-Chebyshev collocation
method for the shallow water equations including shoreline run-up. Applied
Ocean Research 19, 21–34 (1997).
73. Pacific Marine Environmental Laboratory. Tsunami Reseach Program
(2005),
URL: http://www.pmel.noaa.gov/tsunami/research.html.
74. G. Pedersen. Run-up of periodic waves on a straight beach. Preprint Series
in Applied Mathematics, ISBN 82-553-0952-7 1/85, Dept. of Mathematics,
University of Oslo, Norway (1985).
75. G. Pedersen and B. Gjevik. Run-up of solitary waves. J. Fluid Mech. 135,
283–299 (1983).
76. G. Pedersen. A Lagrangian model applied to runup problems. In Advanced
Numerical Models for Simulating Tsunami Waves and Runup, eds. P. L.-
F. Liu, H. Yeh and C. Synolakis (World Scientific Publishing Co., 2008),
p. 311–315.
77. E. Pelinovsky, O. Kozyrev and E. Troshina. Tsunami runup in a sloping
channel. In Long-wave runup models, Eds. H. Yeh, C. E. Synolakis and
P. L.-F. Liu (World Scientific Publishing Co., 1996), p. 332–339.
78. D. H. Peregrine. Long waves on a beach. J. Fluid Mech. 77, 417–431 (1967).
79. D. H. Peregrine. Equations for water waves and the approximation behind
them. In Waves on beaches, Ed. E. Meyer (Academic Press, New York,
1972), p. 357–412.
80. D. H. Peregrine. Breaking waves on beaches. Ann. rev. of Fluid Mech. 15,
149–178 (1983).
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

40 G. Pedersen

81. D. H. Peregrine and S. M. Williams. Swash overtopping a truncated plane

beach. J. Fluid Mech. 440, 391–399 (2001).
82. J. Petera and V. Nassehi. A new two-dimensional finite element model
for the shallow water equations using a Lagrangian framework constructed
along fluid particle trajectories. Int. J. Num. Meth. in Eng. 39, 4159–4182
(1996).
83. M. Petti and S. Longo. Turbulence experiments in the swash zone. Coastal
Engineering 43, 1–24 (2001)
84. R. S. Prasad and I. A. Svendsen. Moving shoreline boundary condition for
nearshore models. Coast. Eng. 49(4), 239–261 (2003).
85. J. A. Puleo, R. A. Beach, R. A. Holman and J. S. Allen. Swash zone sediment
suspension and transport and the importance of bore-generated turbulence.
J. Geophys. Res. 105(C7), 17021–17044 (2000).
86. R. O. Reid and B. R. Bodine. Numerical model for storm surges in Galveston
Bay. J. Waterw., Port, Coast., Ocean Engrg. Div. ASCE 94(WW1), 33–57
(1968)
87. M. C. Shen and R. E. Meyer. Climb of a bore on a beach 3, run-up. J. Fluid
Mech. 16, 113–125 (1963).
88. F. Shi and W. Sun. A variable boundary model of storm surge flooding in
generalized curvilinear grids. Int. J. Numer. Methods Fluids 21, 641–651
(1995).
89. N. Shuto. Run-up of long waves on a sloping beach. Coastal Engng. in
Japan 10, 23–38 (1967).
90. A. Sielecki and M. G. Wurtele. The numerical integration of the nonlinear
shallow-water equations with sloping boundaries. J. Comp. Phys. 6, 219–236
(1970).
91. P. A. Sleigh, P. H. Gaskell, M. Berszins and N. G. Wright. An unstructured
finite-volume algorithm for predicting flow in rivers and estuaries. Comput-
ers & Fluids 27(4), 479–508 (1998).
92. L. Q. Spielvogel. Single wave run-up on sloping beaches. J. Fluid Mech. 74,
685–694 (1976).
93. J.J. Stoker. Water waves. (Interscience, 1957).
94. C. E. Synolakis. The run-up of solitary waves. J. Fluid Mech. 185, 523–545
(1987).
95. C. E. Synolakis, M. K. Deb and J. E. Skjelbreia. The anomalous behaviour
of the run-up of cnoidal waves. Phys. Fluids 31, 3–5 (1988).
96. C. E. Synolakis and J. E. Skjelbreia. Evolution of maximum amplitude
of solitary waves on plane beaches. J. Waterw., Port, Coast., Ocean En-
grg. 119(3), 323–342 (1993).
97. S. Tadepalli and C. Synolakis. The run-up of N-waves on sloping beaches.
Proc. R. Soc. Lond. A 445, 99–112 (1994).
98. T. Takagi. Finite element analysis in bench mark problems 2 and 3. In Long-
wave runup models, Eds. H. Yeh, C. E. Synolakis and P. L.-F. Liu (World
Scientific Publishing Co., 1996), p. 258–264.
99. H. Takeda. Numerical simulation of run-up by variable transformation. J. of
the Oceanographical Society of Japan 40, 271–278 (1984).
August 8, 2008 11:12 WSPC/Trim Size: 9in x 6in for Review Volume 01˙Pedersen

Modeling Runup with Depth Integrated Equation Models 41

100. W. C. Thacker. Some exact solutions to the nonlinear shallow-water wave

equations. J. Fluid Mech. 107, 499–508 (1981).
101. V. V. Titov and C. E. Synolakis. Modeling of breaking and nonbreaking
long-wave evolution and runup using VTCS-2. J. Waterw., Port, Coast.,
Ocean Engrg. 121(6), 308–316 (1995).
102. V. V. Titov and C. E. Synolakis. Numerical modeling of tidal wave runup.
J. Waterw., Port, Coast., Ocean Engrg. 124(4), 157–171 (1998).
103. E. O. Tuck and L.-S. Hwang. Long wave generation on a sloping beach.
J. Fluid Mech. 51(3), 449–461 (1972).
104. T. Umetsu. A boundary condition technique of moving boundary simulation
for broken dam problem by three-step explicit finite element method. In
Advances in Hydro-Science and Engineering, Vol. II, (Tsinghua University
Press, Beijing, 1995), p. 394–399.
105. A. Walters and T. Takagi. Review of finite element methods for tsunami
simulations. In Long-wave runup models, Eds. H. Yeh, C. E. Synolakis and
P. L.-F. Liu (World Scientific Publishing Co., 1996), p. 43–87.
106. G. Watson, D. H. Peregrine and E. F. Toro. Numerical solution of the shal-
low water equations on a beach using the weighted average flux method. In
Computational Fluid Dynamics ’92 Eds. Ch. Hirsch, J. Pèriaux and W. Ko-
rdulla (Elsevier Science Publishers, 1992), p. 495–502.
107. G. Wei, J. T. Kirby, S. T. Grilli and R. Subramanya. A fully nonlinear
Boussinesq model for surface waves. Part 1. Highly nonlinear unsteady
waves. J. Fluid Mech. 294, 71–92 (1995).
108. J.-K. Woo and P. L.-F. Liu. Finite element model for modified Boussi-
nesq equations. I: Model development. J. Waterw., Port, Coast., Ocean
Engrg. 130(1), 1–16 (2004).
109. D.J. Wood, G. Pedersen and A. Jensen. Modelling of run-up of steep non-
breaking waves. Ocean Engineering 30, 625–644 (2003).
110. T. Y. Wu. Long waves in ocean and coastal waters. Proc. ASCE, J. Eng.
Mech. Div. 107, 501–522 (1981).
111. H. Yeh, A. Ghazali and I. Marton. Experimental study of a bore run-up.
J. Fluid Mech. 206, 563–578 (1989).
112. H. Yeh, P. L.-F. Liu, M. Briggs and C. E. Synolakis. Propagation and am-
plification of tsunamis at coastal boundaries. Nature 372, 353–355 (1994).
113. H. Yeh, C. E. Synolakis and P. L.-F. Liu, Eds., Long-wave runup models.
(World Scientific Publishing Co., 1996).
114. J. A. Zelt. The run-up of nonbreaking and breaking solitary waves.
Coast. Eng. 15, 205–246 (1991).
115. J. A. Zelt and F. Raichlen. A Lagrangian model for wave-induced harbour
oscillations. J. Fluid Mech. 213, 203–225 (1990).
116. J. A. Zelt and F. Raichlen. Overland flow from solitary waves. J. Wa-
terw., Port, Coast., Ocean Engrg. 117, 247–263 (1991).
117. X. Zhou, Y. C. Hon and K. F. Cheung. A grid-free, nonlinear shallow-
water model with moving boundary. Engineering Analysis with Boundary
Elements 28(8), 967–973 (2004).
This page intentionally left blank
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

CHAPTER 2

HIGH-RESOLUTION FINITE VOLUME METHODS FOR

THE SHALLOW WATER EQUATIONS WITH
BATHYMETRY AND DRY STATES

Randall J. LeVeque1 and David L. George2

Department of Applied Mathematics, University of Washington
Box 352420, Seattle, WA 98195-2420
E-mail: 1 [email protected], 2 [email protected]

We give a brief review of the wave-propagation algorithm, a high-

resolution finite volume method for solving hyperbolic systems of conser-
vation laws. These methods require a Riemann solver to resolve the jump
in variables at each cell interface into waves. We present a Riemann solver
for the shallow water equations that works robustly with bathymetry and
dry states. This method is implemented in clawpack and applied to
benchmark problems from the Third International Workshop on Long-
Wave Runup Models, including a two-dimensional simulation of runup
during the 1993 tsunami event on Okushiri Island. Comparison is made
with wave tank experimental data provided for the workshop. Some pre-
liminary results using adaptive mesh refinement on the 26 December
2004 Sumatra event are also presented.

1. Introduction
We will present a brief introduction to a class of high-resolution finite vol-
ume methods for hyperbolic problems and discuss the application of these
methods to long-wave runup problems using the shallow water equations. To
solve the benchmark problems for this workshop we have used such meth-
ods in one and two space dimensions that work robustly with bathymetry
(bottom topography) and dry states, and that automatically handle the
moving interface between water and land. Some results on the benchmark
problems are presented in Secs. 6 and 8, and more results, along with some
animations, may be found at the website15 .
We use a mathematical framework known as the wave-propagation al-
gorithm that has been implemented in the software package clawpack
(Conservation Laws Package) in 1, 2, and 3 space dimensions, and which

43
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

44 R. J. LeVeque and D. L. George

also includes adaptive mesh refinement capabilities. This algorithm gives

a general formulation of a class of finite volume methods known as “high-
resolution shock-capturing Godunov-type methods” that are second order
accurate in space and time on smooth solutions while automatically cap-
turing discontinuities in the solution (including shocks or hydraulic jumps)
with minimal numerical smearing and no spurious oscillations. More details
on these algorithms and their application to hyperbolic problems may be
found in LeVeque12,14 and the clawpack software documentation, avail-
able at www.clawpack.org.
The fact that these are finite volume methods means that, rather than
pointwise approximations to the solution, the numerical solution consists
of approximations to the cell averages of the solution over grid cells. Here
the grid cells are assumed to be intervals [xi−1/2 , xi+1/2 ] of uniform length
∆x in one dimension or rectangles [xi−1/2 , xi+1/2 ] × [yj−1/2 , yj+1/2 ] in two
dimensions, but more general nonuniform grids can also be used.
Finite volume methods are particularly appropriate when solving sys-
tems of conservation laws, in which case the integral of the solution over
each grid cell is modified only due to fluxes through the edges of the grid
cell. Dividing this statement by the cell area leads to an update formula for
the cell averages based on numerical approximations to the flux through
each edge, as written out below. Such a finite volume method is based di-
rectly on the integral form of the conservation law and can be applied to
problems with discontinuous solutions more reliably than finite difference
approximations to the differential equation form of the conservation law,
which does not hold at a discontinuity.
In one space dimension the integral form of a conservation law is
Z
d x2
q(x, t) dx = f (q(x1 , t)) − f (q(x2 , t)) ∀x1 , x2 , (1)
dt x1
where q(x, t) ∈ lRm is the vector of conserved quantities and f (q) is the
flux function. This states that the total mass of q in any interval [x1 , x2 ]
changes only due to fluxes through the edges.
The shallow water equations on a flat bottom have this form with m = 2
and
   
h hu
q= , f (q) = , (2)
hu hu2 + 21 gh2
where h is the fluid depth, u is the horizontal velocity, and g is the grav-
itational constant. These equations express the conservation of mass and
momentum.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 45

If the solution is sufficiently smooth, then the integral conservation law

(1) can be manipulated to yield
Z x2
qt (x, t) + f (q(x, t))x = 0 ∀x1 , x2 ,
x1

and hence
qt + f (q)x = 0, (3)
which is the PDE form of the conservation law (with subscripts denoting
partial derivatives). This equation is called hyperbolic if the Jacobian matrix
f 0 (q0 ) ∈ lRm×m is diagonalizable and has real eigenvalues for any physically
relevant state q0 . Hyperbolic problems typically model wave propagation
and the eigenvalues correspond to the propagation velocities if we linearize
about the state q0 . For the shallow water equations given by (2),
 
0 0 1
f (q) = , (4)
−u2 + gh 2u
with eigenvalues
p p
λ1 = u − gh, λ2 = u + gh, (5)
and corresponding eigenvectors
       
1 1 1 1
r1 = √ = , r2 = √ = . (6)
u − gh λ1 u+ gh λ2
A finite volume method in conservation form updates the cell average
Qni of the solution over the grid cell using an expression
∆t n
Qn+1 = Qni − [F n
− Fi−1/2 ], (7)
i
∆x i+1/2
where
Z xi+1/2
1
Qni ≈ q(x, tn ) dx,
∆x xi−1/2
Z tn+1
(8)
n 1
Fi−1/2 ≈ f (q(xi−1/2 , t)) dt
∆t tn

are the numerical approximations to the cell average and interface flux,
respectively. The update (7) comes directly from integrating (1) in time
from tn to tn+1 and dividing by ∆x. Equation (7) can also be viewed as
n
a direct discretization of the PDE (3), but viewing the value Fi−1/2 as
an approximation to the interface flux is key in developing high-resolution
methods for nonlinear problems.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

46 R. J. LeVeque and D. L. George

We will generally drop the superscript n on quantities at time tn to

simplify the notation below, particularly since other superscripts will be
needed.

1.1. Godunov’s Method

The methods we use are extensions of Godunov’s method, a first-order
method for gas dynamics developed in the 1950s in which the interface flux
Fi−1/2 is computed by solving a Riemann problem between the states Qi−1
and Qi . This is simply the conservation law with piecewise constant data
(as in a shock tube or dam break problem). Such a problem naturally arises
at the cell interface if the solution at time tn is approximated by a piecewise
constant function with values Qj at all points in the jth grid cell.
Godunov’s method is applicable to any hyperbolic conservation law. For
a linear problem in which f (q) = Aq for some matrix A (so f 0 (q) = A),
the solution to the Riemann problem for any states Qi−1 and Qi consists
of m discontinuities (waves), each proportional to an eigenvector r p of A,
and propagating at speed equal to the corresponding eigenvalue λp (for
p = 1, 2, . . . , m). Since A must be diagonalizable, we can write

A = RΛR−1 ,

where Λ = diag(λ1 , . . . , λm ) and R = [r1 · · · rm ] is the invertible matrix

of eigenvectors.
Solving the Riemann problem between states Qi−1 and Qi is then easily
accomplished by decomposing ∆Qi−1/2 = Qi − Qi−1 into eigenvectors of
A, i.e., writing ∆Qi−1/2 as a linear combination of the vectors r p ,
m
X m
X
Qi − Qi−1 = αpi−1/2 rp ≡ p
Wi−1/2 . (9)
p=1 p=1

We use W p to denote the pth wave in this Riemann solution. The vector of
coefficients αpi−1/2 is given by αi−1/2 = R−1 ∆Qi−1/2 . Godunov’s method
in the linear case is then defined by setting
| |
Fi−1/2 = f (Q∨ ∨
i−1/2 ) = AQi−1/2 ,

|
where Q∨ i−1/2 denotes the value at the interface xi−1/2 in the Riemann
solution,
| X p
Q∨i−1/2 = Qi−1 + Wi−1/2 .
p:λp <0
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 47

Multiplying by A gives
m
X
Fi−1/2 = AQi−1 + αpi−1/2 (λp )− rp
p=1 (10)
= AQi−1 + A− ∆Qi−1/2 ,

where

A− = R diag((λp )− ) R−1 , with λ− = min(λ, 0). (11)

Alternatively, we can write

| X p
Q∨
i−1/2 = Qi − Wi−1/2
p:λp >0

and obtain
m
X
Fi−1/2 = AQi + αpi−1/2 (λp )+ rp
p=1 (12)
= AQi + A+ ∆Qi−1/2 ,

where

A+ = R diag((λp )+ ) R−1 , with λ+ = max(λ, 0). (13)

For a linear problem the resulting method is simply the first-order upwind
method, extended from the scalar advection equation to a general system
by diagonalizing the system and applying the upwind method to each char-
acteristic component in the appropriate direction based on the propagation
velocity.
For nonlinear problems, such as the shallow water equations, the exact
solution to the Riemann problem is harder to calculate but can still be
worked out (see, e.g. LeVeque14 and Toro19 ) and the resulting interface flux
used for Fi−1/2 . In practice, however, it is usually more efficient to use some
approximate Riemann solver to obtain Fi−1/2 . One popular choice is to use
a “Roe solver” following the work of Roe17 for gas dynamics, in which the
data Qi−1 , Qi is used to define a “Roe-averaged” Jacobian matrix Ai−1/2
by a suitable combination of f 0 (Qi−1 ) and f 0 (Qi ). The numerical flux is
then determined by solving the Riemann problem for the linear problem
qt + Ai−1/2 qx = 0. The Roe average is chosen to have the property that

f (Qi ) − f (Qi−1 ) = Ai−1/2 (Qi − Qi−1 ). (14)

July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

48 R. J. LeVeque and D. L. George

This leads to nice properties in the approximate solution. The Roe matrix
for the shallow water equations is easily computed (see, e.g. LeVeque14 ),
and is simply the Jacobian matrix (4) evaluated at the Roe-averaged state
p √
hi + hi−1 ui−1 ghi−1 + ui ghi
ĥi−1/2 = , ûi−1/2 = p √ . (15)
2 ghi−1 + ghi

The eigenvalues, or “Roe speeds”, are therefore

q q
ŝ1i−1/2 = ûi−1/2 − g ĥi−1/2 , ŝ2i−1/2 = ûi−1/2 + g ĥi−1/2 , (16)

and the Roe eigenvectors are

" # " #
1 1 2 1
r̂i−1/2 = 1 , r̂i−1/2 = . (17)
ŝi−1/2 ŝ2i−1/2

In one dimension the clawpack software requires a Riemann

solver that, for any states Qi−1 and Qi , returns a set of Mw
1 Mw
waves Wi−1/2 , . . . , Wi−1/2 , and propagation speeds for the waves,
s1i−1/2 , . . . , sM
i−1/2 . The number of waves Mw may be equal to m, the
w

dimension of the system, but could be different. The Riemann solver must
also return the fluctuations A− ∆Qi−1/2 and A+ ∆Qi−1/2 , two vectors that
are used to update the solution according to
∆t +
Qn+1
i = Qi − (A ∆Qi−1/2 + A− ∆Qi+1/2 ). (18)
∆x
These fluctuations should have the property that

A− ∆Qi−1/2 + A+ ∆Qi−1/2 = f (Qi ) − f (Qi−1 ), (19)

so that they represent a “flux difference splitting”. The fluctuations may

be defined in terms of the interface fluxes as
A+ ∆Qi−1/2 = f (Qi ) − Fi−1/2 ,
(20)
A− ∆Qi+1/2 = Fi+1/2 − f (Qi ).

Then (19) is satisfied (note the shift in index) and (18) reduces to the
flux-differencing update formula (7). The form (18) is used in clawpack
and the general formulation of the wave-propagation algorithms because it
is more flexible and allows the extension of these methods to hyperbolic
problems that are not in conservation form, in which case there is no “flux
function” (see LeVeque14 ).
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 49

The notation A± ∆Qi−1/2 used for the fluctuation vectors is suggested

by the fact that, for a linear problem, the natural choice is
A− ∆Qi−1/2 = A− (Qi − Qi−1 ),
(21)
A+ ∆Qi−1/2 = A+ (Qi − Qi−1 ),
where the matrices A± are defined by (13) and (11). In general they are
often defined by
Mw
X
A− ∆Qi−1/2 = (spi−1/2 )− Wi−1/2
p
,
p=1
(22)
Mw
X
+
A ∆Qi−1/2 = p
(spi−1/2 )+ Wi−1/2 .
p=1

For a linear problem or a nonlinear problem when the Roe solver is used,
this agrees with the definition (20).
The first-order method (18) only uses the fluctuations returned from
the Riemann solver, and does not make explicit use of the waves W p or
speeds sp . These quantities are used in the high-resolution correction terms
discussed in the next section.

1.2. High-Resolution Corrections

The method (18) is only first-order accurate. The second-order Lax-
Wendroff method for the linear problem can be written as a modification
to (18), as
∆t + ∆t
Qn+1
i = Qi − (A ∆Qi−1/2 +A− ∆Qi+1/2 )− (F̃i+1/2 −F̃i−1/2 ), (23)
∆x ∆x
where the fluctuations are still given by (21) for the linear problem and the
correction fluxes are
   
1 ∆t 2 1 ∆t
F̃i−1/2 = |A| − A ∆Qi−1/2 = I− |A| |A|∆Qi−1/2 ,
2 ∆x 2 ∆x
(24)
where |A| = A+ −A− . Inserting this in (23) and manipulating yields a more
familiar form of the Lax-Wendroff method,
∆t ∆t2 2
Qn+1
i = Qi − A(Qi+1 − Qi−1 ) + A (Qi+1 − 2Qi + Qi−1 ). (25)
2∆x 2∆x2
This method is second-order accurate on smooth solutions but is highly
dispersive and nonphysical oscillations arise near steep gradients or discon-
tinuities in the solution, which can completely destroy the accuracy. These
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

50 R. J. LeVeque and D. L. George

oscillations can be avoided by using the form (23) and applying appropriate
limiters to the correction terms. To this end we rewrite (24) as
m  
1X ∆t p fp
F̃i−1/2 = I− |λ | |λp |W i−1/2 , (26)
2 p=1 ∆x
p
where Wi−1/2 = αpi−1/2 rp are the waves obtained from the Riemann solu-
tion, as in (9), and Wfp
i−1/2 represents a “limited” version of the wave. Each
wave is limited by comparing it to the wave in the same family arising from
the Riemann problem at the neighboring interface in the upwind direction,
i.e. we set
fp
W p p
i−1/2 = limiter(Wi−1/2 , WI−1/2 ), (27)

where

i−1 if λp > 0
I=
i+1 if λp < 0.
p p
If WI−1/2 and Wi−1/2 are “comparable” in some sense then this compo-
nent of the solution is presumed to be smoothly varying. In this case the
corresponding term in (26) can be expected to give a useful correction that
will help improve accuracy and the limiter should return W fp p
i−1/2 ≈ Wi−1/2 .
p p
However, if WI−1/2 and Wi−1/2 quite different then this component is not
smoothly varying and attempting to add an additional term from the Tay-
lor series may make things worse rather than better. In this case the limiter
should modify the wave, typically be reducing its magnitude. There is an
extensive theory of limiters that will not be further discussed here.
The method (23) with correction fluxes (26) is easily extended to non-
linear problems. Recall that the (approximate) Riemann solver returns fluc-
p
tuations A± ∆Qi−1/2 , waves Wi−1/2 , and speeds spi−1/2 . The only change in
the formulas required in order to apply (23) to a general nonlinear problem
is to replace λp in (26) by the local speed spi−1/2 , obtaining

Mw  
1X ∆t p
F̃i−1/2 = I− |s | |spi−1/2 |W
fp
i−1/2 . (28)
2 p=1 ∆x i−1/2

2. The f-Wave Approach

The method described above can be reformulated in a way that will prove
particularly useful when source terms are added to the equations, as needed
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 51

p
to handle variable bathymetry. Recall that the waves Wi−1/2 correspond to
a splitting of the jump in Q at the interface xi−1/2 ,
Mw
X p
Qi − Qi−1 = Wi−1/2 .
p=1

In a linear problem, or a nonlinear problem that has been locally lin-

earized using a Roe matrix Ai−1/2 , these waves are obtained by expressing
p
∆Qi−1/2 as a linear combination of the eigenvectors ri−1/2 of the matrix,
p p p p
i.e., Wi−1/2 = αi−1/2 ri−1/2 for some scalars αi−1/2 , as is done in (9). Al-
ternatively, we could split the jump in f (Q) into eigencomponents as
Mw
X Mw
X
p p p
f (Qi ) − f (Qi−1 ) = βi−1/2 ri−1/2 ≡ Zi−1/2 .
p=1 p=1

If the matrix Ai−1/2 satisfies Roe’s condition (14), then we simply have
p
Zi−1/2 p
= spi−1/2 Wi−1/2 . For other approximate Riemann solvers it is nec-
essary to determine an appropriate splitting of f (Q) based on the splitting
of Q
The vectors Z p are called f-waves because they carry jumps in f rather
p
than jumps in q. Since Zi−1/2 = sgn(spi−1/2 )|spi−1/2 |Wi−1/2
p
for linearized
Riemann solvers, the natural generalization of the correction flux (28) for
the f-wave formulation is
Mw  
1X ∆t
F̃i−1/2 = I− sgn(spi−1/2 ) Zei−1/2
p
, (29)
2 p=1 ∆x
p p
where Zei−1/2 is a limited version of Zi−1/2 calculated using the same limiter
p
as previously applied to W .
One advantage of the f-wave approach is that any linearly independent
p
set of vectors ri−1 can be used to define the splitting of ∆f into waves
p
Zi−1/2 and the method remains conservative. Of course a reasonable choice
is required in order to maintain consistency with the differential equation,
but for example the eigenvectors of any reasonable approximate Jacobian
matrix based on Qi−1 and Qi could be used without needing to impose
the Roe condition (14). For the shallow water equations this suggests using
vectors
   
1 1
r1 = 1 , r2 = 2 , (30)
s s
where s1 and s2 are some approximations to the wave speeds of the two
waves in the Riemann solution. Taking s1 and s2 to be the Roe speeds
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

52 R. J. LeVeque and D. L. George

(16) recovers the Roe solver, but in some cases this is not a good choice.
In particular the Roe solver can fail when dry states are expected in the
solution. In Sec. 4 we present a different choice of speeds that can be used
much more robustly.

3. Bathymetry and Source Terms

We now consider the shallow water equations over non-constant bathymetry
with elevation y = B(x). In this case h(x, t) represents the depth of the
water above the bathymetry. A sloping bottom can accelerate the fluid and
gives rise to a source term in the momentum equation proportional to the
slope Bx (x). The shallow water equations now take the form
qt + f (q)x = ψ(q, x), (31)
where q and f are as in (2) and the source term is
 
0
ψ(q, x) = . (32)
−ghBx
One way to tackle (31) numerically is is to use a fractional step procedure.
In each time step one first solves the homogeneous conservation law (3) to
advance Qn by ∆t to obtain an intermediate state Q∗ , and then solves
qt = ψ(q, x) (33)
over time ∆t to advance Q∗ to Qn+1 . This often works well, but is subject
to numerical difficulties, particularly in situations where there is a steady
state with f (q) = ψ(q, x) and the desired dynamic solution is a relatively
small perturbation of this steady state. In this case solving (3) and (33)
may each lead to significant changes in the solution that should nearly
cancel out. Numerically, the use of distinct numerical methods in separate
steps can lead to errors that swamp the desired solution when the waves of
interest are small perturbations of the steady state.
Instead of using a fractional step method, we modify the f-wave formu-
lation of the hyperbolic solver and incorporate the source term into the flux
difference before decomposing this into waves, i.e. we decompose
Mw
X p
f (Qi ) − f (Qi−1 ) − ∆xΨi−1/2 = Zi−1/2 , (34)
p=1

where Ψi−1/2 is a discretization of the source term. For the shallow

water equations with bathymetry the source term ∆xghBx at xi−1/2
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 53

is approximated by 21 g(hi−1 + hi )(Bi − Bi−1 ), resulting in the vector

f (Qi ) − f (Qi−1 ) − ∆xΨi−1/2 taking the form
" #
hi ui − hi−1 ui−1


.
hi u2i + 21 gh2i − hi−1 u2i−1 + 12 gh2i−1 + 21 g(hi−1 + hi )(Bi − Bi−1 )
(35)
This vector is decomposed into f-waves, for example by writing it as a linear
p
combination of the eigenvectors ri−1/2 , (p = 1, 2) of the Roe matrix or of
(30).
Of particular importance is the special case of a motionless body of
water over variable bathymetry, in which case u ≡ 0 and h(x, t) + B(x) is
initially constant and should remain so. In this case the vector (35) becomes
" #
0
1 2 2

1 . (36)
2 g hi − hi−1 + 2 g(hi−1 + hi )(Bi − Bi−1 )

If hj + Bj is constant in j then Bi − Bi−1 = hi−1 − hi and (36) is the

1
zero vector, resulting in Zi−1/2 2
= Zi−1/2 = 0 and A± ∆Qi−1/2 = 0 in (34).
n+1
Hence Qi = Qi and the steady state is exactly maintained numerically.
Moreover, if we are modeling waves in which perturbations in h + B are
small compared to variations in B, then this vector captures the pertur-
bations after canceling out the steady state portion of the variation. The
wave limiters and high resolution correction terms are applied to these per-
turbations rather than to waves that include the large variation in h due to
the bathymetry. As a result, this method is much more accurate in mod-
eling the propagation of small amplitude perturbations than a fractional
step method. This may be quite important in calculating the long-range
propagation of small-amplitude tsunamis through the ocean.
The f-wave approach and its use for source terms is discussed further in
Bale et al.1 and LeVeque14 . A variety of other approaches have also been
developed recently for this problem, for example Gosse7 , Greenberg et al.8 ,
Jenny and Müller10 , Kurganov and Levy11 , LeVeque13 .

4. Dry States
It is well known that if the Roe solver is used to solve a Riemann problem
in which ui−1 < ui with sufficient difference in velocities, then the approxi-
mate Riemann solution will have a negative depth in the intermediate state
(see Fig. 15.3 in LeVeque14 for an illustration of this). This nonphysical be-
havior often causes the computation to crash. In reality a dry state should
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

54 R. J. LeVeque and D. L. George

form if the velocity difference is sufficiently large, although the Roe solver
can fail even for smaller velocity differences (see Einfeldt et al.6 for some
discussion of related issues for the Euler equations). Similar problems arise
when solving a Riemann problem with one state initially dry, hi−1 = 0 or
hi = 0, as happens at some cell interfaces for any problem involving wave
motion at the shore.
This difficulty can be avoided by using the f-wave approach discussed
in Secs. 2 and 3 using eigenvectors r 1 and r2 from (30) with a better choice
of speeds than the Roe speeds. In most cases we use the “Einfeldt speeds”
p p
s1E = min(ui−1 − ghi−1 , ŝ1 ), s2E = max(ui + ghi , ŝ2 ), (37)

obtained by comparing the Roe speeds with the characteristic speeds λ1i−1
and λ1i (the eigenvalues of the Jacobian matrices in states Qi−1 and Qi ).
This choice is adapted from the suggestion of Einfeldt5 that speeds cor-
responding to these be used in the HLL method for gas dynamics. The
HLL approximate Riemann solver (after Harten et al.9 ) simply uses two
waves to approximate the Riemann solution (regardless of the dimension
m of the system) with speeds s1 and s2 approximating the minimum and
maximum wave speeds arising in the system. The HLLE method, using the
Einfeldt speeds, chooses these speeds by comparing the Roe speed, a rea-
sonable choice if the wave is a shock, and the extreme characteristic speed,
which may be faster if the wave is instead a spreading rarefaction wave.
Since m = 2 in the one-dimensional shallow water equations, the method
we use is closely related to the HLLE method, though not the same and
will be modified further to handle source terms and dry states below. (See
LeVeque and Pelanti16 for some more discussion of the relation between the
HLL and Roe solvers and the f-wave approach.)
Using the f-wave approach with the choice of speeds (37) and corre-
sponding eigenvectors (30) nearly always maintains non-negative depth. In
fact, it can be shown that the total mass in the intermediate Riemann
solution is always positive given these speeds and eigenvectors. However,
as explained below, it is sometimes necessary to further modify the wave
speeds to maintain non-negativity, at least in the subcritical case when
s1E < 0 < s2E . In the supercritical case when both speeds have the same
sign, no modification is needed.
We first consider the case of preserving non-negativity in a cell that has
positive depth initially, hi−1 > 0 or hi > 0, and later consider the case of
preserving non-negativity in a cell that is already initially dry. Even in the
case where both cells have a positive depth initially, hi−1 > 0 and hi > 0,
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 55

a negative depth can be generated if s1E < 0 < s2E when the choice (37)
is used. Although the total mass in the intermediate Riemann solution is
positive, it may happen that the approximate Riemann solution leads to
the mass going negative on one side of the interface. In this case it can
be shown that the mass must be increasing on the other side by at least
the same amount, and so negativity can be avoided by a transfer of mass
that can be accomplished by increasing the speed on the side losing mass.
Working out the formula to increase this speed just to the point where the
negative state reaches h = 0, we find that in general the following speeds
can always be used:
 p 
s1 = min s1E , s2E /2 − (s2E /2)2 + max(0, s2E ∆1 − ∆2 )/hi−1 , (38)
 p 
s2 = max s2E , s1E /2 + (|s1E |/2)2 + max(0, ∆2 − s1E ∆1 )/hi . (39)

That is, given hi−1 > 0 initially, (38) will maintain that hi−1 ≥ 0, and
given hi > 0 initially (39) will maintain that hi ≥ 0. Here ∆1 and ∆2 are
the components of the modified flux difference

f (Qi ) − f (Qi−1 ) − ∆xΨi− 12 , (40)

which takes into account the bathymetry. Each speed (38) and (39) always
corresponds to the Einfeldt speed unless a negative state would arise on
that side, so for a given Riemann problem, at most one of s1 and s2 is
different from the Einfeldt speed, and only when necessary to maintain
non-negativity.
Maintaining the non-negativity in a cell with a depth that is initially 0
is handled somewhat differently. If Bi−1 = Bi , and only one of hi−1 or hi
is positive, then the true solution to this Riemann problem consists only of
√
a rarefaction wave with the leading
p edge propagating at speed ui − 2 ghi
if hi−1 = 0 or speed ui−1 + 2 ghi−1 if hi = 0. We use these speeds as s1
or s2 if hi−1 = 0 or hi = 0 respectively. As stated above, using the speed
(38) will preserve hi−1 ≥ 0 if it is initially positive, and (39) will preserve
hi ≥ 0 if it is initially positive. Using the speed of the leading edge of a
rarefaction wave for the complimentary speed however, will not necessarily
prevent the dry state from becoming negative. It can be shown however
that negativity is only
p possible if the true rarefaction wave is transonic
√
(s1 < 0 < ui−1 + 2 ghi−1 if hi = 0 or ui − 2 ghi < 0 < s2 if hi−1 = 0).
Transonic rarefactions are a more general problem, and are discussed in the
following section.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

56 R. J. LeVeque and D. L. George

5. Entropy Conditions
Integral conservation laws can be satisfied by discontinuous weak solutions
subject to the Rankine-Hugoniot jump conditions. This results in possible
non-uniqueness of weak solutions — an initial value problem might be sat-
isfied by both a smooth solution and a discontinuous one. Determining the
physically relevant solution requires additional admissibility conditions, of-
ten taking the form of an “entropy” function that is conserved except across
a discontinuity. The name arises from the Euler equations of gas dynam-
ics, where the entropy must increase when gas passes through a shock. For
the shallow water equations, the “entropy” function is actually mechanical
energy, which must decrease when passing through a discontinuity.
Since the integral conservation laws alone do not guarantee a unique
solution to an initial value problem, a numerical method based on the
conservation laws alone might converge to an entropy violating weak so-
lution. An “entropy fix” is therefore often needed. Determining such a fix
for Godunov-type methods requires a careful look at the particular Rie-
mann solver being used. If an approximate solver is used, true solutions to
Riemann problems which consist of m waves — any combination of rar-
efactions and shock waves, are replaced by m jump discontinuities locally
at each grid cell. These discontinuities might approximate a physically rel-
evant shock wave, or perhaps a smooth rarefaction. In the latter case, the
jump discontinuity still approximates the conservation law, however it more
closely resembles the entropy violating shock than the physically relevant
rarefaction.
With the wave propagation methods, the waves arising from a particu-
lar Riemann problem at a grid cell interface are averaged onto the adjacent
cells. Therefore, the local discrepancy between an entropy violating dis-
continuity and a smooth rarefaction may have no effect on the numerical
solution, if both produce the same average within a grid cell. This will be the
case if the wave structure of the rarefaction remains entirely within a grid
cell. The exceptional case is a transonic rarefaction — a rarefaction where
one of the eigenvalues passes through zero. This type of rarefaction has a
wave structure that should overlap a cell interface, yet it is approximated
by a jump discontinuity moving either to the left or the right. This does
affect the numerical solution, and can cause a method to converge globally
to an entropy violating weak solution. The fix is to determine when the
correct entropy solution to a Riemann problem corresponds to a transonic
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 57

rarefaction, and then split the entropy violating single wave, apportioning
it to the adjacent grid cells appropriately.
An alternative and numerically more useful formulation of the entropy
condition for the shallow water equations is that a physically correct shock
in the pth (p = 1, 2) characteristic family must have the p-characteristics
impinging on it. That is,
λp (ql ) > s > λp (qr ), (41)
p th
where s is the shock speed and λ is the p eigenvalue evaluated at ql
and qr — the states directly to the left and right of the shock respectively.
Therefore, if the entropy solution to a Riemann problem contains a shock
connecting the left state Qi−1 to the middle state, denoted Qm , then
λ1 (Qi−1 ) > λ1 (Qm ). (42)
Similarly if a shock connects the right state Qi to the middle state Qm ,
then
λ2 (Qm ) > λ2 (Qi ). (43)
If (42) or (43) is violated, then in fact a rarefaction connects the corre-
sponding states in the entropy solution. As explained above, an entropy
violating Riemann solution will not affect the numerical solution, except in
the case of a transonic rarefaction. Therefore the only cases in which an
entropy fix is required are when
λ1 (Qi−1 ) < 0 < λ1 (Qm ) (44)
or
λ2 (Qm ) < 0 < λ2 (Qi ), (45)
which indicate a transonic rarefaction in the first or second families
respectively.
It is easy to check λ1 (Qi−1 ) and λ2 (Qi ). However, with the f-wave ap-
proach, Qm is never explicitly computed so we cannot simply evaluate
λ1 (Qm ) and λ2 (Qm ) directly. In fact the f-wave approach is not equiv-
alent to using a single value for Qm , but two middle values, one to the
left and one to the right of the cell interface. The approach we’ve taken
is to instead compare the Roe speeds, ŝ1i− 1 and ŝ2i− 1 , with the right and
2 2
left speeds, λ1 (Qi−1 ) and λ2 (Qi ). The motivation for this approach is that
the Roe speeds can serve as an estimate for the shock speeds, allowing
us to estimate whether (41) is satisfied. For instance, to detect the pres-
√
ence of a transonic rarefaction in the second family, ui + ghi is compared
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

58 R. J. LeVeque and D. L. George

√
to ŝ2i− 1 . If ui + ghi > ŝ2i− 1 then most likely the true Riemann solu-
2 2
tion has a rarefaction in the second family. Furthermore ŝ2i− 1 serves as
2
an estimate for the characteristic speed at the center of the rarefaction
fan, ŝ2i−1/2 ≈ 21 (λ2 (Qm ) + λ2 (Qi )) and hence λ2 (Qm ) ≈ 2s2i−1/2 − λ2 (Qi ).
Therefore if
p p
2(ŝ2i− 1 ) − (ui + ghi ) < 0 < ui + ghi , (46)
2

then the true Riemann solution is likely to have a transonic rarefaction in

the second characteristic family. The second f-wave Z 2 should then be split
into two waves, one moving to the right the other to the left. A similar test
is done for the first characteristic family.
The entropy fix when one state in the Riemann problem is initially dry
is somewhat different, and also acts to ensure that the depth in the corre-
sponding cell remains non-negative. As mentioned in the previous section,
the true Riemann solution in such a case consists of a single rarefaction.
The speeds of the edges ofpthe rarefaction wave are p given by quantities in
1 2
the wet cell (s = ui−1 − ghi−1 and s = ui−1 + 2 ghi−1 if hi−1 > 0 or
√ √
s1 = ui − 2 ghi and s2 = ui + ghi if hi > 0). In the event of a single
transonic rarefaction, s1 < 0 < s2 , an entropy fix such that the f-waves
simply carry an appropriate proportion of the true single wave is necessary.
We simply apportion to each f-wave an amount based on the proportion of
the rarefaction in each cell.
This approximate Riemann solver works quite robustly in the context
of the first-order accurate Godunov’s method, in the f-wave formulation
with bathymetry source terms and dry states. Addition of the second-order
correction terms (28) adds new potential problems when dry states are
present, as these waves can lead to a negative depth even when the first or-
der method would not. Standard wave limiters devised to avoid nonphysical
oscillations may not completely avoid this problem, and so we have added
an additional limiting procedure to maintain nonnegative depth.
The procedure is best illustrated by writing the second-order method
as the sum of the first-order Godunov update
∆t +
QGi = Qi −
n
(A ∆Qi−1/2 + A− ∆Qi+1/2 ), (47)
∆x
and the second-order correction fluxes (29),
∆t h i
Qn+1
i = QGi − F̃i+ 12 − F̃i− 21 . (48)
∆x
Note that the correction flux at a cell interface takes mass away from one
cell and adds mass to the adjacent cell, with the direction depending on
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 59

the sign of its first component. The gross mass-flux out of the ith grid cell,
due to these correction fluxes, is therefore
h i
1 1
Mi = max(0, F̃i+ 1 ) − min(0, F̃
i− 1 ) . (49)
2 2

If ∆tMi is larger than the mass present after the Godunov update,
∆x(QG 1
i ) , the correction fluxes could potentially create a negative depth
in this cell. This is prevented by re-limiting the correction fluxes based on
which cell they take mass away from
F̃i− 12 → ϕi− 12 F̃i− 21 , (50)
where
(
min(1, ∆x(QG 1
i ) /∆tMi ) if 1
F̃i− 1 < 0
ϕi− 12 = 2
(51)
min(1, ∆x(QG 1
i−1 ) /∆tMi−1 ) if
1
F̃i− 1 > 0.
2

This procedure is consistent with the standard limiting strategy, in that

second-order accuracy is achieved throughout most of the domain, and lim-
iting only occurs where there are problematic features, such as shock waves
or near dry-states.

6. Results for Benchmark Problem 1

The first benchmark problem consists of a linear sloping beach, b(x, t) =
−x, and initially motionless water with a shoreline at x = 0. An incoming
wave is induced by a non-zero initial surface elevation, η(x, 0), specified
by data provided from x = 0 to 50000 meters in increments of 50 meters.
We compare the resulting motion of the shoreline, the surface elevation,
and the velocity field of our numerical solution with a provided analytical
solution. The surface elevation and velocity data for the analytical solution
were provided at three separate times, t = 160, 175, and 220 seconds. The
position and velocity of the shoreline were provided for t ∈ [0, 300] seconds.
We computed this problem using clawpack, on a series of grids of
varying resolution (from 500 to 5000 grid cells), each over a domain x ∈
[−500, 50000] meters. The computational grid cells were clustered densely
near the shoreline, using a piecewise linear mapping. This was required to
efficiently capture the fine-scale motion of the shoreline, which occurred
over a small fraction of the entire domain. Convergence to the analytical
solutions was observed as the grids were refined.
Some of our computational results are shown with the analytical solu-
tions in Figs. 1 and 2. We have chosen to show the results on a grid of
 July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

60 R. J. LeVeque and D. L. George

η(x, t) at time t = 160 seconds

80
beach
60 computed
40
exact

20
0
m

−20
−40
PSfrag replacements
−60
−80
200 300 400 500 600 700 800
m
u(x, t) at time t = 160 seconds

15 computed
exact
10

5
u (m/s)

−5

−10
PSfrag replacements
−15
m/s
200 300 400 500 600 700 800
m
Fig. 1. Top: Water surface elevation at t = 160s, shown near the beach. Bottom: Veloc-
ity field at t = 160s in the same region. Both computations were done on a 1000-point
grid.

1000 points so that the numerical results are still distinguishable from the
analytical solution. The surface elevation and the velocity field are shown
at t = 160s in Fig. 1. Note that the figures show only a small portion of
the domain near the beach. Figure 2 shows the motion of the shoreline
for the same computation. For additional results at other times and grid
resolutions, see the website15 .
 July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 61

Shoreline Position Shoreline Velocity

250 10
Computed Solution Computed Solution
Exact Solution Exact Solution
200
5
150

100 0

50
meters

m/s
−5
0

−50 −10

−100
−15
−150

PSfrag replacements −200

0 50 100 150
PSfrag
200 250
replacements
300 350 400
−20
0 50 100 150 200 250 300 350 400
sec. sec.

Fig. 2. Left: Position of the shoreline as a function of time, computed on a 1000-point

grid. Right: Velocity of the shoreline for the same computation.

7. Extension to Two Space Dimensions

In two space dimensions the shallow water equations with bathymetry take
the form
qt + f (q)x + g(q)y = ψ(q, x, y), (52)
where
    
h hu hv
     
q =  hu  , f (q) =  hu2 + 12 gh2  , g(q) =  huv , (53)
hv 2 1 2
huv hv + 2 gh

and the source term is ψ = ψ1 + ψ2 with

   
0 0
   
ψ1 (q, x, y) =  −ghBx(x, y)  , ψ2 (q, x, y) =  0 , (54)
0 −ghBy (x, y)
The general form of the wave-propagation algorithm is now
∆t +
Qn+1
ij = Qij − (A ∆Qi−1/2,j + A− ∆Qi+1/2,j )
∆x
∆t +
− (B ∆Qi,j−1/2 + B − ∆Qi,j+1/2 ) (55)
∆y
∆t ∆t
− (F̃i+1/2,j − F̃i−1/2,j ) − (G̃i,j+1/2 − G̃i,j−1/2 ).
∆x ∆y
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

62 R. J. LeVeque and D. L. George

The fluctuations A± ∆Q and B ± ∆Q are determined by solving one-

dimensional Riemann problems normal to each cell interface, using the
f-wave formulation described above to incorporate the appropriate portion
of the source term. In the x-direction, for example, we solve the Riemann
problem qt + f (q)x = ψ1 . The solution to this Riemann problem is exactly
the same as the one-dimensional case with the addition of a contact discon-
tinuity wave that passively advects the jump in the orthogonal velocity v.
Using only these terms in (55) without the correction fluxes F̃ or G̃ gives
a two-dimensional generalization of Godunov’s method sometimes called
donor-cell upwind. A better first-order method (corner transport upwind)
is obtained by splitting the waves in each one-dimensional Riemann solu-
tion into waves moving in the transverse direction, so that wave motion
oblique to the grid is more properly modeled. In clawpack this requires
the specification of a “transverse Riemann solver” that takes as input a
fluctuation, say A+ ∆Q, and returns a splitting of this vector into up-going
and down-going portions that affect the fluxes G̃i,j+1/2 and G̃i,j−1/2 re-
spectively. This is typically based on the eigenvalues and eigenvectors of
some approximate Jacobian matrix g 0 (q) in the transverse direction, and
is described in more detail for the shallow water equations in LeVeque14 .
Second-order correction terms can also be incorporated as in one space di-
mension, based on the waves obtained from the one-dimensional Riemann
solution normal to each cell edge.
The presence of dry states leads to new complications when the algo-
rithm is extended to two space dimensions. A wave moving transversely
into a dry or nearly dry cell can produce a negative depth unless spe-
cial care is taken by incorporating a more fully multidimensional limiter.
Currently we do not use transverse propagation or second-order correction
terms in two dimensions and are further developing this approach. Our goal
is to ultimately be able to use transverse propagation and high-resolution
correction terms for dry state problems in two dimensions. However, even
without these terms reasonable results are obtained, as demonstrated in
the next section.

8. Results for Benchmark Problem 2

In the second benchmark problem, we compare our computational solutions
to laboratory data collected from a wave tank experiment. The wave tank
was built as a 1:400 scale model, approximating coastline bathymetry near
Monai, Japan, a region that suffered inundation from the 1993 Okushiri
 July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 63

PSfrag replacements
η

Fig. 3. Three-dimensional view of the numerical solution at one time.

tsunami. In the experiment, an incoming wave was induced by mechanical

wave paddles along one side of the tank, and the resulting motion of the
surface elevation was measured by gages at three separate locations. Ad-
ditionally, a movie was made showing the tank from overhead during the
experiment. This data and movie are available at the workshop website.
We computed this problem with clawpack, originally using a uniform
single grid of 275×175 cells. The computational domain modeled a subre-
gion of the tank, (x, y) ∈ [0, 5m] × [0, 3.5m], which includes an inlet region
that experienced large wave runup in the tsunami. We used the provided
data that specifies the measured surface elevation of an incoming wave along
one side of a subregion of the tank, x = 0, for the period t = [0, 22.5s]. (Note:
the computational domain is rotated in the figures to match the orientation
of the movie, so x runs upward and y from left to right.)
The computed surface elevation is shown at various times in Figs. 3
through 6. Figure 3 shows a three-dimensional view of the computed so-
lution at one time, Fig. 4 shows the solution during the primary runup
period, and Fig. 5 compares the solution with several close-up snapshots
from the wave tank movie, demonstrating similarities in the zone where the
maximum runup occurred.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

64 R. J. LeVeque and D. L. George

Fig. 4. The computed solution during the primary runup of the wave. The primary
runup occurred in the first 30 seconds. Contour lines show the topography that was
initially above the water surface, including an island. Gray scale indicates elevation
above the original water surface.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 65

Fig. 5. Comparison of the numerical solution (right) with snapshots from the overhead
movie of the laboratory experiment (left). Frames 11, 26, 41, 56 from the movie are shown
and computed results are shown at corresponding 30 second intervals. (The movie shows
30 frames per second.)
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

66 R. J. LeVeque and D. L. George

Surface Elevation at Channel 5 (4.521,1.196) Surface Elevation at Channel 7 (4.521,1.696)

0.04 0.05
Computed Solution Computed Solution
Experimental Solution Experimental Solution

0.03 0.04

0.02 0.03
meters

meters
0.01 0.02

0 0.01

−0.01 0

−0.02 −0.01
0 10 20 30 40 50 0 10 20 30 40 50
sec. sec.

Surface Elevation at Channel 9 (4.521,2.196)

Surface Elevation at Channel 7 (4.521,1.696)
0.05
Computed Solution 0.04
Experimental Solution Computed Solution
Experimental Solution
0.035
0.04
0.03

0.03 0.025

0.02
meters

meters

0.02 0.015

0.01
0.01
0.005

0
0
−0.005

−0.01 −0.01
0 10 20 30 40 50 0 50 100 150 200
sec. sec.

Fig. 6. Comparison of the surface elevation, with the laboratory measurements at the
three wave gages (Channels 5, 7, 9). The numerical solution is comparable during the
primary runup in the first 50 seconds. The bottom right figure shows the comparison for
the duration of the laboratory measurements, at one of the gages (Channel 5). Sustained
oscillations in the laboratory measurements are not seen in the numerical solution, as
discussed in the text. The other gages showed similar patterns at later times.

Laboratory measurements at three gage locations were provided for the

workshop, denoted Channel 5, 7, and 9. This data and the numerical so-
lution (surface elevation as a function of time) at these three locations are
shown in Fig. 6. The numerical solution was comparable to the laboratory
measurements for the first 50 seconds of the experiment, during which the
primary runup of the wave and several reflected waves are seen. For the
last 150 seconds of the experiment, the wave tank measurements exhibited
large oscillations that were not evident in the numerical solution. This dis-
crepancy is shown in the last graph of Fig. 6. Based on discussions at the
workshop, it is believed that this is due to the fact that the wave maker
in the experiment did not generate a perfectly clean wave. The incoming
wave data provided for the benchmark problem was only over 22.5 seconds,
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 67

and was followed by other waves in the wave tank for which no data was
provided.

9. Adaptive Mesh Refinement and Large-Scale Tsunami

Propagation
In two-dimensional calculations over large spatial domains it may be crucial
to use nonuniform grids to achieve the desired resolution in some regions
without an excessive number of grid cells overall. For tsunami modeling
there are two types of nonuniformity that may be desirable. First, we may
want to have a finer grid near the shore where the details of the runup must
be computed, or in other regions where the bathymetry has a large impact
on the wave propagation behavior and eventual runup. In principle this
can be achieved with “static refinement” of some spatial regions relative to
others, though ideally computations on the finer grid would only be done at
times when the wave is present. Second, we may want to use an adaptively
refined grid in which the region of refinement moves along with the tsunami
to provide good resolution of the wave at all times, with a minimal number
of grid points in regions where nothing is happening.
Since the Catalina workshop took place, the tragic Sumatra tsunami
of December 26, 2004 has prompted increased study of the all aspects of
tsunamis. Our own efforts in recent months have focused on developing an
adaptive mesh refinement code that works well on the scale of the Indian
Ocean (or other oceans), with the ability to capture tsunami propagation
across the ocean and couple it to the study of runup on much smaller scales
along particular stretches of coastline. In the course of this work we have
also improved the Riemann solver beyond what is described in this paper to
make it even more robust. The new version no longer requires increasing the
wave speeds above the Einfeldt speeds, and instead is based on introducing
additional waves in the Riemann solution using ideas from LeVeque and
Pelanti16 . This work will be reported in more detail elsewhere. See the
webpage15 for pointers.
The clawpack software includes amrclaw, an adaptive mesh refine-
ment version of the code developed with Marsha Berger and described in
more detail in Berger and LeVeque2 . More recently the clawpack formula-
tion has also been incorporated into ChomboClaw by Donna Calhoun at the
University of Washington, allowing the Chombo package4 of C++ routines
for adaptive refinement to be applied. This package uses similar algorithms
to amrclaw but has additional features such as an MPI implementation
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

68 R. J. LeVeque and D. L. George

with load balancing for adaptive refinement on parallel computers, and sup-
port for implicit algorithms that may be needed if dissipative or dispersive
terms are added to the equations.
A rectangular grid is refined by covering a portion of the domain by one
or more rectangular patches of grids that are finer by a factor of k, some
integer. This process can be repeated recursively, with each grid level fur-
ther refined by grid patches at a higher level or refinement. The maximum
number of levels is specified along with the refinement factor at each level.
The grid patches are chosen by flagging grid cells at each level that “need
refining” (see below) and then clustering these cells into a set of rectangular
patches that cover these cells and a limited number of other cells. This is
done by solving an optimization problem (using the algorithm of Berger
and Rigoutsos3 ) that takes into account the trade-off between refining too
many cells unnecessarily and creating too many grid patches, since there is
some overhead associated with applying the algorithm on each patch. The
boundary data (ghost cell values) on a patch must be generated by space-
time interpolation from data on the coarser grid. A time step of length
∆t is first taken on the coarse grid, boundary data is then generated from
the fine grid patches, and k 0 time steps of length ∆t/k 0 are then taken on
the fine grids to reach the same time. For most AMR applications on hy-
perbolic problems we take k 0 = k, refining in time by the same factor as
in space in order to maintain a comparable Courant number at all levels.
However, for tsunami propagation and runup applications where we only
have the finest grids near the shore, it may be desirable to refine in time
√
by a smaller factor k 0 < k. Recall that the wave speeds are roughly gh,
which differs by more than an order of magnitude between the deep ocean
and the coastal regions. The smaller wave speed near shore allows a larger
time step.
After updating to the same time on the fine grids, the coarse grid so-
lution on any grid cell covered by the finer grid is then replaced by the
average of the fine grid values in this cell in order to transfer the more ac-
curate solution to the coarse grid. Additional modifications to the values are
needed near the patch edge to maintain global conservation as described in
Berger and LeVeque2 . This is done recursively at each level. While the inner
workings are somewhat complicated, the general formulation in amrclaw
allows extension of most clawpack computations directly to adaptively
refined grids. The computational time required for the overhead associated
with multiple grids is often negligible compared to the savings achieved by
concentrating fine grids only where needed.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 69

We can flag the cells that “need refinement” however we wish. For the
calculation presented below, we have allowed some refinement everywhere
based on a measure of the variation in the solution, so that the propagating
wave is well resolved. In addition, we allow additional levels of refinement
near the shore in particular regions of interest, though only once the wave
is approaching. Regridding is performed every few time steps to modify the
region of refinement as the wave propagates.
In the course of this work, we ran into several difficulties at the interfaces
between grids that have not been observed in other applications of the
AMR codes. These arise from the representation of the bathymetry and
shore on a Cartesian grid. One problem, for example, is that a coarse grid
cell that is dry (if the bathymetry value in this cell is above sea level) may
be refined into some cells that are above sea level and others below. Even
though h = 0 on the coarse cell we cannot initialize h to zero on all the
fine cells without generating nonphysical wave motion. The fine cells below
sea level must be initialized with a positive h to maintain the constant
sea level. This problem and related difficulties could only be solved by
some substantial reworking of the amrclaw code. The result is a special-
purpose program that incorporates these algorithmic modifications and can
now be applied to many tsunami scenarios. It is currently being tested by
comparing predictions with measurements made at various places around
the Indian Ocean in the wake of the 26 December 2004 Sumatra earthquake.
Some preliminary results are shown in Fig. 7, and additional results and
movies are linked from the webpage15 and will be reported more fully in
future publications.
The top two frames of Fig. 7 show the Bay of Bengal at two early
times. A coarse grid is used where nothing is yet happening and the grid
cells are shown on this “Level 1 grid”, which has a mesh width of one
degree (approximately 111 km). The rectangular region where no grid lines
are shown is a Level 2 grid with mesh width 8 times smaller, about 14
km. Red represents water elevation above sea level, dark blue is below the
undisturbed surface. Figure 7(c) shows a zoomed view of the southern tip of
India and Sri Lanka at a later time. The original Level 1 grid is still visible
along the left-most edge, but the rest of the region shown has been refined
by a Level 2 grid. Due north from Sri Lanka, along the coast of India, there
is a region near Chennai where two additional levels of refinement have
been allowed at this stage in the computation. The grid lines on Level 3
are not shown; the mesh width on this level is about 1.7 km, a factor of 8
finer than Level 2. A Level 4 grid is also visible in the center of this region
and appears as a small black rectangle.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

70 R. J. LeVeque and D. L. George

(a) Time 01:07:10 (550 seconds) (b) Time 01:34:40 (2200 seconds)

(c) Time 03:13:12.5 (8112.5 seconds) (d) Time 03:15:30 (8250 seconds)

(e) Time 03:24:40 (8800 seconds) (f) Time 03:35:30 (9450 seconds)

Fig. 7. Propagation of the 26 December 2004 tsunami across the Indian Ocean, using
adaptive mesh refinement with refinement by a factor of 4096 from the coarsest grid
shown in (a)–(b) to the finest grid shown in (d)–(f). The latter figures show zoomed
views of the region near the harbors of Chennai, India. Times are GMT (and seconds
since initial rupture at 0:58 GMT). See the text for more details.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 71

Figure 7(d) shows a further zoomed view of the coast near Chennai. In
this figure the grid lines show the Level 3 grid. The Level 4 grid is refined
by a factor of 64 relative to the Level 3 grid, so the mesh width is about
27 meters. Grid lines on this finest level are not shown. The rectangle in
this figure shows a region that is expanded in Figs. 7(e) and 7(f) to show
the two harbors of Chennai. The fine-scale bathymetry used in this com-
putation was obtained by digitizing navigational charts. In this simulation
the commercial harbor to the south is inundated, with the tsunami over-
topping the surrounding sea wall, while the fishing harbor to the north is
largely untouched. This appears to agree with what was actually observed,
although we are still investigating this. Moreover, we do not yet have suffi-
ciently accurate data on the height of the sea walls enclosing each harbor.
One data point known accurately from tide gage data is the arrival time
for the initial wave in Chennai, which was at 9:05 local time (3:35 GMT).
This is well matched by our simulation: this is essentially the time shown
in Fig. 7(f).
A more careful study of this region will be performed in the future and
presented elsewhere, including comparisons with runup data collected by
Harry Yeh as part of the Tsunami Survey Team that visited this area in
February, 2005. See the webpage15 for more recent results and movies of
the simulations. In the future we plan to also compare with data collected
by other teams at various other points around the Indian Ocean. We will
also make our computer code available to the community for others to use,
in a form that should allow the local bathymetry from other regions to be
easily incorporated. See the webpage for more details.
Adaptive mesh refinement is crucial for this simulation. The calculation
shown here was run on a single-processor 3.2 GHz PC under linux. Figure
7(c) was obtained after about 15 minutes of running time, Fig. 7(d) about
2.5 hours later, indicating that most of the grid cells are concentrated on
the Level 4 grid, which is introduced only when the wave approaches the
shore in this one region. Were it possible to use the finest grid over the
entire domain, the result of Fig. 7(c) would require more than 4000 years
of computing on the same machine.

10. Conclusions and Future Work

Finite volume methods using an approximate Riemann solver devised to
deal with bathymetry and dry states have been successfully used for Bench-
mark Problems 1 and 2. Benchmark problems 3 and 4 at the workshop
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

72 R. J. LeVeque and D. L. George

model landslide-induced tsunamis, and involve bathymetry that changes

with time. In principle our code can handle this situation but we have not
yet done extensive tests of this.
Adaptive mesh refinement is crucial for large scale problems and we are
further developing and testing the AMR version of our method. We are
also working on formulating this method on the full earth, by switching to
latitude-longitude coordinates on the sphere and calculating on a compu-
tational rectangle in these coordinates using periodic boundary conditions
in longitude. The AMR code is already capable of handling this situation
and expect to soon be able to study the global propagation and compare
with tide gage data (see, for example, Titov et al.18 ) that is available from
many points around the world following the 26 December 2004 event.

Acknowledgments
This work was supported in part by NSF grants DMS-0106511 and CMS-
0245206, and by DOE grant DE-FC02-01ER25474. The authors would like
to thank Marsha Berger and Donna Calhoun for assistance with the adap-
tive refinement aspect of this work.

References
1. D. Bale, R. J. LeVeque, S. Mitran, and J. A. Rossmanith, A wave-propagation
method for conservation laws and balance laws with spatially varying flux
functions, SIAM J. Sci. Comput., 24 (2002), pp. 955–978.
2. M. J. Berger and R. J. LeVeque, Adaptive mesh refinement using wave-
propagation algorithms for hyperbolic systems, SIAM J. Numer. Anal., 35
(1998), pp. 2298–2316.
3. M. J. Berger and I. Rigoutsos, An algorithm for point clustering and grid
generation, IEEE Trans. Sys. Man & Cyber., 21 (1991), pp. 1278–1286.
4. P. Colella et al., chombo software, http://seesar.lbl.gov/anag/chombo/,
2005.
5. B. Einfeldt, On Godunov-type methods for gas dynamics, SIAM J. Num.
Anal., 25 (1988), pp. 294–318.
6. B. Einfeldt, C. D. Munz, P. L. Roe, and B. Sjogreen, On Godunov type
methods near low densities, J. Comput. Phys., 92 (1991), pp. 273–295.
7. L. Gosse, A well-balanced flux-vector splitting scheme designed for hyperbolic
systems of conservation laws with source terms, Comput. Math. Appl., 39
(2000), pp. 135–159.
8. J. M. Greenberg, A. Y. LeRoux, R. Baraille, and A. Noussair, Analysis
and approximation of conservation laws with source terms, SIAM J. Numer.
Anal., 34 (1997), pp. 1980–2007.
July 24, 2008 10:23 WSPC/Trim Size: 9in x 6in for Review Volume 02˙leveque

High-Resolution Finite Volume Methods for the Shallow Water Equations 73

9. A. Harten, P. D. Lax, and B. van Leer, On upstream differencing and

Godunov-type schemes for hyperbolic conservation laws, SIAM Review, 25
(1983), pp. 35–61.
10. P. Jenny and B. Müller, Rankine-Hugoniot-Riemann solver considering
source terms and multidimensional effects, J. Comp. Phys., 145 (1998),
pp. 575–610.
11. A. Kurganov and D. Levy, Central-upwind schemes for the Saint-Venant
system, Math. Model. and Numer. Anal., 36 (2002), pp. 397–425.
12. R. J. LeVeque, Wave propagation algorithms for multi-dimensional hyper-
bolic systems, J. Comput. Phys., 131 (1997), pp. 327–353.
13. R. J. LeVeque, Balancing source terms and flux gradients in high-resolution
Godunov methods: The quasi-steady wave-propagation algorithm, J. Com-
put. Phys., 146 (1998), pp. 346–365.
14. R. J. LeVeque, Finite Volume Methods for Hyperbolic Problems (Cambridge
University Press, 2002).
15. R. J. LeVeque and D. L. George,
http://www.amath.washington.edu/~rjl/catalina04/.
16. R. J. LeVeque and M. Pelanti, A class of approximate Riemann solvers
and their relation to relaxation schemes, J. Comput. Phys., 172 (2001),
pp. 572–591.
17. P. L. Roe, Approximate Riemann solvers, parameter vectors, and difference
schemes, J. Comput. Phys., 43 (1981), pp. 357–372.
18. V. Titov, A. B. Rabinovich, H. O. Mofjeld, R. E. Thomson, and F. I.
Gonzalez, The global reach of the 26 December 2004 Sumatra tsunami,
Science, 309 (2005), pp. 2045–2048.
19. E. F. Toro, Shock-Capturing Methods for Free-Surface Shallow Flows (Wiley
and Sons Ltd., UK, 2001).
This page intentionally left blank
 CHAPTER 3

SPH MODELING OF TSUNAMI WAVES

Benedict D. Rogers and Robert A. Dalrymple*

Department of Civil Engineering, The Johns Hopkins University
3400 N Charles St, Baltimore, Maryland, 21218, U.S.A.
E-mail: *[email protected]

Smoothed particle hydrodynamics (SPH) is used to simulate both two

and three-dimensional cases of tsunami generation and propagation. A
combined SPH-LES type scheme or sub-particle scale (SPS) scheme is
used to describe viscous effects. Results for 2-D compare well with other
nonlinear numerical models in the case of a 2-D landslide-generated
tsunami. The application of the 2-D model to the 3-D landslide
problem is shown to be inappropriate, necessitating a fully 3-D model.
However, the resolution required for 3-D is beyond the current
capabilities of the JHU-SPH code necessitating high performance
parallel computing.

1. Introduction
While the mechanisms that generate the tsunami flows are generally
understood, the ability to predict the flows they produce at coastlines still
represents a formidable challenge due to the complexities of coastline
formations and the presence of numerous coastal structures that interact
and alter the flow. Indeed, overland flows kill more people alone than
any other aspect of tsunamis but are still not well understood or
accurately predictable. Numerous experimental investigations have
focused on different instances of tsunami runup such as circular islands
(Briggs et al. 1994). However, the accurate prediction of runup onto
real coastlines with structures and human habitation still remains
incomplete.

75
76 B. D. Rogers and R. A. Dalrymple

Theoretical approaches to the problem are extremely difficult to

apply due to strong nonlinearities of tsunamis, the essential three-
dimensionality of the flow, and the dramatic turbulent flow that results
from a breaking tsunami. A few numerical models, such as shock-
capturing schemes (Toro et al. 1999) and Reynolds-Averaged Navier-
Stokes (RANS) (Lin et al. 1999), have been used to predict the flow of
tsunamis and runup/rundown in highly simplifed cases. However, as yet,
such numerical schemes have been unable to capture the flow following
the passage of a tsunami when it breaks onto a beach and interacts with
structures in a destructive manner. Furthermore, the use of depth-
averaged techniques, which also includes Boussinesq models (Lynett
et al. 2002), does not adequately predict the full structure of the flow,
which is evidently important given that upon landfall tsunamis are
propagating over dry land and are often interacting with structures on
the same scale as the depth. Hence, fully 3-D numerical models are
essential.
To date however, the numerical study of three-dimensional
phenomena has been very difficult due to their complexity, the physical
scales of the flow regimes, and the presence of the free surface. As part
of this development, 3-D boundary element numerical models are in
development to predict the initial propagation of the surface disturbance
(Grilli et al. 2002). Techniques such as Eulerian grid-based Large Eddy
Simulation (LES) and RANS (Christensen and Deigaard 2001, Lin et al.
1999) offer one approach to solve for these flows since they can resolve
important components of the flow such as mean flow and turbulence.
Alternatively, the development of meshfree methods for CFD (Li and
Liu 2002) offers a different realizable approach to tackle the simulation
of such difficult problems. Of the meshfree techniques developed over
the past decade, Smoothed Particle Hydrodynamics (SPH) has been the
most popular and successful when applied to free-surface hydrodynamics.
Smoothed Particle Hydrodynamics (SPH) is a Lagrangian technique
originally developed for the simulation of non-axisymmetric problems in
astrophysics that predicts the motion of discrete particles with time.
Since its initial development in the late 1970s by Gingold and Monaghan
(1977) and Lucy (1977) the technique has been extended to many
other applications including free-surface hydrodynamics, primitive
 SPH Modeling of Tsunami Waves 77

impact-fracture problems (Randles and Libersky 1996) and modelling

solid-solid impact (Johnson and Beissel 1996).
The core of the SPH formalism lies in the discrete summation over
disordered points as an approximation to integrals. A major attraction of
the SPH technique is that the need for fixed computational grids is
removed when calculating spatial derivatives. Instead, estimates of
derivatives are provided by analytical expressions. Thus, it has a distinct
advantage over other grid-based methods of solving for turbulent
tsunami propagation problems in that there is no meshing required.
Furthermore, the domain can be multiply connected, and there is no need
for special treatment of the free surface; thus wave breaking, and other
flows that result in fluid separation, are modeled as readily as other
flows.
SPH has been used to describe a variety of free-surface flows
including solitary wave propagation over a planar beach (Monaghan and
Kos 1999), plunging breakers (Tulin and Landrini 2000), solid bodies
impacting on the surface (Monaghan 2000) and dam break simulations
(Monaghan 1994). The method predicts fluid pressure, velocities, energy
and particle trajectories for many types of flows making it ideal for
identifying/elucidating formation mechanisms of complicated flow
phenomena that were hitherto intractable. Additionally, using a similar
particle technique known as the Moving Particle Semi-implicit (MPS)
method, Koshizuka et al. (1998) and Gotoh et al. (2002) present two
dimensional results for the movement of passively floating bodies in the
surf zone and an open channel, respectively. Such an approach can be
incorporated into SPH allowing one to examine the effect of tsunami
propagation into coastal areas that have debris within its domain and
whose buildings are subject to destruction.
At Johns Hopkins University (JHU), Dalrymple and Knio (2001) and
Dalrymple et al. (2001) have shown that SPH has great potential for a
wide range of coastal wave propagation problems such as the greenwater
overtopping of offshore platforms, the waves generated by subaerial
landslide impact, and the impact of waves on structures. Gómez-
Gesteira and Dalrymple (2004) have been the first to develop a three-
dimensional free-surface SPH flow model applying it to the impact and
hydrodynamics of a large wave striking and flowing around a simple
78 B. D. Rogers and R. A. Dalrymple

square building — essentially a simple model of a tsunami at the

shoreline. The model is able to track particles, determine forces on
structures, and replicate velocities accurately when compared to
laboratory data. Thus, particle methods now also represent a viable
method for simulating flow around buildings and complicated fluid-
structure interaction problems. More recently, Panizzo and Dalrymple
(2004) applied SPH to the two-dimensional modeling of tsunamis using
the waves generated by underwater sliding block experiments as
verification.
In this paper, Sec. 2 introduces the basic SPH discretization. This is
followed by a description of an LES-type sub-particle-scale formulation
that enhances the basic SPH formulation to model turbulent flow
features. The formulation is then completed by a presentation of the
boundary conditions used within the JHU-SPH scheme. Section 3
presents the numerical results, which are compared with an analytical
solution for tsunami generation due to a two-dimensional landslide, and
experimental data for tsunami generation due to a three-dimensional
landslide. Finally, Sec. 4 draws some conclusions and looks at the future
of applying SPH to the simulation of tsunamis.

2. Numerical Discretization

2.1. Basic SPH Formulation

Smoothed Particle Hydrodynamics (SPH) is based on the approximate

representation of continuous interpolations or integrals by a discrete
particle representation. In the SPH formalism, the value of a flow
quantity, f, at a position vector x is approximated by
f (x ) = ∑ f jW (x − x j )V j , (2.1)
j

where Vj is the volume of the jth particle located at xj with scalar quantity
( )
fj, and W x − x j is the weighting function referred to as the smoothing
kernel. Prior to simulation, the kernel is specified by an analytical
expression making its evaluation in Eq. (2.1) straightforward. In all
results presented, a quadratic kernel has been used for computational
 SPH Modeling of Tsunami Waves 79

speed and to avoid the problems associated with inflexion points in the
derivatives of spline kernels (Liu et al. 2003):
C
W (x − x j ) =  N
( 1
4 )
q2 − q +1 0≤q≤2
(2.2)
 0 otherwise ,
where q = r h , and r = x − x j . In 1-D, C N = C 1 = 3 4 h , in 2-D
C N = C 2 = 3 (2 π h 2 ) and in 3-D C N = C 3 = 15 (16 πh ) . The kernel has a
3

characteristic smoothing length, h, which defines the region of influence,

and is preset to be 1.3∆x for the duration of all simulations where ∆x is
the initial particle spacing.
One of the attractions of SPH is that function derivatives can be
expressed as another summation by simply using a derivative of the
smoothing kernel, i.e.
∂f (x i ) ∂W (x i − x j )
= ∑ ( f j − fi ) Vj . (2.3)
∂xi j ∂xi
This avoids complicated expressions for calculating derivatives since an
analytical expression is known for the specified kernel. Expression (2.3)
also ensures that there is zero divergence for a uniformly constant field.
When simulating fluid flow, SPH is based on a discrete particle
representation of the Navier-Stokes equations in a Lagrangian
formulation
dρ
= − ρ∇.u , (2.4a)
dt
du ∇P 1
=− + g + (∇ ⋅ µ∇ )u , (2.4b)
dt ρ ρ
where ρ is density, µ is laminar viscosity, u is velocity, P is pressure, g is
gravity and t is time. Following Monaghan (1994), to avoid solving the
Poisson equation every timestep for an incompressible system in
simulations involving water, the pressure is given by an approximate
equation of state (Batchelor 1967)
 ρ  γ 
P = B   − 1 , (2.5)
 ρ o  
80 B. D. Rogers and R. A. Dalrymple

where ρo = 1000kg/m3 is a reference density for water, γ = 7 is a

polytropic index, and the factor B is chosen so that water is virtually
incompressible where the speed of sound, c = ∂ P ∂ ρ , does not dictate
a time step that is prohibitively small, that is c ≈ 10u max . Written in
particle form, Eqs. (2.4) become
d ρi
= ∑ m j (u i − u j )⋅ ∇ iWij , (2.6a)
dt j

d ui P Pj 
= −∑ m j  i2 + 2 + Π ij ∇ iWij + g , (2.6b)
ρ 
dt j  i ρj 
where mj is the mass of the jth particle, ∇iWij = ∇iW (x − x j ) , the first
subscript i referring to the derivative of W with respect to the coordinates
of particle i, and Πij is an artificial viscosity term. The artificial viscosity
term (see Monaghan 1992) is used to account for the viscous terms in the
governing equations. For free-surface hydrodynamics, the artificial
viscosity term also prevents the simulation from becoming unstable and
is given by (Monaghan 1992)
αµ c ij ij
Π ij = − , (2.7)
ρ ij
where α is an empirical coefficient (usually taken as 0.01 – 0.1),
( ) ( )
cij = 12 ci + c j , ρ ij = 12 ρ i + ρ j and
hu ij ⋅ x ij
µ ij = . (2.8)
rij2 + η 2
Here, u ij = u i − u j , rij = x i − x j and η 2 = 0.01h2 to prevent
singularities.

2.2. Sub-Particle Scale (SPS) Turbulence Model

It is now recognized that the original formulation of viscous effects

in SPH using the conventional form in Eq. (2.7) suggested by
Monaghan (1992) does not provide an adequate or complete description
of the physics in free-surface hydrodynamics. The empirical coefficient,
 SPH Modeling of Tsunami Waves 81

α, is needed for numerical stability for free-surface flows, but, in

practice, is too dissipative. Thus, in recent years there has been a steady
shift towards improved formulations. An important step in this direction
was the work by Morris et al. (1997) who developed an improved
formulation to account for laminar stresses
1 ν o (ρ i + ρ j ) rij ⋅ ∇ iWij
(∇ ⋅ µ∇ )u = ∑mj u ij , (2.9)
ρ i j ρ ij2 rij2 + 0.01h 2
where νo = 10−6 m2/s is the kinematic viscosity. This avoids using
second-order derivatives of the smoothing kernel as these can be
inaccurate under certain circumstances (Morris et al. 1997).
In other areas of CFD, the concept of Large-Eddy Simulation (LES)
has proved useful in capturing coherent turbulent structures within
turbulent flows (see Meneveau and Katz 2000 for a review of LES
models). In LES, the Navier-Stokes equations are filtered spatially over
a length scale similar to the grid size to account for turbulent motions
occurring on scales smaller than the resolved grid resolution. Closure
submodels are then required to model the sub-grid-scale (SGS)
turbulence that completes the system of equations.
Hence, given the unphysical damping behavior associated with the
artificial viscosity approach, the next logical step has been to incorporate
the ideas behind LES into particle simulations. Gotoh et al. (2001) were
the first to introduce the enhancement of an LES-type scheme within
their incompressible MPS technique. Shortly after, Lo and Shao (2002)
became the first to implement an LES-type filtering within their
incompressible version of SPH (I-SPH). This was followed by Shao
and Gotoh (2004) and Gotoh et al. (2004) who successfully applied
their I-SPH-LES scheme to wave-structure interaction problems with a
partially submerged cut-off wall.
In this work, we solve the compressible flow equations, and
accordingly, we are required to include the effects of compressibility in
the development of an SPH-LES scheme. This is achieved by Favre-
averaging ( ~f = ρ f ρ ) (where ‘–’ denotes a particle-scale quantity). This
is a useful form of spatial averaging that does not introduce extra
terms into the continuity equation. Applying an implicit (or flat-top)
82 B. D. Rogers and R. A. Dalrymple

spatial-filter over the particle-scale governing equations, Eqs. (2.4)

become (Yoshizawa 1986)
dρ
= − ρ ∇.uɶ , (2.10a)
dt
~
du 1 1 ~ + 1 ∇.τ ,
= − ∇P + g + (∇ ⋅ ρ ν o ∇ )u (2.10b)
dt ρ ρ ρ

where τ is the sub-particle scale (SPS) stress tensor with elements (in
tensor notation):
~ ~
( )
τ ij = ρ 2ν t S ij − 23 S kk δ ij − 23 ρ C I ∆2δ ij . (2.11)
~
Following Blin et al. (2002), CI = 0.00066. The strain tensor, S ij , is
given by
~
~ 1  ∂u~ ∂u j 
S ij =  i + . (2.12)
2  ∂x j ∂xi 

As the name suggests, sub-particle scale (SPS) schemes are very similar
to sub-grid scale (SGS) schemes used in Eulerian grid-based LES
schemes. The final stress-gradient term in Eq. (2.10) represent the effect
of turbulent motions occuring on length scales smaller than the particle-
scales of the resolved flow. Herein, it is accepted that more sophisticated
viscosity models have been developed for grid-based LES schemes, but,
in this work the standard Smagorinsky model (1963) for the eddy
viscosity appears sufficient. This is given by
ν t = (C s ∆l )2 S , (2.13)

where the Smagorinsky constant, Cs = 0.12, ∆l is the initial particle

spacing and the local strain rate is given by
S = (2S ij S ij )
1/ 2
. (2.14)

Within our SPH scheme we discretize the SPS stresses according to the
symmetric formulation given by Lo and Shao (2002)

1   τ i τj 
∇ ⋅ τ = ∑ m j 2 + 2  ⋅ ∇ iWij .
 (2.15)
ρ ρ ρ j 
i j  i
 SPH Modeling of Tsunami Waves 83

2.3. Boundary Conditions

Boundaries are described by discrete particles that interact with the

interior water particles. In the test cases presented in Sec. 3, open
boundaries are modeled by placing a solid boundary far enough away
from the domain of interest so that it does not interact with the flow. At
present in the JHU-SPH scheme, there are two options to model the
boundary-water particle interaction.

(i) Repulsive force boundary condition

One method to model solid boundaries is by means of a repulsive force

with a singularity as the distance between an approaching water particle
and the wall tends to zero (Monaghan and Kos 1999). The force
experienced by a water particle, f, acts normal to the wall and is given by
f = nR( y ) P( x)ε ( z , u⊥ ) , (2.16)
where n is the normal of the solid wall. The distance y is the
perpendicular distance of the particle from the wall, while x is the
projection of interpolation location xi onto the chord joining the two
adjacent boundary particles. The repulsion function, R(y), is evaluated in
terms of the normalized distance from the wall, q = y 2h , as
1
R( y ) = A (1 − q ) , (2.17)
q
where the coefficient A is
1
A= 0.01ci2 . (2.18)
h
The function P(x) is chosen so that a water particle experiences a
constant repulsive force as it travels parallel to the wall
1  2πx  
P(x ) = 1 + cos   , (2.19)
2  ∆b  
where ∆b is the distance between any two adjacent boundary particles.
Finally, the function ε ( z , u ⊥ ) is a modification to Monaghan and Kos’s
original suggestion and adjusts the magnitude of the force according to
84 B. D. Rogers and R. A. Dalrymple

the local water depth and velocity of the water particle normal to the
boundary
ε ( z, u ⊥ ) = ε ( z ) + ε (u ⊥ ) , (2.20)
where
 0.02 z≥0

ε ( z ) =  z ho + 0.02 0 > z ≥ − ho (2.21)
 1 z ho > 1 ,

and
 0 u⊥ > 0

ε (u ⊥ ) =  20u ⊥ co 20u ⊥ < co (2.22)
 1 20u ⊥ > co .

In Eqs. (2.20)–(2.22), z is the elevation above the local still-water level
( )
ho, u ⊥ = u WaterParti cle − u BoundaryParticle ⋅ n and c o = Bγ ρ o .

(ii) Double-Particle boundary

This technique uses conventional SPH particles that remain part of the
prescribed boundary. While the position and velocity of these particles is
known a priori, the density of these particles is predicted by Eqs. (2.6a)
or (2.10a). This density is then inserted into the equation of state (2.5)
to give the pressure that a boundary particle exerts on a fluid particle
when evaluating the momentum equations (2.6a) or (2.10a). This type of
boundary condition has proved robust for three-dimensional SPH
simulations (Gómez-Gesteira and Dalrymple 2004). However, when
this boundary condition was used for three-dimensional simulations in
Sec. 3.2, it was found to have too much wall friction which became
critical near the shoreline.

3. Numerical Test Cases

The test cases carried out here are chosen from those offered for this
Tsunami workshop. The other test cases involved the modeling of large
domains, which would require far more descretization than feasible for
 SPH Modeling of Tsunami Waves 85

our present model. This highlights one of the current limitations of SPH
in that its application to huge domains remains inappropriate and beyond
practical use. In all cases, the governing equations were advanced using
a second-order predictor-corrector time integration scheme.

3.1. Test 3 — Tsunami Generation and Runup Due to a Two-

Dimensional Landslide

In this test, a Gaussian-shaped landslide, initially at the shoreline

moves down a sloping bed generating a free-surface displacement. An
analytical solution for the linear shallow water equations is provided by
Liu et al. (2003), so that the objective is to compare numerical results
from our SPH scheme with the solution. In dimensional form, the
seafloor is described by
h( x, t ) = H ( x ) + ho ( x, t ) , (3.1)
where
H(x) = tan β , (3.2a)
  2

  xµ 2 g 2  
ho ( x, t ) = δ exp − 2 − µt , (3.2b)
  δ tan β δ  
   
where δ is the maximum slide thickness, µ is the thickness to slide length
ratio, β is the beach slope, and x is the distance from the still-water level.
For the two test cases (A and B) with different bed slopes, domain
lengths and landslide height-to-length ratios, numerical results will only
be presented for one case.
Test A involves a domain of length 10km, maximum depth 997m
where the analytical solution predicted a free-surface displacement
of 0.07m. This ratio between the domain depth and free-surface
displacement necessitates a prohibitively large number of equi-sized
particles both in the vertical and horizontal directions. To the authors’
knowledge, no SPH scheme for free-surfaces that includes variable
resolution has been developed which would make the simulation of such
a problem quicker and more efficient. Therefore, the JHU-SPH code
was not run for this case. Test case B, however, involves a problem
86 B. D. Rogers and R. A. Dalrymple

domain length of 130m and depth 13m, respectively. According to the

analytical solution, the movement of this landslide produces a free-
surface displacement of ~1.6m for which the application of SPH is
straightforward.
Figure 1 shows a snapshot of the simulation with 34466 particles
(∆x = 0.196m), while in Fig. 2, four plots show the comparison at
different times (non-dimensionalized by µ −1 δ g ) of the free-surface
given by SPH and the linear analytical solution. The free-surface for the
SPH scheme was estimated by dropping a density-depth gage from above
the free-surface every metre. The density was evaluated as
ρ (x ) = ∑ ρ jW (x − x j )V j . (3.3)
j

The free-surface was defined as the location at which ρ(x) exceeds

1
ρ = 500 kg/m 3 .
2 o

−2

−4
y (m)

−6

−8

−10

−12
0 20 40 60 80 100 120
x (m)
Fig. 1. Test 3 — Tsunami generation by a two-dimensional landslide.
 SPH Modeling of Tsunami Waves 87

tND = 0.5
2
SPH
1.5 Analytical
free-surface (m)

0.5

0
0 20 40 60 80 100 120
-0.5

-1
x (m)

tND = 1.0
2
SPH
1.5 Analytical
free-surface (m)

0.5

0
0 20 40 60 80 100 120
-0.5

-1
x (m)

Fig. 2. (a and b) Test 3 — Tsunami generation by a two-dimensional landslide,

comparison with analytical solution of linear shallow water equations.
88 B. D. Rogers and R. A. Dalrymple

tND = 2.5
2
SPH
1.5 Analytical
free-surface (m)

0.5

0
0 20 40 60 80 100 120
-0.5

-1
x (m)

tND = 4.5
2
SPH
1.5 Analytical
free-surface (m)

0.5

0
0 20 40 60 80 100 120
-0.5

-1
x (m)

Fig. 2. (c and d) Test 3 — Tsunami generation by a two-dimensional landslide,

comparison with analytical solution of linear shallow water equations.
 SPH Modeling of Tsunami Waves 89

In plots (a and b) we can see that as the solution initially develops,

the discrepancy is greatest near the shoreline where the Lagrangian
nature of the solution gives a different answer from that predicted by the
analytical solution which uses a fixed shoreline position with the
movement of the bed. As the solution develops in plots (c and d), we can
see that SPH differs considerably from the linear analytical solution in
that the left-to-right traveling crest is lower and further advanced for
SPH. Similarly, the trough in SPH is nearer the shoreline than predicted
analytically. These results are in qualitative agreement with results from
nonlinear depth-averaged approaches (e.g. Liu et al. 2003). Thus, it is
very encouraging that SPH captures inherently the nonlinearity of the
flow in this test case even when using a quadratic kernel.

3.2. Test 4 — Tsunami Generation and Runup Due to a

3-Dimensional Landslide
This problem requires the modeling of a sliding mass down a 1:2 plane
beach slope and compares the predictions from the JHU-SPH scheme
with laboratory data. Large-scale experiments were conducted in a wave
tank with a length 104m, width 3.7m, depth 4.6m and with a plane slope
(1:2) located at one end of the tank (Raichlen and Synolakis 2003).
A solid wedge was used to model the landslide. The triangular face has
the following dimensions: a horizontal length of b = 91cm, a vertical face
a = 46cm high and a width of w = 61cm. The wedge was instrumented
with an accelerometer to define accurately the acceleration-time history
and a position indicator to independently determine the velocity- and
position-time histories.
In our simulations, we moved the wedge according to the motion of
the wedge as measured experimentally. We conducted both two and
three-dimensional simulations of the experiment.
3.2.1. Two-Dimensional Simulation

(i) Wedge initially partially submerged – Run 30

Figure 3 shows a snapshot of the particle positions. One can see that
following the passage of the wedge down the slope, a large volume of
90 B. D. Rogers and R. A. Dalrymple

water has been drawn down in generating the offshore propagating wave.
This is confirmed in Fig. 4 which displays a comparison of the free-
surface data at a wave gage located 1.83m offshore from the still-water
line. The 2-D SPH-LES scheme vastly overpredicts both the initial wave
and the following drawdown of the free-surface. This simulation was
performed using 8600 particles where ∆x = 0.09m taking 3 hours of cpu
time. A convergence study showed no change in the predicted numerical
behavior allowing us to conclude that the cause of such a large
discrepancy is that important three-dimensional effects of the drawdown
and violent splash collisions are neglected.

1
frame= 65 t= 2.6s
0
−1
y (m)

−2
−3
−4
−5
−1 0 1 2 3 4 5 6 7 8 9 x (m)
Fig. 3. Test 4 — Tsunami generation by a three-dimensional landslide, run 30 (initially
partially submerged): 2-D SPH-SPS simulation snapshot.

0.2

0.1

0
free-surface (m)

0 0.5 1 1.5 2 2.5 3

-0.1

-0.2

-0.3 experimental data

SPH 2-D
-0.4

-0.5
tim e (s)

Fig. 4. Test 4 — Tsunami generation by a three-dimensional landslide, run 30:

Comparison of 2-D SPH-SPS with wave gage 1 data.
 SPH Modeling of Tsunami Waves 91

(ii) Wedge initially completely submerged – Run 32

When the wedge is placed just under the water surface so that it is only
just completely submerged, the simulation exhibits the same behavior as
for run 30 above. Figure 5 shows the comparison of the water surface at
the same wave gage. The same large initial wave and drawdown are
visible. A similar convergence study again confirmed the lack of three
dimensionality.

0.2

0.1

0
free-surface (m)

0 0.5 1 1.5 2 2.5 3

-0.1

-0.2

-0.3

-0.4
experimental data
2-D SPH
-0.5
time (s)

Fig. 5. Test 4 — Tsunami generation by a three-dimensional landslide, run 32: (initially

partially submerged): Comparison of 2-D SPH-SPS with wave gage 1 data.

3.2.2. Three-Dimensional Simulation

The two-dimensional simulations demonstrate the necessities and

requirements of a three-dimensional simulation. The extra spatial
dimension brings a further demand on the computational cost of using
SPH for flows to which as a numerical technique it is otherwise ideally
suited.
Our experience of using the double-particle boundary described in
Sec. 2.3 was that this boundary condition caused too much wall friction.
This had the undesirable effect that in the wake behind the sliding block,
the particles would not occupy the space left behind by the block.
Thus, a 3-D version of the adjusted Monaghan boundary condition
was used. The simulation was performed on a single processor machine
92 B. D. Rogers and R. A. Dalrymple

with 512MB RAM with 62377 particles taking approximately 80 hours

of cpu. The computer resources available limited the discretization to
∆x = ∆y = ∆z = 0.1m. Figures 6–8 display three snapshots from the
simulation. Each snapshot shows four particle plots from different
views: a front view (with the block moving towards the viewer), a
side view, a 3-D view and a top view looking down on the domain. In
Fig. 6, we see the simulation just after the block has entered the water:
from the side view, there is a very clear surface wave being generated by
the initial push of the block. In Fig. 7, the block has continued to slide
down the slope with the initial wave continuing to propagate offshore;
from the top view, we can see that the water from the either side of the
wedge is now beginning to flow and occupy the space vacated by the
moving block. In Fig. 8, the water flowing to occupy the space behind
the block has now collided with itself generating a “rooster tail” or
violent upward splash event (see side view and front view) as observed
in the experiments. There is then a wave that propagates out from the
splash-up which is not shown here.
If we identify the free-surface position at the wave-gage located
1.83m offshore of the initial shoreline, we see the comparison of the 3-D
SPH results with the 2-D and experimental data in Fig. 9. There is a
clear improvement moving from 2-D to 3-D in that the size of the wave
and drawdown generated is smaller. However, there is still a very large
discrepancy between SPH and experimental data.
This is for several reasons. Primarily, the resolution of ∆x = ∆y = ∆z
= 0.1m is clearly insufficient to capture free-surface displacements and
flow features whose scale is on the order of 0.02m. Other contributing
factors include the size of the domain that has been dictated by
computational costs. Furthermore, at present, no method in SPH exists
to model offshore non-reflecting boundaries properly for free-surface
flows. Hence, here, periodic lateral boundaries were used in the
longshore direction, while a wall was placed offshore. Neither of these
boundaries was at a sufficient distance from the moving block, (i.e. if
computationally possible, we would have placed the offshore wall much
further from the moving block). Other sources of error include the non-
dynamic sub-particle scale turbulence model and boundary conditions.
 SPH Modeling of Tsunami Waves
Fig. 6. Test 4 — Tsunami generation by a three-dimensional subaerial landslide, run 30: Monaghan boundary condition, 3-D SPH

93
simulation, snapshot 1.
 94
B. D. Rogers and R. A. Dalrymple
Fig. 7. Test 4 — Tsunami generation by a three-dimensional subaerial landslide, run 30: Monaghan boundary condition, 3-D SPH
simulation, snapshot 2.
 SPH Modeling of Tsunami Waves
Fig. 8. Test 4 — Tsunami generation by a three-dimensional subaerial landslide, run 30: Monaghan boundary condition, 3-D SPH
simulation, snapshot 3.

95
96 B. D. Rogers and R. A. Dalrymple

0.2

0.1
free-surface (m)

0
0 0.5 1 1.5 2 2.5 3
-0.1

-0.2
experimental data
-0.3
SPH 2-D
-0.4 SPH 3-D
-0.5

time (s)
Fig. 9. Test 4 — Tsunami generation by a three-dimensional sub-aerial landslide, run 32:
Comparison of 3-D and 2-D SPH-SPS with wave gage 1 data.

Overall however, the results from this case clearly demonstrate one
of the essential drawbacks of SPH. The resolution needed to perform an
accurate simulation of this test case must be on the order of O (∆x, ∆y,
∆z) ≈ 0.005m. For the dimensions of the current case, this would mean ~
32×107 particles, which can only be accomplished using high-
performance parallel computing, this is unavailable to the authors at the
present time.

4. Conclusions and Future Work

Smoothed particle hydrodynamics has been presented as method that can
be applied to free-surface hydrodynamics. The method can be used
straightforwardly for two-dimensional problems where the ratio of the
free-surface displacement to the depth and length of the domain is not
prohibitively small. For the 2-D landslide generated tsunami, the SPH-
LES method captures inherently nonlinear effects.
Looking to the future, clearly, due to its high computational cost,
improving computer power and resources will make SPH an increasingly
viable method for simulating tsunami flows. In the meantime, SPH
would be best used in conjunction with other models that predict the
 SPH Modeling of Tsunami Waves 97

far-field flow which are then linked to SPH which models the
complicated 3-D fluid-structure interaction. Hence, of particular
engineering merit will be the patching together of different numerical
models with SPH whereby flow information is passed between each
model. At JHU, work is already underway investigating how SPH can
be linked to shallow water and Boussinesq-type models (FUNWAVE
from the University of Delaware) that will predict the far-flow field
utilizing SPH to focus on localised flow features. Boussinesq-type
models are now a well established technique for predicting the
propagation of dispersive and nonlinear waves in shallow water. Recent
advances (e.g. Madsen et al. 2003) have greatly extended the accuracy of
the linear dispersion characteristics and nonlinear properties of
Boussinesq-type models. In the depth-integrated approach, the modeling
of wave breaking itself has been difficult requiring approximate
submodels such as modifications of Svendsen’s (1984) decay of a
turbulent bore. SPH will clearly be able to handle the turbulent flow
field with the propagation of the breaking wave.
Following tsunami propagation into shallow water, SPH will also be
ideally suited to examining the three-dimensional flow fields of tsunami-
interaction with coastal structures and inundation of dry land. This will
allow modelers to address the important issues of tsunami-structure
interaction and modeling debris flow following the passage of the
tsunami wave.
A natural consequence of being able to model tsunami-structure
interaction is that we also envisage using SPH to model the rupture of
safety critical buildings such as oil storage tanks. In contrast to other
formulations of fluids flow and intense wave breaking, SPH facilitates
easily the modeling of multi-fluid flow enabling us to examine the
transport and dispersal of contaminants and pollutants following such
catastrophic failures.

Acknowledgment
This work was supported by the Office of Naval Research (ONR).
98 B. D. Rogers and R. A. Dalrymple

References
1. Batchelor, Sir. G. K., An Introduction to Fluid Dynamics (Cambridge University
Press, 1967) pp. 635.
2. Blin, L., A. Hadjadj and L. Vervisch, Large eddy simulation of turbulent flows in
reversing systems, Selected Proceedings of the 1st French Seminar on Turbulence
and Space Launchers, CNES-Paris, 13–14 June, 2002, eds. P. Vuillermoz, P. Comte
and M. Lesieur, 2002.
3. Briggs, M. J., C. E. Synolakis and G. S. Harkins, Tsunami runup on a conical
island, Proc. Waves — Physical and Numerical Modeling, University of British
Columbia, Vancouver, Canada, 446– 455, 1994.
4. Christensen, E. D. and R. Deigaard, Large eddy simulation of breaking waves,
Coastal Eng. 42, 53–86, 2001.
5. Dalrymple, R. A. and O. Knio, SPH modeling of water waves, Proc. Coastal
Dynamics 2001, ASCE, 779–787, Lund, Sweden, 2001.
6. Dalrymple, R. A., O. Knio, D. T. Cox, M. Gesteira, and S. Zou, Using a Lagrangian
particle method for deck overtopping, Proc. Ocean Wave Measurement and
Analysis, ASCE, 1082–1091, 2001.
7. Gingold, R. A. and J. J. Monaghan, Smoothed particle hydrodynamics: theory and
application to non-spherical stars, Mon. Not. R. Astron. Soc., 181, 375–389, 1977.
8. Gómez-Gesteira, M. and R. A. Dalrymple, Using a 3D SPH method for wave
impact on a tall structure, J. Waterways, Port, Coastal, and Ocean Engineering,
ASCE, 130(2), 63–69, 2004.
9. Gotoh, H., T. Sakai and M. Hayashi, Lagrangian model of drift-timbers induced
flood by using moving particle semi-implicit method, J. Hydroscience and Hyd.
Eng. 20(1), 95–102, 2002.
10. Gotoh, H., T. Shibihara and T. Sakai, Sub-particle-scale model for the MPS method
— Lagrangian flow model for hydraulic engineering, Computational Fluid
Dynamics J., 9(4), 339–347, 2001.
11. Gotoh, H., S. Shao and T. Memita, SPH-LES model for numerical investigation of
wave interaction with partially immersed breakwater, Coastal Engineering J., 46(1),
39–63, 2004.
12. Grilli, S. T., S. Vogelmann and P. Watts, Development of a 3D Numerical Wave
Tank for modeling tsunami generation by underwater landslides, Eng. Analysis
Boundary Elemt. 26(4), 301–313, 2002.
13. Johnson, G. R. and S. R. Beissel, Normalized smoothing functions for SPH impact
computations, Int. J. Num. Meth. Eng. 39, 2725–2741, 1996.
14. Koshizuka, S., A. Nobe and Y. Oka, Numerical analysis of breaking waves using
the moving particle semi-implicit method, Int. J. Numer. Meth. Fluids, 26, 751–
769, 1998.
15. Li, S. and W. K. Liu, Meshfree and particle methods and their applications, Appl.
Mech. Rev. 55(1), 1–34, 2002.
 SPH Modeling of Tsunami Waves 99

16. Li, Y. and F. Raichlen, Energy balance model for breaking solitary wave runup, J.
Waterways, Port, Coastal and Ocean Engineering, 47–59, 2003.
17. Lin, P., K.-A. Chang and P. L.-F. Liu, Runup and Rundown of Solitary Waves on
Sloping Beaches, Journal of Waterways, Port, Coastal and Ocean Engineering,
125(5), 247–255, 1999.
18. Liu, P. L.-F., P. Lynett and K. Synolakis, Analytical solutions for forced long waves
on a sloping beach, Journal of Fluid Mechanics, 478, 101–109, 2003.
19. Lo, E. Y. M. and S. Shao, Simulation of near-shore solitary wave mechanics by an
incompressible SPH method, Applied Ocean Research, 24, 275–286, 2002.
20. Lucy, L. B., A numerical approach to the testing of fusion process, Astronomical J.,
88, 1013–1024, 1977.
21. Lynett, P., T.-R. Wu and P. L.-F. Liu, Modeling Wave Runup with Depth-
Integrated Equations, Coastal Engineering, 46(2), 89–107, 2002.
22. Madsen, P. A., H. B. Bingham and H. A. Schaffer, Boussinesq-type formulations
for fully nonlinear and extremely dispersive water waves: derivation and analysis,
Proc. Roy. Soc. London, Series A, 459(2033), 1075–1104, 2003.
23. McGuire, B., Apocalypse: a Natural History of Global Disasters, Cassell, London,
1999.
24. Monaghan, J. J., Smoothed particle hydrodynamics, Annual Review of Astronomy
and Astrophysics, 30, 543–574m, 1992.
25. Monaghan, J. J., Simulating free surface flows with SPH, J. Computational Physics,
110, 399–406, 1994.
26. Meneveau, C. and J. Katz., Scale-invariance and turbulence models for large-eddy
simulation, Annual Rev. Fluid Mechanics, 32, 1–32, 2000.
27. Monaghan, J. J., SPH without a tensile instability, J. Computational Physics, 159,
290–311, 2000.
28. Monaghan, J. J. and A. Kos, Solitary waves on a Cretan beach, J. Waterway, Port,
Coastal, and Ocean Eng, ASCE, 125, 3, 1999.
29. Morris, J., P. Fox and Y. Zhu, Modeling low Reynolds number incompressible
flows using SPH, Journal of Computational Physics, 136, 214–226, 1997.
30. Panizzo, A. and R. A. Dalrymple, SPH Modelling of Underwater Landslide
Generated Waves, Proc. 29th Intl. Conference on Coastal Engineering, Lisbon,
World Scientific Publishing Co., 1147–1159, 2004.
31. Raichlen, F. and C. E. Synolakis, Run-up from three dimensional sliding mass,
Proceedings of the Long Wave Symposium 2003, (Briggs, M, Coutitas, Ch.) XXX
IAHR Congress Proceedings, 247–256, 2003.
32. Randles, P. W. and L. D. Libersky, Smoothed Particle Hydrodynamics: Some
recent improvements and applications, Comput. Methods Appl. Mech. Eng., 139,
375–408, 1996.
33. Shankar, N. J. and M. P. R. Jayaratne, Wave run-up and overtopping on smooth and
rough slopes of coastal structures, Ocean Engineering, 30(2), 153–295, 2003.
100 B. D. Rogers and R. A. Dalrymple

34. Shao S. D., and H. Gotoh, Simulating coupled motion of progressive wave and
floating curtain wall by SPH-LES model, Coastal Engineering Journal, 46(2), 171–
202, 2004.
35. Smagorinsky, J., General circulation experiments with the primitive equations. I.
The basic experiment, Mon. Weather Rev., 91, 99–164, 1963.
36. Svendsen, I., Wave heights and set-up in a surf zone, Coastal Engineering, 8, 303–
329, 1984.
37. Toro, E. F., M. Olim and K. Takayama, Unusual increase in tsunami wave
amplitude at the Okushiri island: Mach reflection of shallow water waves, Proc.
22nd Int. Symp. on Shock Waves, Imperial College, London, UK, July 18–23 1999.
38. Tulin, M. P. and M. Landrini, Breaking waves in the ocean and around Ships, Proc.
23rd ONR Symposium on Naval Hydrodynamics, Val de Reuil, France, 2000.
39. Yeh, H. and M. Petroff, Bore in a box experiment, http://engr.smu.waves.edu/solid.html,
2003.
40. Yoshizawa, A., Statistical theory for compressible turbulent shear flows, with the
application to subgrid modeling, Phys. Fluids A, 29, 2152–2164, 1986.
 CHAPTER 4

A LARGE EDDY SIMULATION MODEL FOR TSUNAMI AND

RUNUP GENERATED BY LANDSLIDES

Tso-Ren Wu* and Philip L.-F. Liu

School of Civil and Environmental Engineering, Cornell University
Ithaca, NY, 14850, U.S.A.
E-mail: *[email protected]

A numerical model is developed to simulate landslide-generated

tsunamis. This numerical model is based on the Large Eddy Simulation
(LES) approach, solving three-dimensional filtered Navier-Stokes
equations. While the Smagorinsky subgrid scale (SGS) model is
employed for turbulence closure, the Volume of Fluid (VOF) method is
used to track the free surface and shoreline movements. A numerical
algorithm describing the movements of a solid landslide is also
developed. The numerical results are compared with laboratory
experimental data (The benchmark problem 4). A very good agreement
is shown for the time histories of runup and generated waves. Based on
the numerical results, the detailed 3D complex flow patterns, free
surface and shoreline deformations are further analyzed.

1. Introduction
Tsunamis could be generated by many different geophysical phenomena,
such as earthquake, landslide, and volcano eruption. Aerial and
submarine landslides have been documented as a source of several
destructive tsunamis. For example, in 1958 a rockslide, triggered by an
8.3 magnitude earthquake, occurred in Lituya Bay, Alaska. An estimated
volume of 30.6 million m3 of amphibole and biotite slide down into the
Gilbert Inlet at the head of Lituya Bay, generating large tsunamis. The
runup height on the opposite side of the slide in the Gilbert Inlet was
estimated as 524 m. On November 29, 1975, a landslide was triggered by

101
102 T.-R. Wu and P. L.-F. Liu

a 7.2-magnitude earthquake along the southeast coast of Hawaii. A 60

km stretch of Kilauea’s coast subsided 3.5 m and moved seaward 8 m.
This landslide generated a local tsunami with a maximum run-up height
of 16 m at Keauhou (Cox and Morgan, 1977). More recently, the Papua
New Guinea (PNG) tsunami in 1998 is thought to have been generated
by a submarine landslide, triggered by an earthquake of magnitude 7
(Bardet et al., 2003; Kawata et al., 1999; Synolakis et al., 2002). The
PNG tsunami killed more than 4000 people and was considered as one of
most devastating tsunamis in recent history.
Modeling the landslide generated tsunamis faces many challenges,
such as the complex three-dimensional interfacial flow structures
between water and slide materials, the moving free surface and shoreline
boundaries, and the strong local turbulence in the vicinity of the slide and
near the free surface. So far most of the numerical simulation models for
landslide-generated waves have been built upon the depth-integrated
continuity and momentum equations (Jiang and LeBlond, 1994; Lynett
et al., 2002). However, the basic assumptions employed in the depth-
integrated models, such as the long wave approximation with weak
vertical acceleration, limit their applications to small and smooth slides.
An alternative to the depth-integrated model is to use the Boundary
Element Method (BEM) to solve the three-dimensional fully nonlinear
potential flow (Grilli et al., 2002). The advantage of the BEM is its
ability in solving three-dimensional flows with both moving solid
boundaries and free surface. However, the potential flow assumption
limits its applications to mostly non-breaking wave problems.
In order to study landslide-generated waves more precisely, it is
desirable to solve the three-dimensional Navier-Stokes (N-S) equations.
Theoretically, the direct numerical simulation (DNS) can be performed
to resolve both large eddy motions and the smallest turbulence
(Kolmogorov) scale motion. Therefore, DNS requires very fine spatial
and temporal resolutions and most of DNS applications can only be
applied to relatively low Reynolds number flows within a small
computational domain (Kim et al., 1987). Landslide generated breaking
waves are high Reynolds number flows with strong free surface
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 103

deformation. With the current available computing resources, the DNS

does not seem to be a feasible approach for solving this type of problems.
The alternatives to the DNS approach for computing the turbulent
flow characteristics include the Reynolds Averaged Navier-Stokes
(RANS) equations method and the Large Eddy Simulation (LES)
method. In the RANS equations method, only the ensemble-averaged
(mean) flow motion is resolved. The turbulence effects appear in the
momentum equations for the mean flow and are represented by Reynolds
stresses, which are modeled by an eddy viscosity model. The eddy
viscosity is hypothesized as a function of the turbulence kinetic energy
( k ) and the turbulence dissipation rate ( ε ), for which balance equations
are constructed semi-empirically. Therefore, this method is also referred
to as the k − ε turbulence model. Recently, Lin and Liu (1998a, 1998b)
have successfully applied the k − ε turbulence model in their studies of
wave breaking and runup in the surf zone. Lin and Liu’s model has been
extended and applied to many different coastal engineering problems,
including the wave-structure interaction [Liu et al. (1999, 2001)]. In the
LES method, the three-dimensional turbulent motions are directly
simulated and resolved down to a pre-determined scale, and the effects of
smaller-scale motions are modeled by closures. In terms of the
computational expense, LES lies between RANS and DNS. Comparing
to DNS in solving high-Reynolds-number flows, LES avoids explicitly
representing small-scale motions and therefore, the computational cost
can be greatly reduced. Comparing to RANS models, because the large-
scale unsteady motions are computed explicitly, LES can be expected to
provide more statistical information for the turbulence flows in which
large-scale unsteadiness is significant (Pope, 2001). In this paper, a LES
approach will be presented for solving the landslide-generated water
waves and runup.
The flow equations for LES are derived from the NS equations by
applying a low-pass spatial filter. Similar to the RANS approach, a
residual-stress tensor or the sub-grid-scale (SGS) Reynolds stress tensor
appears in the filtered N-S equations. Thus, a closure model is also
required to relate the residual-stress tensor to the filtered velocity field.
104 T.-R. Wu and P. L.-F. Liu

The traditional Smagorinsky model (Smagorinsky, 1963) is probably the

simplest LES SGS model and has been used in several breaking wave
studies (Watanabe and Saeki, 1999; Christensen and Deigaard, 2001; Lin
and Li, 2003). In this study, we shall adopt the traditional LES approach
with the Smagorinsky SGS model.
The landslide-generated wave problem is further complicated due to
the appearance of a free surface. The location of the free surface is a part
of solution to be determined. The boundary fitting method (Lin and Li,
2002) is one of many ways to locate the free surface, in which the free
surface is treated as a sharp interface. The boundary fitted grids are
employed and the computational domain must be re-meshed each time as
the free surface moves. This method gives highly accurate results of
tracking the free surface. However, due to the difficulty of fitting the free
surface associated with breaking waves, where air bubbles are entrained
into water body, this method is not practical for the present problem.
Other available methods for tracking free surface movements include
the Marker-and-Cell (MAC) method (Harlow and Welch, 1965) and the
Volume-of-Fluid (VOF) (Hirt and Nichols, 1981) method. Different from
the boundary fitting method, these two schemes do not define the
interface as a sharp interface and computations are performed on a
stationary grid. In the MAC method, massless particles are introduced
initially on the free surface. The free surface is determined by following
the motions of these massless particles. The MAC method is appealing
because they can accurately treat the complex phenomena such as wave
breaking. However, the computing effort of the MAC scheme could be
daunting, especially in three dimensions because in addition to solving
the equations governing the fluid flow, one has to trace the motions of a
large number of marker particles.
In the VOF method, along with the conservation equations for mass
and momentum, one only needs to solve one additional equation for the
fraction factor f assigned for each numerical cell, where f is defined
as follows: f = 1 in a water cell and f = 0 in an air cell, and 0 < f < 1
indicates a free surface cell. Beside this additional transport equation, a
free surface reconstruction algorithm is used to determine the orientation
of the free surface. The VOF method has been successfully implemented
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 105

for the simulation of various breaking wave related problems [e.g., Lin
and Liu (1998a, 1998b)]. In their studies, the interface between air and
water was reconstructed by either a horizontal or a vertical line. This
piecewise constant interface reconstruction scheme is relatively crude.
Rider and Kothe (1998) provided a second-order reconstruction scheme,
in which the interface is approximated as piecewise linear. In this paper,
we will use the VOF method and adopt the piecewise linear interface
reconstruction scheme in a LES model.
In addition to the treatment of free surface, modeling the landslide-
generated waves faces another difficulty: the treatment of moving solid
boundaries (i.e., landslides). One possible method is to use curvilinear
boundary-fitted coordinates. However, for a problem with moving or
deforming boundaries the boundary-fitted grids must be either
regenerated or deformed as the boundary geometry changes, adding
considerable complexity to the computations. Another approach is the
immersed boundary method (IBM). This method introduces a body-force
field so that a desired velocity distribution can be assigned over a solid
boundary (e.g., Mohd-Yusof, 1997; Goldstein et al., 1993). This method
allows the imposition of the boundary conditions on a given surface
which does not coincide with the computational grids. The IBM has been
successfully applied to the internal combustion (IC) piston simulations
(Fadlun et al., 2000; Verzicco et al., 2000). The IBM has the advantage
of providing highly accurate results for moving boundary problems.
However, since this method depends on the interpolated velocity on the
boundary to evaluate the immersed body-force, it requires fine resolution
on the solid boundary.
Heinrich (1992) developed a partial cell method, in which a source
function is added to the continuity equation to represent the solid
boundary movement. He applied this method to simulate waves
generated by a two-dimensional landslide and obtained very good
numerical results for water surface profiles as compared to the laboratory
data. This moving solid algorithm does not require any interpolation
scheme and is independent from the grid system. Thus, it is extended to
three-dimensional problems in the current study.
The present numerical examples concern the solid landslide motion
as well as the landslide generated breaking wave runup and rundown on
106 T.-R. Wu and P. L.-F. Liu

a slopping beach. A subaerial and a submerged landslide cases will be

addressed. The submerged landslide case is the benchmark problem #4
discussed in the Third International Workshop on Long-Wave Runup
Models held at Wrigley Marine Science Center, Catalina Island,
California (2004).
The numerical results will be compared with the experimental data
(Raichlen and Synolakis, 2003) for time histories of free surface
fluctuations. The numerical results will be used to illustrate the complex
3D flow patterns in terms of the velocity field, shoreline movement, and
free surface profiles.

2. Governing Equations and Boundary Conditions

2.1. The Navier-Stokes Equations

The fluid motions of an incompressible Newtonian fluid can be described

by the N-S equations:
∇ ⋅ u = 0, (2.1)

∂u 1 1
+ ∇ ⋅ (uu ) = − ∇p + g + ∇ ⋅τɶ, (2.2)
∂t ρw ρw
where u represents the velocity vector, ρw water density, τɶ the stress
tensor, g the gravity force vector, t time, and p pressure. The viscous
stress tensor is a function of the molecular viscosity µw and the strain
rate ∇u :

τɶ = µ w (∇u + ∇T u ). (2.3)

The above equations can be also written in the tensor format:

∂ui
= 0, (2.4)
∂xi
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 107

∂u i ∂uiu j 1 ∂p 1 ∂τij
+ =− + gi + , (2.5)
∂t ∂x j ρw ∂xi ρw ∂x j

where i, j = 1, 2, 3 for three-dimensional flows, and τ ij is the

molecular viscous stress tensor:

 ∂u ∂u 
τij = µw  i + .
j
(2.6)
 ∂x j ∂xi 

In order to solve the NS equations, the initial and boundary

conditions have to be applied. Along a solid boundary, Γ1 , (Fig. 2.1) the
fluid velocity must be the same as the boundary velocity, usi . Thus the
boundary condition is written as:

ui = usi . (2.7)

Γ1 Γ2
0
Γ1
u si

Ω
usi0

Fig. 2.1. A sketch of the flow domain and boundaries. The gray parts indicate the solid
material.

Another type of boundary is the free surface, Γ2 , which is the

interface between air and water. Along the free surface, both dynamic
and kinematic boundary conditions are needed. Without considering the
surface tension, the normal and tangential stress components must be
continuous, which can be expressed as:
108 T.-R. Wu and P. L.-F. Liu

∂u n 
− p + 2µw = S n 
∂n 
 ∂uT , on Γ 2 (2.8)
 ∂un  
µ w  k
+ 
 = STk 
 ∂n ∂Tk  

where the subscripts n and Tk denote the outward normal direction and
two tangential directions ( k =1, 2), respectively, on the free surface. Sn
and STk are the normal and tangential stress components specified by the
air flow on the free surface. The kinematic boundary condition requires
that the free surface be a sharp interface separating the two fluids so that
no flow is allowed to go through it. The mathematical expression of the
kinematic boundary can be derived from the equation that describes the
free surface. If the free surface is expressed as J (x, t ) = 0 , the total
derivative of J ( x, t ) must be zero on the surface:

DJ ( x, t )
=0 (2.9)
Dt on J (x ,t )=0 or on Γ 2

or
∂J
+ u ⋅∇J = 0 on J (x, t ) = 0, or on Γ2 . (2.10)
∂t
For the initial conditions, the whole flow field has to be prescribed:

ui ( x, 0) = ui0 (x) , (2.11)

and the initial value of J can be represented as:

J (x, 0) = J 0 (x). (2.12)

Therefore, the initial pressure, p 0 , can be calculated from Eq. (2.5)

based on the initial velocity field, ui0 (x) . Finally, the initial velocity and
pressure fields must satisfy the following conditions (see Fig. 2.1 for a
sketch of the fluid domain and boundaries):
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 109

∂ui0
= 0, in Ω
∂xi
∂ui0 ∂ui u j 1 ∂τij
0 0 0
1 ∂p 0
+ =− + gi + , in Ω
∂t ∂x j ρw ∂xi ρw ∂x j
ui0 = usi0 , on Γ1
∂un0 (2.13)
− p 0 + 2µ w = S n0 , on Γ2
∂n
 ∂uT0 ∂u 0 
µw  k + n  = ST0k , on Γ2
 ∂n ∂Tk 
∂J 0 ∂J 0
+ ui0 = 0, on Γ2 .
∂t ∂xi

2.2. Large Eddy Simulation

The flow motions generated by a landslide movement can be described

by the NS equations with appropriate initial and boundary conditions as
shown in the previous section. However, the size of the physical domain
for this type of problem is usually very large and makes the DNS nearly
impossible. In this section we briefly summarize the LES model
(Deardorff, 1970), which solves the large-scale eddy motions and models
the small-scale turbulent fluctuations. The Smagorinsky model will be
adopted to close the residual stress in the filtered NS equation.

2.2.1. Filtering

In the LES, a low-pass filter is performed so that the filtered velocity

field can be resolved on a relatively coarse grid. A filtered variable
(denoted by an overbar) is defined by (Leonard, 1974):

φ ( x ) = ∫ φ ( x ') G ( x, x ') dx ', (2.14)

110 T.-R. Wu and P. L.-F. Liu

where G is the filter function, which can be a Gaussian, a box (a local

average), or a cutoff (used in the spectrum method) filter. Each filter has
a specified filter width ∆ associated with it. Roughly speaking eddies of
size smaller than ∆ are small eddies and need to be modeled. The
specified filter function G satisfies the normalization condition:

∫ G ( x, x ') dx ' = 1. (2.15)

In this paper, since the finite-volume method (FVM) is used, the

FVM provides a natural filtering operation:
1
φ ( x) = ∫ φ ( x ') dx ', x ' ∈ V , (2.16)
V V

where V is the volume of a computational cell. Therefore, the filter

function G here is specified as:

1/ V for x ' ∈ V

G ( x, x ') =  . (2.17)
0 otherwise

2.2.2. Filtered Navier-Stokes Equations

The filtered N-S equations can be derived by applying the filter function
to the N-S equations. The filtered continuity and momentum equations
are:

 ∂u  ∂u
 i  = i = 0, (2.18)
 ∂x  ∂x
i i

∂ (ui ) ∂ (uiu j ) 1 ∂p
+ = − + gi
∂t ∂x j ρw ∂xi
(2.19)
1 ∂   ∂ui ∂u j 
+ µw  + ,
ρw ∂x j   ∂x j ∂xi 
 
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 111

where u and p are filtered velocity and pressure fields, respectively. It

is important to reiterate here that, the filtered product, ui u j , is different
from the product of the filtered velocity, ui u j :

ui u j ≠ ui u j . (2.20)

The difference is also called the residual-stress tensor:

τijR ≡ ui u j − ui u j . (2.21)

In LES, τijR is also called the SGS Reynolds stress. The residual
kinetic energy is

1
kr ≡ τiiR , (2.22)
2
and the anisotropic residual-stress tensor is defined by

2
τijr ≡ τijR − kr δij . (2.23)
3
The isotropic residual stress is included in the modified filtered pressure

2
p ≡ p + kr . (2.24)
3
Substituting (2.23) and (2.24) into (2.19), the filtered momentum
equation can be rewritten as:

D (ui ) 1 ∂p
=− + gi
Dt ρw ∂xi
(2.25)
1 ∂   ∂ui ∂u j  1 ∂τij
r

+ µw  + − ,
ρw ∂x j   ∂x j ∂xi  ρw ∂x j
 
where the substantial derivative based on the filtered velocity is:
112 T.-R. Wu and P. L.-F. Liu

D ∂
≡ + u ⋅∇. (2.26)
Dt ∂t

2.2.3. Smagorinsky Model

In (2.25), the SGS Reynolds stress contains the local average of the small
scale field, therefore the SGS model should be based on the local
velocity field. The most commonly used SGS model is the Smagorinsky
model (Smagorinsky, 1963). It is essentially a linear eddy viscosity
model:
 ∂u ∂u 
τijr = −ν t  i +  = −2ν t Sij .
j
(2.27)
 ∂x j ∂xi 
The Smagorinsky model relates the residual stress to the filtered
strain rate. The coefficient ν t ( x, t ) is the subgrid-scale eddy viscosity of
the residual motions. Based on the dimensional analysis, the subgrid-
scale eddy viscosity is then modeled as:

ν t = ℓ 2S S
(2.28)
= (CS ∆) S,
2

where ℓ s is the Smagorinsky length scale, which is a product of the

Smagorinsky coefficient CS and the filter width ∆ ; S is the
characteristic filtered rate of strain:

S ≡ (2 Sij Sij ) .
12
(2.29)

In the finite volume discretization, ∆ can be represented by the

order of magnitude of grid size and is defined as:

∆ = (dx1 × dx2 × dx3 ) ,

1/ 3
(2.30)

where dx1 , dx2 , and dx3 are the three components of the grid lengths.
Under the isotropic turbulence condition, the Smagorinsky coefficient
CS ≈ 0.2 . However, in general, CS is not a constant; its value varies
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 113

from 0.1 to 0.2 in different flows. Following the suggestion made by

previous work for wave-structure interactions (Lin and Li 2003), the
present simulations have used a value of 0.15. We remark here that
existing literatures have also suggested that the Smagorinsky SGS model
needs to be modified in the region close to the free surface (i.e., Shen and
Yue 2001). A mixed dynamic SGS model has been developed for
complex flows (Dommermuth et al., 2002; Hendrikson et al., 2003). In
present simulations we have decided to use the Smagorinsky SGS model
for its simplicity ( Cs is the only empirical coefficient used in the entire
simulation model). The dynamic SGS model approach will be explored
in the future work.
The momentum equations for the LES with the Smagorinsky SGS
model can be expressed as follows:

∂ui ∂ui u j 1 ∂p
+ =− + gi
∂t ∂x j ρw ∂xi
(2.31)
1 ∂   ∂u ∂u j 
+  µw,eff  i +  ,
ρ w ∂x j   ∂x j ∂xi 
 
where µw,eff = µw + µt is the sum of molecular and eddy viscosities.

2.2.4. Near-Wall Treatment

In the near-wall viscous sub-layer region, the largest local turbulent eddy
sizes are limited by the viscous scales. The well resolved LES requires
grids nearly as fine as those used in the DNS. This restriction should be
applied not only in the wall-normal direction but also in the streamwise
direction. The near-wall resolution requirement clearly limits the
application of LES in simulating high Reynolds number flows.
Therefore, a modeling strategy is required to solve the practical
applications. Instead of modeling every detail in the near-wall region,
this study uses a wall function approach to reduce the number of
computational cells. Cabot and Moin (2000) derived a set of near-wall
damping functions and used them to approximate the eddy viscosity in
114 T.-R. Wu and P. L.-F. Liu

the first cell adjacent to the wall. The eddy viscosity ν t is obtained from
a mixing-length eddy viscosity model with near-wall damping:
νt
( )
+ 2
= κ yw+ 1− e− yw / A , (2.32)
ν
where yw+ = ywuτ / ν is the distance to the wall in wall units, κ = 0.41
and A = 19 (Cabot and Moin, 2000). The filtered velocity field is
assumed to satisfy the no-slip condition on the wall.

3. Numerical Implementation
The filtered continuity and momentum equations are solved by using the
finite-volume two-step projection method (Bussmann et al., 2002). The
forward time difference method is used to discretize the time derivatives.
In order to track the free-surface, the VOF method is used. The VOF
method was originally developed by Hirt and Nichols (1981) and
improved by Kothe et al. (1999) to second-order accuracy on the free
surface reconstruction. The current numerical model is modified from
Truchas 1.8.4 developed by Doug Kothe, Jim Sicilian and their Telluride
team members at Los Alamos National Laboratory. The original program
solves N-S equations by using the projection method with the finite
volume discretization. A few new algorithms have been added in order to
simulate turbulent free surface flows generated by a moving landslide.
The VOF method as well as the moving boundary algorithm will be
described in Section 3.2.
In order to simulate the free-surface motion in a fixed grid system,
which include both air and water, a volume of fluid function, f ,
representing the volume fraction of water within a computational cell is
introduced. The f value equals to one if the cell is full, zero if empty,
and 0 < f < 1 if the cell is partially filled with water (Fig. 3.1). Since f
is conserved and governed by the transport equation, it can be described
by:
∂f ∂ui f
+ =0. (3.1)
∂t ∂xi
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 115

The f value represents the spatial position of the fluid body. It also
provides the kinematic information on the fluid.

f =

Fig. 3.1. A sketch of the volume fraction f , where f = 1 indicates that the cell is filled with
water, f = 1 filled with air, and 0 < f < 1 contains an air-water interface.

The momentum equation (2.25) needs to be modified as:

∂ρui ∂ρui u j ∂p
+ =− + ρ gi
∂t ∂x j ∂xi
(3.2)
∂   ∂u ∂u 
+  µ  i + j  ,
 eff  ∂x 
∂x j   j ∂xi 

where the cell density ρ is defined as:

ρ = f ρw (3.3)

or in the vector form:

∂ρu
+ ∇⋅ (ρ u u ) = −∇p + ρg + ∇⋅  µeff (∇u + ∇T u)  ,
n
(3.4)
∂t  
where µeff is the cell viscosity.

3.1. Finite Volume Discretization Method

Consider a general conservation equation for an arbitrary quantity φ :

116 T.-R. Wu and P. L.-F. Liu

∂φ
+ ∇⋅ (uφ ) = S (φ ) , (3.5)
∂t
where S is a source term which is generally dependent upon φ . The
finite volume method starts from the volume integral form of the partial
differential equation, e.g., (3.5), we have,
 ∂φ 
∫  + ∇⋅ (uφ ) = S (φ ) dV . (3.6)
 ∂t 
By adopting the Gauss divergence theorem, (3.6) can be expressed as
∂φ
∫ ∂t  dA ⋅(uφ) = ∫ S (φ) dV ,
dV + ∫ (3.7)

where V is the volume, and A is the area vector

A = nA , (3.8)
with n being the unit normal vector and A the face area (Fig. 3.2).

n1

n2 dA1
dA2
dV
n4
dA4
dA3
n3

Fig. 3.2. A sketch of the control volume and face normal vectors. Each face has a unique
normal vector n and face area dA.

The physical domain then can be divided into a number of discrete

volume elements (“cells”) denoted by subscript i . In (3.7), the volume
integral can be viewed as the integral over each cell, and the surface
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 117

integral is over the bounding surface of the cell. Equation (3.7) can then
be approximated by a discrete numerical scheme:

φin+1 − φin 1
dt
+
Vi
∑  A ⋅ u
F
n  [φ ]n = Sin ,
 F F (3.9)

where superscript n indicates the nth time step, dt is the time step size,
Vi is the i th cell volume, and subscript F denotes the cell face.
On the left-hand side of (3.9), the surface integral has been
approximated as a sum over discrete cell faces F with an area vector
A F ; Vi is the control volume, and φι is a local volume-averaged value,

∫ φdV = 1 φdV ,
φi = ∫ (3.10)
∫ dV V i

and similarly Si denotes the volume-averaged source function.

In (3.9), a first-order forward time-differencing scheme has been
adopted. For spatially second order schemes, φ is assumed to vary
linearly in space. Hence, φi is expressed as:

φi = φ (xc ) , (3.11)

where x c is the geometric centroid of the control volume. Similarly, the

face quantities φ F are given at the geometric centroid of each face.
Because of the usage of FVM, it is natural to adopt the collocated
arrangement that defines all the fluid properties at the cell centroids,
including density, pressure, velocities, and viscosity. However, from
(3.9), the FVM requires information not only at the cell centroid but also
on the face centroid. The face centroid quantities may be a scalar (e.g.,
density), a vector (e.g., velocity vector), or a gradient (e.g., pressure
gradient or velocity gradient). Therefore, an algorithm is required to
convert cell centroid data to face centroids or to evaluate the quantity
gradients at each cell centroid. The gradient on the cell face is calculated
based on the cell-centered values which adjoin the cell face.
118 T.-R. Wu and P. L.-F. Liu

3.2. Interface Kinematics

In order to implement the free surface boundary conditions, the position
of the free surface must be determined first. The VOF algorithm,
developed by Kothe et al. (1999), is robust and computational efficient
and will be adopted in this study.
The VOF algorithm involves two major steps: the transport of the
volume fraction and the interface reconstruction.

3.2.1. The Transport of the Volume Fraction

In the VOF algorithm, besides the mass and momentum equations, an

additional equation, (3.1), for the volume fraction, f , must be solved.
Equation (3.1) is also called the VOF equation and can be discretized,
based on FVM formulation:
1
f n+1 = f n −
Vi
∑ dt  A ⋅ u
F
n  [ f ]n .
 F F
(3.12)

The second term in the right-hand side of (3.12) denotes the

volume fluxes across the cell faces. However, unless small dt has
been used, the error could be large if the volume fluxes are
estimated by the volume fraction on cell faces (Fig. 3.3). The
volume fluxes can be estimated more precisely based on the
geometric calculation:
1
f n+1 = f n −
Vi
∑ dt  A ⋅ u
F
n  f Fn ,
 F (3.13)

where f Fn denotes the volume fraction of dVFn associated with water

through a cell face:

dVwn, F
f Fn = . (3.14)
dVFn
Once the solution to (3.13) is found, the fluid volumes are marched
forward in time.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 119

uF dt uF dt

[ f 2 ]F
n
dV2,F dV2,F f 2,nF

uF uF

[ f1 ]F
n
dV1,F dV1,F f1,nF

Fig. 3.3. The advected mass through a cell face. The left one tends to have larger error
than the right one.

3.2.2. Volume Tracking Algorithm

To solve the volume fraction f n+1 at the new time-step from (3.13), we
need the information on f Fn as well as the free surface reconstruction. In
this paper, a multidimensional PLIC (piecewise linear interface
calculation) (Rider and Kothe, 1998) is utilized to construct the plane in
each free surface cell. The plane is then used to estimate f Fn .
The PLIC algorithm consists of two steps: the planar reconstruction
and the calculation of the volume of water across a cell face. As for the
first step, the free surface within a cell is assumed to be a plane. Using
the information on f n , the orientation of the free surface can be
estimated as the gradient of f n . As for the second step, the volume of
water across a cell face can be evaluated by multiplying the volume flux
by dt . This volume is then used to update the volume fraction in the cell
and to transport other quantities in the momentum equations. We shall
discuss the detailed algorithm in the following sections.

3.2.2.1. Estimation of normal vector on the free surface

A normal vector to the free surface is defined as the gradient of the

volume fraction, f, i.e.,
n = ∇f , (3.15)
120 T.-R. Wu and P. L.-F. Liu

and can be normalized as:

∇f
nˆ = . (3.16)
∇f

The cell center and cell face gradient calculations are carried out via the
least square algorithms.
By assuming that the free surface geometry is piecewise linear
(planar), the free surface in each free surface cell can be described as:

nˆ ⋅ x − C p = 0 , (3.17)

where x is the position vector of a point on the plane and C p is the plane
constant to be determined.

3.2.2.2. Reconstruction of the free surface

After determining the normal vector, n̂ , on the free surface, the free
surface locations can be calculated iteratively. For each free surface cell,
the free surface plane divides space into regions inside and outside of
water body, depending upon the direction chosen by n̂ . Based on the
definition expressed in (3.16), n̂ points to the interior of the water body.
Hence, nˆ ⋅ x − C p will be positive for any point inside the water body;
zero for any point x on the free surface plane; and negative for any point
x outside the water body. The volume Vtr is the truncation volume lying
within the water body and separated by (3.17). By varying C p , Vtr will
also be changed. Our goal is to find the constant C p iteratively (see
Fig. 3.4) so that a nonlinear function defined as:

E (C p ) = Vtr (C p ) − f ∗ V (3.18)

becomes zero or smaller than a convergence criteria. The algorithm of

geometric calculation of the truncation volume Vtr can be found in Rider
et al. (1998) and Kothe et al. (1999).
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 121

Xp

Fig. 3.4. Locating the interface. The interface can be moved up and down by changing C
(Eq. (3.17)) and followed the direction of n̂ . This is constrained by the volume
conservation.

When E (C p ) is equal to zero or smaller than a certain tolerance, the

free surface plane is declared “reconstructed” in that cell. There are many
root-finding algorithms, e.g., Bisection, Newton’s method, and Brent’s
method, are available to find the zero of this nonlinear function, but
Rider and Kothe (1998) have found that Brent’s method (Press et al.,
1986) gives the best results in practice.
After the free surface plane is reconstructed, a geometric calculation
of volume fluxes of water across cell faces can be done based on the
plane equation (3.17) of each cell.
Now our volume tracking algorithm template is summarized as
follows:
1. Estimate the interface normal n̂ from discrete f data.
2. Find the plane constant C p in (3.17) by solving (3.18) to reconstruct
the interfaces within the cell in a volume conservation manner.
3. A geometric calculation of the volume fluxes of different materials
crosses cell faces based on f Fn .
4. Update the cell center volume fraction from (3.13).
122 T.-R. Wu and P. L.-F. Liu

3.2.3. Treating Air as Void

In our simulations, because both the pressure gradient and the

momentum flux are often much smaller in the air region than those in the
water region, a void model is used to simplify the equations in the air
region. Voids are used to represent regions where the cells are fully
occupied by air. If a cell is void, the N-S equations are not solved and the
pressure in the void regions is a constant or zero.
Once the void model is applied, there is no momentum contribution
from air. Therefore, the there is no momentum exchange between the
void and non-void cells. Consequently, the face flux, e.g., the momentum
flux through the cell face, will be set to zero, if this cell face is adjacent
to a void cell. If a non-void cell is connected to a void cell, the velocity
gradient on the cell face is then calculated from the neighboring cells
with fluid in them.

3.3. Projection Method

The projection method has been widely used to solve the NS equations.
This method was first proposed by Chorin (1968) and then used by Bell
and his coworkers to solve the constant-density (Issa, 1986) or variable-
density (Rider and Kothe, 1998) incompressible N-S equations. In this
study, the projection method will be used to solve the filtered N-S
equations.
In the projection method, momentum equations, (3.2), can be
described by two fractional steps:

ρ n+1u* − ρ n u n
dt
n n
(
= −∇⋅ (ρ n u u) + ∇⋅ µeff (∇u + ∇T u) ,
n
) (3.19)

ρ n+1u n+1 − ρ n+1u*

= −∇p n+1 + ρ n+1g , (3.20)
dt
where (3.19) is an explicit expression for the interim velocity u* ,
referred to as the predictor step. In (3.19), all forces except gravity and
pressure gradient are included. On the other hand, (3.20) is called the
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 123

projection step. Combining (3.19) with (3.20) produces the time

discretization of (3.2):

ρ n+1u n+1 − ρ n u n
= −∇⋅ (ρ u u ) −∇p n+1 + ρ n+1g
n

dt (3.21)
+∇⋅ µ ( n
eff (∇u n
+ ∇ u) .
T n
)
No additional approximation results from this decomposition.
Equation (3.20) relates u n+1 to u* . By adopting the solenoidal
condition, (2.18), we have:

∇p n+1  u* 
∇⋅ = ∇⋅  + g . (3.22)
ρ n+1  dt 

The above equation is also called the Poisson Pressure Equation (PPE).
The cell-centered pressure p n+1 at the new time step can be obtained by
solving (3.22), and cell-face pressure gradient ∇pFn+1 is calculated from
the same stencil as that of the density interpolation to faces. Equation
(3.20) is then used to calculate the solenoidal face velocity field:

 ∇pFn+1 
u n+1
= u − dt  n+1 − g .
*
(3.23)

F F
 ρ F

Finally, the pressure gradient on cell faces ∇pFn+1 ρ Fn+1 is

interpolated to cell centers in order to obtain the velocity field u n+1
at new time-step, which satisfies the divergence free condition.

3.3.1. Momentum Advection

After obtaining the volume tracking results, the momentum advection

∇⋅ (ρ u u ) can be evaluated by the face-centered velocities.
n

Discretizing the advection term of (3.19), we obtain:

124 T.-R. Wu and P. L.-F. Liu

∇⋅ρ uρ u u dVdV ≈≈∑ dVdV

dtdt ∫∇⋅
n n n n n
F ρu
ρuF
F
(3.24)
= ∑ M Fn u nF ,
F

where dV f is the advected volume through each cell face, and M Fn is

the advected mass through each cell face. Figure 3.1 illustrates the
advection of water and air across the right face of a cell containing the
free surface. The advected volume is:

dVF = dtA F ⋅ u F = dtAF u F ⋅ nˆ F , (3.25)

where u F ⋅ nˆ F is the normal component of the solenoidal face velocity

u F on the face, and AF is the face area.
The mass of water leaving a cell across face F can be expressed by
multiplying dVw, F and the corresponding densities:
M F = ρw dVw, F , (3.26)

and with (3.14):

M F = ρw dVF f F . (3.27)

Equation (3.24) can then be written as:

) dV ≈ ∑
(ρuudV
∫ ∇⋅uu
n
dt dt∇⋅ ≈ M FnM
u nF u
F
(3.28)
= dt ∑ ρw ( A F ⋅ u nF ) f Fn u nF ,
F

where ( A F ⋅ u ) f is directly obtained from (3.13). Thus, (3.24) is

n
F F
n

consistent with mass advection. Note that (3.28) is an upwind scheme

and is a first-order accurate approximation.

3.3.2. Momentum Diffusion

The net viscous stress on the control volume is calculated by applying

the divergence theorem to the volume integral of the local stress. This
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 125

reduces to a sum of the dot product of the face normal vector with the
local velocity gradient multiplied by the face area:

(
∇⋅ µeff
n
(∇u + ∇T u)
n
)
(3.29)
, F AF  n F ⋅ (∇u F + ∇ u F ) .
= ∑ µeff ˆ 
n
n T

F
 

The velocity gradient is calculated by the least squares method. The

effective cell center viscosity µeff is evaluated by using volume
weighted technique.

µeff = f µw,eff . (3.30)

The cell center viscosity is then harmonically averaged to face centroids

by using an orthogonal approximation:

2
µeff , F = , (3.31)
1 µnb +1 µ p

where µnb is the neighbor cell viscosity and µ p is the cell center
viscosity.

3.3.3. Projection

After obtaining the momentum advection and diffusion, the “predicted”

velocity u* can be calculated from (3.19). The predicted velocity u* and
f n+1 are then interpolated to cell faces to obtain u*F and ρFn+1 ,
respectively. Since we assume velocity does not vary discontinuously
near the free surface, the velocity interpolation can be done by the least
square linear reconstruction method. However, because density varies
discontinuously across the free surface, we have to limit the size of the
stencil. The cell face densities are calculated by averaging the cell
densities on either side of the face.
126 T.-R. Wu and P. L.-F. Liu

The discretization of (3.22) is:

 n +1   
 A ⋅ ∇pF  =  A ⋅  u F + g  ,
*

∑  F ρ n+1  ∑  F  dt F  (3.32)

F F F

where the cell face pressure gradient ∇pFn+1 is calculated from the same
stencil as that of the density interpolation to faces:
n+1
n +1
pnb − p pn+1
∇p F = . (3.33)
dx
Here pnb n +1
is the neighbor cell pressure, p np+1 the cell pressure, and δ x
the centroid-to-centroid distance.
Equation (3.32) is first solved for the cell-centered pressure p n+1 .
Equation (3.23) is then used to calculate the solenoidal face velocity field
u nF+1 . The last step is to interpolate ∇pFn+1 / (ρ Fn+1 ) − g to cell centers,
which, along with (3.20) and u* , is used to obtain cell center velocity
field u n+1 .
Equation (3.32), the PPE, is a set of linear algebraic equations for the
pressure field that can be solved by iterative methods. In this study, the
conjugate gradient (CG) algorithm with symmetric successive over-
relation (SSOR) (Varga, 1962) pre-conditioner (Golub and Van Loan,
1989) will be used to solve the linear algebraic equations.

3.4. Computational Cycle

We now summarize the computing cycle to update the field variables
within one time step. Given values of u n , u nF , and p n at the time step n,
we advance the solutions to the time step n + 1 in the following
procedures:
1. Solve (3.13) for f n+1 using u nF and utilizing the multidimensional
PLIC volume tracking algorithm. New time step cell density ρ n+1 is
obtained via ρ n+1 = f n+1ρw and a geometric calculation of volume
fluxes of water crosses cell faces.
2. Obtain u* by (3.19) where ∇⋅ (ρ u u ) is evaluated by using the
n

volume flux information obtained directly from the previous step, and
the viscous force is obtained via (3.29).
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 127

3. u*F and ρFn+1 are obtained by interpolating u* and ρ n+1 to cell

faces.
4. Solve PPE, (3.22) and (3.32), to obtain cell center pressure p n+1 .
5. A solenoidal face velocity field u nF+1 is obtained by (3.23).
6. Finally, ( ∇ pFn+1 / ρFn+1 − g F ) is interpolated to cell center to obtain
u n+1 from u* via (3.20).

3.5. Moving Solid Algorithm

In order to simulate the movement of a solid body through fixed mesh,

the partial grid algorithm and an internal source function are added to the
program. Consider a volume V containing an obstacle (Fig. 3.5), which
is defined by its volume Vobst(t ) and its surface Aobst(t ) . If the obstacle
volume increases, i.e. dVobst dt > 0 , the volume of the fluids decreases,
and vice versa. The conservation of mass in the volume V can be
expressed as below:

dVobst(t)
∫
u ⋅ ndA =
A dt
. (3.34)

The new continuity equation for the volume V can be expressed as:

1 dVobst(t )
∇⋅ u = = φ ( x, y , z , t ) , (3.35)
V dt
where φ = internal source function. Substituting the new continuity
equation, (3.35), into the original governing equations, the momentum
equation becomes:

∂ (ρui ) ∂ (ρ ui u j ) ∂p
+ = − + ρ gi
∂t ∂x j ∂xi (3.36)
∂   ∂ui ∂u j 
+ µeff  +  + ρu iφ.
∂x j   ∂x j ∂xi 
 
128 T.-R. Wu and P. L.-F. Liu

Obstacle Volume Vobst(t )

Surface Aobst(t )
Volume V
Surface A

Fig. 3.5. A solid body in the fluid domain.

The main procedure of the moving solid algorithm can be expressed as:
1. Update the new time step solid body VOF (Fig. 3.6).
2. Add a positive source function to the region where water will be
pushed out, and a sink term to the wake zone where water will be
sucked in (Fig. 3.7).
3. Solve the new time step velocities.

t1

t2

Fig. 3.6. Update the solid body VOF from old time step (t1) to new time step (t2).

The difference between present moving solid algorithm and

Heinrich’s (1992) is that the present algorithm updates the new time step
solid VOF first, and then adds additional source functions to the areas
where solid materials change. In Heinrich’s method, the source functions
are added before updating the solid VOF. The present algorithm yields
more accurate moving boundary effects (Fig. 3.8).
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 129

One of the advantages of using the moving solid algorithm is that the
grid system does not need to fit the solid boundary. Therefore, a part of
the internal cells will be occupied by the solid material. In order to deal
with a cell occupied partially by solid materials and water, a simple
partial cell treatment will be applied.

t1

add a negative source

t2 (sink) term in the
pressure Poisson solver

t3 (=t2 + dt)

add a positive source

(source) term in the
pressure Poisson solver

Fig. 3.7. Add a positive source term in front of the solid body, and a negative source term
(sink term) in the wake zone. Where dt = t2 – t1, t3 = t2 + dt.

t1 t1

t2
t2

t3 (=t2 + dt)

Heinrich's Method Present Algorithm

Fig. 3.8. The comparison between Heinrich’s method and present moving solid
algorithm.
130 T.-R. Wu and P. L.-F. Liu

3.5.1. Partial Cell Treatment

The purpose of the partial cell treatment is to adopt the moving solid
algorithm and to deal with unstructured interior obstacles. Since the
obstacle inside the computational domain occupies a part of the interior
fluid volume, the effective cell volume will be a fraction of the original
cell volume:
Veff = (1− f solid )V
(3.37)
= θV ,
where Veff is the effective cell volume, f solid is the volume fraction
occupied by solid material in each cell, V is the original cell volume,
and θ is the effective volume fraction (Fig. 3.9).

Veff = V Veff = θV θV uF = 0

uF = 0

θV θV uF = 0

uF = 0

θV uF = 0
(Solid material)
uF = 0

Fig. 3.9. A sketch of the partial cell treatment. If a cell contains partial solid material, the
effective cell volume will be a fraction of the original cell volume. The face velocity will
be set to zero if at least one side of the cell is fully filled with solid material.

If a cell contains partially water and solid material, the flow solver
has to deal with it. Cell faces are defined either to be entirely closed or
opened. Cell faces are “closed” only if at least one of the two
immediately neighboring cells is entirely occupied by solid material. If
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 131

the cell faces are “closed”, the face velocity of the cell is set to zero, and
the face pressure is no longer calculated in the pressure solution. On the
other hand, if any face between two cells, containing at least a partial cell
volume of fluid, is “open”, the code solves the velocities and pressure
gradients.
Compared to the conventional way to treat the irregular solid
material that creates the “saw-tooth” boundary, the partial cell treatment
introduced here is a better choice to represent the smooth real geometry
of the boundary.

4. Landslide Simulations and Benchmark Problems

z
η
−δ x

Fig. 4.1. A definition sketch for the physical variables employed in this study.

4.1. Channel Description

A series of landslide experiments was carried out in a wave tank with a

length of 104 m, width of 3.7 m, and depth of 4.6 m at Oregon State
University (Synolakis and Raichlen, 2003; Raichlen and Synolakis,
2003). A 1:2 (vertical to horizontal) uniform slope was installed near one
end of the tank. A solid wedge was used to represent the landslide and to
generate waves (Fig. 4.1). The triangular face has the following
dimensions: a length of b = 91.44 cm, a front face a = 45.72 cm and a
width of w = 61 cm. The wedge moved down the slope by gravity. The
132 T.-R. Wu and P. L.-F. Liu

experiments provided several pieces of information, such as the initial

wedge elevation δ , specific weight (defined as γ = ρwedge ρw ), the
trajectory of the wedge movements, time-history of runup, and time-
history of free-surface elevations. Figure 4.2 shows the positions of
runup gauges and wave gauges. One of the submerged landslide cases
has been used as the benchmark problem #4 discussed in the Third
International Workshop on Long-Wave Runup Models held at Wrigley
Marine Science Center, Catalina Island, California (2004).

pe
Slo
1:2

Fig. 4.2. The numerical setup and gauge positions. Unit is in meter.

4.2. Numerical Simulations

The numerical simulations are carried out in a numerical wave tank and
the landslide is modeled by the moving solid algorithm. Because this
study focuses on the runup and water waves generated in the near field,
the length of the computational domain is shorter than the laboratory
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 133

setup (6.6 m). Furthermore, since the vortex shading behind the moving
wedge is small, we assume that the flow field is symmetric with respect
to the centerline and thus, only half of the physical domain is simulated
(Fig. 4.2). The number of grid point is 60 in the streamwise (the direction
of landslide movement) direction, 50 in the span-wise direction, and 60
in the vertical direction. A non-uniform grid is used in the streamwise
and vertical directions. The finest cell is located at the onshore corner to
have a finer resolution for the runup simulations. Coarser resolution is
applied to the offshore deep water corner. The side walls as well as the
end-wall are treated as impermeable free-slip walls. Therefore, the
landslide generated waves will be reflected by the end-wall, which does
not exist in the laboratory experiments. However, since the maximum
and minimum runup will occur before the reflected waves reach the
domain of interest, the reflecting end-wall boundary condition is still a
good choice to simplify the numerical setup. Other types of boundary
conditions such as numerical sponge layer and advective open boundary
condition (Liu et al., 2005) can also be used to avoid the reflection
from the end-wall. The boundary condition on the ceiling of the
computational domain is the zero-pressure open boundary condition. The
1:2 slope is designed to extend from upper-right corner to lower-left
corner (Fig. 4.2), so the solid material will occupy 50% of the cell
volume on the slope. This design ensures that the effective cell volume
will not be smaller than 50% of the total cell volume, therefore the time
marching will not be limited by certain cells with extremely small
effective cell volumes due to the Courant number restriction. This design
also helps us to clearly identify the shoreline location without being
interfered with the irregular effective cell volume. The time step is
determined dynamically based on the Courant number criteria, which is
0.45 in this study.
Similar to the laboratory setup, three runup gauges are installed close
to the moving landslide; four wave gauges numbered from #4 to #7 are
installed in the offshore direction; and one wave gauge array with six
wave gauges numbered from #8 to #11 is installed on the lateral side of
the moving wedge (Fig. 4.2). However, the gauge locations might not
coincide with grid centroid. Thus, the nearest neighbor interpolation
134 T.-R. Wu and P. L.-F. Liu

method is used to interpolate gauge data. Finally, the displacement of

the moving landslide is obtained from the laboratory measurements.
Figure 4.3 shows the time history of the displacement (S) and speed of
the slide motion for the subaerial landslide case.

4
S (m)

0
0 0.5 1 1.5 2 2.5 3
time (sec)
3
Moving speed (m/sec)

0
0 0.5 1 1.5 2 2.5 3
time (sec)

Fig. 4.3. The displacement (S) and the speed of the moving slide. In this subaerial case,
the initial wedge position ( δ ) is 0.454 m above the still water surface. The specific
weight ( γ ) of the wedge is 2.64.

In this subaerial landslide case, the initial elevation of the wedge is

0.454 m above the initial still water level ( δ = 0.454m ). The specific
weight is γ = ρsoild ρwater = 2.64. The numerical simulations are
conducted in a domain with size ( x, y, z )Domain size = (6.6,1.85,3.3) ,
where the unit is meter and ( x, y, z ) represent the directions of off-shore
( x) , lateral ( y ) , and vertical ( z ) respectively. Non-uniform grids are
used in x and z directions with the finest grid size
(dx, dz ) finest = (0.0391, 0.0196) , located at the corner near the initial
shoreline. A uniform grid was adopted in the lateral direction. The grid
has been specially designed so that dx = 2dz . It took about 10 hours
(wall clock) to finish the simulation from time = 0.0 sec to 4.0 sec on a
personal computer with an Intel 2.8 GHz CPU.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 135

4.3. Results and Discussions

4.3.1. The Runup Gauges

In the laboratory setup, two runup gauges were installed on the slope to
provide the time history of the runup information. The first runup gauge
(Gauge #2) is located at the edge of the sliding wedge, the other one is
located at one wedge width away from the centerline. Figure 4.4 shows
the comparison between the numerical results and laboratory data. The
solid lines are the numerical results and the broken lines are the
experimental measurements. The numerical runup heights are determined
by the contour line where the water occupies 50% of the effective cell
volume. A very good agreement is shown at Gauge #2. However, at
Gauge #3 the numerical solutions slightly over-predict the maximum
runup heights and the disagreement is about 10%. Overall, the numerical
simulation is able to capture the maximum runup heights in the near field
region, and most importantly the model is able to accurately predict the
arrival time of the maximum runup height.

0.15

0.1 Gauge #2

0.05
R (m)

−0.05

−0.1

0 0.5 1 1.5 2 2.5 3 3.5 4

Time (Sec)

0.15

0.1 Gauge #3

0.05
R (m)

−0.05

−0.1

0 0.5 1 1.5 2 2.5 3 3.5 4

Time (Sec)

Fig. 4.4. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time history runup height at Gauge #2 and Gauge #3. δ = 0.454 m.
γ = 2.64.
136 T.-R. Wu and P. L.-F. Liu

4.3.2. Four Fixed Wave Gauges

There are four fixed wave gauges installed in front of the landslide and
they are labeled as gauge #4 ~ #7 (see Fig. 4.2). Figure 4.5 shows the
comparison of numerical results and wave gauge data. Again, the solid
lines are the numerical solutions and the broken lines are the
measurements. Gauge #4 and #6 are the ones closer to the shoreline. The
comparison shows that the numerical simulation successfully predicts
the leading wave height as well as the phase speed. Gauge #5 and #7
show that the numerical solutions under-predict the leading wave
heights. However, the numerical model is able to predict the phase of the
propagating waves.

0.15 0.15
Gauge #4 Gauge #5
0.1 0.1

0.05 0.05
η (m)

η (m)

0 0

−0.05 −0.05

−0.1 −0.1

0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec)

0.15 0.15
Gauge #6 Gauge #7
0.1 0.1

0.05 0.05
η (m)

η (m)

0 0

−0.05 −0.05

−0.1 −0.1

0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec)

Fig. 4.5. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time histories of free surface fluctuations at wave gauge #4 ~ #7;
δ = 0.454 m, γ = 2.64. The coordinates for gauges are: Gauge #4: (x, y) = (1.83, 0);
Gauge #5: (x, y) = (2.74, 0); Gauge #6: (x, y) = (1.83, 0.61); Gauge #7: (x, y) = (2.74,
0.61). The unit is in meter.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 137

4.3.3. Wave Gauge Array

Wave gauges #8 ~ #13 are installed on the side of the moving slide to
record the lateral movement of landslide-generated waves. Figure 4.6
shows both the numerical results and laboratory data. The comparisons
show that the numerical solutions successfully predict the wave heights
of the leading and secondary waves. The numerical solutions also
accurately capture the phase of the waves. However, from gauge #8, #9,
and #11, around time = 1.75 sec, the comparisons show that the
numerical solutions cannot predict some small scale motions. This might
be cause by the lack of fine resolution in numerical simulations. It could
also be the experimental errors because there are significant wave
breaking and air bubble entrainment. Fortunately, these errors do not
seem to affect the predictions of the secondary waves.

0.15 0.15 0.15

Gauge #8 Gauge #9 Gauge #10
0.1 0.1 0.1

0.05 0.05 0.05

η (m)

η (m)

η (m)

0 0 0

−0.05 −0.05 −0.05

−0.1 −0.1 −0.1

0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec) Time (Sec)

0.15 0.15 0.15

Gauge #11 Gauge #12 Gauge #13
0.1 0.1 0.1

0.05 0.05 0.05

η (m)

η (m)

η (m)

0 0 0

−0.05 −0.05 −0.05

−0.1 −0.1 −0.1

0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec) Time (Sec)

Fig. 4.6. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time histories of free surface fluctuations at wave gauge #8 ~ #13;
δ = 0.454 m, γ = 2.64. The coordinates for gauges are: Gauge #8: (x, y) = (0.4826,
1.092); Gauge #9: (x, y) = (0.8636, 1.092); Gauge #10: (x, y) = (1.2446, 1.092); Gauge
#11: (x, y) = (0.635, 0.4826); Gauge #12: (x, y) = (0.635, 0.8636); Gauge #13: (x, y) =
(0.635, 1.2446). The unit is in meter.
138 T.-R. Wu and P. L.-F. Liu

4.3.4. Free-Surface Profiles

Figure 4.7 shows the snapshots of the free surface profiles. The initial
shoreline is at the intersection of x = 0 and z = 0 . The landslide first
pushes the water in front and generates the outgoing waves (time = 0.0 ~
0.8 sec). Then the landslide submerges into water and the free surface
caves in, generating strong lateral converging flows (time = 0.8 ~ 1.2
sec). While the landslide keeps submerging, the converging lateral flows
in the wake zone generate strong reflecting (rooster-tail) waves (time =
1.2 ~ 3.0 sec). These rooster-tail waves are the key source to the
maximum runup. After time = 3.0 sec, the solution is gradually
contaminated by the waves reflected from the side walls and will be
excluded from this discussion.

4.3.5. Velocities

Figure 4.8 presents the detailed velocity vectors at the centerline cross-
section. From time = 0.0 ~ 0.8 sec, the velocities are generated by the
entry of the landslide. After time = 0.8 sec, the elevation of the top of the
slide is lower than the still water surface and water starts to flood into the
wake area above the slide. At time = 1.2 sec, the slide is fully submerged.
A very complex three-dimensional flow pattern can be observed right
after the slide has fully submerged into water (time = 1.2 ~ 1.5 sec). The
depth-integrated equation models are not suitable for modeling this type
of flows. From time = 2.1 to 2.7 sec, the velocity distributions show that
a strong flow current has been generated by the slide motion, which has a
thickness about 1.5 ~ 2.0 times the front face height ( H slide ) of the slide.
Above this current, there exists a returning current and generates a
negative wave. Between these two currents, there exists an eddy that can
be clearly seen from time = 2.1 sec to 2.7 sec. The size of the eddy
depends on the moving speed of the slide. In this case, the eddy size is
about one front face height ( H slide ) of the slide. After t = 2.7 sec, the
slide stops. However, the current behind the slide still keeps on moving
with the decreasing strength.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 139

Fig. 4.7. Snapshots of free-surface profile for the sliding wedge with δ = 0.454 m, and
γ = 2.64. The unit is in meter.
140 T.-R. Wu and P. L.-F. Liu

Fig. 4.7. (Continued )

 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 141

Fig. 4.8. Snapshots of velocity vectors on the centerline vertical plane for the sliding
wedge with δ = 0.454 m, and γ = 2.64. The unit is in meter.
142 T.-R. Wu and P. L.-F. Liu

4.3.6. Shoreline Movement

Figure 4.9 shows the snapshots of the shoreline movement. The snapshot
at time = 1.2 sec shows that the slide has fully submerged into water
and an air bubble is trapped there. After time = 1.5 sec, the shoreline
starts to move up, spreads out, and then reaches the maximum runup
height at time = 2.7 sec. From the sequence of the snapshots we
observe that at time = 2.7 sec, both the centerline and the near field
region (−1 m ≤ y ≤ 1 m) reaches the maximum runup. Therefore, the
intersection of the shoreline and z = 0 m at time = 2.7 sec can be treated
as the maximum inundation area.

4.3.7. x-Cross-Section

Figure 4.10 provides the velocity distribution from x = −2.5 m to x =

−0.25 m, where the initial shoreline is located at ( x, z ) = (0, 0) . On each
plot, the vectors are the fluid particle velocity vectors, and the dot lines
are the water boundaries. The upper solid lines indicate the interface
between air and water, and the lower solid lines indicate the interface
between water and solid walls. Starting from x = −2.5 m, the water is
pushed away and moves into the ambient water body. The region
affected by the moving slide has a length scale about 3 times the width of
the slide ( Wslide ) laterally and 2 times the height of the slide ( H slide ).
From x = −2.0 m to x = −1.5 m, the cross-sections are interacting with the
slide. An important feature is that a strong negative wave has been
generated by the slide. At these cross-sections, water is generally moving
away from the slide. However, starting from x = −1.75 m, the converging
currents at both sides of the slide can be observed. These converging
currents become significantly and dominate the flow domain from x =
−1.25 m to x = −0.25 m. These converging currents collide into each
other and generate a strong rooster-tail wave.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 143

Fig. 4.9. Snapshots of shoreline movement for the sliding wedge with δ = 0.454 m, and
γ = 2.64. The unit is in meter.
144 T.-R. Wu and P. L.-F. Liu

Fig. 4.10. Velocity vectors on the vertical x-planes at time = 1.5 sec for the sliding wedge
with δ = 0.454 m, and γ = 2.64. The unit is in meter. The magnitude of the reference
vector indicates the speed of the wedge.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 145

4.3.8. y-Cross-Section

Figure 4.11 shows the cross-section from y = 0.0 m to y = 0.8 m, where

y = 0.0 m denotes the centerline cross-section. From y = 0.0 m to y =
0.3 m, the velocity distributions clearly show how the water has been
pushed away from the moving slide, and how the water has been dragged
by the slide in the wake zone. At y = 0.6 m and y = 0.8 m, the converging
currents can be also observed above the slope.

4.3.9. z-Cross-Section

Figure 4.12 shows the cross-section from z = −1.1 m to z = 0.0 m.

Starting from z = −1.1 m, the plot shows the region affected by the slide
movement. Clearly, the largest velocity happens in front of the slide and
within one slide-width, Wslide . From z = −0.8 m to z = −0.2 m, the details
of converging flows in the wake area are evident. At z = −0.1 m and z =
0.0 m, the complex flow patterns can be seen and the fluid motion is
mainly toward the offshore direction.

5. A Submerged Landslide Simulation

In the previous section, we have verified the accuracy of a 3D subaerial
landslide simulation. In this section, a 3D submerged landslide
simulation is going to be inspected and discussed. Because there are
many similarities between the subaerial and submerged landslide cases,
we shall focus the descriptions and discussions on the differences
between these two cases. We note that this is also the benchmark
problem #4 discussed in the workshop.

5.1. Setup

The initial elevation of the submerged landslide is 0.05 m below the

still water level ( δ = −0.1 m ). The specific weight γ is 2.64. The
numerical simulation is conducted in a domain with size
( x, y, z )Domain size = (5.1,1.85, 2.55) . In order to save computational
146 T.-R. Wu and P. L.-F. Liu

Fig. 4.11. Velocity vectors on the vertical y-planes at time = 1.5 sec for the sliding wedge
with δ = 0.454 m, and γ = 2.64. The unit is in meter. The magnitude of the reference
vector indicates the speed of the wedge.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 147

Fig. 4.12. Velocity vectors on the horizontal z-planes at time = 1.5 sec for the sliding
wedge with δ = 0.454 m, and γ = 3.24. The unit is in meter. The magnitude of the
reference vector indicates the speed of the wedge.
148 T.-R. Wu and P. L.-F. Liu

efforts, we apply the symmetric boundary condition at y = 0.0m . Non-

uniform grids are applied in x and z directions with the finest grid size
(dx, dz ) finest = (0.02050, 0.01025) located at the corner near the initial
shoreline. A uniform grid is adopted in the lateral direction. Figure 5.1
shows the displacement (S) and the speed of the moving slide measured
from the laboratory experiment.
The submerged landslide case has the same initial and boundary
conditions as those in the subaerial case.

4
S (m)

0
0 0.5 1 1.5 2 2.5 3
time (sec)
3
Moving speed (m/sec)

0
0 0.5 1 1.5 2 2.5 3
time (sec)

Fig. 5.1. The displacement (S) and the speed of the moving slide. In this subaerial case,
the initial wedge position ( δ ) is −0.1 m above the still water surface. The specific weight
( γ ) of the wedge is 2.64.

5.2. Results and Discussions

5.2.1. The Runup Gauges

Figure 5.2 shows the comparisons of the numerical solutions and

laboratory data. From the comparison of Gauge #2 and #3, the numerical
results appear to be in good agreement with the laboratory data in terms
of the runup height and phase. Again, this comparison validates the
accuracy of the numerical model.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 149

0.15

0.1 Gauge #2

0.05
R (m)

−0.05

−0.1

0 0.5 1 1.5 2 2.5 3 3.5 4

Time (Sec)

0.15

0.1 Gauge #3

0.05
R (m)

−0.05

−0.1

0 0.5 1 1.5 2 2.5 3 3.5 4

Time (Sec)

Fig. 5.2. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time history runup height at Gauge #2 and Gauge #3. δ = −0.1 m.
γ = 2.64.

5.2.2. Four Fixed Wave Gauges

Figure 5.3 shows the comparisons of wave gauge data. Gauge #4 ~

Gauge #7 are placed in the offshore region. The comparisons show a
very good agreement between the numerical solutions and the laboratory
data. The numerical model successfully predicts the maximum wave
height in Gauge #6 and Gauge #7, and captures the phases of waves in
all measurements.
From Gauge #4 to Gauge #7, the maximum wave height does not
happen at the leading but at the secondary wave. This observation is
different to that in the subaerial case. This shows that in the subaerial
landslide, a large portion of energy is transferred from slide to water by
the initial “pushing” process. However, in the submerged landslide case,
the momentum generated by the initial pushing process will be transferred
to the wake area, and has fewer effects on the free-surface waves.

5.2.3. Wave Gauge Array

Figure 5.4 shows the comparison of Gauge #8 ~ Gauge #13. These six
wave gauges mainly monitor the energy transfer in the lateral direction.
150 T.-R. Wu and P. L.-F. Liu

0.15 0.15
Gauge #4 Gauge #5
0.1 0.1

0.05 0.05
η (m)

η (m)
0 0

−0.05 −0.05

−0.1 −0.1

0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec)

0.15 0.15
Gauge #6 Gauge #7
0.1 0.1

0.05 0.05
η (m)

η (m)
0 0

−0.05 −0.05

−0.1 −0.1

0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec)

Fig. 5.3. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time histories of free surface fluctuations at wave gauge #4 ~ #7;
δ = −0.1 m, γ = 2.64. The coordinates for gauges are: Gauge #4: (x, y) = (1.83, 0);
Gauge #5: (x, y) = (2.74, 0); Gauge #6: (x, y) = (1.83, 0.61); Gauge #7: (x, y) = (2.74,
0.61). The unit is in meter.
0.15 0.15 0.15
Gauge #8 Gauge #9 Gauge #10
0.1 0.1 0.1

0.05 0.05 0.05

η (m)

η (m)

η (m)

0 0 0

−0.05 −0.05 −0.05

−0.1 −0.1 −0.1

0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec) Time (Sec)

0.15 0.15 0.15

Gauge #11 Gauge #12 Gauge #13
0.1 0.1 0.1

0.05 0.05 0.05

η (m)

η (m)

η (m)

0 0 0

−0.05 −0.05 −0.05

−0.1 −0.1 −0.1

0 1 2 3 4 0 1 2 3 4 0 1 2 3 4
Time (Sec) Time (Sec) Time (Sec)

Fig. 5.4. The comparison between numerical results (solid lines) and experimental data
(broken lines) for the time histories of free surface fluctuations at wave gauge #8 ~ #13;
δ = −0.1 m, γ = 2.64. The coordinates for gauges are: Gauge #8: (x, y) = (0.4826,
1.092); Gauge #9: (x, y) = (0.8636, 1.092); Gauge #10: (x, y) = (1.2446, 1.092); Gauge
#11: (x, y) = (0.635, 0.4826); Gauge #12: (x, y) = (0.635, 0.8636); Gauge #13: (x, y) =
(0.635, 1.2446). The unit is in meter.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 151

The comparisons show that before time = 3 sec, the numerical solution
has a good agreement with the laboratory data. After time = 3 sec, the
solution and data are contaminated by the reflected wave from the side
walls, and will be excluded from current discussion.

5.2.4. Snapshots

We present the snapshots of free-surface elevation, velocities, and

shoreline movements from Figs. 5.5–5.7. The basic physical phenomena
are very similar to those in the subaerial cases. In Fig. 5.5, we can see
that the free surface has smaller free-surface displacement than that in
the subaerial one. This is because that the submerged landslide has a
lower initial slide elevation and a smaller specific weight.
Figure 5.6 show the velocities at the centerline cross-section.
Compared to the subaerial case, the main processes to generate waves
and velocities are similar. One important observation is that the velocity
on top of the slide has a speed faster than that in the moving slide. This is
caused by the three-dimensional effect. We have found that the moving
slide generates a three-dimensional eddy, and this eddy transfers the
momentum to the centerline location and makes the high velocity jet on
top of the slide. This phenomenon can be clearly seen in Fig. 5.6.
Figure 5.7 shows the snapshots of the shorelines. Because of the
slide movement, from time = 0.0 sec to 0.9 sec, the runup height has
negative values. After time = 1.2 sec, a rebounding wave causes the
shoreline to rise up. The maximum runup at the centerline happens at
time = 1.75 sec. Compared to the subaerial case (Fig. 4.9, time = 2.7
sec), different shoreline curves can be observed. In the subaerial
simulation, the shoreline curve looks like a hump. However, in this
subaerial case, the shoreline curve is more like a concave.

5.2.5. Cross-Sections

In this section, the detailed velocity fields (time = 1.5 sec) at different
cross-sections will be presented from Figs. 5.8–5.10. Because the major
observations are close to that of the subaerial case, only the differences
and important phenomena will be addressed.
152 T.-R. Wu and P. L.-F. Liu

Fig. 5.5. Snapshots of free-surface profile for the sliding wedge with δ = −0.1 m, and
γ = 2.64. The unit is in meter.
Eddy Simulation Model for Tsunami and Runup Generated by Landslides 153

Fig. 5.5. (Continued )

154 T.-R. Wu and P. L.-F. Liu

Fig. 5.6. Snapshots of velocity vectors on the centerline vertical plane for the sliding
wedge with δ = −0.1 m, and γ = 2.64. The unit is in meter.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 155

Fig. 5.7. Snapshots of shoreline movement for the sliding wedge with δ = −0.1 m, and
γ = 2.64. The unit is in meter.
156 T.-R. Wu and P. L.-F. Liu

Fig. 5.8. Velocity vectors on the vertical x-planes at time = 1.5 sec for the sliding wedge
with δ = −0.1 m, and γ = 2.64. The unit is in meter. The magnitude of the reference
vector indicates the speed of the wedge.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 157

Fig. 5.9. Velocity vectors on the vertical y-planes at time = 1.5 sec for the sliding wedge
with δ = −0.1 m, and γ = 2.64. The unit is in meter. The magnitude of the reference
vector indicates the speed of the wedge.
158 T.-R. Wu and P. L.-F. Liu

Fig. 5.10. Velocity vectors on the horizontal z-planes at time = 1.5 sec for the sliding
wedge with δ = −0.1 m, and γ = 2.64. The unit is in meter. The magnitude of the
reference vector indicates the speed of the wedge.
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 159

From Fig. 5.8, x = −2.0 m ~ −1.25 m, the three-dimensional effect

which makes the momentum concentrate into the centerline can be
observed. The velocities on the centerline cross-section are higher than
those in the off-center cross-section (Fig. 5.9, y = 0.0 m and y =
0.2 m). The three-dimensional movement can be also observed from
Fig. 5.10, at z = −0.7 m and z = −0.6 m.

6. Concluding Remarks
In this paper we have presented a LES model for 3D free surface flows.
Although the applications shown in the paper are for landslide generated
waves and shoreline movements, the model can be used as a tool for
studying other practical coastal problems. The advantage of the
numerical simulation approach is its ability in providing a large amount
of information in the entire flow domain of interest, which is not possible
for most of laboratory observations. This is especially true for a large
scale experiment.
The success in developing a moving solid algorithm greatly extends
the model capability to simulate wave-structure interaction problems.
However, further improvement on the turbulent closure modeling for
flow separation and wave breaking is essential.

Acknowledgment
This work was supported by research grants from National Science
Foundation to Cornell University.

References
1. Bardet, J.-P., Synolakis, C. E., Davies, H. L., Imamura, F., and Okal, E. A. (2003)
Landslide Tsunamis: Recent Findings and Research Directions. Pure Appl. Geohys.
160, 1793–1809.
2. Barth, T. J. (1995) Aspects of unstructured grids and finite-volume solvers for Euler
and Navier Stokes equations, VKI/NASA/AGARD Special Course on Unstructured
Grid Methods for Advection Dominated Flows, AGARD Publication R-787.
160 T.-R. Wu and P. L.-F. Liu

3. Bussmann, M. Kothe, D. B., and Sicilian J. M. (2002) Modeling high density ratio
incompressible interfacial flows. Proceedings of FRDSM’02, 2002 ASME Fluids
Engineering Division Summer Meeting. FEDSM2002-31123, 1–7.
4. Cabot, W. and Moin, P. (2000) Approximate wall boundary conditions in the large-
eddy simulation of high Reynolds number flow. Flow Turb. Combust. 63, 269–291.
5. Chorin, A. J. (1968) Numerical solution of the Navier-Stokes equations. Math.
Comp., 22, 745–762.
6. Christensen, E. D., and Deigaard R. (2001) Large eddy simulation of breaking
waves, Coast. Eng. 42, 53–86.
7. Cox, D. C. and Morgan, J. (1977) Local tsunamis and possible local tsunamis in
Hawaii, HIG 77–14, Hawaii Institute of Geophysics, University of Hawaii, pp. 118.
8. Deardorff, J. W. (1970) A numerical study of three-dimensional turbulent channel
flow at large Reynolds numbers. J. Fluid Mech., 41(2), 452–480.
9. Dommermuth, D. G., Rottman, J. W., Innis, G. E., and Novikov, E. A. (2002)
Numerical simulation of the wake of a towed sphere in a weakly stratified fluid, J.
Fluid Mech. 473, 83–101.
10. Fadlun, E. A., Verzicco, R., Orlandi, P., and Mohd-Yusof, J. (2000) Combined
immersed-boundary finite-difference methods for three-dimensional complex flow
simulations. J. Comp. Phys. 161, 35–60.
11. Goldstein, D., Handler, R., and Sirovich, L. (1993) Modeling a no-slip flow
boundary with an external force field, J. Com. Phys. 105, 354.
12. Golub, G. H. and Van Loan, C. F. (1989) Matrix Computations. Johns Hopkins
University Press.
13. Grilli, S. T., Vogelmann, S., and Watts, P. (2002) Development of a 3D numerical
wave tank for modeling tsunami generation by underwater landslides, Eng Analysis
Boundary Elemt. 26(4), 301–313.
14. Harlow, F. H. and Welch, J. E. (1965) Numerical calculation of time-dependent
viscous incompressible flow. Phys. Fluids, 8, 2182–2189.
15. Heinrich, P. (1992) Nonlinear water waves generated by submarine and aerial
landslides. J. Waterw., Port, Coast., Ocean Eng. ASCE, 118(3), 249–266.
16. Hendrikson, K., Shen, L., Yue, D. K. P., Dommermuth, D. G., and Adams, P.
(2003) Simulation of steep breaking waves and spray sheets around a ship: The last
frontier in computational ship hydrodynamics. The Resource Newsletter, U.S. Army
Engineer Research and Development Center Information Technology Laboratory,
Fall, 2–7.
17. Hirt, C. W. and Nichols, B. D. (1981) Volume of fluid (VOF) method for the
dynamics of free boundaries. J. Comp. Phys., 39, 201–225.
18. Issa, R. I. (1986) Solution of implicitly discretized fluid flow equations by operator-
splitting. J. Comput. Phys., 62, 40–65.
19. Kawata, Y., Benson, B. C., Borrero, J. C., Borrero, J. L., Davies, H. L., DE Lange,
W. P., Imamura, F., Letz, H., Nott, J., and Synolakis, C. E. (1999) Tsunami in
 Eddy Simulation Model for Tsunami and Runup Generated by Landslides 161

Papua New Guinea was as Intense as First Thought, EOS. Trans. Am. Geophys.
Union 80, 101, 104–105.
20. Jiang, L., and LeBlond, P. H. (1994) Three-dimensional modeling of tsunami
generation due to a submarine landslide, J. Phys. Oceanog., 24, 559–572.
21. Kim, J., Moin P., and Moser R. (1987) Turbulence statistics in fully developed
channel flow at low Reynolds number. J. Fluid Mech., 177, 133–166.
22. Kothe, D. B., Williams, M. W., Lam K. L., Korzewa, D. R., Tubesing, P. K., and
Puckett E. G. (1999) A Second-order accurate, linearity-preserving volume tracking
algorithm, for free surface flows on 3-D unstructured meshes, Proceedings of the
3rd ASME/JSME Joint Fluids Engineering Conference. FEDSM99-7109, July 18–
22.
23. Leonard, A. (1974) Energy cascade in large eddy simulation of turbulent fluid flow.
Adv. Feopgys. 18A, 237–248.
24. Li, C.-W. and Lin, P. (2001) A numerical study of three-dimensional wave
interaction with a square cylinder. Ocean Eng., 28, 1545–1555.
25. Lilly, D. K. (1967) The representation of small-scale turbulence in numerical
simulation experiments. In H. H. Goldstine (Ed.), Proc. IBM Scientific Computing
Symposium on Environmental Sciences, IBM Form No. 320-1951, Yorktown
Heights, New York, pp. 195–210.
26. Lin, P. and Li, C. W. (2002) A σ -coordinate three-dimensional numerical model for
surface wave propagation. Int. J. Numer. Meth. Fluids, 38: 1048–1068.
27. Lin, P. and Li, C. W. (2003) Wave-current interaction with a vertical square
cylinder. Ocean. Eng., 30, 855–876.
28. Lin, P. and Liu, P. L.-F. (1998a) A numerical study of breaking waves in the surf
zone. J. Fluid Mech., 359, 239–264.
29. Lin, P. and Liu, P. L.-F. (1998b) Turbulence transport, vortices dynamics, and
solute mixing under plunging breaking waves in surf zone. J. Geophys. Res.,
103(C8), 15677–15694.
30. Liu, P. L.-F., Lin, P. Z., and Chang, K. A. (2001) Numerical modeling of wave
interaction with porous structures – Closure, J. Waterw., Port, Coast., and Ocean
Eng., ASCE 127(2): 124–124.
31. Liu, P. L.-F., Lin P. Z. and Chang, K. A. (1999) Numerical modeling of wave
interaction with porous structures, J. Waterw., Port, Coast., and Ocean Eng., ASCE
125(6): 322–330.
32. Liu, P. L.-F., Wu, T.-R., Raichlen, F., Synolakis, C. E., and Borrero, J. C. (2005)
Runup and rundown generated by three-dimensional sliding masses, J. Fluid
Mechanics 536: 107–144.
33. Lynett, P., Wu, T.-R. and Lim P. L.-F. (2002) Modeling Wave Runup with Depth-
Integrated Equations, Coastal Engineering, 46(2), 89–107.
34. Mohd-Yusof, J. (1997) Combined immersed boundaries/B-splines methods for
simulations of flows in complex geometries, CTR Annual Research Briefs, NASA
Ames/Stanford University.
162 T.-R. Wu and P. L.-F. Liu

35. Pope, S. B., Turbulent Flows (Cambridge University Press, 2001).

36. Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P., Numerical
recipes in Fortran (Cambridge, 1986).
37. Raichlen, F. and Synolakis, C. E. (2003) Run-up from three dimensional sliding
mass, in Proceedings of the Long Wave Symposium 2003, (eds. Briggs, M, Coutitas,
Ch.) XXX IAHR Congress Proceedings, ISBN-960-243-593-3, 247–256.
38. Rider, W. J. and Kothe, D. B. (1998) Reconstructing Volume Tracking, J. Comp.
Phys., 141, 112–152.
39. Shen, L. and Yue, D. K. P. (2001) Large-eddy simulation of free-surface
turbulences, J. Fluid Mech., 440, 75–116.
40. Smagorinsky, J. (1963) General circulation experiments with the primitive
equations: I. The basic equations, Mon. Weather Rev. 91, 99–164.
41. Synolakis, C. E., Bardet, J. P., Borrero, J. C., Davies, H., Okal, E. O., Silver, E. A.,
Sweet, S., and Tappin, D. R. (2002) Slump origin of the 1998 Papua New Guinea
Tsunami, Proc. Royal Society, London, Ser. A, 458, 763–789.
42. Synolakis, C. E., Raichlen, F. (2003) Waves and runup generated by a three-
dimensional sliding mass, in Submarine mass movements and their consequences,
Advances in Natural Hazards, 19, eds. Locat, J. and Mienert, J. (Kluwer Academic
Publishers, Dordrect, 113–120).
43. Varga, R., Matrix Iterative Analysis (Prentice-Hall, Englewood Cliffs, NJ, 1962).
44. Verzicco, R., Mothod-Yusof, J., and Orlandi, P. (2000) Large eddy simulation in
complex geometric configurations using boundary body forces, AIAA Journal 38(3),
427– 433.
45. Watanabe, Y. and Saeki, H. (1999) Three-dimensional large eddy simulation of
breaking waves, Coast. Eng., 41(3, 4), 281–301.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

CHAPTER 5

FREE-SURFACE LATTICE BOLTZMANN MODELING

IN SINGLE PHASE FLOWS

J. B. Frandsen
School of Civil Engineering
The University of Sydney, NSW 2006, Australia
Email: [email protected]

The mathematical framework of kinetic theory allows for investigations

of micro structures in the flow at the mesoscopic level and thus, in theory,
offers further insight into underlying mechanisms of nonlinear processes
at the free surface than traditional Navier-Stokes and Non-Linear Shal-
low Water (NLSW) solvers. This is of course only true if the lattice
spacing is sufficiently small. The Lattice Boltzmann (LB) method simu-
lates fluid flow by tracking particle distributions in a Lagrangian manner.
The particles are constrained to move on lattices. The Boltzmann equa-
tion relates the time evolution and spatial variation of a collection of
molecules to a collision operator that describes the interaction of the
molecules. Mathematically, the collision integral of the LB Equation
(LBE) poses difficulties when solution of the equation is sought. Investi-
gators overcome this through descriptions of models with different levels
of accuracy in the approximations of the integral. We consider a model in
which the collision assumptions are simplified to a single-time relaxation
form. This is the simplest form of the LBE. It is referred to as the Lat-
tice Bhatnagar-Gross-Krook (LBGK) scheme. In the present chapter,
we present solutions primarily based on the standard LBE which herein
approximate the NLSW equations in rotational flow. We show some
preliminary LBGK test cases including weak bore and dam break pre-
dictions. Fairly good agreement was found with Riemann solutions and
model scale experiments. With reference to the test cases of the present
workshop, the tsunami-generated runup onto a plane beach (benchmark
1) was also simulated; presented in Chapter 16 in this volume. We tested
the standard LBE scheme and a second-order Finite-Difference (FD) LB
model. The LB schemes with shoreline algorithm of Lynett et al. (2002)
worked well whereas the thin film approach would result in inaccurate
velocities. We highly recommend the second-order FD LB scheme to
form a basis for free-surface water wave developments.

163
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

164 J. B. Frandsen

1. Introduction and Overview

There are several kinetic or mesoscopic methods, e.g. the lattice gas cellu-
lar, the lattice Boltzmann equation, the gas kinetic schemes, the smoothed
particle hydrodynamics, and the dissipative particle dynamics method. The
Lattice Boltzmann (LB) method with roots in the Boltzmann equation
stems from the ideas of Ludwig Boltzmann (1844–1906) who made advances
in electromagnetism and thermodynamics. We seek to utilize his gas dy-
namics theory (Boltzmann, 1964). Lattice gas models and LB models have
been used in fluid mechanics since the 1990s. About 1000 articles have been
published over the last fourteen years. The majority of these publications
are related to physics and computer sciences (e.g. soft condensed matter,
porous media, microfluidics, electrokinetic flows, magnetohydrodynamics,
etc.). The applications are broad. Sukop and Thorne (2006) have pro-
vided an overview of a number of papers published (1992–2004), divided
into various areas of science, in the introductory chapter of their recent
book.
The purpose of this chapter is to introduce the LB Equations (LBE),
the simplified Boltzmann Equation, with application to free-surface water
waves. Applying the LB approach will allow us to solve the fluid flow at
the mesoscopic scale level, i.e. length scales ranging between the atomistic
or micro and the continuum mechanics level (assuming the grid refinement
is adequate). The latter is hereafter referred to as the macro scale level.
Somewhere between the atomistic and macro scale level (Fig. 1(a)), the
continuum model approach breaks down where part of the physical sys-
tems cannot be assumed to be continuous (e.g. breaking waves). This is
one of the main motivations behind exploring the LB approach. Herein, we
shall, however, focus on single phase LB models only, as this is our first ex-
perience with LB modeling. We have highlighted some questions which we
asked ourselves before we chose to explore the LB approach. Some of these
modeling features and issues are shown in Fig. 1(b). We have further indi-
cated our progress on what we have/have not tested. The first important
question to ask was whether or not we could predict nonlinearities at the
free-surface. We have got some promising indication that the LB method
may work, as shown in the test case studies in §6. Our investigations have
so far been limited to shallow water depths. We do not yet know how accu-
rate the method would be regarding wave breaking predictions. One feature
which motivated us to chose the LB approach is that we can model bub-
ble break-up. Regarding Fluid-Structure interaction (FSI) we have so far
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 165

Newton Lattice
Gas

Liouville Lattice
Liouville

Boltzmann Lattice
Boltzmann

Navier Stokes (a) (b)

Fig. 1. Description of Fluid Motion. (a)“Length scale” hierarchy. The microscopic
details reduce when moving toward the Navier-Stokes equation. (b) Lattice Boltzmann
Model (LBM) properties.

found that the LB approach can provide a relatively easy means of simulat-
ing flow around single and multiple bluff bodies. The standard LB method
is based on uniform lattices. This presents the usual difficulties as to grid
refinement requirement, and thus yields a host of questions on accuracy
and efficiency. This is on-going research, as described in the review part of
this chapter.
Further, we should point out that our intention with the LB modeling
approach is to use it in local areas where physical details are needed such
as the nearshore regions and in the breaking wave process itself. Therefore
a realistic goal and potential improvement to the current tsunami model
literature would be to propose an LB model coupled with a macro-scale
level model for large domain tsunami model predictions.
The key issues to bear in mind when reading through this chapter
are that the standard LBE scheme/approach has the following model
properties:
• the fluid mass is collected in discrete particles, located on corner
points of the lattice site;
• the velocity in a lattice node is discretized (in addition to time and
space);
• the particles move according to a finite, discrete set of velocities;
• the viscosity (ν) is introduced via the single time relaxation param-
eter (τ ) and is a discretized equation with uniform lattice spacing
2
∆x and time step ∆t, ν = c3 (τ − ∆t 2 ) where c = ∆x/∆t. In the
incompressible flow simulation herein, we must ensure that ν is
constant;
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

166 J. B. Frandsen

• the formulation contains a linear convection operator;

• the solutions to the Navier-Stokes equations are in the nearly in-
compressible limit;
• the solutions herein are based on the depth integrated shallow water
equations in rotational flows;
• the solutions to the LBE can be shown to satisfy the Navier-Stokes
equations (§4.4);
• the fluid pressures
p p are calculated
√ from the equation of state: p =
2
cs ρ where cs = ∂p/∂ρ (= c/ 3) is the speed of sound and ρ is
the fluid density;
• the transformation between the microscopic level to the macro-
scopic variables consists of simple relationships;
• the standard LBE is second-order accurate in space and first-order
in time (§4.6);
• the fundamental differential equations of gas dynamics are similar
to those of shallow water wave theory.

From our viewpoint, the LB model is mainly proposed due to the molecular
properties of the formulation and its ease of handling large data sets in rela-
tion to understanding underlying mechanisms of breaking ocean waves. As
outlined above, there are of course several other important reasons an LB
modeling approach appears to be a promising simulation method in fluid dy-
namics. Some of the numerical model advantages of the LB method are: (1)
no Poisson equation, and (2) a reduction from second-order to first-order
partial differential equations, and therefore a simplification of nonlinear
free-surface numerical modeling. However, as mentioned, the major motiva-
tion of the present work is driven by the mathematical framework allowing
molecular level modeling of fluid flow which differs from any conventional
free-surface models. The locality of the formulation is another advantage,
essential for CPU efficiency and thus practicality. As for most fluid dynam-
ics problems, high CPU requirement is a barrier in advancing free-surface
models which typically involve high CPU time near the moving boundary,
especially when discontinuities occur. The LB method has been demon-
strated to show good parallel computing performance (e.g. Aidun et al.
(1998); Thürey et al. (2006a); Körner et al. (2006)). Furthermore, com-
pared to traditional continuum mechanics solvers, the LB solver has also
shown promise when simulating fluid interaction with multiple bluff-bodies
in terms of low CPU time and an easy means of rearranging bluff-bodies to
study optimum solutions for flow-induced vibration problems with/without
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 167

a free surface. An example of bore propagation through a coastal city is

shown in Fig. 2.

Fig. 2. Bore propagation through a city.

The standard LB method is based on uniform grids. Although it is an

advantage that no conventional mesh generation is required, it is usually in-
adequate for most fluid dynamics problems. Compared to traditional CFD,
few initiatives have been done on unstructured LB lattices, as mentioned
later in the review.
In this chapter, we have assumed that ocean and coastal engineers are
not familiar with LB-based models and therefore, this chapter is designed to
be as self-contained as possible. Having said this, the reader is encouraged
to carry out a self-study on physical gas dynamics to understand basic
kinetic theory and the core idea of connecting microscopic and macroscopic
physics before reading about Boltzmann modeling concepts, e.g. Vincenti
and Kruger, Jr. (1965); Burgers (1969); Gombosi (1994); Laney (1998);
Struchtrup (2005). It is also recommended to acquire some knowledge
in statistical mechanics, e.g. Penrose (1970); Kalikmanov (2001). Among
other useful books in the field of LB Equations and modeling are those
of Cercignani (1988); Succi (2001); Karniadakis and Beskok (2002); Harris
(2004); Zhou (2004); Sukop and Thorne (2006). We should also point out
that the LBE was developed in the wake of Lattice Gas Cellular Automata
(LGCA). The basic ideas of LGCA have been extensively described by, for
example, Rothman and Zaleski (1997); Rivet and Boon (2001).
The first part of the present chapter starts with a short review of LB
models. The description of fluid motion is divided into an introduction to
kinetic theory followed by the Boltzmann equation. Then a presentation
of the approximate form of the collision integral is given; leading into the
numerical treatment. The second part of the chapter focuses on the LB
development with application to free-surface water waves. The solver per-
formance is first tested for weak bores in tanks, then a dam-break problem
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

168 J. B. Frandsen

and finally the tsunami runup simulation is undertaken (presented in this

chapter). We should point out that the case studies presented represent
preliminary work. Further work is required to address moving shore line
dynamics, etc., in a well documented manner.

2. Review of Lattice Boltzmann Models

This brief review is meant to provide the reader with insight into the work
and progress on development and applications of LBE, in particular, with
flow around bluff-bodies and free-surface flows in mind. We emphasize that
this approach has not been applied and validated in the area of free-surface
ocean wave predictions. Therefore, a review in a more general format is
presented. We outline the development of various branches of LB modeling.
In doing so we highlight ideas of expansion and development of LB-based
approaches relevant to the ocean and coastal engineering communities.
The LB formulation offers an alternative treatment for description of
fluid particle motion. It is based on statistical mechanics concepts and
recovers the Navier-Stokes equations in the nearly incompressible limit, as
mentioned by Chen and Doolen (1998). The approach has been used exten-
sively in molecular dynamics and physics but little attention has been given
to the modeling of free-surface water waves. The LB method was devel-
oped as an improvement of the method of Lattice Gas Automata (LGA),
as described by Rothman and Zaleski (1997). It was first introduced by
McNamara and Zanetti (1988). Comprehensive reviews of the LB method
have been given (Benzi et al. (1992); Chen and Doolen (1998); Nourgaliev
et al. (2003)). The LB method belongs to a class of the pseudocom-
pressible solvers of the Navier-Stokes equations and can be classified as a
Lagrangian, local equilibrium, finite-hyperbolicity approximation. The nu-
merical formulation typically (but not necessarily) separates the equations
into the dissipative and non-dissipative parts. The most popular form of LB
equations is the Lattice-BGK (LBGK) incorporating a single-time approxi-
mation of the Boltzmann equation, as described by Bhatnagar et al. (1954),
hereafter the LBGK model. In this model the mean free path between par-
ticles is assumed to be the same. It should also be noted that the governing
LB formulation is second-order accurate space and first-order accurate in
time, as shown by Junk et al. (2005). The disadvantages inherent in the
LBGK model can be reduced by using a Multi-Relaxation-Time (MRT)
approach. It separates the relaxation times for different kinetic modes and
improves numerical stability and accuracy. For more detailed discussions
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 169

on MRT-LBE method in general, we refer the reader to D’Humières et al.

(2002). It is notable that the trend in progress of LB modeling is reported
in relation to the MRT approach. Stability of the LBE method is further
addressed through the generalized Lattice Boltzmann equation which is
based on moment space rather than discrete velocity space, e.g. Lallemand
and Luo (2000); Luo (2000).

2.1. Grid Refinement and Curved Solid Boundaries

The majority of LBE model predictions are based on uniform regular spaced
lattices. Compared to traditional CFD, few initiatives have been done on
unstructured LB lattices. Although an increase of contributions is evident,
a literature review still reveals only relatively few publications on this topic,
e.g. He et al. (1996); Tölke et al. (1998); Peng et al. (1999); Kandhai et al.
(2000); Van der Sman and Ernst (2000). Mesh refinement techniques are
also discussed by Crouse et al. (2002); Ubertini et al. (2003); Crouse et al.
(2003). Recent progress on grid refinement method using a nested adap-
tive grid approach are described by Tölke et al. (2006), in which numerical
examples with rising bubble simulations are presented. Geller et al. (2006)
simulates flow around multiple cylinders based on unstructured grids at
Re=200. Good agreement was found with traditional CFD solvers. Several
investigators have made contributions regarding the treatment of curved
boundaries on uniform grids in 2-D, e.g. Filippova and Hänel (1997); Mei
et al. (1999). Expansion into 3-D flows have also been developed, as de-
scribed by Mei et al. (2000). We should also mention that “stair-step”
effects on the boundary can be overcome by using mapping techniques,
like in traditional finite difference models. For example, He and Doolen
(1997) developed a curvilinear formulation to simulate flow past a circular
cylinder.
Today, the research group directed by Professor Krafczyk at the Insti-
tute for Computer Applications in Civil Engineering, Technical University
of Braunschweig, Germany, is at the forefront on LB-grid method devel-
opment. We should also mention that they are in the forefront in many
other LB developments as well, including contributions to civil/mechanical
engineering applications. The reader is invited to visit their web site which
has many publications available for downloading including a nice article
search engine (www.cab.bau.tu-bs.de).
With respect to LB-model development with application to steep and
breaking free-surface water waves, it would be necessary to have models
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

170 J. B. Frandsen

with non-uniform grids, just like it is when using traditional CFD finite
volume and finite element solvers. In contrast to the macro mechanics
models, it should be noted that non-uniform grid implementation in the
LB context is non-trivial.

2.2. Turbulence Modeling

Capturing turbulence accurately and identification of appropriate subgrid

scale laws are topics of research in the LB community. From our view
point, one central idea behind proposing the mesoscopic LB formulation is
to capture smaller scales naturally, and therefore postpone/avoid the need
to apply empirical turbulence models. Of course this highly depends on
grid resolution and on the application in mind. However, it may be needed
in civil engineering applications with high Reynolds numbers (Re > 106 )
or high Froude numbers (F r > 1). Note that the reader will often in the
fundamental LBE literature come across the mention of “high” Re in the
order of one. Indeed, this is very misleading for researchers involved with
civil engineering hydro-and aerodynamics applications. Although turbu-
lence modeling in a Lattice Boltzmann framework is a relatively young re-
search field, some research has been carried out. The limitations of an LB
approach in the context of turbulence modeling are not well defined since
few engineering related studies have been undertaken. Some studies in fun-
damental theoretical LBE do indicate additional terms needed to capture
smaller scales. Having said this, traditional mesoscopic model applications
usually involve length scales in the order of 10–100 nm. The investigations
undertaken typically involve combining traditional subgrid models with the
LB method. Most of the investigations are typically done with a LBGK
or MRT-LBE collison model combined with a Smagorinsky model (Lesieur
et al. (2005)). The particle distribution function and hydrodynamic mo-
ments will subsequently have to be modified from the standard form,
Eq. (11), including a total viscosity ν = ν0 + νSGS , where ν0 and νSGS
are the basic kinematic viscosity and the eddy viscosity, respectively. Some
examples can be found in the book by Zhou (2004) and in the articles
by Krafczyk et al. (2003) and Yu et al. (2006). An example is also given
by Hou et al. (1996) who developed an LB subgrid model, proposing to
incorporate space-filtered particle distribution functions. Hou et al. (1996)
present high Re flows 100 < Re < 1 × 106 for driven cavity flows. Another
interesting study was done by Derksen (2005) who compared the perfor-
mance of different turbulence models including a standard Smagorinsky
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 171

and a Sagaut mixed-scale model with application to flow in a swirl tube at

Re ≈ 2000. It is notable that Derksen (2005) successfully made use of an
immersed boundary method to avoid stair-step surfaces combined with wall
damping functions to treat flow at solid curved walls accurately. Krafczyk
et al. (2003) undertook MRT-LBE large eddy simulation studies of flow
around a cube at Re = 40,000 using the standard no-slip bounce back
rules at boundaries. Direct Numerical Simulations (DNS) of turbulence in
the LB framework (the grid Re is less than 1) has also been undertaken,
e.g. Luo et al. (2002); Yu et al. (2005); Lee et al. (2006). Multi-block grid
refinement for turbulence model predictions were recently developed by Yu
and Girimaji (2006).
It should also be mentioned that a 3-D commercial solver,
“P owerF LOW ”, exists including subgrid scale modeling capabilities (Chen
et al. (2004); Li et al. (2004)). It has 3-D modeling capabilities but with-
out free-surface algorithms. The code is distributed by Exa Corporation
(www.exa.com) with application to the aerodynamics/automobile indus-
try, allowing the user to model relatively high Re flow past fixed bluff-
bodies. Extensive evaluation of the code has been undertaken, as described
by, for example, Lockard et al. (2002). Other publications based on the
P owerF LOW code are listed on the web site of Exa.

2.3. Key Contributions

We should mention that the annual meeting, International Confer-
ence for Mesoscopic Methods in Engineering and Science (ICMMES),
www.icmmes.org, provides an excellent forum of LBM discussions. The
meeting is a relatively new effort initiated to bridge the gap between the-
ory and applications. The first ICMMES was held in 2004. The ICMMES
proceedings are usually published in two journals: one in statistical and/or
computational physics in general, and the other in computational fluid dy-
namics (CFD). Currently following special issues have been published:

• 2005: Journal of Statistical Physics, Vol. 121(1/2), 2005.

• 2006: Computers and Fluids, Vol. 35 (8/9); Intl. Journal of CFD,
Vol. 20(6).
• 2008: Progress in Computational Fluid Dynamics, Vol. 8 (1/2/3/4).

Studying papers in these volumes alone would bring the reader quickly up
to date in the area of Lattice Boltzmann modeling (we have referenced sev-
eral articles herein). We should also mention that fundamentals on LBE
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

172 J. B. Frandsen

(and beyond) can be found in a series of publications and lecture notes

by Professor L.-S. Luo and co-workers, Dept. of Mathematics & Statis-
tics, Old Dominion University, USA. These works are available on the web
site: www.lions.odu.edu/∼lluo. It is notable that most (if not all!) refer-
ences on fundamental LBE development and progress can be tracked down
through the publications of Luo. The same can be said about the research
of Professor Krafczyk and co-workers which also has a focus on engineering
applications (www.cab.bau.tu-bs.de).
The book of Sukop and Thorne (2006) is also recommendable as an
introduction to the LB method. The authors share many LB experiences
(including code material!). The code is available online (www.fiu.edu/∼
sukopm/) amongst many links to LB investigators, lecture notes, books,
on-line LB codes, etc.

3. Review of Free-surface Lattice Boltzmann Models

Several research contributions have been helpful in guiding the present

work of long water wave runup. Amongst these that can be mentioned are
the free-surface and internal wave model development of Salmon (1999a);
He et al. (1999); Ghidaoui et al. (2001); Deng et al. (2001); Xu (2002);
Ginzburg and Steiner (2003); Buick and Greated (2003); Krafczyk and
Tölke (2004); Zhou (2004); Körner et al. (2005); Ghidaoui et al. (2006);
Thürey et al. (2006b). We shall highlight some of these free-surface model
developments.

3.1. Non-Overturning Surfaces in Shallow Liquid Depths

Common for all of these works of kinetic model approximation of the

depth-integrated Navier-Stokes equations is that there is no need to for-
mulate a free-surface algorithm when assuming non-overturning waves.
The reason is that the form of the kinetic scheme allows one to solve for
the variables of the non-homogeneous shallow water equations automat-
ically. The kinetic scheme utilizes the microscopic particle distribution
function as the basis to construct the fluxes. Linear and more importantly
non-linearities at the surface are introduced and captured through the
collision term. Further, it means that the NLSW LB scheme does not
include the kinematic and dynamic free-surface boundary prescribed, as
usually required and known from traditional free-surface numerical model-
ing approaches. To understand this further, one first needs to recognize the
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 173

similarity between the equations for shallow water theory and the funda-
mental differential equations of gas dynamics (see the references listed in
§4.7). We shall later introduce kinetic theory which forms the background
of LB models (§4.1).
In the context of kinetic modeling of water waves, the approximation of
the collision integral in the Boltzmann equation becomes very important.
We should say that the LBGK approximation is the crudest approximation
of the possibilities the LB model series would offer. Since, the LB free-
surface water wave research is not an established area, the LBGK approach,
the simplest form, would however be a natural starting point for model
development. As far as we know, there exists no attempt yet which includes
a LB Navier-Stokes formulation for non-overturning, as well as breaking
waves, in shallow water depth which has undergone validation (the same can
be said for LB deep water predictions which also has not yet been explored).
We believe that this has a simple explanation, that is, LB investigators come
with very different backgrounds and interests, not to mention the numerical
approach is a relative new research area.
The first attempt on LB non-overturning free-surface flows was under-
taken by Salmon (1999a). The single phase LBGK model of Salmon obeys
the shallow water equations in rotating flows. Salmon’s interest was to use
the LB model for 3-D ocean circulation models [Salmon (1999b)]. There-
fore, a relatively large flow domain in the order of 4000×4000 km2 was
prescribed. It is also notable that the numerical results were based on rel-
atively coarse models of 100×100 lattice points. Dellar (2002) undertakes
a critical review of the stability of the shallow water LBE. Zhong et al.
(2005) is also interested in a large scale rotational flow field with applica-
tion to ocean circulation. They expanded the model of Salmon (1999a) to
include a fully explicit second-order accurate model. Ghidaoui et al. (2001)
and Deng et al. (2001) developed a finite volume BGK model for open chan-
nel flows and contaminant flows, respectively. Xu (2002) investigated the
performance of a BGK scheme, especially the treatment of shocks. Sim-
ilar to the formulation of Salmon (1999a), Zhou (2002) explored LBGK
solutions including an eddy viscosity model which represented shallow wa-
ter equations with rotating flows. Zhou (2004) accumulated many of his
investigations in book format with a focus on flow effects in channels on
rough beds with application to river mechanics problems.
Recent progress in development and applications are by Que and
Xu (2006); Ghidaoui et al. (2006); Frandsen (2006a); Thömmes et al.
(2007). Thömmes et al. (2007) applied their LBGK solver to complex
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

174 J. B. Frandsen

geometries with irregular bathymetry with application to flow in the Strait

of Gibraltar. From the view point of river mechanics and bathymetry im-
plementation, it is encouraging to learn about LB model usages. However,
the free-surface solutions were somewhat non-physical; some results are also
similar to Zhou (2004). Ghidaoui et al. (2006) have developed a Finite
Volume (FV) BGK solver for shallow water flows around circular cylinders.
They reported on wake flow physics for Re < 200,000. Que and Xu (2006)
also developed FV BGK to model roll and solitary waves. Relevant to the
present workshop, they simulated runup. To handle the wet-dry interface,
they prescribed a thin film layer on the beach slope. Comparison was done
with the experimental data of Synolakis (1987). Reasonable agreements
were found in terms of the free-surface elevations. However, the velocity
components were not reported on. We found that this variable drives the
resolution required. In other words, we found that while the free-surface
elevation would compare well with experimental data, this would not mean
that the velocity components would. See further details on this in bench-
mark test case 1, as reported in Chapter 16 in this volume.
Common to the above publication is the missing adequate time evolu-
tions of the free-surface elevation and velocities. Without report on the
time histories of the important variables, one cannot make any judgement
as to whether or not a numerical free-surface solver performs adequately.
We initiated some simulations based on a standard LBGK solver (Frandsen
(2006a)). Promising free-surface results were found in weak bore formation
studies. The weak jump in the surface agreed fairly well with the physical
model scale tests. The single phase LB model has demonstrated promise
and competitive performance compared to a Riemann solver and other free-
surface solutions. However, these solutions may be viewed as finite differ-
ence discretizations of the depth-averaged Navier-Stokes equations, and not
mesoscopic level flow solutions.

3.2. Breaking Surfaces in Any Liquid Depths

Common for all of these works is that a free-surface algorithm is embedded

in the LB kinetic frame work. Although we have not seen LB literature
for linear and steep waves free-surface algorithm, there are developments
of LB models with allowance for break-up of the free-surface. Ginzburg
and Steiner (2003) developed an immiscible LB free-surface model in single
phase flow. They proposed to treat the LB free-surface with a Volume of
Fluid (VOF) approach within the framework of the generalized LBE. The
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 175

test cases on regular grids had application to filling processes in casting

and were treated as one fluid (air particle interactions on the melt metal
were ignored) in 2-D and 3-D predictions. Filling simulations included
dense fluids of Reynolds no. <717 and Froude no. <10.7. The free-surface
solutions were based on a first-order Chapman-Enskog expansion of the dis-
tribution functions and a first-order explicit upwind scheme. Undesirable
fluctuations were dampened by inclusion of a Smagorinsky model. Since
the free-surface model of Ginzburg and Steiner (2003) was approximated
using the VOF type of approach at the free-surface, it should in principle
work for two phase flows. However, this remains to be explored and how
well this approach works for LB approximate water wave behavior remains
unanswered too. Obviously liquid with greater viscosity than water ex-
hibits quite different behavior and is in some sense easier to model, as the
free-surface motion is dampened. Later, Körner et al. (2005) developed a
laminar free-surface LBGK model with application to foaming processes,
especially foaming of metal. The free-surface was treated with a VOF
type of approach and regarded as a surface capturing method (Eulerian).
Their bubble simulation results covered 2-D advection-diffusion with D2 Q4
velocity sets. Krafczyk and Tölke (2004) extended the model of Körner
et al. (2005) by including a large eddy simulation model to describe tur-
bulent flows. Their test cases included an adaptive 3D simulation of the
classical dam-break problem and interactions with various structures. The
most recent research is undertaken by Thürey (2006) who has developed
various free-surface LB solvers, allowing for wave breaking based on VOF
(Thürey et al. (2006)) and level set algorithms (Thürey and Rüde (2004)).
Thürey recommends the VOF algorithm due to the CPU time efficiency.
Solvers of Thürey’s includes the discretization the NLSW equations and
the Navier-Stokes equations, respectively. Thürey et al. (2006b) present
VOF simulations with application to ship hydrodynamics, based on an LB
solver which couples a 2-D shallow water formulation, after Zhou (2004),
with 3-D Navier-Stokes flow. It should be mentioned that the research of
Thürey has a focus on computer graphic development. The visual effects
shown in images and animations are very impressive (www.ntoken.com).
Some source code is also made available (elbeem.sourceforge.net/).

3.3. Bubbles and Drops

Most LB publications assume that the molecules behave as hard spheres

and collide elastically. However, we should mention that a strength of the
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

176 J. B. Frandsen

LB model approach and our attraction to this approach is that one can
take the collision approach much further to include soft sphere modeling,
as highlighted in Fig. 1(b). For example, Professor Yeoman and co-workers
(Oxford, Theoretical Physics) have undertaken work on this aspect of LB
modeling in the general area of binary fluids, wetting/spreading, and the ap-
plication of droplet impact predictions (funded by the inkjet/laserjet printer
industry). The group has also made advances made in liquid crystal dy-
namics, e.g. Denniston et al. (2004). Abraham and co-workers (Purdue
University, Mechanical Engineering) have conducted a variety of bubble in-
teraction dynamics experiments, e.g. McCracken (2004); Premnath (2004);
Premnath and Abraham (2005, 2007). These experiences are useful start-
ing points for the free-surface breaking-wave application. Furthermore,
Professor Aidun and co-workers (Georgia Tech, Mechanical Engineering)
have simulated stokes flow with deformable particles amongst a long list of
contributions in the area of suspended particles hydrodynamics, e.g. Ding
and Aidun (2003). Some of the fundamental research have been applied
to biomedical problems, e.g. Ding and Aidun (2006). Recently, Aidun
et al. have introduced an LB-fluid model with deformable finite-elements
particles to simulate mesoscale blood flow. The work, in preparation for
publication, was presented at the 2006 meeting of the American Physical
Society, Division of Fluid Dynamics.
Finally we note that it is still an unresolved LB problem as to how to
tackle the high density ratio difference at the free-surface interface in two
phase flow in ocean and costal engineering applications. Some work on
this has been initiated in relation to bubble and droplets, as described by,
e.g. Inamuro et al. (2004); Lee and Lin (2005).
These LB experiences mentioned in this section, although not obvious at
first glance, are all very useful starting points for the free-surface breaking-
wave application in ocean and coastal engineering.

3.4. Status on LB Free-Surface Water Wave Validation

In our review on free-surface LB solver development, we did not come across

validation material necessary to make judgement as to whether or not an
LB approach would be a reliable type of approach to predict linear and/or
nonlinear water wave behavior. Recently, we did some bore prediction
studies based on a standard LBE solver which included relatively long time
evolutions of the free-surface. The time histories of the free-surface com-
pared fairly well with experimental results, as shown in §6.2 and in Frandsen
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 177

(2006a). The LBGK solver did not reveal any sign of numerical dissipa-
tion but an excessive resolution was required to simulate the weak jump at
the surface. We have done some further stability assessment through the
wave runup studies herein; see benchmark test case 1 reported in Chapter
16 in this volume. These studies suggest that high-order finite difference
LB schemes improve the instability problems typically found in the evo-
lution of the velocity components. Having said this, we cannot do direct
comparisons between the bore studies and wave runup problems, as the
latter test case is governed and influenced by the treatment of the wet-dry
interface.
Our observations based on the current progress do suggest advantages
when using high-order finite-difference LB schemes. Some of our LB ex-
periences indicate that this kind of LB model should be a prerequisite for
breaking LB wave models. In our free-surface review, it was notable that
some investigators have applied the standard LBE with turbulence models
and a breaking-wave algorithm (VOF). Some investigators have used the
means of turbulence models to suppress instabilities. Some LB solvers have
also gone to the extent of tuning the viscosity, i.e. so that the fluid is not
a water representation. From a numerical water wave mechanics develop-
ment view point, none of these are appropriate steps to take. Although
the above-mentioned research forms some of the background necessary for
the development of the LB free-surface framework, it is still uncertain if
an LB approach can simulate water wave behavior accurately. Little re-
search on water waves using LB modeling have been undertaken. The LB
research is very much in the beginning of entering the group of numerical
solver options available to ocean and coastal ocean engineering applications.
On a final note, we conclude that it has not yet been demonstrated that
non-overturning water waves based on standard LBE formulations or other
LB models are an adequate means of describing free-surface water wave
behavior.

4. The Governing Formulation

The present model is based on kinetic theory which attempts to describe

the macroscopic fluid behavior using the laws of mechanics and probability
theory. Provided that the fluid is near a state of equilibrium and the hydro-
dynamics moments of the equilibrium distribution functions are conserved,
kinetic theory can be shown to satisfy the Navier-Stokes equations.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

178 J. B. Frandsen

4.1. Kinetic Description of Fluid Motion

The LB Equation (LBE) originates from the kinetic theory of gases and is
a minimal form of the Boltzmann kinetic equation in which all the details
of molecular motion are removed, except those that are strictly needed
to recover hydrodynamic behavior at the macroscopic level such as mass
and momentum conservation on sufficient resolution of the lattices (Succi,
2001). The LB method discretizes kinetic theory.
Kinetic theory assumes that the fluid is described by a large number
of molecular constituents whose motions obey Newtonian mechanics. The
objective is not to know the motion of every individual molecule, but the
collective behavior for which we need a statistical description of the system,
as illustrated in Fig. 3.

Molecular Meso Continuum

Fig. 3. The LB model mimics collective molecular behavior at the mesoscopic level
(i.e. at length scales in between the atomistic and continuum mechanics levels), if the
lattice spacing is sufficiently small.

The statistical description of a fluid at or near equilibrium is contained

in the single-particle distribution function, fi (x, ci , t), where x represents
spatial coordinates, ci represents microscopic velocity of molecules and t
denotes time. This function resides in a time-evolving phase space. The
goal of kinetic theory is to formulate and solve a transport equation gov-
erning the time and spatial evolution of the distribution function with dif-
ferent collision processes dictated by the nature of the interactions between
molecules, as described by Gombosi (1994). The flow field is represented
by particles which are constraint to move from lattice node to lattice node,
i.e. in a Lagrangian manner. The particles naturally stream/propagate
and collide when meet. After collision, the particles stream to adjacent
lattice nodes. In contrast to continuum based numerical models, where
only space and time are discrete, the discrete variables of the LB model
are space, time and particle velocity. The fluid mass is collected in discrete
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 179

Table 1. Velocity set of the D2 Q9 lattice.

i 0 1 2 3 4 5 6 7 8

ci,x 0 +c 0 -c 0 +c -c -c +c
ci,y 0 0 +c 0 -c +c +c -c -c

(lattice) particles. The particles are by default located at the corner nodes
of a regular lattice. The particles move according to a finite, discrete set
of velocities. The standard LB model notation follows a Dn Qm reference
where n is dimension and m denotes number of particles. A simple example
is the 1-D diffusion model or following the notation a D1 Q2 lattice model.
A 2-D convection-diffusion model would be referred to as a D2 Q9 lattice
model and the 3-D counter part as a D3 Q19 lattice model. The velocity
components of the D2 Q9 model are illustrated in Fig. 4.

17

14 6
13
2 10 9
6 5 2
18
0
4 3

16
3 0 1 k 1
7 8
11 12
γ 5

y z β j
7
4 8 y α 15

x (a) x i (b)

Fig. 4. (a) D2 Q9 and (b) D3 Q19 velocity components.

The propagating direction of the discrete velocities vectors of the D2 Q9

model is given by

ci=1−4 = (cos[(i − 1)π/2], sin[(i − 1)π/2])c and

√ (1)
ci=5−8 = (cos[(2i − 9)π/4], sin[(2i − 9)π/4]) 2c ,

where the lattice velocity c = ∆x/∆t, ∆x is the lattice size (uniform) and
∆t is the duration of the time step. The rest particle corresponds to i=0
(ci=0 = 0) and others represent lattice vectors in the direction of the nearest
neighbors. The complete velocity set of a D2 Q9 model is listed in Table 1.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

180 J. B. Frandsen

4.2. The Boltzmann Equation

The Boltzmann equation relates the time evolution and spatial variation of
a collection of molecules to a collision operator that describes the interaction
of the molecules,

∂f 1
+ ci · ∇f = J(f, f ) , (2)
∂t 

where  = Kn = l/L and f ≡ fi (x, ci , t). The Knudsen number Kn is

defined by the length scale ratio, where l is length scale in the mean free
path and L is the characteristics macroscopic length. A continuous system
satisfies small values of  whereas the system is no longer continuous if 
is large. The form of the collision function, J(f, f ), can be derived under
the variety of different assumptions. Conventionally, it is assumed that
the density is relatively low so only binary collisions need be considered.
In general, the molecules are also assumed to be uncorrelated before the
collision takes place. Considering two particles, as originally suggested by
Boltzmann, the collision integral can be written as

π 2π
1
Z Z Z
J(f, f ) = dφ dΦ dV B(φ, V ) (f10 f20 − f1 f2 ) , (3)
2m 0 0

where m denotes mass of particle. The variables f 0 and f represent post-

and pre-collision distribution functions where φ is the angle of an incoming
particle and Φ is the angle in the plane perpendicular to the relative speed
V (measured with respect to some arbitrary origin), as shown in Fig. 5.

2 r0 f1
2
+c
f2
1 f1 −c
+c
φ V
1
r0 Φ
−c

f2
(a) (b)

Fig. 5. Elastic collision between two hard-sphere molecules.

August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 181

The quantity B(φ, V ) is given by

2
 2 k  s−1 (s−5)
b
B(φ, V )dφ dΦ = V (s−1) βdβ dΦ , (4)
m
where kb denotes the Boltzmann constant, β = (mV 2 /4 kb )1/s−1 2r0 sinφ
and r0 denotes the radius of the sphere. We see that for the case s =
5, we have B(φ, V ) = B(φ). This leads to considerable simplifications in
the theory of the Boltzmann equation. This result was first obtained by
Maxwell, and molecules which are supposed to interact in this manner are
referred to as Maxwell molecules. Therefore, for Maxwell molecules, the
relaxation frequencies are independent of relative speed V .
In view of the mathematical difficulties posed in the solving the Maxwell-
Boltzmann collision integral (3), investigators have suggested different lev-
els of approximations of the collision processes. We shall spend some time
looking into these approximations. However, our first point of investiga-
tions into applying the LB modeling approach will be the most popular
one; the single time relaxation approximation (Bhatnagar et al. (1954)).
Herein, we shall further assume that the molecules behave as hard spheres
and collide elastically, as illustrated in Fig. 5.

4.3. Collision Model: Single-Time Relaxation

Computationally, the LB method belongs to a class of the pseudo-

compressible solvers of the Navier-Stokes equations and can be classified as
a Lagrangian, local equilibrium, finite-hyperbolicity approximation. The
LBE method is originally derived from hyperbolic equations. The numeri-
cal formulation, as presented in the literature, often separates the equations
into dissipative and non-dissipative parts when solution of the equation is
sought. In the following and in the test cases presented herein, we consider
a model in which the collision assumptions are simplified to a single-time
relaxation form. The idea was first published by Bhatnagar et al. (1954).
This is the simplest form of the LBE, and as mentioned, is referred to as
the Bhatnagar-Gross-Krook (BGK) approximation. Since we focus on the
Lattice BGK approach, we shall refer to this as the LBGK model. For de-
tails on work done on BGK modeling, the reader is referred to the research
of Ghidaoui et al. (2006) who modeled shallow water flows. We should also
mention the recent work of Liang et al. (2007) who describe surface water
problems based on a BGK numerical model without need for any operator
splitting.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

182 J. B. Frandsen

The LBGK model implements a single time-relaxation-approximation

collision model. The Boltzmann equation with the BGK collision model is
discretized in velocity space by introduction of a finite set of velocities, ci ,
and associated distribution functions, fi (x, ci , t), yielding
∂fi fi − fieq
Z
0 0
+ ci · ∇fi = [f1 f2 − f1 f2 ] ≈ − , (5)
∂t τ
where the first term represents the effect of the local change of the fluid
motion in time and the second term describes the convection, that is, a
linear advection operator. The first term on the right-hand side is the
non-equilibrium distribution function which describes the effect of colli-
sions (Succi (2001)). Discretizing equation (5) in space and time yields
the following form, commonly known as the Lattice Boltzmann Equation
(LBE):
1 
fi (~x + ∆~x, t + ∆t) = fi (~x, t) − fi (~x, t) − fieq (~x, t) , (6)
τ
where the velocity space is described through the finite set of velocities
(1) and associated distribution functions (fi ) and equilibrium distribution
functions (fieq ). The factor 1/τ implements the kinematic viscosity,
c2
 
∆t ∆x
ν= τ− where c = , (7)
3 2 ∆t
and is the inverse of the relaxation time, i.e. the collision frequency of the
molecules or in other words the relaxation time represents the mean free
time between collisions. The standard LBGK scheme of (5) is conditionally
stable for τ > ∆t/2 and must be satisfied to ensure numerical stability.
The definition of the the relaxation time also allows for the possibility of
discretization error in kinetic viscosity. A useful discussion on viscosities
in LBE can be found in the article by Dellar (2001). We should also recall
that LBE actually are compressible but when the Mach number M a = u/cs
is small, then density (and temperature) fluctuations are O(M a2 ) and the
flow is approximately isothermal and incompressible. The dynamic viscos-
ity is independent of density and a function of temperature only. One should
also bear in mind always to show that density fluctuations are minimized in
LBE simulation tests. Fluid density time series, at carefully selected loca-
tions in the flow domain, can simply be used as a measure of demonstrating
that numerical error has been minimized in an incompressible flow simula-
tions, as it is constant for our ocean and coastal engineering applications
of interest, that is, herein with focus on non-overturning waves. We should
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 183

mention that recent discussions have been undertaken on the importance

of compressibility in water waves during runup and impact on sea walls.
Presentations on this matter was given by Prof. Peregrine, University of
Bristol (U.K.), and Prof. Dalrymple, Johns Hopkins (USA), at the ICCE
2006 held in San Diego, USA. We are not entering this discussion herein
but it should be mentioned that an LB model may be a useful contributing
candidate on this matter. For now the reader is referred to the discus-
sion by Dalrymple and Rogers (2006) who, in a series of publications, have
demonstrated usage of the smoothed particle hydrodynamics approach to
predict water wave behavior.
The equilibrium distribution function fieq represents the invariant func-
tion under collision (no gradients are involved) and is dependent on the
microscopic velocity of the molecule (ci ) and the macroscopic velocity (u).
In classical kinetic theory the equilibrium distribution function feq is de-
scribed by the Maxwellian equilibrium (collisions do not contribute)
ρ h (c − ū)2 i
f (0) = exp , (8)
(2πθ)D/2 2θ
√
where c ≡ ∆x/∆t = 3θ; or c2s = θ = c2 /3. Dimension of space is D, ρ,
ū and θ = kB T /m are the macroscopic density of mass, the velocity and
the normalized temperature where T is the absolute temperature, kB is
the Boltzmann constant and m is the particle mass. The Maxwell velocity
distribution function (8), published in 1859, is considered to be a key result
in developing a connection between microscopic and macroscopic physical
processes. Note that Maxwell’s distribution law (8) has identical form to
a Gaussian distribution with average velocity u and variance cs . In the
LBM framework cs is referred to as the speed of sound. However, the
physical interpretation of c2s , should this be attempted in our water wave
calculations, is only a measure of “the width of the bell” of the equilibrium
distribution function. The choice of c2s = c2 /3 is rooted in recovery of the
Navier-Stokes equation. It is a free parameter in the LB schemes and can
be tuned to a particular application of interest. We note that an expansion
of f (0) and f up to O(u2 ) is required to derive the Navier-Stokes equation,
ρ h c2 i c · u (c · u)2 u2 
f eq = exp − 1 + + − + O(u3 ) .
(2πθ)D/2 2θ θ 2θ2 2θ
The general form of the equilibrium distribution functions can be expressed
as a power series in macroscopic velocity (Rothman and Zaleski (1997)),
fieq = A + B ci ui + C ci cj ui uj + D ui uj δij , (9)
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

184 J. B. Frandsen

where the constants (A−D) are found by imposing conservation constraints

(11) and δij is the Kronecker delta. There are some degrees of freedom in the
expressions of the equilibrium distribution but the moments must represent
conservation of mass and momentum and static/dynamic pressures. It
should also be noted that fi is by definition an averaged, smooth, quantity.
This stems from the original development of the nonlinear LBE in which
noise is erased in the microdynamics of the LGCA scheme. The standard
equilibrium distributions (without a free-surface) for the D2 Q9 model are
defined as
4h 3 i
f0eq = ρ − u · u ,
9 2
eq 1 h 9 3 i
fi=1−4 = ρ + 3ci · u + (ci · u)2 − u · u , (10)
9 2 2
eq 1h 9 3 i
fi=5−8 = ρ + 3ci · u + (ci · u)2 − u · u ,
36 2 2
where ρ is the fluid density. The equilibrium distributions have important
conservation constraints, yielding zeroth to second-order moments. For
second-order accurate solutions, the conservation constraints have to be
satisfied at each lattice site for each cartesian component. The moments
represent conservation of mass and momentum and static/dynamic pres-
sures, as follows:
8
fieq = ρ ;
X

i=0
8
ci fieq = ρ u ;
X
(11)
i=0
8
1 2
ci cj fieq =
X
g ρ δij + ρ u u ,
i=0
2
where g is the acceleration of gravity. The density in (11) becomes the total
water depth h in the shallow water applications, as shown later. Satisfying
the equilibrium distribution functions is essential in obtaining valid physics.
Note that the form of the functions would be dependent on the application
of interest. For example, the deep water formulation would have different
functions compared to the shallow water equation counter-part. These
functions would typically have to be derived through Taylor expansions
and Chapman-Enskog analysis from which it can be shown that the Navier-
Stokes equation can be derived from the Boltzmann equation. This kind of
analysis will be described in the following §4.4.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 185

4.4. LB and Navier-Stokes Flows

There are mainly two ways to truncate an infinite series of transport equa-
tions. As mentioned, one technique is based on the Chapman-Enskog
method and another is introduced by Grad (1949). Herein we highlight
the former method. Chapman (1888–1970) in England and Enskog (1884–
1947) in Sweden described independently the general solution to the Boltz-
mann equation in a series of papers published 1911–21, e.g. Chapman (1916,
1918); Enskog (1921). The primary use is the calculation of transport
equations. As we know these quantities are unknown in the macroscopic
theories. The Chapman-Enskog procedure considers an expansion of the
equations for the moments of fi and assumes that the time dependence of
fi occurs through mass density, velocity and temperature. The Chapman-
Enskog expansion corresponds to a multi-scale expansion to first-order in
space and second-order in time and assumes that the diffusion time scale
is much slower than the convection time scale.
Below we shall briefly demonstrate that the Boltzmann equation (2) sat-
isfies the Navier-Stokes equation using the second-order Chapman-Enskog
expansion technique. We assume the moments of the equilibrium distri-
bution is equal to the Maxwell-Boltzmann distribution. The analysis is
performed for the Lattice BGK scheme. The Chapman-Enskog expansion
parameter is the Knudsen number (). Our definition of the coordinate
system and the subscript indices follow the velocity set of the LB cube, as
shown in Fig. 4(b).
To perform the Chapman-Enskog expansion, a Taylor expansion of the
discretized Boltzmann equation (6) is first undertaken,
 1 1 
fi (~x + ∆~x, t + ∆t) ≈ 1 + D + D2 + D3 + ..... fi (~x, t) . (12)
2 6

The differential operator D = ∆t(ci,α ∂α + ∂t ) where the first term denotes

the spatial derivatives and the second term is the time derivative. Inserting
the expression of D into (12) up to second-order, the Taylor expansion is

fi (~x + ∆~x, t + ∆t) − fi (~x, t)

 
1 1
≈ ∂t + ci,α ∂α + ci,α ∂α (ci,β ∂β + ∂t ) + ∂t (ci,α ∂α + ∂t ) fi (~x, t) ∆t .
2 2
(13)
For simplicity reasons, the notation (∂1x )α = ∂1α . This also applies to the
y- and z-directions.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

186 J. B. Frandsen

Expanding the distribution functions, the time and space derivatives in

terms of  about equilibrium gives
(0) (1) (2)
fi = f i +  fi +  2 fi + ... ;
2 (14)
∆t ∂t =  ∂1t +  ∂2t + .... ;
∆t ∂α =  ∂1α .
Note that weak deviations from equilibrium is assumed ( << 1;  ≈ ∆t).
(0)
The zeroth-order approximation, fi , is taken to be the equilibrium distri-
eq
bution, fi . We note that convection and diffusion processes are assumed
to operate at different time scales; t1 denotes convection and is the fast
time scale whereas the slower time scale t2 relates to diffusion.
We can perform the Chapman-Enskog expansion by substituting (14)
into (6) and (13), resulting in the following expression,
h 1
(∂1t + 2 ∂2t ) + ci,α ∂1α + ci,α ∂1α (ci,β ∂1β + (∂1t + 2 ∂2t ))
2
1 i 
(0) (1) (2)

+ (∂1t +  ∂2t )(ci,α ∂1α + (∂1t + 2 ∂2t )) × fi +  fi + 2 fi
2
2
1  (0) (1) (2)

≈ − fi +  fi + 2 fi − fieq .
τ
(15)
P (a)
The conservation of mass and momentum requires f
i i = 0 and
P (a)
i fi ci,α = 0 where a = 1, 2. To first-order in , (15) reduces to
(0) (0) 1 (1)
∂1t fi + ci,α ∂1α fi ≈ − fi . (16)
τ
Summing equation (16) using (11) gives
∂1t ρ + ∂1α ρ uα = 0 . (17)
Multiply equation (16) by ci,β to get
(0) (0) 1 (1)
ci,β ∂1t fi + ci,β ci,α ∂1α fi ≈ − fi ci,β . (18)
τ
Summing (18) and using (11) gives
1 − d0 2
∂1t ρ uβ + ∂1α ρ uα uβ ≈ −∂1α ρ c δα β , (19)
Dn
where d0 is a weighting coefficient related to the rest particle. It is defined
as d0 = ρ0 /(q + 1) where q is the total number of states on a lattice site
and ρ0 is the average particle density, as introduced by Chen et al. (1992)
(for example, D1 Q3 : d0 = 1/3ρ; D2 Q9 : d0 = 4/9ρ; D3 Q19 : d0 = 1/3ρ).
Dn denotes the dimensional space. The term δα β is the Kronecker delta
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 187

function where δα β = 1 if α = β or δα β = 0 if α 6= β. Expansion up to

second-order in , we get
(0) (1) (1) 1 h (0) (0)
i
∂2t fi + ∂1t fi + ci,α ∂1α fi + ∂1t ∂1t fi + ci,α ∂1α fi
2 (20)
1 h
(0) (0)
i 1 (2)
+ ∂1α ci,α ∂1t fi + ci,β ci,α ∂1β fi ≈ − fi .
2 τ
Summing (20) over i, it can be observed that several terms are zero due
to the mass and momentum conservation and due to (17) and (19). This
leaves
∂2t ρ = 0 . (21)
Multiplying (20) by ciγ and summing over i gives
X (0) (1) (1)
ci,γ ∂2t fi + ci,γ ∂1t fi + ci,γ ci,α ∂1α fi
i
1 h (0) (0)
i
+ ∂1t ci,γ ∂1t fi + ci,γ ci,α ∂1α fi (22)
2

1 h
(0) (0)
i 1 (2)
+ ∂1α ci,α ci,γ ∂1t fi + ci,γ ci,β ci,α ∂1β fi ≈ − fi .
2 τ
Several terms are can be found to be zero, yielding
(1)
X
∂1α ci,α ci,γ fi
i
(1 − d0 )c2 ρc2
 
= −τ ∂1t ∂1α ρδαβ + ∂1α ∂1β [uα δβγ + uβ δαγ + uγ δαβ ] .
Dn Dn + 2
(23)
One can further show that
∂2t ρ uγ = ν ∂1α ∂1α ρ uγ + ∂1γ (Υ ∂1α ρ uα ) , (24)
2
where ν = c (τ − 1/2)/(Dn + 2) denotes the kinematic viscosity and Υ =
(τ − 1/2)[2c2 /(Dn + 2) − c2 (1 − dn )/Dn ] is the bulk viscosity. Now the
continuity equation reads
∂t ρ + ∂ α ρ uα = 0 , (25)
and the Navier-Stokes equation is,
h ρ(1 − d ) i
0 2
∂ t ρ u α + ∂ β ρ u β uα = ∂ β c δαβ + ν ∂β ∂β ρ uα + ∂α Υ ∂β ρ uβ .
Dn
(26)
We should also emphasize that the Chapman-Enskog technique applied to
show that LBE satisfies the Navier-Stokes equation is widely published. The
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

188 J. B. Frandsen

reader is invited to extract further details on this topic in the literature: in

books, e.g. Chapman and Cowling (1958); Vincenti and Kruger, Jr. (1965);
Gombosi (1994); Struchtrup (2005); in papers, e.g. Koelman (1991); Chen
et al. (1992); Qian et al. (1992).

4.5. Boundary Treatments

In the LB method, the boundary conditions are described through distribu-

tion functions fi . There exists a variety of schemes used for implementation
of wall/bed boundary treatments. The most popular one is “the bounce-
back scheme”, e.g. Zhou (2004) which includes cells outside the computa-
tional domain (ghost cells). In this scheme, particles bounce against the
wall and the particle distribution are scattered back to the node it came
from. An example of a free-slip boundary is shown in Fig. 6. Sugges-
tions to improve the bounce-back scheme are described by Ginzburg and

∆x

∆y 6 2 5

∆y /2 8 4 2
7
Wall
6 5
Ghost cells

∆x /2 ∆x

8 2 5
∆y 7
6 6
3 1
∆y /2 1 7
7 Wall
5 2 6 8
4 8
Ghost cells
(a) (b)
Fig. 6. Free slip boundary conditions at wall and corner (a) before and (b) after
streaming.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 189

D’Humières (2003); Noble et al. (1995). Recent investigations on boundary

conditions treatment are discussed by Pan et al. (2006).
It is notable that the present depth integrated LBGK model does not in-
clude the conventional dynamic and kinematic boundary conditions at the
free-surface. Instead, the free-surface dynamics are accounted for through
the non-equilibrium particle distribution function (f neq = fi − fieq ). No ad-
ditional algorithm or surface boundary conditions are necessary to be pre-
scribed in the present numerical model. We should stress that the present
model can only expect to work well for a continuous surface in shallow
water. Discontinuous surfaces and/or other water depths would involve
entirely new LB formulations. In particular, new sets of equilibrium dis-
tribution functions would be needed. A numerical Boltzmann formulation
in arbitrary water depth could include a free-surface treatment as known
from traditional solvers, e.g. a volume of fluid algorithm, a level set, etc.
to account for breaking. Therefore a host of models could potentially be
developed to bridge the gap between micro and macroscopic models and
thus potentially advance the way we currently model free-surface flows. A
further strength is that the standard LBE scheme is known not to introduce
numerical viscosity.
Regarding open boundary and wet-dry interface treatments, please see
the description related to the wave runup test case presented in Chapter
16 in this volume.

4.6. Standard LBE and Higher-Order Schemes

With reference to the list of LBM properties in the introduction section,

we are now ready to make some further important observations. First we
recall that Eq. (6) is referred to as the standard LBE. We observe that this
is the results of replacing the time derivatives (5) by a first-order time dif-
ference and a first-order upwind discretization for the convective term. We
observe that (6) represents a finite difference equation. Therefore, viewing
the LBM as a finite-difference model for solving the discrete-velocity Boltz-
mann equations, it becomes necessary to address numerical instability and
accuracy. It is important to note that the equation is first-order accurate
in time and space. This contradicts the former derivation (§4.4) as it was
shown that the LBE satisfies the Navier-Stokes equation in the near in-
compressible limit. Sterling and Chen (1996) mentioned that the LBE is
second-order accurate both in time and space due to the Lagrangian nature
of the spatial discretization of (6) embedded in the viscous term. Recently,
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

190 J. B. Frandsen

Junk et al. (2005) undertook a detailed analysis of the LBE. They related
the LBE to the finite discrete-velocity model of the Boltzmann equation
with diffusive scaling and obtained the incompressible Navier-Stokes equa-
tions as opposed to the compressible Navier-Stokes equations obtained by
the Chapman-Enskog analysis with convective scaling. Junk et al. proved
that the LBE is second-order accurate in space. For the incompressible
Navier-Stokes, they also proved that the LBE is first-order accurate in
time asymptotically. Gou et al. (in press) also recently contributed to this
concern. Through unsteady flow predictions, Guo et al. observed that LBE
is second-order accurate in time.
Further, we note that the standard bounce-back condition, the most
applied boundary condition, is first-order in numerical accuracy at the
boundaries. This, of course, degrades the LB method. As mentioned in
the review, several investigators have contributed to second-order accurate
boundary conditions. New contributions to the LBM literature are for this
reason (and others) still on-going research.
In this chapter, we present results based on the standard LBE. The wave
runup test case, presented in Chapter 16 in this volume, has undergone fur-
ther tests including high-order finite-discretization schemes, thus improving
the standard LBE. We should mention that it did not originally occur to in-
vestigators that the LBE could be expanded this way (like traditional CFD
methods). It is still not common to consider a particular discretization for
the discrete Boltzmann equation. However, some progress on high-order
LB discretization schemes have been reported. Cao et al. (1997) were the
first investigators to view the LBM as a special finite difference (FD) dis-
cretization. Further elaborations on this topic have been undertaken since,
e.g. Kandhai et al. (2000); Junk (2001); Seta and Takahashi (2002); Sofonea
and Sekerka (2003); Van der Sman (2006). Some Finite Volume (FV) LB
models have also been developed, e.g. Ghidaoui et al. (2006); Que and Xu
(2006). As far was we are aware of, there has been no comparison under-
taken for FDLB and FVLB model, so the difference in solutions accuracy
is not clear to us. Furthermore, we should also mention that various dis-
cretization schemes for gas dynamics computations are described in details
in the book by Laney (1998).

4.7. Gas Dynamics Analogy

Since it is not obvious why a numerical free-surface model with roots

in gas dynamics theory would work for water waves, we have gathered
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 191

Fig. 7. Kinetic modeling approach to water wave predictions with roots in the gas
dynamics theory of Boltzmann. Comments and equations derivations on this idea can
be found in the listed summary of references. (The photograph is a wave runup snapshot
from the Northshore of Oahu, Hawai’i.)

some interesting comments and observations made by other investiga-

tors, as summarized in Fig. 7. It is believed that Dimitri Riabouchinsky
(1882–1962), a Muscovite fluid dynamicist, was the first investigator to
publish a note on recognizing the similarity between the equations for
shallow water theory and the fundamental differential equations of gas
dynamics (Riabouchinsky (1932)). Stoker (1957) elaborates on the
French notes and has a couple of pages in his book (pp. 25–26) on this
subject. The book of Mei (1992) also has some related notes (pp. 512–
513) on this topic and one can find a quote: “Knowledge and meth-
ods established in gas dynamics can be transferred directly to long water
waves.”
Although the LB results herein simulate long waves, we should say that
the reason why we have proposed the LBM originates from independent and
different ideas. It is our goal and intent to apply the method to breaking
waves in any water depth. It was nevertheless very encouraging to find
these references which are relevant to the present contribution.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

192 J. B. Frandsen

5. LBGK Modeling with Application to Water Waves

We shall in the remaining part of the chapter focus on applying and testing
a Lattice Boltzmann modeling approach to free-surface water waves. We
assume that the collisions of particles are represented through the single
time relaxation scheme (LBGK approximation). It should be noted that
there is no deeper reason for starting the Boltzmann implementation this
way other than this is the first logical modeling development step. How-
ever, this is not necessarily the most accurate and stable approach due
to the BGK collision approximation. Our LB model contains equilibrium
distribution functions which are based on particle motions in shallow wa-
ter and therefore our test cases are limited to shallow water depths. The
discretized equations are based on viscous free-surface waves. Therefore,
flows with horizontal velocity dominating free-surface waves is applicable
and is our focus only. Furthermore, the present LBGK model explores the
free-surface behavior of non-overturning water waves on uniform grids. For
validation purposes, we first investigate the dynamics of viscous liquid and
water waves in tanks with flat and slope beds; in 1-D and 2-D, respectively.
Then, we test the tsunami runup on sloped beaches (workshop benchmark
test 1). The latter test case is presented in Chapter 16 in this volume.
We consider the flow of water with a free-surface under gravity in a 3-D
domain where x − y denotes a horizontal plane while z defines the vertical
direction. The free-surface elevation coincides with the z-axis. Assuming
that the vertical component is negligible in the field of interest, the fluid
motion can be described by the shallow water equations with rotating flows,

∂h ∂(h ux ) ∂(h uy )
+ + = 0;
∂t ∂x ∂y

∂(h ux ) ∂(h u2x ) ∂(h ux uy )

+ +
∂t ∂x ∂y

∂  h2  ∂  ∂ux  ∂  ∂ux 
= −g + hν + hν + Fx ; (27)
∂x 2 ∂x ∂x ∂y ∂y

∂(h uy ) ∂(h u2y ) ∂(h ux uy )

+ +
∂t ∂y ∂x

∂  h2  ∂  ∂uy  ∂  ∂uy 
= −g + hν + hν + Fy ,
∂y 2 ∂x ∂x ∂y ∂y
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 193

where h = h0 + ζ, h0 is the still water depth (or initial water depth) and ζ
denotes the free-surface elevation measured vertically above still water level,
t is time, g = 9.81 m/s2 denotes the acceleration of gravity, ρ = 1000 kg/m3
is the water density, ν = 1 × 10−6 m2 /s is the kinematic viscosity, ux , uy
are the horizontal depth-averaged velocity components, and F denotes force
terms.
Equivalent to (27), the proposed single phase LBGK formulation for
shallow liquid rotating flows can be written as

2
∂fi ∂fi (fi − fieq ) 1 1 X
+ ci =− + c i Fi , where Ni = ci ci ,
∂t ∂x τ Ni c 2 c2 i=0
(28)

with Ni as a constant which depends on the lattice geometry and is defined

by Ni = 2 and Ni = 6 for the D1 Q3 and D2 Q9 velocity set, respectively.
F denotes force terms. Herein, it is assumed that, at any time, the LB
fluid is characterized by the populations of three discrete velocities (D1 Q3
model) in 1-D. The nine discrete microscopic velocity model (D2 Q9 model)
represents the physics of waves in a 2-D horizontal plane (x − y) based on
the depth-average equations (27), as shown in Fig. 2.
Equation (28) can be viewed as a special finite-difference discretiza-
tion of the single time relaxation approximation of the Boltzmann
equation for discrete velocities. One approach to solving the discrete
Boltzmann equations is to use a first-order Euler time difference scheme
and a first-order upwind space discretization for the convection term in
a uniform lattice spacing ∆x. One can then obtain following algebraic
relation,

∆t   ∆t
fi (~x + ∆~x, t + ∆t) − fi (~x, t) = − fi (~x, t) − fieq (~x, t) + c i Fi .
τ Ni c 2
(29)

This equation is commonly referred to as the Lattice-BGK equation

or Lattice-Boltzmann Equation (LBE). As mentioned, it can be shown
that the LBE is second-order accurate in space and first-order accu-
rate in time (Junk et al. (2005)). See also further discussion of this
in §4.6.
The LBGK-based equilibrium distribution functions in shallow water
for a D2 Q9 model are (after Salmon (1999a)),
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

194 J. B. Frandsen

eq 5g(h0 + ζ)2 ∆t2 2(h0 + ζ)∆t2

fi=0 = h0 + ζ − − (ui ui ) ,
6∆x2 3∆x2

eq g(h0 + ζ)2 ∆t2 (h0 + ζ)∆t2

fi=1−4 = + (ci ui )
6∆x2 3∆x2
(h0 + ζ)∆t4 (h0 + ζ)∆t2
+ (c i c j u i u j ) − (ui ui ) , (30)
2∆x4 6∆x2

eq g(h0 + ζ)2 ∆t2 (h0 + ζ)∆t2

fi=5−8 = + (ci ui )
24∆x2 12∆x2
(h0 + ζ)∆t4 (h0 + ζ)∆t2
+ 4
(ci cj ui uj ) − (ui ui ) ,
8∆x 24∆x2
where the subscript i denotes the propagating direction of the discrete
velocities, as described in Eqs. (1). Important to water waves simulations
are the zeroth to second moments which represent conservation of mass and
momentum and static/dynamic pressures, as follows:
X eq
ci fieq = h(x, t) ui ;
X
fi = h(x, t) ;
i i
(31)
1
ci cj fieq
X
= g h(x, t)2 δij + h(x, t) ui uj .
i
2

An elastic-collision scheme assuming no flow through slip walls and bed

have been prescribed using the well-known bounce-back rules (e.g. Zhou
(2004)).
The present mesoscopic numerical solutions are solved on uniform lat-
tices. Initially we typically prescribe the non-physical values of the velocity
u = 0 and the distribution functions fi = fieq . In the test cases of the
present chapter, we have used a splitting operator scheme, as outlined in
the solution procedure of Fig. 8. So the standard LBE scheme herein sep-
arates into advection and diffusive parts before computing fi . Finally, the
microscopic properties are transformed into macroscopic variables of the
free-surface (ζ) and the depth averaged velocities (ux , uy ) are calculated as
the first and second moments of the distribution function,
8 8 8
X 1 X 1 X
ζ= fi −h0 ; ux = cix fi and uy =
ciy fi .
i
(h0 + ζ) i (h0 + ζ) i
(32)
The procedure is repeated at each time step ∆t and the surface elevation
and velocity is updated from (32). The hydrodynamic moments of the
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 195

Fig. 8. Solution procedures of standard LBGK model.

equilibrium distribution functions (30) are satisfied at every time step.

Further, it should be noted that it is assumed that the Mach number
M a = u/cs << 1, where cs is the speed of sound. The Peclet number
P e=u ∆x/ν < 2 and the Courant number Cr=u ∆t/∆x < 1 are also obeyed
(u = ux or u = uy ). It should be noted that the P e and Cr constraints are
borrowed from continuum level models and thus may not necessarily apply
to LB models. Establishment of LB constraints requires more validation
work than presented herein.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

196 J. B. Frandsen

Finally, we should emphasize that the standard LBGK solutions of (28)

are second-order accurate in space and first-order accurate in time (see
discussion in §4.6). Further, it should be noted that the LBGK schemes
are highly suited for vectorization and parallelization. However, the test
cases herein are generated using a single CPU of about 3 GHz.

6. Case Studies

We have undertaken various test cases to examine the performance of the

LBGK free-surface solver. These are preliminary test cases which do not
involve the benchmark test cases of the workshop. We do this because
the LB modeling approach for water waves has not undergone any thor-
ough validation other than the 1-D LBGK studies in tanks undertaken by
Frandsen (2006a). We should note that these test cases proved very helpful
in the model development of the tsunami-generated wave runup modeling
problem. An overview of the various numerical experiments undertaken so
far are illustrated in Figs. 9–11.
Prior to the free-surface simulation test case initiation, we tested
the LBGK solver performance without a free-surface, as described in

Fig. 9. LBGK bluff-body Navier-Stokes flows without a free-surface.

August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 197

zoom for time No 1 zoom for time No4

180
180

170
170

160 160

150

y
150
y

140 140

130 130

120 120
180 200 220 240 260 180 200 220 240 260
x x

Fig. 10. LBGK flow past circular cylinder. Velocity snapshots (Re = 200).

Frandsen (2006b). This particular article also showed some initial at-
tempts on predictions of free-surface bluff-body interactions (e.g. Fig. 2).
These studies included single and multiple bluff-bodies studies at Re ≤ 250
(Fig. 9). In general, we found that the LB model provided an easy means of
handling multiple bluff-bodies with sharp edges. We are currently modeling
flow around circular cylinders (Fig. 10). The curved boundary is modeled
through the second-order accurate boundary condition proposed by Filip-
pova and Hänel (1997). Good agreement with other investigators have been
achieved for Re ≤ 250. We have also implemented a curvilinear formula-
tion, as developed by He and Doolen (1997). This approach seems to work
equally well for flow around circular cylinder.
The free-surface test cases, as illustrated in Fig. 11, have so far in-
cluded (1) weak bore simulations; (2) simulations of wave interactions with
surface-piercing bodies; (3) dam-break predictions; (4) force term assess-
ment (e.g. beach slope); and (5) wave runup. We have done extensive
studies of bore predictions (without bluff-bodies) and wave runup.
We shall in the following, present some of the bore studies and dam-
break predictions. The wave runup studies follow the description of the
benchmark test 1 set-up problem, which is reported in Chapter 16 in this
volume.

6.1. Case Study: Viscous Based Bores in Fixed Tanks

Viscous free sloshing in fixed rectangular tanks is chosen as a first bench-

mark validation test of the LBGK solver. The numerical model is based
on the NLSW in rotational flow (27). Our results are based on a D2 Q9
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

198 J. B. Frandsen

Fig. 11. LBGK free-surface flows tests in shallow water.

model in the horizontal plane (x-y). The bounce-back scheme for slip walls
have been prescribed. We should emphasize that the solutions in this first
test series are based on highly viscous liquid. This is not only motivated
by our comparison source (Wu et al (2001)) but also because we were at
first not certain if we could simulate free-surface behavior at all, including
both highly viscous liquid and water. Therefore, we began our studies with
highly viscous liquid.
The numerical predictions are compared with the linearized Navier-
Stokes solution derived by Wu et al. (2001) who showed that the wave
elevation at an arbitrary depth is given by
ζ0 g k2 tanh(k2 h0 ) tanh(α2 h0 ) λ t
Z
ζ = ζ0 + e dλ , (33)
2π ν i Γ C2
p
where k2 = 2π/b is the wave number for n2 and αn=2 = kn2 + λ/ν,
where λ is a complex number and b denotes the width of the tank. The
denominator in (33) is complex and is presented in detail in the article of
Wu et al. (an analytical solution exists for λ = 0 only). The deep water
limit of (33) is explored by Wu et al. (2001) for Re = 2, 20, 200, with
√
Re = h0 g h0 /ν, in a rectangular tank with an aspect ratio h0 /b = 0.5.
In this paper, it is the shallow liquid limit of (33) which is used in the
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 199

comparison studies. Assuming that the liquid depth is sufficiently small

such that tanh(kn h0 ) → kn h0 and tanh(αn h0 ) → αn h0 , it can be shown
that the denominator simplifies to
λ α 2 h0  2 
C2 = − λ + 4 k22 νλ + g k22 h0 , (34)
ν
and (33) simplifies to

ζ0 g k22 h0 ν eλ t
Z
ζ = ζ0 + × eλ t dλ . (35)
2π ν i Γ −λ3 − 16π 2 νλ2 + g 4π 2 h0 λ
In this form, the denominator has three poles in the complex plane which
allows the integral to be calculated analytically. The integration path Γ is
the path for calculating inverse Laplace transforms. Letting ωn2 = g kn2 h0
where ωn represents the sloshing frequency in shallow liquid, (35) reduces
to
ζ0 ωn2 −ν kn2 t
ζ = ζ0 + e f (ν kn2 t) , (36)
kn2 ν

where f (ν kn2 t) is the inverse Laplace transform which solution can be found
in Poularikas (1996). The derivation leads to following solution for the
viscous time evolution of the free-surface in shallow liquid,
ζ0 h
ζ= p 1 − kn2 ν
2 4 − ωn2 /(kn2 ν)2
   √ 
2 /(k2 ν)2 ν k2 t
p −2− 4−ωn (37)
− k 2 ν 2 + 4 − ω 2 /(k 2 ν)2 e
n n n
n n

   √ 
2 /(k2 ν)2 ν k2 t
i
p −2+ 4−ωn
+ kn2 ν 2 − 4 − ωn2 /(kn2 ν)2 e n n
.

Initially the velocity in the flow domain is zero, and the free-surface
ζ0 (x, n) = a cos(kn x), where a is the amplitude of the initial wave profile
and x is the horizontal distance from the left wall. Nonlinear free-surface
motions are investigated by varying the wave steepness, defined herein as
 = a ωn2 /g and is a measure of non-linearity at the free-surface in the
fixed-tank studies. The results presented are for a tank of aspect ratio h0 /b
= 0.05 and for Re = 20, 200 with corresponding ν = 1.75×10−3 m2 s−1 ,
1.75×10−4 m2 s−1 . The Re = 20 corresponds to glycerine type of liquid
whereas Re = 200 is oil-based liquid (SAE 20W-20). The time histories of
the free-sloshing motions are presented in non-dimensional form using the
sloshing frequency ωn and the initial amplitude a.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

200 J. B. Frandsen

1 1
ε = 0.1972 ε = 0.1972
ε = 0.0296 ε = 0.0296
ε = 0.002 ε = 0.002
Wu et al. Wu et al.

ζ/a
ζ/a

0 0

−1 −1
0 20 40 0 25 50
txω t x ω2
2

Fig. 12. Free-surface at wall at Re = 20 and Re = 200.

Figure 12 shows the time histories of the free-surface elevation for the
second sloshing mode (n = 2) for the liquids with Re = 20 and 200, re-
spectively. Each of these Re solutions are based on a grid resolution of
100 and 500 nodes, respectively. Solutions are presented for time steps ∆t
= 0.01 s, 0.002 s and ∆x = 0.01 m, 0.002 m yielding τ = 1.025 for low-
est grid density/Re case and τ = 0.7625 for highest grid density/Re case.
The increase of nodes and associated reduced time step were necessary in
order to maintain accuracy of the explicit solver for higher Re. The free-
surface time histories at the left wall are shown for small to moderate waves
( = 0.002, 0.197). The Re = 20 result shows one oscillation cycle only.
Hereafter, due to the high viscosity, the amplitude decays to zero. This is
also true for all of the other wave steepness cases considered shown. The
small initial wave  = 0.002 case agrees fairly well with the linear solution
of Wu et al. The increasing wave steepness introduces a phase-shift with
larger peaks and lower troughs. The higher Re = 200 test case generates
larger wave steepness (k2 × ζmax = 0.43) with associated amplitude skew-
ness and an increase in oscillation cycles, which eventually decays toward
zero, representing the physics well. We should point out that the skewness
of the amplitudes is the characteristic of a bore as opposed to non-skewed
amplitudes of sloshing motion. Bores are known to form at shallow liquid
depths for a relatively large forcing frequency. The next validation study
is on water bores.

6.2. Case Study: Water Based Bores in Moving Tanks

Recently, the author was involved in setting-up small scale wave tank exper-
iments to investigate sloshing motions. The test facility is shown in Fig. 13.
The set of test series described below were highly motivated by having this
experimental facility. The present model explores the free-surface behavior
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 201

Sway

b/2
y

3 x Surge

b /4
2

1m

b /4
1m
1
l /2 l /2

Fig. 13. Shake table with shallow water sloshing in a square tank with base dimensions
of 1×1 m2 . Definition of excitation modes and plan view of tank with wave gauge
locations (1–3). (The photograph is from the laboratory of Frandsen located at Louisiana
State University, USA.)

of the LBGK model, as described above. We have tested the performance

of the LBGK free-surface solver in tanks on flat beds (Fx = 0; Fy 6= 0 in
Eq. 27). The forced horizontal acceleration of the tank in the y-direction
(sway) is Y 00 = ah ωh2 cos(ωh t) where ah , ωh are forcing amplitude and fre-
quency, respectively (F = h × Y 00 in Eq. 28). The case study presented
includes free-surface behavior with an initial water depth h0 /b = 0.05 where
b = 1 m is the tank width. At this shallow water depth, it is expected that
the free-surface has horizontal dominating velocity components, and thus
is suitable for the present model formulation. Moreover, we emphasize that
in shallow water depth, a traveling wave or a bore may be generated, as
reported by, e.g. Wu et al. (1998) and Mei (1992), as opposed to sloshing
motion occurring in moderate to deep water depths. Herein we discuss the
free-surface behavior in relation to waves generated due to sway (and/or
surge) base excitation in a square tank, as illustrated in the sketches of
Figs. 13 and 14. Special attention is given to cases that generate bores.
Although the bore formation is confined in tanks, the physics at the lo-
cation of the bore is similar to bores observed on beaches, as researched
by, e.g. Keller et al. (1960); Shen and Meyer (1963); Hibberd and Peregrine
(1979); Yeh et al. (1989). Moreover, fully developed tank bores resemble the
tongue of a tsunami wave. However, due to the non-overturning numerical
formulation, our numerical study is limited to weak bores.
The effects on the free-surface when forcing the tank to move in mod-
erate sway motion, are investigated. The forcing frequency is varied to
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

202 J. B. Frandsen

z
ζ U0
h0
uy

h1
α

h2
y
b

Fig. 14. Definition sketch.

capture off- and near-resonance free-surface water waves while the forc-
ing amplitude are ah /b = 0.006 or ah /b = 0.02. Initially, the free-surface
elevation and the velocity field are prescribed to be zero. The non-
dimensional horizontal forcing parameter κh = ah ωh2 0 /g and the Froude
√
number F r = U0 / g h1 (where U0 is the propagating speed and h1 is the
developed water depth) are used as a measure of non-linearity and bore
strength, respectively.
The numerical results are compared with experimental data, as de-
scribed elsewhere (Frandsen (2006a)). We also compare the LB results
with a low-order Riemann solution. Herein we use a first-order upwind
method with simple linear interpolations, as described by Toro (1999). We
realize this is a shortcoming in our comparison study as it is well known that
schemes which are first-order accurate in space and time can produce arti-
ficial viscosity. Indeed, this was experienced. Recognized shock-capturing
schemes typically contain high-order interpolations with second-order accu-
racy methods including limiters. Extensive theories of these schemes exist
in terms of flux functions (e.g. Roe, Godunov, Van Leer, Steger-Warming,
etc.), reconstruction methods and limiters, however, this will not be fur-
ther discussed here. Instead, the reader is referred to the comprehensive
descriptions of the host of Riemann solvers in the books by Toro (2001);
LeVeque (2002); and Chapter 2 in this volume by LeVeque.
All results are presented in dimensionless form. The free-surface (ζ) is
non-dimensionalized by the still water depth (h0 ) and the non-dimensional
time T = ω1 t and the non-dimensional time step ∆T = ω1 ∆t where the
first initial
p linear natural sloshing frequency in shallow water is defined as
ω1 = g (π/b)2 h0 . The first numerical tests carried out were designed to
check the sensitivity of the LBGK scheme to the time step and the grid res-
olution (Frandsen, 2006a). The 1-D simulations are based on D1 Q3 lattices.
The bore formation was found to occur at Fr1 =0.93, showing two distinct
developed water depths h1 /h2 = 1.34. All grid resolutions used yielded
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 203

0.4 0.2
T =128.80 T =129.43 T
6 7 8
T =130.06
8
T =130.69
9

0
ζ/h
ζ/h0

−0.4 −0.15
−0.5 0 0.5 −0.086 −0.085
y/b y/b

Fig. 15. LB wave profiles (Fr1 = 0.93; h1 /h2 = 1.34) near resonance (ωh /ω1 = 1; κh =
0.003). Grid resolution, nodes: −−, 20,000 (τ =0.547); − · −, 30,000 (τ =0.578); − + −,
40,000 (τ =0.591); − ◦ −, 50,000 (τ =0.643); −, 60,000 (τ =0.553).

similar wave profiles and thus represent grid converged solutions except at
the bore tip, as shown in Fig. 15. Higher grid resolution is required for
the near resonance cases compared to outside resonance, as expected. It
was found that the LB solutions for κh = 0.003 required a minimum of
about 10,000 nodes outside resonance and 20,000 nodes near resonance,
respectively, to capture the physics experienced experimentally. However,
in order to represent the bore tip adequately, this would not yield accurate
solutions. The LBGK and Riemann solutions agree fairly well in terms of
the free-surface elevation and velocity predictions. However, a phase lag
exists which appears to be more pronounced in the wave profiles than in
the time evolutions at various locations in the tank. The phase lag is caused
by numerical viscosity related to the first-order upwind scheme of the Rie-
mann solution. The low-order Riemann solution produced a slower bore
propagation compared to the LBGK bores. We found that, to completely
suppress the bore tip oscillations, a 60,000 node resolution for the standard
LBGK model would be required (Fig. 15). The first-order Riemann solution
would require about a four times smaller time step to achieve equivalent
accuracy. Further, it should be noted that the reason for the requirement of
the high resolution is due to the vertical jump in the surface. Lesser resolu-
tion would yield a sloped jump. Moreover, the Riemann solution with 2,000
nodes would not be able to capture a bore whereas the LB solution would.
In general, the low-order Riemann solution would have a tendency to pre-
dict sloped jumps whereas the LB solution would have vertical jumps for
the same grid resolution (analogous to conservative versus non-conservative
methods). However, the LB bore front showed numerical oscillations at the
tip (Fig. 15) which could be suppressed with an increase in resolution, or
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

204 J. B. Frandsen

0.6
a h /b = 0.02

ζmax/h0

a h /b = 0.006

0
0.8 1.0 1.2 1.4
ωh/ω1

Fig. 16. Maximum free-surface elevation at the wave gauge 2 (x, y) = (0, -b/4) location.
−, LB solution; −−, Experimental solution. •, time evolution comparison.

perhaps better using a high-order LB finite-difference scheme. We fur-

ther compare the LBGK solutions with the experimental tests series. The
test series are summarized in Fig. 16 showing comparisons of the max-
imum free-surface elevation of the time series at forcing frequency ratio
ωh /ω1 ∈ [0.85, 1.2]. The numerical near resonance test series are shown
with distinct data points for the two forcing amplitudes. Figure 16 rep-
resents surface locations at the quarter locations. Fairly good agreement
is achieved away from the wall whereas, at the wall itself, large discrepan-
cies exist (not shown) due to the limitations of the numerical formulation.
Furthermore, it can also be observed that the water motion exhibits a hard-
ening spring oscillator behavior in shallow water, as the maxima occurs at
ωh /ω1 > 1. The hysteresis effect becomes more pronounced with increas-
ing κ, as expected. In these comparison studies, the LB solutions are based
on 40,000 nodes (τ =0.0.591) and ∆T = 8.8 × 10−5 whereas the Riemann
solutions have 40,000 nodes and ∆T = 4.4 × 10−5 . Figures 17 and 18 show
a comparison of the free-surface elevation for the two forcing amplitudes.
The close-up look (T ∈ [190, 210]) represents steady state evolutions. The
near resonance cases showed small amplitude waves in a first sloshing mode
initially (T < 30), both experimentally and numerically. The experimental
and numerical horizontal (sway) base excitation test series captured similar
weak bore features. However, the forcing amplitude ah /b = 0.006 required
numerically to generate a weak bore was smaller than the experimental
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 205

0.3 0.3
ζ / h0

ζ / h0
0 0

−0.3 −0.3
0 15 30 0 15 30
t × ω1 t × ω1

0.4 0.4
ζ / h0

ζ / h0
0 0

−0.4 −0.4
190 200 210 190 200 210
t × ω1 t × ω1

(a) wave gauge 2 (l/2,b/4). (b) wave gauge 3 (0,0).

Fig. 17. Small forcing ah /b = 0.006. Comparisons of free-surface evolutions at transient
and steady state phases. −, LB solution; − · −, Riemann solution; −−, experimental
solution.

equivalent (ah /b = 0.02). It was found that the experimental tank required
greater initial momentum than the numerical tank, simply occurring in
the physical process of activating the movement of the tank. Therefore,
it is not possible to compare test series for the exact same forcing am-
plitudes. Although the dominating horizontal flow maintained its overall
one-dimensionality inside the tank, away from the bore tongue and walls,
discrepancies were found at the bore location and at the wall due to vertical
components present in the experimental data. This could also be a reason
the bore is developed for smaller forcing amplitudes numerically. Further,
it should be noted that the skewness of the numerical free-surface is not
typical. It is caused by the bore which occurred periodically in the tank.
The numerical simulations predicted a bore (Fr1 = 0.93, h1 /h2 = 1.34)
whereas the physical experiments captured traveling waves (Fr0 = 1.0,
h1 /h2 ≈ 1). Although the numerical solutions agree overall fairly well
with the experimental data, as shown in the close-up look (e.g. Fig. 18), it
should be emphasized that the vertical velocity components occurring at
the walls and at the location of the bore are not captured by the governing
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

206 J. B. Frandsen

0.8 0.8

0.4 0.4
ζ / h0

ζ / h0
0 0

−0.4 −0.4

−0.8 −0.8
190 200 210 190 200 210
t×ω t×ω
1 1

(a) wave gauge 2 (l/2,b/4). (b) wave gauge 3 (0,0).

Fig. 18. Moderate forcing ah /b = 0.02. Free-surface evolutions and snapshots of wave
profile comparisons (time instances labelled with vertical dashed lines). −, LB solution;
− · −, Riemann solution; −−, experimental solution.

equations, and are thus a source of error inherently built into the numerical
formulation. Further details and discussions on the 1-D solution can be
found in Frandsen (2006a). Two-dimensional simulations with sway base
excitations have also been undertaken. We found that the 1-D and 2-D
weak bore solution compares fairly well.

Fig. 19. Snapshots of wave profile comparisons for sway and surge base excitation.

Our third numerical test to be presented involves sway and surge base
excitation for ahx = ahy =0.006 (expanding the above pure sway case). The
fluid flow is characterized by D2 Q9 lattices. These preliminary test series
include low resolution grids 500×500 (∆t × ω1 =0.002). Figure 19 shows
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 207

numerically and experimentally generated time instances of the propagat-

ing bores. Two consecutive bores were observed to form, propagating di-
agonally at F r1 =1.78 (h1 /h2 =2.27). The bores were also observed to form
at ah /b=0.006 similar to the numerical experiments. It should be noted
that similar observations have been reported by Wu et al. (1998) who used
a fully nonlinear 3-D potential flow solver.
The 1-D results presented herein have been computed on a single pro-
cessor of 3.4 GHz. The near resonance cases were the most intensive with
average LB CPU time in the order of 6 h for 60,000 node resolutions. The
combined sway and surge test cases took about 8 h (low resolution grids
compared to the pure sway cases).

7. Case Study: Dam-Break Simulations

The following test series represents a classical dam-break problem on a wet

bed. The present test case is carried out to test the solver’s further ability
to handle these kind of shocks. The model is discretized with D2 Q9 lattices
(Fig. 20(b)).
The domain size is l × b = 200 × 200 m2 . The initial still water level is
h1 =10 m when y < b/2 and h2 =5 m when y > b/2. A flat smooth bed is
assumed and the shear stress is assumed negligible. The outer tank walls
are modeled with a slip condition whereas the dam-break wall has no-slip

Cl
y
lw1

x
Flow 2
6 5
lb

Cl
l

3 0 1
h1 h2
lw2

y
7 8
4
b (a) x (b)

Fig. 20. (a) Definition sketch of the dam-break test case. (b) D2 Q9 velocity set.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

208 J. B. Frandsen

Fig. 21. Snapshots of dam-break surface flow through asymmetric wall configuration
at 1.6 s, 4.9 s, 7.2 s.

(a) (b)

10
10
] ][m]

1s 3s 5s 7.2s

5200
5
0
0 y
y/b [m] 200
(c)
Fig. 22. Snapshots of the free-surface elevation contours of dam-break flow through the
asymmetric wall configuration (in meters). (a) LBGK, (b) BGK, after Ghidaoui et al.
(2001), (c) LBGK free-surface profiles at selected instances (x = lw1 + lb /2).

conditions prescribed. The standard first-order accurate bounce-back tech-

nique is applied. The walls are located at y = b/2. The wall lengths lw1
and lw2 are 30 m and 95 m, respectively. The preliminary test case is set-up
with a grid resolution of 400×400 and the time step is ∆t=0.0025 s. The
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 209

asymmetric dam-break problem has a breach lb = 75 m, as shown in Fig. 20.

The dam is assumed to break instantaneously. The Froude number for this
study is approximately F r ≈0.34. The free-surface time evolutions of the
dam-break are shown in Fig. 21. The magnitude of the elevations are shown
in Fig. 22. Good agreement is found with similar studies of Fennema and
Chaudhry (1990) and Ghidaoui et al. (2001) who used a finite discretization
of the shallow water equations and a finite-volume BGK scheme, respec-
tively. Future studies shall investigate supercritical flows in which an order
of magnitude higher grid resolutions are expected if using the standard
LBGK scheme. Values of h1 /h2 larger than 2 also would require higher
resolution and/or scheme improvement over the present reported test case.
In summary, the LBGK appears to yield promising results. It is notable
that the resolution for the dam-break test case is an order of magnitude
less compared to the bore studies presented in previous test cases. We are
currently undertaking further investigations.
We have done a variety of other test cases in the step-by-step implemen-
tation and validation process of the LBGK solver aiding towards modeling
the tsunami-generated runup on sloped beaches. Some of the work is still

(a)

(b)

(c)
Fig. 23. (a) Dam-break problem including (b) wall movement and (c) bed slope (α = 4);
5 s between snapshots.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

210 J. B. Frandsen

in progress. However, we would like to present some preliminary test cases

to the reader just to provide some insight into what would be possible with
this type of solver. For example, we briefly looked at comparing the effects
of adding bluff-bodies to the above dam-break problem, as shown in Fig. 23.
No-slip boundaries are prescribed at the surfaces of the bluff-bodies. We
also included a bed slope α (Fx = 0; Fy 6= 0 in Eq. (27)). The force term
Fy = −g h ∂h∂y where hb is the height of the slope. Obviously many issues
b

need to be verified here, especially high F r and boundary layer physics,

but overall the physics of the flow generated is promising.

8. Conclusions and Future Directions

We have investigated an LB scheme with the simplest approximation to the

collision integral. We have explored the standard LBGK scheme for single
phase flows with application to free-surface water waves. For incompressible
Navier-Stokes equations, the scheme is second-order accurate in space and
first-order accurate in time.
The 1-D and 2-D LBGK numerical solutions presented approximate the
NLSW in rotational flows and can be shown to satisfy the depth-integrated
Navier-Stokes equations. Fairly good agreement with Riemann solutions
and model-scale experiments have been achieved for weak bore propagation
in tanks. Dam-break simulations were also undertaken. The free-surface
predictions compared fairly well with other investigators. The tsunami
runup problem (benchmark 1) was also simulated, applying two LB schemes
(1) the standard LBGK and (2) a second-order Finite Difference (FD) LB
solvers. We found fairly good agreement with the NLSW solutions of Peder-
sen (Chapter 17 in this volume) and the semi-analytical solution of Carrier
et al. (2003) when we used a shoreline algorithm of Lynett et al. (2002).
The FD LB solver is, however, recommended as it most accurate and effi-
cient for the free-surface solutions presented in the benchmark test 1 study
herein. The other test cases presented in this chapter are based on the
standard LBGK scheme only. In future studies,we recommend a compar-
ison with the FD LB scheme and expect major improvements in terms of
accuracy, stability and efficiency, similar to what we observed for the wave
runup problem. The CPU time for all tests was competitive, especially the
FD LB scheme for tsunami predictions. There are, however, many issues
which can be improved. First, we recommend that a multi-time-relaxation
scheme be applied instead of the single-time relaxation scheme as it would
provide the stability, accuracy and improve the description of the details of
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 211

the physics of the flow processes. Second, instead of the regular standard
lattice scheme one could test the LB performance using adaptive irregular
lattices.
Based on our current knowledge, we highly recommend that that a
second-order FD LB scheme form the basis for free-surface water-wave de-
velopments. We are currently developing an LB model applicable at any
water depth. This involves an entirely new LB model formulation which
satisfies the Navier-Stokes equations. Our focus is on breaking wave pre-
dictions and we are investigating free-surface algorithms which will allow
for steep and overturning waves. Closely related, we are also undertaking
detailed bubble break-up studies based on soft-sphere collision models. An
LB sediment-transport model is also part of the LB development. At some
point in time, we should be able to simulate the wave transformation from
deep to shallow water to runup.
On a final note, it is our hope that this book chapter represents a useful
starting point for those interested in learning about and/or applying LB
models. In our opinion, LB simulations are a promising route in calculation
methods and may be a helpful method in discovering new physics.

Acknowledgments

I wish to thank Professor P. L.-F. Liu, Cornell University (USA), for giving
me the opportunity to write about the first time experiences with the Lat-
tice Boltzmann generated water waves. Thanks are also due to Professor
G. Pedersen, University of Oslo (Norway), who provided the NLSW and
Boussinesq data allowing us to assess the Lattice Boltzmann solutions in
detail. I would also like to mention that I met Professor H. Yeh, Oregon
State University, in Spring 2004, who actually introduced me to the wave
runup problem. In many ways, this was a turning point for the Lattice
Boltzmann model development and other new research directions. I am
also thankful to Professor C. Dalton, University of Houston (USA), for his
many valuable comments and corrections to the manuscript.
Further thanks to the many students, including under-represented mi-
nority and early career researchers, who got partly involved in this project
of LB model learning. Especially, I would like to thank visiting scholar, Dr.
J. Zhang, University of Hawai’i Manoā (UHM), who assisted producing the
results for the benchmark test case 1 at the final phase of the chapter devel-
opment. Finally, many thanks to the organizers of the faculty ambassadors
program of UHM, for giving me the opportunity to share with middle and
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

212 J. B. Frandsen

high school student the importance of practical experiences, research and

education in ocean and coastal engineering including the present tsunami
application and studies.
The first thoughts of exploring an LB approach to ocean waves were
compiled in proposals to the US National Science Foundation (NSF) in
Summer 2003, and sponsored Fall 2004 by NSF, the Division of Civil,
Mechanical and Manufacturing Innovation (grant no. CMMI-CAREER-
0625423) and the Division of Chemical, Bioengineering, Environmental,
and Transport Systems (grant no. CBET-0622388).
Finally, but not least, the opportunity to participate in the Catalina
Island NSF workshop (2004) was in so many ways extremely valuable to
me. It was a highly efficient, educational and well planned meeting, thanks
to the organizers, Professors P. L.-F. Liu, H. Yeh and C. Synolakis.
I am very grateful that all these happenings came my way.

References

Aidun, C. K., Lu, Y. and Ding, E.-J. (1998) Direct analysis of particulate sus-
pensions with inertia using discrete Boltzmann equation, Journal of Fluid
Mechanics 373, pp. 287–311.
Benzi, R., Succi, S. and Vergassola, M. (1992) The Lattice Boltzmann equation:
theory and applications, Physics Reports (review section of Physics Letters)
222(3), pp. 145–197.
Bhatnagar, P., Gross, E. and Krook, M. A. (1954) A model for collision processes
in gases i: small amplitude process in charged and neutral one component
systems, Physics Review A 94, pp. 511–526.
Boltzmann, L. (1964) Lectures on Gas Theory (Dover Publications).
Buick, J. M. and Greated, C. A. (2003) Lattice Boltzmann modelling of interfacial
gravity waves, Physics of Fluids 10(6), pp. 1490–1511.
Burgers, J. M. (1969) Flow Equations for Composite Gases (Academic Press).
Cao, N., Chen, S., Jin, S. and Martinez, D. (1997) Physical symmetry and lattice
symmetry in the Lattice Boltzmann method, Physical Review E 55(1),
pp. R21–R24.
Carrier, G. F., Wu, T. T. and Yeh, H. (2003) Tsunami run-up and draw-down on
a plane beach, Journal of Fluid Mechanics 475, pp. 79–99.
Cercignani, C. (1988) The Boltzmann Equation and its Applications (Springer-
Verlag).
Chapman, S. (1916) On the law of distribution of molecular velocities, and
on the theory of viscosity and thermal conduction, in a non-uniform
simple monatomic gas, Philosophical Transactions Royal Society A 216,
pp. 279–348.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 213

Chapman, S. (1918) On the kinetic theory of gas. part ii: a composite monatomic
gas: diffusion, viscosity, and thermal conduction, Philosophical Transac-
tions Royal Society A 217, pp. 115–197.
Chapman, S. and Cowling, T. G. (1958) The Mathematical Theory of Non-
Uniform Gases (Cambridge University Press).
Chen, H., Chen, S. and Matthaeus, W. H. (1992) Recovery of the Navier-Stokes
equations using a lattice-gas Boltzmann method, Physical Review A 45(8),
pp. R5339–R5342.
Chen, H., I. Saroselsky, I., Orszag, S. and Succi, S. (2004) Expanded analogy
between boltzmann kinetic theory of fluids and turbulence, Journal of Fluid
Mechanics 519, pp. 301–314.
Chen, S. and Doolen, G. D. (1998) Lattice Boltzmann method for fluid flows,
Annual Review of Fluid Mechanics 30(4), pp. 329–364.
Crouse, B., Rank, E., Krafczyk, M. and Tölke, J. (2003) A LB-based approach
for adaptive flow simulations, International Journal of Modern Physics B
17(1 & 2), pp. 109–112.
Crouse, M., B. Krafczyk, Kühner, S., Rank, E. and Van Treeck, C. (2002) In-
door air flow analysis based on Lattice Boltzmann methods, Energy and
Buildings 34, pp. 941–949.
Dalrymple, R. A. and Rogers, B. (2006) A note on wave celerities on compressible
fluid, in Proceedings, 30th International Conference on Coastal Engineering
(San Diego, USA).
Dellar, P. J. (2001) Bulk and shear viscosities in Lattice Boltzmann equations,
Physics Review E 64, 031203.
Dellar, P. J. (2002) Non-hydrodynamic modes and a priori construction of shallow
water lattice boltzmann equations, Physics Review E 65, 036309.
Deng, J. Q., Ghidaoui, M. S., Gray, W. G. and Xu, K. (2001) A Boltzmann-based
mesoscopic model for contaminant transport in flow systems, Advances in
Water Resources 24, pp. 531–550.
Denniston, C., Marenduzzo, D., Orlandini, E. and Yeomans, J. M. (2004) Lattice-
Boltzmann algorithm for three-dimensional liquid crystal hydrodynamics,
Philosophical Transactions, Royal Society of London Series A Mathematical
Physical and Engineering Sciences 362(1821), pp. 1745–1754.
Derksen, J. J. (2005) Simulations of confined turbulent vortex flow, Computers
and Fluids 34(3), pp. 301–318.
D’Humières, D., Ginzburg, I., Krafczyk, M., Lallemand, P. and Luo, L.-S. (2002)
Multiple-relaxation-time for the Lattice Boltzmann models in three di-
mensions, Philosophical Transactions, Royal Society London A. 360(1),
pp. 437–451.
Ding, E.-J. and Aidun, C. K. (2003) Extension of the Lattice Boltzmann method
for direct simulation of suspended particles near contact, Journal of Statis-
tical Physics 112(314), pp. 685–708.
Ding, E.-J. and Aidun, C. K. (2006) Cluster size distribution and scaling for
spherical particles and red blood cells in pressure-driven flows at small
reynolds number, Physical Review Letters 96, 204502.
Enskog, D. (1921) The numerical calculation of phenomena in fairly dense gases,
Arkiv Mat. Astr. Fys. 16(1).
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

214 J. B. Frandsen

Fennema, R. J. and Chaudhry, M. H. (1990) Explicit methods for 2-D transient

free-surface flows, Journal of Hydraulic Engineering 116(8), pp. 1013–1034.
Filippova, O. and Hänel, D. (1997) Lattice Bolzmann simulation of gas-particle
flow in filters, Computers and Fluids 26(7), pp. 697–712.
Frandsen, J. B. (2006a) Free surface water wave 1-D LBGK predictions, Interna-
tional Journal of Computational Fluid Dynamics 20(6), pp. 427–437.
Frandsen, J. B. (2006b) A Lattice Boltzmann bluff body model for VIV sup-
pression, in Proceedings, 25th International Offshore Mechanics & Arctic
Engrg., Paper no. 92271 (Hamburg, Germany).
Frandsen, J. B. (2008) “A 1-D Lattice Boltzmann model applied to tsunami runup
on a plane beach,” in Advanced Numerical Models for Simulating Tsunami
Waves and Runup, Advances in Coastal and Ocean Engineering, Vol. 10,
eds. Liu, P. L.-F., Yeh, H. and Synolakis, C. (World Scientific Publishing
Co.).
Geller, S., Krafczyk, M., Tölke, J., Turek, S. and Hron, J. (2006) Benchmark
computations based on Lattice-Boltzmann, finite element and finite volume
methods for laminar flows, Computers and Fluids 35(8), pp. 888–897.
Ghidaoui, M. S., Deng, J. Q., Gray, W. G. and Xu, K. (2001) A Boltzmann based
model for open channel flows, International Journal for Numerical Methods
in Fluids 35, pp. 449–494.
Ghidaoui, M. S., Kolyshkin, A. A., Liang, J. H., , Chan, F., Li, Q. and Xu,
K. (2006) Linear and nonlinear analysis of shallow wakes, Journal of Fluid
Mechanics 548, pp. 309–340.
Ginzburg, I. and D’Humières, D. (2003) Multireflection boundary conditions for
the lattice Bolzmann models, Physical Review E 066614 68(1), pp. 1–30.
Ginzburg, I. and Steiner, K. (2003) Lattice Boltzmann model for free-surface flow
and its application to filling process in casting, Journal of Computational
Physics 185, pp. 61–99.
Gombosi, T. I. (1994) Gaskinetic Theory (Cambridge University Press).
Gou, Z., Liu, H., Luo, L.-S. and Xu, K. (in press) A comparative study of the
LBE and GKS methods for 2D near incompressible flows, Journal of Com-
putational Physics.
Grad, H. (1949) On kinetic theory of rarefied gases, Communications on Pure
and Applied Mathematics 2, pp. 331–407.
Harris, S. (2004) An Introduction to the Theory of the Boltzmann Equation (Dover
Publications).
He, X., Chen, S. and Zhang, R. (1999) A Lattice Boltzmann scheme for in-
compressible multiphase flow and its application in simulation of Rayleigh-
Taylor instability, Journal of Computational Physics 152, pp. 642–663.
He, X. and Doolen, G. (1997) Lattice Boltzmann method on curvilinear coordi-
nates system, Journal of Computational Physics 134, pp. 306–315.
He, X., Luo, L.-S. and Dembo, M. (1996) Some progress in Lattice Boltzmann
method. Part I. nonuniform mesh grids, Journal of Computational Physics
129, pp. 357–363.
Hibberd, S. and Peregrine, D. H. (1979) Surf and run-up on a beach: a uniform
bore, Journal of Fluid Mechanics 95, pp. 323–345.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 215

Hou, S., Sterling, J., Chen, S. and Doolen, G. D. (1996) A Lattice Boltzmann
subgrid model for high Reynolds number flows, Fields Institute Communi-
cations 6, pp. 151–165.
Inamuro, T., Ogata, S., Tajima, S. and Konishi, N. (2004) A Lattice Boltzmann
method for incomressible two-phase with large density differences, Journal
of Computational Physics 198, pp. 628–644.
Junk, M. (2001) A finite difference interpretation of the Lattice Boltz-
mann method, Numerical Methods Partial Differential Equations 17(4),
pp. 383–402.
Junk, M., Klar, A. and Luo, L.-S. (2005) Asymptotic analysis of the Lattice
Boltzmann equation, Journal of Computational Physics 210, pp. 676–704.
Kalikmanov, V. I. (2001) Statistical Physics of Fluids. Basic Concepts and
Applications (Springer).
Kandhai, D., Soll, W., Chen, S., Hoekstra, A. and Sloot, P. (2000) Finite-
difference Lattice-BGK methods on nested grids, Computer Physics Com-
munications 129, pp. 100–109.
Karniadakis, G. E. and Beskok, A. (2002) Micro Flows. Fundamentals and Sim-
ulations (Springer-Verlag).
Keller, H. B., Levine, D. A. and Whitham, G. B. (1960) Motion of a bore over a
sloping beach, Journal of Fluid Mechanics 7, pp. 302–316.
Koelman, J. M. V. A. (1991) A simple Lattice Boltzmann scheme for Navier-
Stokes fluid flow, Europhysics Letters 15(6), pp. 603–607.
Körner, C., Pohl, T., Rüde, U., Thürey, N. and Zeiser, T. (2006). Parallel Lat-
tice Boltzmann Methods for CFD Applications (Numerical Solution of Par-
tial Differential Equations on Parallel Computers. Eds. A. M. Bruaset, A.
Tveito. Lecture Notes in Computational Science and Engineering, Vol. 51,
Springer).
Körner, C., Thies, H., M., T. Thürey, N. and Rüde, U. (2005) Lattice Boltzmann
model for free surface flow for modeling foaming, Journal of Statistical
Physics 121(1-2), pp. 179–196.
Krafczyk, M., Tölke, J. and Luo, L.-S. (2003) Large-eddy simulations with a mul-
tiple relaxation time LBE model, International Journal of Modern Physics
B 17(1), pp. 33–39.
Krafczyk, M. and Tölke, J. (2004) Lattice-Boltzmann methods — basics and
recent progress, in NAFEMS CFD Workshop on Simulation of Complex
Flows (Germany).
Lallemand, P. and Luo, L.-S. (2000) Theory of the Lattice Boltzmann method:
dispersion, dissipation, isotropy, galilean invariance, and stability, Physical
Review E 61, pp. 6546–6562.
Laney, C. B. (1998) Computational Gasdynamics (Cambridge University Press).
Lee, K., Yu, D. and Girimaji, S. S. (2006) Lattice Boltzmann DNS of decay-
ing compressible isotropic turbulence with temperature fluctuations, Inter-
national Journal of Computational Fluid Dynamics 20, pp. 401–413, doi:
doi:10.1080/10618560601001122, URL http://www.ingentaconnect.com/
content/tandf/gcfd/2006/0000002%0/00000006/art00006.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

216 J. B. Frandsen

Lee, T. and Lin, C.-L. (2005) A stable discretization of the Lattice Boltzmann
equation for simulation of incomressible two-phase flows at high density
ratio, Journal of Computational Physics 206, pp. 16–47.
Lesieur, M., Métais, O. and Comte, P. (2005) Large-eddy Simulations of Turbu-
lence (Cambridge University Press).
LeVeque, R. J. (2002) Finite Volume Methods for Hyperbolic Problems (Wiley).
LeVeque, R. J. and George, D. L. (2008) “High-resolution finite volume meth-
ods for the shallow water equations with bathymetry and dry states”, in
Advanced Numerical Models for Simulating Tsunami Waves and Runup,
Advances in Coastal and Ocean Engineering, Vol. 10, eds. Liu, P. L.-F.,
Yeh, H. and Synolakis, C. (World Scientific Publishing Co.).
Li, Y., Shock, R., Zhang, R. and Chen, H. (2004) Numerical study of flow past
an impulsively started cylinder by the Lattice-Boltzmann method, Journal
of Fluid Mechanics 519, pp. 273–300.
Liang, J. H., Ghidaoui, J. Q., M. S. Deng and Gray, W. G. (2007) A Boltzmann-
based finite volume algorithm for surface water flows on cells of arbitrary
shapes, Journal of Hydraulic Research 45(2).
Lockard, D. P., Luo, L.-S., Milder, S. D. and Singer, B. A. (2002) Evaluation of
PowerFLOW for aerodynamic applications, Journal of Statistical Physics
107(1/2), pp. 423–478.
Luo, L.-S. (2000) The lattice-gas and Lattice Boltzmann methods: Past, present,
and future, in Proceedings of the International Conference on Applied Com-
putational Fluid Dynamics (Beijing, China), pp. 52–83.
Luo, L.-S., Qi, D. and Wang, L. P. (2002) Applications of the Lattice Boltzmann
method to complex and turbulent flows (Eds. M. Breuer, F. Durst, and C.
Zenger. Lecture Notes in Computational Science and Engineering, Vol. 21,
pp. 123-130.).
Lynett, P. J., Wu, T.-R. and Liu, P. L.-F. (2002) Modeling wave runup with
depth-integrated equations, Coastal Engineering 46, pp. 89–107.
McCracken, M. E. (2004) Development and Evaluation of the Lattice Boltzmann
Models for Investigations of Liquid Break-up, PhD Thesis, Purdue Univer-
sity, USA.
McNamara, G. R. and Zanetti, G. (1988) Use of the Boltzmann equa-
tion to simulate lattice gas automata, Physics Review Letter 61(20),
pp. 2332–2335.
Mei, C. C. (1992) The Applied Dynamics of Ocean Surface Waves (World
Scientific Publishing Co.).
Mei, R., Luo, L.-S. and Shyy, W. (1999) An accurate curved boundary treatment
in the Lattice Bolzmann method, Journal of Computational Physics 155,
pp. 307–330.
Mei, R., Shyy, W., Yu, D. and Luo, L.-S. (2000) Lattice Bolzmann method for
3-D flows with curved boundaries, Journal of Computational Physics 161,
pp. 680–699.
Noble, R., Chen, S., Georgiadis, J. and Buckius, R. (1995) A consistent hydrody-
namic boundary condition for the Lattice Boltzmann method, Physics of
Fluids 7(1), pp. 203–209.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 217

Nourgaliev, R. R., Dinh, D. N., Theofanous, T. G. and Joseph, D. (2003)

The Lattice Boltzmann equation method: theoretical interpretation, nu-
merics and implications, International Journal of Multiphase Flows 29,
pp. 117–169.
Pan, C., Luo, L.-S. and Miller, C. (2006) An evaluation of Lattice Boltzmann
schemes for porous medium flow simulation, Computer and Fluids 35(8/9),
pp. 898–909.
Pedersen, G. (2008) “A Lagrangian model applied to runup problems”, in Ad-
vanced Numerical Models for Simulating Tsunami Waves and Runup, Ad-
vances in Coastal and Ocean Engineering, Vol. 10, eds. Liu, P. L.-F., Yeh,
H. and Synolakis, C. (World Scientific Publishing Co.).
Peng, G., Xi, H. and Duncan, C. (1999) Lattice Boltzmann scheme method
on unstructured grids: further developments, Physical Review E 59(4),
pp. 4675–4682.
Penrose, O. (1970) Foundations of Statistical Mechanics. A Deductive Treatment
(Dover Publications).
Poularikas, A. D. (1996) The Transforms and Applications Handbook (CRC
Press).
Premnath, K. N. (2004) Lattice Boltzmann Models for Simulations of Drop-Drop
Collisions, PhD Thesis, Purdue University, USA.
Premnath, K. N. and Abraham, J. (2005) Simulations of binary drop collisions
with a multi-relaxation-time Lattice Boltzmann model, Physics of Fluids
17, p. 122105.
Premnath, K. N. and Abraham, J. (2007) Three-dimensional multi-relaxation-
time (mrt) Lattice Boltzmann models for multiphase flow, Journal of Com-
putational Physics 224, pp. 539–559.
Qian, Y. H., D’Humières, D. and Lallemand, P. (1992) Lattice BGK models for
Navier-Stokes equation, Europhysics letters 17(6), pp. 479–484.
Que, Y.-T. and Xu, K. (2006) The numerical study of roll-waves in inclined
open channels and solitary wave run-up, International Journal for numer-
ical methods in Fluids 50, pp. 1003–1027.
Riabouchinsky, D. (1932) Sur l’analogie hydraulique des movements d’un fluide
compressible. (Institute de France, Academie des Sciences, Comptes Ren-
dus, 195(998–999)).
Rivet, J.-P. and Boon, J. P. (2001) Lattice Gas Hydrodynamics (Cambridge Uni-
versity Press).
Rothman, D. H. and Zaleski, S. (1997) Lattice-Gas Cellular Automata. Simple
Models of Complex Hydrodynamics (Cambridge University Press).
Salmon, R. (1999a) The Lattice Boltzmann method as a basis for ocean circulation
modeling, Journal of Marine Research 57, pp. 503–535.
Salmon, R. (1999b) Lattice Boltzmann solutions of the three-dimensional plane-
tary geostrophic equations, Journal of Marine Research 57, pp. 847–884.
Seta, T. and Takahashi, R. (2002) Numerical stability analysis of FDLBM, Jour-
nal of Statistical Physics 107(1/2), pp. 557–572.
Shen, M. C. and Meyer, R. E. (1963) Climb of a bore on a beach. part 3: Run-up,
Journal of Fluid Mechanics 16, pp. 113–125.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

218 J. B. Frandsen

Sofonea, V. and Sekerka, R. F. (2003) Viscosity of finite difference lattice Boltz-

mann models, Journal of Computational Physics 184, pp. 422–434.
Sterling, J. D. and Chen, S. (1996) Stability analysis of Lattice Boltzmann meth-
ods, Journal of Computational Fluid Dynamics 123, pp. 196–206.
Stoker, J. J. (1957) Water Waves. The Mathematical Theory with Applications
(John Wiley & Sons).
Struchtrup, H. (2005) Macroscopic Transport Equations for Rarefied Gas Flows.
Approximation Methods in Kinetic Theory (Springer-Verlag).
Succi, S. (2001) The Lattice Boltzmann Equation for Fluid Dynamics and Beyond
(Oxford University Press).
Sukop, M. C. and Thorne, D. T. (2006) Lattice Boltzmann Modeling (Springer).
Synolakis, C. E. (1987) The run-up of solitary waves, Journal of Fluid Mechanics
185, pp. 523–545.
Thömmes, G., Seaïd, M. and Banda, M. K. (2007) Lattice Boltzmann methods
for shallow water flow applications, International Journal for Numerical
Methods in Fluids 55(7), pp. 673–692.
Thürey, N. (2006) Physically Based Animation of Free Surface Flows with the
Lattice Boltzmann Method, PhD Thesis, Der Technischen Fakultät der Uni-
versität Erlangen-Nürnberg, Germany.
Thürey, N. and Rüde, U. (2004) Free surface lattice-boltzmann fluid simulations
with and without level sets, in Proceedings, Vision Modeling and Visualiza-
tion (Stanford, USA).
Thürey, N., Rüde, U., Öchsner, M. and Körner, C. (2006a) Optimization and
stabilization of LBM free surface flow simulations using adaptive parame-
terization, Computers and Fluids 35, pp. 934–939.
Thürey, N., Rüde, U. and Stamminger, M. (2006b) Animation of open wa-
ter phenomena with coupled shallow water and free surface simulations,
in Proceedings, Eurographics/ACM SIGGRAPH Symposium on Computer
Animation.
Tölke, J., Freudiger, S. and Krafczyk, M. (2006) An adaptive scheme using hierar-
chical grids for Lattice Boltzmann multi-phase flow simulations, Computers
and Fluids 35(8), pp. 820–830.
Tölke, J., Krafczyk, M., Schulz, M. and Rank, E. (1998) Implicit discretiza-
tion and nonuniform mesh refinement approaches for FD discretizations
of LBGK models, International Journal of Modern Physics C 9(8),
pp. 1143–1157.
Toro, E. F. (1999) Riemann Solvers and Numerical Methods for Fluid Dynamics.
A Practical Introduction (Springer, p.168–171).
Toro, E. F. (2001) Shock-Capturing Methods for Free-surface Shallow Flows
(Wiley).
Ubertini, S., Bella, G. and Succi, S. (2003) Lattice Boltzmann scheme method on
unstructured grids: Further developments, Physical Review E 68, 016701,
pp. 1–10.
Van der Sman, R. G. M. (2006) Finite Boltzmann schemes, Computer and Fluids
35, pp. 849–854.
August 26, 2008 17:6 World Scientific Review Volume - 9in x 6in 05˙main

Free-Surface Lattice Boltzmann Modeling in Single Phase Flows 219

Van der Sman, R. G. M. and Ernst, M. H. (2000) Convection-diffusion Lattice

Boltzmann scheme for irregular lattices, Journal of Computational Physics
160, pp. 766–782.
Vincenti, W. G. and Kruger, Jr., C. H. (1965) Introduction to Physical Gas Dy-
namics (Krieger Publishing Company).
Wu, G. X., Eatock Taylor, R. and Greaves, D. M. (2001) The effect of viscosity
on the transient free-surface waves in a two-dimensional tank, Journal of
Engineering Mathematics 40, pp. 77–90.
Wu, G. X., Ma, Q. A. and Eatock Taylor, R. (1998) Numerical simulation of
sloshing waves in a 3D tank based on a finite element method, Applied
Ocean Research 20, pp. 337–355.
Xu, K. (2002) A well-balanced gas-kinetic scheme for the shallow-water equations
with source terms, Journal of Computational Physics 178, pp. 533–562.
Yeh, H., Ghazali, A. and Marton, I. (1989) Experimental study of bore run-up,
Journal of Fluid Mechanics 206, pp. 563–578.
Yu, H. and Girimaji, S. S. (2006) Multi-block Lattice Boltzmann method: Ex-
tension to 3D and validation in turbulence, Physica A 362, pp. 118–124.
Yu, H., Girimaji, S. S. and Luo, L.-S. (2005) DNS and LES of decaying isotropic
turbulence with and without frame rotation using the Lattice Boltzmann
method, Journal of Computational Dynamics 209, pp. 599–616.
Yu, H., Luo, L.-S. and Girimaji, S. S. (2006) LES of turbulent square jet
flow using and MRT Lattice Boltzmann model, Computers and Fluids 35,
pp. 957–965.
Zhong, L., Feng, S. and Gao, S. (2005) Wind-driven ocean circulation in shallow
water Lattice Boltzmann model, Advances in Atmospheric Sciences 22(3),
pp. 349–358.
Zhou, J. G. (2002) A Lattice Boltzmann model for the shallow water equa-
tions with turbulence modeling, International Journal of Modern Physics
C 13(8), pp. 1135–1150.
Zhou, J. G. (2004) Lattice Boltzmann Methods for Shallow Water Flows
(Springer).
This page intentionally left blank
April 3, 2008 3:12 WSPC/Trim Size: 9in x 6in for Review Volume divider

Part 2
Extended Abstracts
This page intentionally left blank
 CHAPTER 6

BENCHMARK PROBLEMS

Philip L.-F. Liu1, Harry Yeh2, and Costas E. Synolakis3

1
School of Civil and Environmental Engineering, Cornell University
Ithaca, NY 14853 USA
E-mail: [email protected]
2
Department of Civil Engineering, Oregon State University
3
Department of Civil Engineering, University of Southern California

Four benchmark problems were selected before the workshop so that

numerical models can be compared, evaluated and discussed among the
participants during the workshop. All of the benchmark-problem
descriptions and necessary data were provided to the participants six
months prior to the workshop. Based on the benchmark problems, each
participant was asked to submit his/her written discussion prior to the
workshop. The actual laboratory or physical measurements were only
presented during the workshop in the same format, allowing the
comparisons of predictions with measurements. All of the benchmark
data are archived for download in the NEES Central Data Repository at
https://central.nees.org/?projid=123.

1. Tsunami Runup onto a Plane Beach

This is a simple setup for tsunami runup modeling exercise: a uniformly
sloping beach and no variation in the lateral direction, viz. a 2-D problem
in the vertical plane. The initial-value-problem (IVP) technique
introduced by Carrier, Wu and Yeh (J. Fluid Mech., 475, 79–99, 2003) is
used to produce the benchmark data. For the benchmark problem #1, the
beach slope is fixed at 1/10 and the initial free surface elevation is given
(see below). The workshop participants are asked to compute and present
the snapshots of the free surface and velocity profiles at t = 160 s, 175 s,
and 220 s and describe the algorithm used to calculate the motion of the

223
224 P. L.-F. Liu, H. Yeh and C. E. Synolakis

shoreline (the air-water-beach interface). Moreover, workshop participants

are also required to present the temporal variations of the shoreline
location and shoreline velocity from t = 100 s to 280 s
The problem statement and input data file can be found at
http://www.cee.cornell.edu/longwave/data/benchmark_1.txt.

2. Tsunami Runup onto a Complex Three-Dimensional Beach

The 1993 Okushiri tsunami caused many unexpected phenomena. One of
them was the extreme runup height of 32 m that was measured near the
village of Monai in Okushiri Island. This tsunami runup mark was
discovered at the tip of a very narrow gulley within a small cove. This
benchmark problem is an 1/400 scale laboratory experiment of the Monai
runup, using a large-scale tank (205 m long, 6 m deep, 3.4 m wide) at
Central Research Institute for Electric Power Industry (CRIEPI) in
Abiko, Japan. The figures shown below are the bathymetry (Fig. 1) and
coastal topography (Fig. 2) used in the laboratory experiment. The
incident wave from offshore, at the water depth d = 13.5 cm will be
given. Note that there are reflective vertical sidewalls at y = 0, and 3.5 m
as indicated in bold lines in Fig. 2(a). The primary theme of this
benchmark problem is the temporal and spatial variations of the shoreline
location, as well as the temporal variations of the water-surface
variations at one or two specified nearshore locations. Although it was
not required, animated presentation for the numerical model was
encouraged.

Fig. 1. Offshore profile.

 Benchmark Problems 225

cm
(a) Whole model area

m
(b) Detailed topography near the maximum runup

Fig. 2. Bathymetry and topography reproduced in the tank.

226 P. L.-F. Liu, H. Yeh and C. E. Synolakis

The problem statement, data file, input file, bathymetry file and the
experimental data file can be found in the following websites,
respectively:

http://www.cee.cornell.edu/longwave/data/problem02.doc
http://www.cee.cornell.edu/longwave/data/Benchmark_2.txt
http://www.cee.cornell.edu/longwave/data/Benchmark_2_input.txt
http://www.cee.cornell.edu/longwave/data/Benchmark_2_Bathymetry.txt
http:// www.cee.cornell.edu/longwave/data/benchmark2/output_ch5-7
9.xls

3. Tsunami Generation and Runup Due to a Two-Dimensional

Landslide
For the second benchmark problem, the workshop participants are asked
to predict the free surface elevation and runup associated with a
translating Gaussian shaped mass, initially at the shoreline. In
dimensional form, the seafloor can be described by:

h(x,t) = H(x) + ho (x,t) ,

where

H(x) = x tan β ,

  2  
2
x µ g
ho = δ exp  −  2 − µt   ,
  δ tan β δ  
 

δ = maximum vertical slide thickness, µ = thickness/slide length, and β is

the beach slope. Once in motion, the mass moves at constant
acceleration. The initial position of the block is shown in the figure
below which shows the start of the motion.
 Benchmark Problems 227

The workshop participants are asked to provide snapshots of the free

surface at selected times. The following two setups will be benchmarked:

CASE A: tan β/µ = 10 -> β = 5.7°, δ = 1 m, µ = 0.01

CASE B: tan β/µ = 1 -> β = 5.7°, δ = 1 m, µ = 0.1

Four comparisons should be made for each setup. Spatial snapshots of

the free surface are given at four different non-dimensional times:
228 P. L.-F. Liu, H. Yeh and C. E. Synolakis

g
CASE A: µ t = 0.1, 0.5, 1.0, 1.5
δ
g
CASE B: µ t = 0.5, 1.0, 2.5, 4.5
δ

The data files for this benchmark problem are named as readme1.txt,
bench3A_t05.txt, bench3B_t10.txt, bench3A_t15.txt, bench3B_t05.txt,
bench3B_t10.txt, bench3B_t25.txt, and bench3B_t45.txt; they can be
found in the following websites:

http://www.cee.cornell.edu/longwave/data/benchmark3/readme1.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3A_t05.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3A_t10.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3A_t15.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3B_t05.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3B_t10.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3B_t25.txt
http://www.cee.cornell.edu/longwave/data/benchmark3/bench3B_t45.txt

Included in each data file are four columns of data: x (m); H (m); ho (m);
ζ (m). The free surface elevations, ζ , are those predicted by an analytical
solution. Agreement with CASE B using any model other than the linear
shallow water wave equations may be poor. For more information, please
refer to Liu, Lynett and Synolakis (J. Fluid Mech., 478, 101–109, 2003).

4. Tsunami Generation and Runup Due to a Three-Dimensional

Landslide
This benchmark problem requires the modeling of a sliding mass down
a 1:2 plane beach slope and compares the predictions with laboratory
data. Large-scale experiments have been conducted in a wave tank with
a length 104 m, width 3.7 m, depth 4.6 m and with a plane slope
(1:2) located at one end of the tank. A solid wedge was used to model
the landslide. The triangular face has the following dimensions: a
horizontal length of b = 91 cm, a vertical face a = 46 cm high and a
 Benchmark Problems 229

width of w = 61 cm. The wedge was instrumented with an accelerometer

to accurately define the acceleration-time history and a position indicator
to independently determine the velocity- and position-time histories. The
wedge traveled down the slope by gravity rolling on specially designed
wheels (with low friction bearings) riding on aluminum strips with
shallow grooves inset into the slope. A snapshot of the wedge motion is
shown in the figure below.

A sufficient number of wave gages were used to determine the seaward

propagating waves, the waves propagating to either side of the wedge,
and for the submerged case, the water surface-time history over the
wedge. In addition, the time history of the runup on the slope was
accurately measured. The workshop participants are asked to model the
flow with the wedge starting from two different initial elevations: one
submerged the other subaerial. The recorded block motions are
provided. The workshop participants are asked to produce time histories
230 P. L.-F. Liu, H. Yeh and C. E. Synolakis

of surface elevation at selected locations in the channel and the shoreline

motion. An animation of the results is encouraged.
The data files for this benchmark problem are named as problem
statement, Landslide setup, time history (run30), time history (run32);
they can be found at following websites:

http://www.cee.cornell.edu/longwave/data/Benchmark_4_readme.txt
http://www.cee.cornell.edu/longwave/data/Benchmark_4_landslide_setu
p.pdf
http://www.cee.cornell.edu/longwave/data/Benchmark_4_run30.txt
http://www.cee.cornell.edu/longwave/data/Benchmark_4_run32.txt

The wave gage data and runup data can be located in the following
websites:

http://www.cee.cornell.edu/longwave/data/benchmark4/gage_data_read
me.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run30_wave_gag
e_1.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run30_wave_gag
e_2.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run30_runup_ga
ge_2.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run30_runup_ga
ge_3.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run32_wave_gag
e_1.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run32_wave_gag
e_2.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run32_runup_ga
ge_2.txt
http://www.cee.cornell.edu/longwave/data/benchmark4/run32_runup_ga
ge_3.txt

For more information, please refer to Liu, Wu, Raichlen, Synolakis, and
Borrero (J. Fluid Mech., 536, 107–144, 2005).
August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

CHAPTER 7

TSUNAMI RUNUP ONTO A PLANE BEACH

Zygmunt Kowalik, Juan Horrillo and Edward Kornkven

Institute of Marine Science, University of Alaska Fairbanks
Fairbanks, AK 99775, USA
E-mail: [email protected]

The problem description and setup of Benchmark Problem 1 (BM1) can

be found in Chapter 6. A detailed analytical solution of the problem is
described in Carrier et al.1
To solve BM1 three approaches have been carried out:
(1) First-order approximation in time
(2) Second-order approximation in time (leap-frog)
(3) Full Navier-Stokes (FNS) approximation aided by the Volume of
Fluid (VOF) method to track the free surface.
Approaches (1) and (2) use one-dimensional nonlinear shallow water
(NLSW) wave theory. The finite difference solution of equation of mo-
tion and the continuity are solved on a staggered grid. Both methods
have second-order approximation in space. The FNS-VOF approach has
been used to visualize differences against the NLSW approaches and
analytical solution. The FNS equation includes the vertical component
of velocity/acceleration and some differences are expected. This method
solves two-dimensional transient incompressible fluid flow with free sur-
face. The finite difference solution of the incompressible FNS equations
are obtained on a rectilinear mesh.

1. Brief Description of the Methods and their Numerical

Schemes
1.1. First-Order Method
Equations of motion and continuity read
∂u ∂u ∂ζ 1 ∂ζ ∂uD
+u +g + ru|u| = 0 and + = 0, (1)
∂t ∂x ∂x ρD ∂t ∂x

231
August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

232 Z. Kowalik, J. Horrillo and E. Kornkven

where ρ is the water density, u is the vertically averaged particle velocity,

ζ is the sea level, D = (ζ + H) is the total depth, H is the mean water
depth, r is the friction coefficient and g is the gravity acceleration. The
numerical solution is usually searched by using the one-time-level numeri-
cal scheme, Kowalik and Murty.7 The numerical scheme is constructed as
follows (Kowalik and Murty8 ):
gT m um
p T
um+1
j = um
j −
m
(ζj − ζj−1 )− (um m
j − uj−1 )
h h
um
nT 2T
− (um m
j+1 − uj ) −
m m
m + D m ) ru |u |, (2)
h ρ(Dj−1 j

T
ζjm+1 = ζjm − (f luxm+1 m+1
j+1 − f luxj ),
h
(H +H )
where f luxj = um+1 p
m
ζj−1 + um+1
n ζjm + um+1
j
j
2
j−1
, up = 0.5(uj + |uj |)
and un = 0.5(uj − |uj |). T is the time step, h is the space step. Indices m
and j = 1, 2, 3, ...n−1 stand for the time stepping and horizontal coordinate
points, respectively. For the runup condition, the following steps are taken
when the dry point (jwet − 1) is located to the left of the wet point jwet ,
thus: if (ζ m (jwet ) > −H(jwet − 1)), then um m
jwet = ujwet +1 , see Kowalik and
8
Murty.

1.2. Leap Frog Method

Equation of motion and continuity are expressed in flux form as
∂M ∂M 2 ∂ζ gn2 ∂ζ ∂M
+ + gD + 7/3 M |M | = 0 and + = 0, (3)
∂t ∂x ∂x D ∂t ∂x
where M = uD is the water transport and n is the Manning’s roughness
coefficient. The numerical scheme is constructed on a space-time staggered
grid having second-order of approximation in space and time, see Imamura
et al.4 The two-time-level numerical scheme reads
T m−1/2 m−1/2
ζjm = ζjm−1 − (Mj − Mj−1 ),
h

m+1/2 1 m−1/2 gDr T m
Mj = (1 − µx )Mj − (ζj+1 − ζjm )
(1 + µx ) h
m−1/2 m−1/2 2 m−1/2 
(Mj+1 )2 (Mj ) (Mj−1 )2

T
− λ1 m−1/2
+ λ 2 m−1/2
+ λ 3 m−1/2
, (4)
h DM j+1 DM j DM j−1
August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

Tsunami Runup onto a Plane Beach 233

where µx is a friction term factor, DM is the total depth at M points,

and Dr is the total depth which depends of the sea level and depth of the
neighboring cells. Parameters λ1 , λ2 and λ3 are the up-down wind’s switches
2 m−1/2
used in the nonlinear term. µx and DM are defined as µx = gn7/3 T
|Mj |
2Dr
m−1/2
and DM j = 41 (Djm−1 + Dj+1
m−1
+ Djm + Dj+1
m
), respectively.

1.3. Full Navier-Stokes Approximation and VOF Method

Equation of continuity for incompressible fluid and the momentum equation
∂~u 1
∇ · ~u = 0 and + (~u · ∇)~u = − ∇p + ν∇2 ~u + ~g (5)
∂t ρ
are solved in the rectangular system of coordinates, where ~u(x, y, t) is the
instantaneous velocity vector, ρ is the fluid density, p is the scalar pressure,
ν is the kinematic viscosity, ~g is the acceleration due to gravity and t is
the time. Solution of the equations is searched using the two-step method
(Chorin2 and Harlow and Welch3 ). The time discretization of the momen-
tum equation is given by
~um+1 − ~um 1 m
= −(~u · ∇)~um − m ∇pm+1 + ν∇2 ~u + ~g,
T ρ
and it is broken up into two steps as
~ũ − ~um m
= −(~u · ∇)~um + ν∇2 ~u + ~g, (6)
T

~um+1 − ~ũ 1
= − m ∇pm+1 . (7)
T ρ
Equation (7) and the continuity equation, ∇ · ~um+1 = 0, can be combined
into a single equation (Poisson equation) for the solution of the pressure as
∇ · ~ũ
 
1 m+1
∇ · m ∇p = . (8)
ρ T
The fluid free surface is described by the discrete VOF method, see Nichols
and Hirt5 and Nichols et al.6

2. Discussion and Conclusions

Figures 1, 2 and 3 summarize results of BM1. The simple velocity extrapo-
lation used in the first-order method seems to follow the shoreline evolution
as prescribed by the NLSW analytical solution. Sea level and velocity profile
August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

234 Z. Kowalik, J. Horrillo and E. Kornkven

Analytical
0.03
First Order
Leap−Frog
FNS−VOF
0.02

0.01 t=3.0
815
(220
s)
η (dimensionless)

0 s)
(175 s)
512 (160
t= 2.4 411
t=2.2
−0.01
Slop
e=(
1:10

−0.02
)

−0.03

−0.04

0 0.05 0.1 0.15 0.2

x (dimensionless)

Fig. 1. Snapshots of water-surface.

Analytical
First Order
Leap−Frog
0.1 t=2
.24
11
(16
0s
ec
)
0.05
u (dimensionless)

MWL
0
t=3.081
5 (220
se c)
−0.05
c )
se
75
(1

−0.1 Slop
12

e=(
1:10
45

)
2.
t=

−0.15

−0.2
−0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
x (dimensionless)

Fig. 2. Snapshots of velocity profiles.

August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

Tsunami Runup onto a Plane Beach 235

3.5
Analytical
First Order
Leap−Frog
FNS−VOF
3.0815 220
3
t (dimensionless)

t(seconds)
2.5
2.4512 175

2.2411 160

1.5
−0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14
x (dimensionless)

Fig. 3. Temporal and spatial variation of the shoreline.

match very well with the analytical solution. The extrapolation of the veloc-
ity from the immediate wet cell to the dry cell facilitates runup, improving
the timing. The leap frog method does well in predicting the analytical
solution of the shoreline evolution, sea level and velocity profile as well.
However, due to the small difference in timing, some discrepancy in the
velocity profile can be seen, i.e. at time 220s. The FNS-VOF method gives
a frame of reference to validate the NLSW solutions. Some differences in
wave profile, shore line evolution and timing are quite plausible, since FNS
approximation allows vertical fluid velocity/acceleration while the NLSW
theory does not. From Figs. 1 and 3, it is clear that dispersion effects are
important. NLSW and analytical solutions underestimate the runup and
overestimate the rundown. Timing of maximum runup and rundown occur
slightly earlier in the NLSW solutions.

References
1. G. Carrier, T. T. Wu and H. Yeh, “Tsunami runup and draw-down on a plane
beach”, Journal of Fluid Mechanics, 475:79–99, 2003.
2. A. J. Chorin, Numerical solution of the Navier-Stokes equations, Math. Comp.,
22:745–762, 1968.
August 8, 2008 11:58 WSPC/Trim Size: 9in x 6in for Review Volume 07˙kowa

236 Z. Kowalik, J. Horrillo and E. Kornkven

3. F. H. Harlow and J. E. Welch, “Numerical calculation of time-dependent vis-

cous incompressible flow of fluid with a free surface”, The Physics of Fluids,
8:2182–2189, 1965.
4. F. Imamura, C. Goto, Y. Ogawa and N. Shuto, “Numerical method of tsunami
simulation with the leap-frog scheme”, IUGG/IOC Time Project Manuals,
1995.
5. B. D. Nichols and C. W. Hirt, “Method for calculating multi-dimensional,
transient free surface flow past bodies”, Proc. of the 1st Int. Conf. Num. Ship
Hydrodynamics, Gaithersburg, Maryland, 1975.
6. B. D. Nichols, C. W. Hirt and R. S. Hotchkiss, “SOLA-VOF: A solution al-
gorithm for transient fluid flow with multiple free Boundaries”, LA-8355, Los
Alamos National Laboratory, 1980.
7. Z. Kowalik, and T. S. Murty, Numerical Modeling of Ocean Dynamics, World
Scientific Publishing Co., 1993.
8. Z. Kowalik and T. S. Murty, “Numerical simulation of two-dimensional
tsunami runup”, Marine Geodesy, 16:87–100, 1993.
 CHAPTER 8

NONLINEAR EVOLUTION OF LONG WAVES OVER A

SLOPING BEACH

Utku Kânoğlu
Department of Engineering Sciences, Middle East Technical University
Ankara, 06531, Turkey
[email protected]

The initial value problem solution of the nonlinear shallow water-wave

equations developed by Kânoğlu (2004) is applied to the benchmark
problem — tsunami runup onto a plane beach — and results are
compared. Comparisons with the benchmark solution produce identical
results.

The solution method presented by Kânoğlu (2004) for the initial value
problem solution of the nonlinear shallow water-wave equations over a
sloping beach is applied to the benchmark problem. Here only the
description of the method and how it is applied to the benchmark
problem will be explained. Details of the method summarized here can
be found in Kânoğlu (2004).
The dimensionless nonlinear shallow water-wave equations that
describe a propagation problem over the undisturbed water of variable
depth h( x) = x are
[u (η + h)]x + ηt = 0, ut + u u x + η x = 0.

Here u(x, t), η(x, t), and β are the horizontal depth-averaged velocity, the
free-surface elevation, and the beach angle from the horizontal
respectively. The initial shoreline is chosen at x = 0 with x increasing in
the seaward-direction. The dimensionless variables are defined using
~ ~ ~
l , ( l tan β ), and l /( g~ tan β ) as the characteristic length, height, and time
scales; g~ being the gravitational acceleration. The major analytical

237
238 U. Kânoğlu

advance for the solution of the nonlinear shallow water-wave equations

on a uniformly sloping beach was presented by Carrier and Greenspan
(1958). They outlined a hodograph transformation introducing a potential
ϕ (σ , λ ) :

ϕ ϕ u2 u 2 ϕλ σ 2 λ
u= σ , η= λ − , x= − + , t =u− ,
σ 4 2 2 4 16 2

and reduced the nonlinear shallow water-wave equations into the second-
order linear equation: σϕ λλ − (σϕσ )σ = 0. The instantaneous shoreline is
defined at σ = 0 in the transform space with this transformation. Given
initial condition in terms of an initial wave profile with zero initial
velocity, Carrier and Greenspan (1958) provided the following solution:
∞ ∞
ϕ (σ , λ ) = −
∫ ∫
0 0
(1/ ω ) ζ 2 J 0 (ωσ ) sin(ωλ ) J1 (ωζ ) φ (ζ ) dω dζ ,

where the initial condition implies that φ (σ ) = uλ (σ , 0) = 4ησ (σ , 0) / σ .

The major difficulty with this transformation is to drive an equivalent
initial condition over the transform (σ, λ)-space for a given initial wave
profile in the physical (x, t)-space. Therefore, Carrier and Greenspan
(1958) presented solutions for two very specific initial profiles.
Unresolved difficulty to drive an equivalent initial condition over
(σ, λ)-space for a given initial wave profile in (x, t)-space is overcome
by simply with the linearization of the transformation for the spatial
variable as described by Kânoğlu (2004), i.e. x ≈ σ 2 / 16. This linearized
transformation can be used to represent any initial waveform over the
transform space. Then the nonlinear evolution of the wave, shoreline
velocity, and shoreline runup–rundown motion can be evaluated using
numerical integration.
Carrier et al. (2003) developed the Green function representation of
the solution of the nonlinear shallow water-wave equations using a
slightly different transformation than Carrier and Greenspan (1958).
Carrier et al. (2003) evaluated Green function explicitly and obtained the
highly singular complete elliptic integral of the first kind. Then, they
solved the nonlinear propagation problem for arbitrary initial waveform
employing numerical integration.
 Nonlinear Evolution of Long Waves Over a Sloping Beach 239

The initial waveform used as the benchmark problem is given by

Carrier et al. (2003) as
 − 0.4444 ( x − 4.1209) 2   − 4.0 ( x −1.6384) 2 
η ( x, t = 0) = 0.006 e  
− 0.018 e  
,

in dimensionless form. This initial profile is used to obtain the solution

for benchmark problem through Kânoğlu (2004). However, the results
are converted to the dimensional quantities using a reference length
~
l = 5000 m and a beach slope tan β = 1/10 to compare with the benchmark
solution. Figure 1 compares the initial wave profile resulted from the
present nonlinear solution with the initial profile given in the benchmark
problem. The comparison shows that the linearization of the spatial
variable in the definition of initial condition does produce identical result.
The spatial variation of the water surface elevations and velocities are
compared with the benchmark solution at three different times in Fig. 2
and comparisons are excellent. In Fig. 3, comparisons are presented for
the shoreline position and velocity showing identical results.
A new initial value problem solution to the nonlinear shallow water-
wave equations is developed by Kânoğlu (2004) using the direct
integration — without resorting to singular elliptic integrals — and is
applied to the benchmark problem. Given an initial condition in the

0
η(m)

−5

−10
0 10 20 30 40 50
x(km)

Fig. 1. Comparison of the initial wave profiles resulted from the present nonlinear
solution and given in the benchmark problem. The circles represent the initial wave
profile given in the benchmark problem.
240 U. Kânoğlu

physical space, the derivation of the equivalent initial condition in the

transform space is possible by using linearized form of the hodograph
transformation for the spatial variable at initial time. Comparisons show
that direct integration yields identical results with the benchmark
solution and the proposed analysis appears simpler than Carrier et al.
(2003).

0 9
6
u(m/s)
−10
η(m)

t = 160s 3 t = 160s
−20
0
−30 −3
0 200 400 600 800 0 200 400 600 800
0 0
u(m/s)

−10 −5
η(m)

t = 175s t = 175s
−20 −10

−30 −15
0 200 400 600 800 0 200 400 600 800
20 2
u(m/s)

0
η(m)

10 t = 220s t = 220s
−2

0 −4
−200 0 200 400 600 800 −200 0 200 400 600 800
x(m) x(m)

Fig. 2. Comparison of the present nonlinear solution and benchmark solution for the
spatial variation of the surface elevations and velocities at three specific times. The
circles represent the benchmark solution.
 Nonlinear Evolution of Long Waves Over a Sloping Beach 241

300
(a)
200

100
x(m)

−100

−200
0 50 100 150 200 250 300 350

10
(b)
0
u(m/s)

−10

−20
0 50 100 150 200 250 300 350
t(s)

Fig. 3. Comparison of the present nonlinear solution and benchmark solution for the
shoreline (a) position and (b) velocity. The circles represent the benchmark solution.

References
1. G. F. Carrier and H. P. Greenspan, J. Fluid Mech. 4, 97 (1958).
2. G. F. Carrier, T. T. Wu and H. Yeh, J. Fluid Mech. 475, 79 (2003).
3. U. Kânoğlu, J. Fluid Mech. 513, 363 (2004).
This page intentionally left blank
 CHAPTER 9

AMPLITUDE EVOLUTION AND RUNUP OF LONG WAVES;

COMPARISON OF EXPERIMENTAL AND NUMERICAL DATA
ON A 3D COMPLEX TOPOGRAPHY

Ahmet Cevdet Yalciner1, Fumihiko Imamura2, and Costas E. Synolakis3

1
Middle East Technical University, Civil Engineering Department, Ocean Engineering
Research Center, 06531, Ankara, Turkey
[email protected]
2
Tohoku University, Disaster Control Research, Aoba 980 Sendai Japan
3
University of Southern California, Department of Civil and Environmental Engineering
Los Angeles, 90089 LA, CA, USA

The runup of long waves on the sloping planes is described by the

analytical solutions of the long wave equations with special initial
conditions, proper approximations and boundary conditions. These
studies are also compared with experimental data (Yeh et al. (1996);
Lin et al. (1999); Yalciner et al. (2003)). Similarly the numerical
methods solving governing equations with proper boundary conditions
are also developed and compared with either analytical or experimental
or field data for long wave propagation and runup. In this study
particularly, the experimental set up is applied in the numerical model
with the same wave and boundary conditions. The computed shape and
amplitude evolution of the wave at the complex topography are
compared with the existing experimental results. The performance of
the numerical method is also discussed.

1. Introduction
Tsunami runup onto a complex 3D dimensional beach is one of the
Benchmark Problems of tsunami runup. The extreme runup height of
32 m that was measured near the village of Monai in Okushiri Island by
1993 Okushiri tsunami is one of the examples of this kind of benchmark
problems. There also exists experimental data in a 1/400 scale laboratory

243
244 A. C. Yalciner, F. Imamura and C. E. Synolakis

experiment of the Monai runup, using a large-scale tank

(205 m long, 6 m deep, 3.4 m wide) at Central Research Institute for
Electric Power Industry (CRIEPI) in Abiko, Japan. The figures showing
the bathymetry, coastal topography, incident wave profile at the water
depth d=13.5 cm, and wave profiles at three selected locations are
provided. In this study we have used the computer program TUNAMI-
N2 for this problem and we compared our results of temporal and
spatial variations of the shoreline location, the temporal variations of the
water surface variations at three specified nearshore locations with
experimental results.

2. Modeling
The numerical model, TUNAMI-N2, used for the simulation of the
propagation of long waves is authored and developed by Prof. Imamura
in Disaster Control Research Center in Tohoku University, Japan,
through the Tsunami Inundation Modeling Exchange (TIME) program
(Shuto et al. (1990)). TUNAMI-N2 is one of the key tools for the studies
for propagation and coastal amplification of tsunamis in relation to
different initial conditions. It solves the nonlinear form of long wave
equations with bottom friction by finite difference technique, and
computes water surface fluctuations and depth averaged velocities at all
locations even at shallow and land regions within the limitations of grid
size on the complex bathymetry and topography. Consequently the
program simulates the propagation and coastal amplification of long
waves. Shuto et al. (1990), Nagano et al. (1991), Imamura (1996), Goto
et al. (1997), Imamura et al. (1999), Yalciner et al. (2001a, 2001b, 2002,
2003, 2004), are some of the studies used in TUNAMI-N2.
The computation domain is selected as 5.488 m by 3.402 m (393
nodes × 244 nodes with the grid size of 0.014 m). The bathymetry data is
the same as the one used in laboratory experiments where the minimum
depth is −0.125 m (land) and maximum depth is 0.13535 m. A typical
value of 0.025 is used as Manning’s coefficient (Munson et al. (1998)).
The time step was chosen as 0.05 sec to satisfy the CFD condition. The
propagation of the wave in the basin with respect to the inputted wave
form (from laboratory data) is simulated.
 Amplitude Evolution and Runup of Long Waves 245

3. Comparison of the Results

In order to compare the numerical and experimental results, the water
surface at different times, and time histories of water surface oscillations
at selected locations and the distribution of maximum positive
amplitudes in the simulation duration of 22.5 sec (model time) at all grid
points are computed. The computed snapshots (at t = 12.5 and 18.4 sec.
in the model) of the water surface are shown in Fig. 1. According
to simulation results, the maximum amplitude is achieved at around
18.5 sec. near the shoreline. The distribution of the computed maximum
positive amplitudes in the computational domain is also shown in the
same figure. According to simulation results, the maximum positive
amplitude of the wave at land is found to be 0.061 m. This value is
somewhat less than the maximum runup measured at the site of Monai
Runup of Okushiri tsunami. One of the reasons for this difference is the
local topography in the inundation zone at Monai, which could not be
fully represented in the numerical model.
The measured and computed values of temporal water surface
variations measured at the gauge locations (x, y) = (4.521 m, 1.196 m),
(4.521 m, 1.696 m), and (4.521 m, 2.196 m) are also given in Fig. 2. The
good agreement between the computed and measured water surface
oscillations are obtained when the experimental and computational

Fig. 1. The computed snapshots (at t = 12.5 and 18.4 sec. in the model) of the wave
propagation (left two figures) and maximum positive amplitudes during 22.5 sec
simulation.
246 A. C. Yalciner, F. Imamura and C. E. Synolakis

Fig. 2. The measured and computed values of temporal water surface variations
measured at different gauge locations.

results for three gauge locations are compared. These results show that
the simulation using TUNAMI-N2 provides fairly consistent results with
experiments.

Acknowledgments
The tsunami propagation model TUNAMI-N2 is a registered copyright
of Imamura, Yalciner and Synolakis.

References
Goto, C., Ogawa, Y., Shuto, N. and Imamura, F. (1997) Numerical method of tsunami
simulation with the leap-frog scheme, IOC Manual 35, U.N. Educ., Sci. and Cult. Org.,
Paris.

Imamura, F. (1996) Review of tsunami simulation with a finite difference method, in

Long-Wave Runup Models, eds. Yeh, H., Liu, P. L.-F. and Synolakis, C., pp. 25–42,
(World Scientific Publishing Co.).

Imamura, F., Koshimura, S. and Yalciner, A. C. (1999) Field survey and numerical
modeling of tsunami generated by Turkish earthquake of August 17, 1999 (in Japanese),
Proc. Coastal Engineering in Japan, 47, 331–333.

Lin, P., Chang, K. A. and Liu, P. L.-F. (1999) Runup and Rundown of Solitary Waves on
Sloping Beaches, Journal of Waterway, Port, Coastal, and Ocean Engineering,
September/October 1999.

Munson, B. R., Young, D. F. and Okiishi, T. H. (1998) Fundamentals of Fluid

Mechanics (John Wiley, New York).
 Amplitude Evolution and Runup of Long Waves 247

Nagano, O., Imamura, F. and Shuto, N. (1991) A numerical model for a far field
tsunamis and its application to predict damages done to aquaculture, Nat. Hazards, 4,
235–255.

Shuto, N., Goto, C. and Imamur, F. (1990) Numerical simulation as a means of warning
for near field tsunamis, Coastal Engineering in Japan, 33(2), 173–193.

Yalciner, A. C., Pelinovsky, E., Okal, E. and Synolakis, C. E. (Eds.) (2001a) NATO
ARW, Underwater Ground Failures on Tsunami Generation, Modeling, Risk and
Mitigation, North Atlantic Treaty Org., Istanbul, Turkey.

Yalciner, A. C., Synolakis, A. C., Alpar, B., Borrero, J., Altinok, Y., Imamura, F.,
Tinti, S., Ersoy, Ş., Kuran, U., Pamukcu, S. and Kanoglu, U. (2001b) Field Surveys and
Modeling 1999 Izmit Tsunami, International Tsunami Symposium ITS 2001, Session 4,
Paper 4–6, Seattle, August 7–9, 2001, pp. 557–563.

Yalciner, A. C., Alpar, B., Altinok, Y., Ozbay, I. and Imamura, F. (2002) Tsunamis in
the Sea of Marmara: Historical documents for the past, models for future, Mar. Geol.
190(1–2), 445–463.

Yalciner, A. C., Pelinovsky, E., Okal, E. and Synolakis, C. E. (2003) Submarine

Landslides and Tsunamis, NATO Sci. Ser., Ser. IV, Vol. 21, pp. 327 (Kluwer Academic,
Norwell, Mass).

Yalçiner, A. C., Pelinovsky, E., Talipova, T., Kurkin, A., Kozelkov, A. and Zaitsev A.,
(2004) Tsunamis in the Black Sea: Comparison of the historical, instrumental, and
numerical data, J. Geophys. Res., 109, C12023, doi:10.1029/2003JC002113.

Yeh, H., Liu, P. L.-F. and Synolakis, C. (Eds.) (1996) Long-Wave Runup Models (World
Scientific Publishing Co.).
This page intentionally left blank
August 8, 2008 12:8 WSPC/Trim Size: 9in x 6in for Review Volume 10˙wang

CHAPTER 10

NUMERICAL SIMULATIONS OF TSUNAMI RUNUP ONTO

A THREE-DIMENSIONAL BEACH WITH SHALLOW
WATER EQUATIONS

Xiaoming Wang∗ and Phillip L.-F. Liu†

School of Civil and Env. Eng. Cornell University 14853, Ithaca, NY
∗
[email protected], † [email protected]

Alejandro Orfila
IMEDEA(CSIC-UIB), Miquel Marques, 21.07190 Esporles, Spain
[email protected]

1. Introduction
In this paper we present the numerical results for the benchmark problem
#2 proposed. The model implemented is the Cornell Multigrid Coupled
Tsunami Model, COMCOT (Liu et al., 1998). Although a multi-grid sys-
tem, dynamically coupled up to three regions (either spherical or cartesian)
can be implemented, in the present simulations only one layer has been used.
The governing equations for the model are based on the shallow water wave
equations. The continuity and momentum equations are expressed as,
∂η ∂P ∂Q
+ + = 0, (1)
∂t ∂x ∂y

P2
   
∂P ∂ ∂ PQ ∂η
+ + + gH + τx = 0 , (2)
∂t ∂x H ∂y H ∂x

Q2
   
∂Q ∂ PQ ∂ ∂η
+ + − gH + τy = 0 , (3)
∂t ∂x H ∂y H ∂y
where η is the free-surface displacement, h is the still water depth, H = h+η
is the total water depth and u, v are the depth averaged velocities in the

249
August 8, 2008 12:8 WSPC/Trim Size: 9in x 6in for Review Volume 10˙wang

250 X. Wang, P. L.-F. Liu and A. Orfila

x- and y-directions. The bottom friction in equations (2)–(3) is obtained

from,
gn2
1/2
τx = 7/3
P P 2 + Q2 , (4)
H

gn2
1/2
τy = 7/3
Q P 2 + Q2 , (5)
H
where n is the roughness coefficient. The model solves (1)–(3) with an
explicit modified Leap-Frog finite difference scheme (Cho, 1995).

2. Numerical Results
The initial surface profile provide by the benchmark problem is set as the
initial wave input in the left boundary and the wave propagation is sim-
ulated for 150 s. Solid reflective walls are implemented in the other three
boundaries. Numerical results are compared with data provided at three
gage locations A = (4.521,1.196); B = (4.521,1.696) and C = (4.521,2.196).

2.1. Mesh Size, Bottom Fiction and Linear vs Nonlinear

Model Sensitivity
To test the sensitivity of the model to the grid size, results from two different
meshes with resolutions of ∆x = 0.05 m (393 × 244 grid points) and ∆x =
0.01435 m (1098×681 grid points) are compared. In the coarse grid, the time
increment has set as dt = 0.001 s so that the Courant number is 0.08 while
in the finer, the time increment is dt = 0.0002 s being the Courant number
0.05. Figure 1(a) displays the numerical results for the fine grid (grey) and
for the coarse grid (black) at the gage A. As observed, both configurations
provide similar values for the arrival time of the leading wave as well as for
the maximum amplitude.
The importance of the bottom friction (4)–(5) in the tsunami propa-
gation is tested again with two simulations, one considering the bottom
friction (with a fiction coefficient of n = 0.01) and the other without the
friction terms. Results, displayed in Fig. 1(b), show that in the present case,
the importance of the friction terms (grey line) is almost negligible provid-
ing similar results than those obtained without friction (black line). Since
tsunami wavelengths are much larger than the mean water depth, the lin-
ear long wave theory should be valid providing a faster estimation than the
non-linear equations. To test this hypothesis, finally the non-linear shallow
August 8, 2008 12:8 WSPC/Trim Size: 9in x 6in for Review Volume 10˙wang

Numerical Simulations of Tsunami Runup onto a 3D Beach 251

0.04
Coarser grid
a)
Amplitude (m)

Finer grid
0.02

-0.02
0 50 100 150
time (s)
0.04
No friction
b)
Amplitude (m)

With friction
0.02

-0.02
0 50 100 150
time (s)
0.06
Linear SW
Amplitude (m)

Non-linear SW
0.04
c)
0.02

-0.02
0 50 100 150
time (s)

Fig. 1. Comparison between (a) grid sizes; (b) bottom friction effects and (c) linear vs.
non-linear simulations.

water equations are compared with the linearized version of the equations.
Figure 1(c) shows the numerical results for the linear SW equations (black)
and for the nonlinear SW equations (grey) where the main differences are
obtained at the small scales providing overall similar results.

3. Comparison with Gage Data

As a result from the sensitivity analysis presented in the precious section,
the nonlinear model where the friction terms have been neglected is im-
plemented and results compared with the data provided at locations A,
B and C. Figure 2 displays comparison between numerical results (black)
and experimental data (grey) at locations A (Fig. 2(a)), B (Fig. 2(b)) and
C (Fig. 2(c)). As seen, the model provides good estimations for both the
time arrival of the leading wave as well as for the amplitude of the leading
and secondary waves. The nonlinear model with the coarse grid is a good
August 8, 2008 12:8 WSPC/Trim Size: 9in x 6in for Review Volume 10˙wang

252 X. Wang, P. L.-F. Liu and A. Orfila

0.03 0

0.01
0

0.02

0.04

0 05

Fig. 2. (a) Free-surface elevation at gage A; (b) B and (c) C provided by the numeri-
cal model black and the experiments (grey). (d) Maximum runup superimposed to the
bathymetric contours.

approximation since an agreement between grid size and time consumptions

has to be considered (0.074 s per time step for the coarse grid compared
to 1.193 s for the finer one). The runup, displayed in Fig. 2(d) provides a
maximum value of 10.2 cm.

4. Conclusions
The numerical results show that the problem can be well simulated as a first
attempt with the linear system without including bottom friction. Results
indicate that the grid size reduction does not lead to a much accurate
results. COMCOT results match the experimental data very good for both
the arrival time and amplitude of leading waves. In the near shore region,
the waves becomes very nonlinear and will break. In this zone, the present
model is no longer capable to deal with it and the wave amplitude may be
exaggerated.
August 8, 2008 12:8 WSPC/Trim Size: 9in x 6in for Review Volume 10˙wang

Numerical Simulations of Tsunami Runup onto a 3D Beach 253

References
1. P. L.-F. Liu, S.-B. Woo and Y. S. Cho, Computer Program for Tsunami Prop-
agation and Inundation, Cornell University (1998).
2. Y.-S. Cho, Numerical Simulations of Tsunami Propagation and Runup, Ph.D.
thesis, Cornell University (1995).
This page intentionally left blank
 CHAPTER 11

3D NUMERICAL SIMULATION OF TSUNAMI RUNUP

ONTO A COMPLEX BEACH

Taro Kakinuma
Tsunami Research Center, Port and Airport Research Institute
3-1-1 Nagase, Yokosuka, Kanagawa 239-0826, Japan
E-mail: [email protected]

This paper describes two 3D numerical models to simulate tsunami

phenomena including runup onto a complex beach. The governing
equations are the continuity and Reynolds equations for incompressible
fluids in porous media. In the first model, water surface displacement is
determined by the vertically integrated equation of continuity, while in
the second model, by the 3D-VOF method. Seabed topography can be
smoothly expressed with the porous model. These two models were
applied to reproduce the existing hydraulic-model experiment, which
treated the 1993 Hokkaido Nansei-Oki earthquake tsunami in Okushiri
Island, i.e. Benchmark No. 2. The highest runup calculated by the VOF
method indicates the full-scale value of about 30.6 m at the valley.

1. Introduction
From offshore to coastal zones, phenomena having various scales on
space and time should be solved economically. For this purpose we have
developed a hybrid model, STOC, 1 which consists of 3D models,
multilevel models and connection models. As shown in Fig. 1, the
multilevel models are used for wide-area calculation, where we treat
hydrostatic pressure, while the connection models are applied to overlap
regions for smooth connection between the multilevel models and the 3D
models, where we solve pressure without assumption of hydrostatic
pressure. By local application of the 3D models to narrower areas

255
256 T. Kakinuma

Tab. Calculation condition.

Fig. 1. Diagram of STOC.

 3D Numerical Simulation of Tsunami Runup onto a Complex Beach 257

including structures over steep topography etc., we can perform efficient

and economical computation of 3D characteristics of flow, which is a
blue rose for the traditional nonlinear long-wave models.
STOC includes two 3D models, i.e. STOC-NS and STOC-VF. In this
paper, only these 3D models of STOC have been applied to simulate
tsunami runup of Benchmark No. 2.

2. Runup Calculation Using Two 3D Numerical Simulators

The governing equations are the continuity and Reynolds equations for
incompressible fluids with a porous model describing seabed topography
smoothly. We solve these equations utilizing a finite difference method,
which is stated in detail by Kakinuma and Tomita.1,2
In STOC-NS, we calculate water surface displacement using the
vertically integrated equation of continuity. In runup regions, ‘‘Bucket-
brigade method’’ is used, i.e. there exists water where the water depth is
larger than some reference value. Unfortunately this model is not appli-
cable to cases where the water surface elevation is described by a multi-
valued function of x, because of the vertical integration.
On the other hand, STOC-VF is a 3D-VOF method, adopting a con-
vection equation of the VOF function. Using this model, which can work
also when the water surface elevation is described by a multi-valued
function of x, we can represent breaking of water waves.
The target is to reproduce the 1/400-scale laboratory experiment3 of
the Monai runup, i.e. Benchmark No. 2. The calculation conditions are
given as shown in the table. It should be noted that it is not easy to
compare the CPU time for the calculation with STOC-NS and that with
STOC-VF because ∆ t changes according to Courant number in calcu-
lation, as well as the grid-point numbers are different.
Figures 2(a)–2(c) show time variation of water level at Channels 5, 7
and 9 in Benchmark No. 2, respectively. Although the eddy viscosity is
assumed to be equal to zero in STOC-VF, the results by STOC-VF are
similar to those by STOC-NS, where an LES model is installed, except at
Channel 9 while t = 15 − 16.5 (s). The reason of this difference is that
STOC-NS has no wave-breaking model.
258 T. Kakinuma

(a) Water level at Channel 5. (b) Water level at Channel 7.

(c) Water level at Channel 9.

Fig. 2. Time variation of water level calculated by STOC-NS and STOC-VF in com-
parison with the experimental data.

As shown in Fig. 2, the calculation results of water level under the

slip-bed condition show higher first-peak values than the experimental
data, while those under the non-slip-bed condition show much lower
water level, which suggests that some bottom-friction model is required.
After draw-down, the tsunami crest travels toward the shore, and then
the maximum runup appears at Point A in Fig. 3 when t = 16.5 s. This
runup height, calculated by STOC-VF under the slip-bed condition, is
about 7.65 cm in the model scale, corresponding to 30.6 m in the
prototype scale. Figure 4 shows velocity vectors and isobaric curves on
the x–z plane including Point A at this time. Not only is wave height
important, both fluid velocity and force are important elements in the
consideration of countermeasures against tsunami disasters.
 3D Numerical Simulation of Tsunami Runup onto a Complex Beach 259

Fig. 3. Highest runup at Point A when t = 16.5 s. This water surface profile was calculated
by STOC-VF under the slip condition.

Fig. 4. Velocity vectors and constant-pressure curves on the x–z plane including Point A
in Fig. 3 when t = 16.5 s. These were calculated by STOC-VF under the slip condition.

3. Conclusions
Two numerical, hydrodynamic models, STOC-NS and STOC-VF, have
been applied to Benchmark No. 2, where water surface displacement is
determined by different ways. The calculation results by both models
under the slip-bed condition generally show correspondence with the
260 T. Kakinuma

experimental data, resulting in indication of about 30.6 m runup at the

Monai Valley.

References
1. T. Kakinuma and T. Tomita, Proc. 29th Int. Conf. on Coastal Eng., 1552 (2004).
2. T. Kakinuma and T. Tomita, OCEANS/TECHNO-OCEAN Conf. Proc., 146 (2004).
3. M. Matsuyama and H. Tanaka, Proc. Int. Tsunami Symposium 2001, 879 (2001).
August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

CHAPTER 12

EVALUATING WAVE PROPAGATION AND INUNDATION

CHARACTERISTICS OF THE MOST TSUNAMI MODEL
OVER A COMPLEX 3D BEACH

Arun Chawla∗ , Jose Borrero† and Vasily Titov∗,‡

∗
Joint Institute for the Study of Atmospheres and Oceans
NOAA Center for Tsunami Research
†
University of Southern California
‡
[email protected]

1. Introduction
Runup measurements from the 1993 tsunami off the coast of Okushuri is-
land in Japan showed that the highest wave was measured at the end of a
narrow gully in a cove near the village of Monai. The measured runup mark
at 32 m was significantly higher than other recorded runup values around
the island, and was the largest recorded measurement in Japan over the last
century.2 To determine how the local bathymetric and topographic features
could act as a focusing mechanism for wave energy, a laboratory scale ex-
periment of this region was carried out at the Central Research Institute
for Electric Power Industry (CRIEPI) in Akibo, Japan. This experimental
data set provides a valuable benchmark for evaluating long-wave tsunami
runup models and was part of a recent international workshop on long-wave
runup models.
In this paper the laboratory data set is used to evaluate a 2D long-wave
model (MOST), which is the primary tsunami inundation model used by the
NOAA Center for Tsunami Research at the Pacific Marine and Engineering
Laboratory in Seattle. It is currently being implemented as a forecast model
at the two U.S. Tsunami Warning Centers.4 The model has been developed
by Titov and Synolakis3 and has been validated with analytical solutions,
the conical island experimental studies of Briggs et al.1 and several historic
tsunami events. The aim here is to characterize how accurately the model
can simulate wave propagation and inundation processes in a more complex
but still controlled environment.

261
August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

262 A. Chawla, J. Borrero and V. Titov

2. Laboratory Experiments
Simulations of wave runup around the village of Monai in Okushuri island
were carried out with a 1/400 scale laboratory experiment using a large scale
tank (205 m long, 6 m deep, 3.4 m wide) at the Central Research Institute
for Electric Power Industry (CRIEPI) in Akibo, Japan. This benchmark
problem was part of the “3rd International Workshop on Long-Wave Runup
Models” held on June 17–18, 2004, at the Wrigley Marine Science Center
in Catalina Island California.
Figure 1 shows a side view of the experimental setup. Waves are shoaled
from the deep water using a series of linearly sloping beaches. The area
marked as “the model” in the figure is where the complex bathymetric and
topographic features from the field have been reproduced and marks the
limit of the numerical modeling exercise. The bathymetry for this model
area is shown in Fig. 2. The wave propagates from left to right, with solid
walls along the sides and the right end. The incoming wave is determined
from a gage located near the offshore boundary and is given in Fig. 3.

3. Model
MOST is a 2D long-wave model that uses the method of fractional steps5 to
reduce the 2D shallow water wave equations into a series of 1D equations
along each spatial dimension. Each set of 1D equations is re-written in
characteristic form and solved using an explicit finite difference scheme
(Method of undetermined coefficients). The scheme is second-order accurate
in space and first-order accurate in time. Solving a series of 1D equations
is computationally more efficient than solving the 2D equations. Detailed

Fig. 1. Side view sketch of the experimental setup.

August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

Evaluating Wave Propagation and Inundation Characteristics of MOST 263

3 −0.12

−0.08
2.5
Gage 5
−0.01
2
0.12 0 0 7 −0.08
y(m)

0.08 Gage
0.06 0.04 −0.02
1.5 0.02 0.01 −0.04

Gage 9

0.5

0
0 1 2 3 4 5
x(m)

Fig. 2. Domain bathymetry in the model area. The smallest depth is −0.125 m and
the deepest offshore depth is 0.135 m. Both the cove and the offshore island from the
field have been reproduced in this physical model. The three gage locations used for
model–data comparisons are given by the solid black circles.

0.02

0.015

0.01

0.005
η (m)

−0.005

−0.01

−0.015
0 5 10 15 20 25
time (sec)

Fig. 3. Incoming wave at the boundary x = 0.

August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

264 A. Chawla, J. Borrero and V. Titov

information on the model algorithm and numerical scheme can be obtained

in Titov and Synolakis.3
In its current form the model has been made flexible enough to allow for
multiple nested grids, and can be run in spherical or cartesian coordinate
systems. There is no limit on the number of nested grids allowed and any
particular grid can have a series of nested grids under it. The nested grids
have the option of two-way or one-way (coarse to fine) coupling. The model
also has the option of switching on wetting and drying capability in all
the grids. Earlier versions of the model only allowed wetting and drying in
the finest grids, thus limiting the depth to which the coarser grids could
be extended in shallow water. A manning formulation is used to account
for bottom friction. For these runs the manning coefficient has been set to
zero (no friction). The model does not have a separate treatment for wave
breaking. Breaking appears as a shock wave in the numerical solution.

4. Results
The simulations were carried out using two grids of different resolutions,
with the coarser grid covering the full domain and the finer grid nested
inside. The coarser grid had a resolution of ∆x = 0.028 m and a time step
∆t = 0.02 s. The surface time series in Fig. 3 was provided as the boundary
condition along the left boundary (x = 0). The corresponding horizontal
velocity was determined from shallow water linear wave theory. Reflecting
boundary conditions were used along the remaining three boundaries. The
internal grid had a resolution of ∆x = 0.014 m and a time step ∆t = 0.01 s.
Its boundary conditions were determined from the solutions of the outer
grid. Imposing boundary conditions from the coarser grid on the incoming
waves and allowing the outgoing waves to radiate out are trivial with the
MOST model since it solves a set of characteristic equations along each
dimension. Wetting and drying was switched on for both the grids, with
one-way coupling between the coarse and fine grid. The simulation was
run on an a single processor of a 4 processor Intel Xeon 3.6 GHz machine
running Red Hat Linux and took approximately 2.7 minutes to complete.
Time series comparison of model and data for the three wave gages in
shallow water (Fig. 2) are shown in Fig. 4. The gap in the model time series
at gage 9 corresponds to drying. There was a phase lag of 0.6 s between the
data and the model at all the three gages. This could either be because there
is an offset in the start time of the model runs as compared to the data, or
that the offshore gage is not exactly at x = 0 as in the model. Accounting
August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

Evaluating Wave Propagation and Inundation Characteristics of MOST 265

gage 5
0.04

0.02
η (m)

−0.02
0 5 10 15 20

gage 7

0.04
η (m)

0.02

0 5 10 15 20

gage 9

0.04
η (m)

0.02

0 5 10 15 20
time (sec)

Fig. 4. Model (solid line) and data (dashed line) comparisons at the three gage loca-
tions. See Fig. 2 for gage locations.

for that phase lag, we see from the figure that the model compares very
well with the data. The high frequency waves at the wave crest are not
reproduced in the model simulations. Considering the shallow water depth
at these gages (approximately 1 cm), the high frequency waves are probably
caused by higher order details of wave breaking which are not simulated
in the long wave model. Maximum alongshore (y-direction) wave runup
from the simulations is shown in Fig. 5. The highest value of 10.047 cm is
observed at the cove.
August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

266 A. Chawla, J. Borrero and V. Titov

2.8

2.6

2.4
Gage 5
2.2
y (m)

1.8 Gage 7
−0 −
1.6 0 .02 0.0 −0 −−0.1
4 .060.08
1.4
Gage 9
1.2 0.0
2
1
4 4.2 4.4 4.6 4.8 5 5.2 5.4 2 4 6 8 10 12
x (m) runup (cm)

Fig. 5. Maximum wave runup along the coast.

5. Conclusions
Earlier laboratory validation studies of the MOST tsunami model3 have
shown that the model compares well in simple domains. The model is cur-
rently being used to develop a short-term tsunami forecasting capability4
and has been considerably revised over the last year. Some of the changes
include a generalized methodology for nesting grids and the capability to
have wetting and drying in all the grids. These comparisons with a labora-
tory study in a complex domain show that the model performs reasonably
well. This increases our confidence in the model.

Acknowledgments
This publication is [partially] funded by the Joint Institute for the Study of
the Atmosphere and Ocean (JISAO) under NOAA Cooperative Agreement
No. NA17RJ1232, Contribution number 1315.

References
1. M. J. Briggs, C. E. Synolakis, G. S. Harkins and D. R. Green, Laboratory
experiments of tsunami runup on circular island, Pure and Applied Geophys.
144 (1995) 569–593.
2. N. Shuto and H. Matsutomi, Field survey of the 1993 Hokkaido Nansei-Oki
tsunami, Pure and Applied Geophys. 144 (1995) 649–664.
August 26, 2008 17:11 WSPC/Trim Size: 9in x 6in for Review Volume 12˙titov

Evaluating Wave Propagation and Inundation Characteristics of MOST 267

3. V. Titov and C. Synolakis, Numerical modeling of tidal wave runup, Water-

way, Port, Coastal and Ocean Eng. 124(4) (1998) 157–171.
4. V. Titov, F. I. González, E. N. Bernard, M. C. Eble, H. O. Mofjeld, J. C.
Newman and A. J. Venturato, Real-time tsunami forecasting: challenges and
solutions, Nat. Hazards 35(1), Special Issue, U.S. National Tsunami Hazard
Mitigation Program (2005) 41–58.
5. N. Yanenko, The Method of Fractional Steps, translated by M. Holt (Springer,
New York, Berlin, Heidelberg, 1971).
This page intentionally left blank
August 26, 2008 19:8 WSPC/Trim Size: 9in x 6in for Review Volume 13˙Kornkven

CHAPTER 13

TSUNAMI GENERATION AND RUNUP DUE TO

A 2D LANDSLIDE

Zygmunt Kowalik, Juan Horrillo and Edward Kornkven

Institute of Marine Science, University of Alaska Fairbanks
Fairbanks, AK 99775, USA
E-mail: [email protected]

The objective of this problem is to predict the free surface elevation

and runup associated with translating a Gaussian shaped mass which is
initially at the shoreline. The problem description of Benchmark Problem
3 (BM3) can be found in Chapter 6. A detailed analytical solution of
the problem is described in Liu et al.3
Two problems are given in BM3 to be solved:
A. tan β/µ = 10, where: β = 5.7◦ , δ = 1 m and µ = 0.01
B. tan β/µ = 1, where: β = 5.7◦ , δ = 1 m and µ = 0.1,
where δ is the maximum vertical slide thickness, µ = δ/L is the slide
thickness-length ratio, and β is the slope angle.
To solve BM3 two approaches have been carried out:
(1) First-order approximation in time
(2) Full Navier-Stokes (FNS) approximation aided by the Volume of
Fluid (VOF) method to track the free surface.
Approach 1 uses one-dimensional linear and nonlinear shallow water
wave theories, (LSW) and (NLSW) respectively. The finite difference
solution of equation of motion and the continuity equation is solved on a
staggered grid, Kowalik and Murty.4 This method has second-order ap-
proximation in space and first-order in time. More detailed information
is indicated in Chapter 7.
The two-dimensional FNS-VOF solution has been included to vi-
sualize differences against the shallow water solutions. The FNS-VOF
method uses the full nonlinear Navier-Stokes equation to model two-
dimensional transient and incompressible fluid flow with free surfaces.
The finite difference solution is obtained on a rectilinear mesh. The model
has been extended to deal with rigid moving objects in the computa-
tional domain. Since FNS equation includes the vertical component of

269
August 26, 2008 19:8 WSPC/Trim Size: 9in x 6in for Review Volume 13˙Kornkven

270 Z. Kowalik, J. Horrillo and E. Kornkven

velocity/acceleration, some differences are expected if nonhydrostatic

effects are strong.

1. Full Navier-Stokes Approach and VOF Method

Including Moving Objects
Equation of continuity for incompressible fluid and the momentum
equation,

1 dVo ∂~u 1
∇ · ~u = = φ and + ∇ · (~u~u) − φ~u = − ∇p + ν∇2 ~u + ~g , (1)
V dt ∂t ρ

are solved in the rectangular system of coordinates, where ~u(x, y, t) is the

instantaneous velocity vector, ρ is the fluid density, p is the scalar pressure,
ν is the kinematic viscosity, ~g is the acceleration of gravity and φ(~x, t) is
a moving object internal function needed to force zero divergence in the
computational control volume V . The momentum and continuity equations
take into account the conservation mass due to the incursion or retreat of
a moving object in the domain.
Solution of the equations is searched using the two-step method
(Chorin1 and Harlow and Welch2 ). The time discretization of the equa-
tions is similar to that described in Chapter 7.

2. Discussion and Conclusions

Results obtained using first-order numerical model and LSW analytical
solution are depicted in Fig. 1 for Case A. Agreements of the numerical
model results (NLSW and LSW) with the analytical solution are quite good.
The analytical solution does an excellent job in predicting wave runup and
wave propagation for thin slide, i.e. tan β/µ = 10.
For thicker slide, Case B, omission of nonlinearity leads to disagreements
in later stages of wave propagation, i.e. t > 1.0 (see Figs. 2 and 3). Note
that the LSW numerical result follow very well its analytical solution.
A FNS-VOF solution is presented for case B in order to compare with
the NLSW model solution. The FNS-VOF solution is ideally suited for this
case, where relatively high vertical acceleration occurs. Due to dispersion
effect FNS-VOF method predicts a wave a little bit more elongated, skewed
but slightly less tall than the NLSW model, i.e. at t = 2.5 and t = 4.5. All
numerical models and analytical solution agreed very well at earlier time
(i.e. t = 0.5 and t = 1).
August 26, 2008 19:8 WSPC/Trim Size: 9in x 6in for Review Volume 13˙Kornkven

Tsunami Generation and Runup Due to a 2D Landslide 271

& ! "  '("   *)+, - #./0- '-  .  1 243 &65 $
7 *81"   *)+, - #./0- '9-, .  1 3 &65 $
:  - ;*"  *) <=79" >#.  ;*7" #!" ? . !A@ !-B;*7 * *) .C9D 
E9
F




: 7*  <=)
3C 9    I !- J  " K . 8A!@" 7 '*"  '

   G
C    & "  '17  L 9  - .  . 8 ;*M# #
  C
 NO 7 ;*8A*" !P
 E
C 3C  9
  P GQ
. 7 E$ Q PC 

  *R CS3 

 .C 
RC
K K
. R C#T- .  K Q G$ P .
K

KC9U  V  Q  W
  . C 
G !'" )
'*C  H     $
.C 9
'.C  H
  9 * ;#$
 

. C  

 
 
 
 



    !" #%$

Fig. 1. CASE A: (tan β/µ = 10). Free surface comparison.

2
Solid lines: First Order (NLSW) Case B:
Thin dash lines: First Order (LSW)
L=10m. Horz. length of
Thick dash lines: FNS−VOF landslide
Circles: Analytical solution
1.5 δ = 1m. Slide max.
t=4.5 vert. thickness
β=5.7 Beach slope
µ=δ/L=0.1
tan(β)/µ=1
1 x’=Lx
t=2.5
η’=η
η (Dimensionless)

t’=sqrt(g/δ) µ t
g=9.81 m/s2
t=1.0
0.5
t=0.5

0
Models:
a) First Order
dx= 1e−3 ( 0.01m)
dt=3e−5 (0.0001sec)
−0.5
b) FNS−VOF
dx= 0.025 (0.250m)
dy=0.125 (0.125m)
dt=3e−5 (0.0001sec)
−1
0 2 4 6 8 10 12 14
x (Dimensionless)

Fig. 2. CASE B: (tan β/µ) = 1). Free surface comparison.

August 26, 2008 19:8 WSPC/Trim Size: 9in x 6in for Review Volume 13˙Kornkven

272 Z. Kowalik, J. Horrillo and E. Kornkven

t=0.5 t=1.0
0.6 Analytical
0.6
0.5 First Order LSW
First Order NLSW
0.4 0.4
FNS−VOF
0.3
0.2
y

0.2
0.1 0
0 −0.2
−0.1
−0.2 −0.4
0 0.5 1 1.5 0 0.5 1 1.5 2

t=2.5 0.4 t=4.5

1
0.8 0.2
0.6 0
0.4 −0.2
y

0.2
−0.4
0
−0.6
−0.2
−0.4 −0.8
0 0.5 1 1.5 2 2.5 3 0 1 2 3 4
x x

Fig. 3. Case B: (tan β/µ = 1). Shoreline location comparison.

References
1. A. J. Chorin, Numerical solution of the Navier-Stokes equations, Math. Comp.,
22:745–762, 1968.
2. F. H. Harlow and J. E. Welch, Numerical calculation of time-dependent viscous
incompressible flow of fluid with a free surface, The Physics of Fluids, 8:2182–
2189, 1965.
3. P. L.-F. Liu, P. Lynett and C. E. Synolakis, Analytical solution for forced long
waves on a sloping beach, Journal of Fluid Mechanics, 478, 101–109, 2003.
4. Z. Kowalik and T. S. Murty, Numerical Modeling of Ocean Dynamics, World
Scientific Publishing Co., 1993.
 CHAPTER 14

BOUSSINESQ MODELING OF LANDSLIDE-GENERATED

WAVES AND TSUNAMI RUNUP

Okey Nwogu
Dept. of Naval Architecture and Marine Engineering, University of Michigan
Ann Arbor, MI 48109, U.S.A.
E-mail: [email protected]

1. Introduction
The massive destruction and loss of life associated with recent tsunamis
(Indonesia, 2004; Papua New Guinea, 1998) has underscored the need to
develop and implement tsunami hazard mitigation measures such as
early warning systems in tsunami-prone areas. Given either the seabed
deformation or ocean measurements of water surface elevation by buoys,
hydrodynamic models can be used to forecast in real-time the
propagation and transformation of tsunamis across the world’s ocean
basins and the subsequent inundation of low-lying coastal areas.
Numerical models based on the non-dispersive shallow water
equations are often used to simulate tsunami propagation and runup (e.g.
Titov and Synolakis [6]). The shallow water equations assume that the
vertical fluid motions are much smaller than the horizontal fluid motions
and the resulting fluid pressure is hydrostatic. The hydrostatic pressure
assumption is reasonable for seismic tsunamis since most ocean basins
have depths of O(1km) while earthquake-generated tsunamis typically
have wavelengths of O(100km).
Landslide-generated tsunamis are characterized by shorter
wavelengths [O(10km)], depending on the relative slide speed or depth-
based Froude number. The propagation of tsunamis with water depth to

273
274 O. Nwogu

−
wavelength ratios of O(10 1) can be more accurately modeled by
dispersive water wave evolution equations such as Boussinesq-type
equations (e.g. Nwogu [3]). For dispersive waves, the horizontal
velocities are no longer uniform over the depth and the pressure field is
no longer hydrostatic. The velocity and pressure field is derived by
asymptotically expanding the velocity potential in terms of a frequency
dispersion parameter. The flow field is then substituted into the mass
and momentum equations to yield Boussinesq-type equations.
Boussinesq equations have also been used to simulate landslide-
generated waves by modifying the boundary condition at the seabed (e.g.
Lynett and Liu [2]). This paper documents the extension of a Boussinesq
model to landslide tsunami problems and the validation of the model
with benchmark problems as part of the Third International Workshop
on Long-Wave Runup Models.

2. Description of Numerical Model

For surface gravity waves with wavelengths much longer than the
characteristic water depth, the vertical profile of the velocity field can be
approximated by expanding the velocity potential as a Taylor series
about an arbitrary elevation zα in the water column (e.g. Nwogu [3]). The
velocity potential that satisfies the Laplace equation and moving bottom
boundary condition can be written as:

φ ( x, y, z, t ) = φα ( x, y, t ) + µ 2 ( zα − z ) [ht + ∇φα ⋅ ∇h ]
1 2
+ µ ( zα + h )2 − ( z + h )2  ∇ 2φα , (1)
2 

where h(x, y, t) is the water depth, µ is a frequency dispersion parameter

and φα = φ (x, y, zα, t). The above expansion can be substituted into the
continuity equation and the 3-D Euler equations of motion to yield:
 Boussinesq Modeling of Landslide-Generated Waves 275

  h +η  
ηt + ∇ ⋅ ( h + η ) u  = − ht − ∇ ⋅ (h + η ) ( zα + h ) −  ∇ht  ,
  2  

( )
uα ,t + g∇η + uη ⋅ ∇ uη + wη ∇wη

− ( uα ,t ⋅ ∇h ) + (h + η )∇ ⋅ uα ,t  ∇η

+ ( zα − η ) ∇ ( uα ,t ⋅ ∇h ) + (∇ ⋅ uα ,t )∇h 
(2)

1
( zα + h ) − (h + η )2  ∇ (∇ ⋅ uα ,t )
2
+
2 
1 1
+ ∇ [ν (h + η )∇ ⋅ uα ] + fw ub ub = − ( zα − η ) ∇htt ,
h +η h +η

where η (x, y, t) is the water surface elevation, uα is the horizontal

velocity at z = zα, g is the gravitational acceleration, fw is a bottom
friction coefficient, ν is a wave energy dissipation coefficient, u is the
depth-averaged horizontal velocity, uη and wη are the horizontal and
vertical velocities at the free surface (z = η), and ub is the horizontal
velocity at the seabed (z = − h). The different horizontal and vertical
velocities are obtained from Eq. (1). The equations (2) have been
numerically integrated using a finite-difference method (Nwogu and
Demirbilek [4]). The moving boundary algorithm is described in [5].

3. Benchmark Problems
3.1. Tsunami Runup onto a Complex Three-dimensional Beach

In this benchmark problem, we attempt to reproduce the results of a

physical model study of wave runup over three-dimensional topography
during the 1993 Okushiri tsunami. The bathymetry is shown in Fig. 1.
The complex nature of the bathymetry led to wave focusing and a
maximum runup height of 32m at the tip of a cove.
276 O. Nwogu

Ch. 9
2 -0.01
y(m)

-0.1
-0.12

-0.08

0.02
-0.04 Ch. 7 0

-0.
-0.01

06
Ch. 5
1

0
0 1 2 3 4 5
x(m)
Fig. 1. 2-D map of bathymetry for laboratory model study of Okushiri tsunami.

Numerical simulations were performed with grid size ∆x = ∆y =

0.04m, and time step ∆t = 0.01s. The model was initialized with the
measured water surface elevation along the offshore boundary. Figure 2
shows a comparison of the measured and predicted time histories of the
water surface elevation at Gauge #9. The Boussinesq model does
reproduce the maximum height of the tsunami in shallow water although
there are differences in the phasing of the reflected wave.

0.04 Lab Data

Elevation (m)

Num. Model

0.02

0
0 10 20 30 40 50
Time (s)
Fig. 2. Measured and predicted wave elevation time histories at Gauge #9 for Okushiri
tsunami.

3.2. Tsunami Generation and Runup Due to a Two-dimensional

Landslide
We consider the problem of a Gaussian-shaped mass sliding down a
beach of slope β. The instantaneous seabed elevation is given by:
 Boussinesq Modeling of Landslide-Generated Waves 277


( )  ,
2
h( x, t ) = x tan β + δ exp  − 2 xhµ2 / δ tan β − hµ t g / δ (3)


where δ is the maximum vertical slide thickness and hµ is the

thickness/slide length. Liu et al. [1] derived an analytical solution for the
problem based on the linearized shallow water equations. Numerical
simulations were carried out for both the nonlinear shallow water
(NLSW) and Boussinesq equations for β = 5.7°, δ = 1m with hµ = 0.01
(Case A) and hµ = 0.1 (Case B). For Case A, the NLSW and Boussinesq
solutions were identical to the analytical solution and are not presented
here. For Case B with a larger slide thickness to length ratio, differences
can be seen between the three solutions as shown in Fig. 3.
Elevation (m)

0.8 tnd = 1.0 NLSW

Boussinesq
0.4 Analytical

0
0 10 20 30 40 50
x(m)
Elevation (m)

1.6
tnd = 4.5 NLSW
0.8 Boussinesq
Analytical
0
-0.8
0 50 100 150 200
x(m)
Fig. 3. Numerical and analytical profiles of waves generated by sliding Gaussian mass.

References
1. P.L.-F. Liu, P. Lynett, and C. Synolakis, J. Fluid. Mech., 478, 101 (2003).
2. P. Lynett and P.L.-F. Liu, Proc. R. Soc. Lond., 458, 2885 (2002).
3. O. Nwogu, J. Waterw. Port Coastal Ocean Eng., 119, 618 (1993).
4. O. Nwogu and Z. Demirbilek, Tech. Rep. ERDC/CHL 01-25, U.S. Army Engineer R
& D Center, Vicksburg, MS (2001).
5. O. Nwogu and Z. Demirbilek, Coastal Engineering, accepted (2005).
6. V. Titov and C.E. Synolakis, J. Waterw. Port Coastal Ocean Eng., 121, 308 (1995).
This page intentionally left blank
August 26, 2008 17:30 WSPC/Trim Size: 9in x 6in for Review Volume 15˙yasuda

CHAPTER 15

NUMERICAL SIMULATION OF TSUNAMI RUNUP

ONTO A COMPLEX BEACH
WITH A BOUNDARY-FITTING CELL SYSTEM

Hiroyasu Yasuda
River Engineering Division, Civil Engineering Research Institute of Hokkaido
Hiragishi 1-3, Toyohira, Sapporo, 062-8602, Japan
E-mail: [email protected]

Tsunami runup onto a complex beach was modeled and analyzed us-
ing a method that allows accurate reproduction of complex topographic
features. The analysis produced accurate results and confirmed that to-
pography greatly affects the flow regime.

1. Introduction
The tsunami caused by the 1993 Hokkaido-Nansei-Oki Earthquake off
southwestern Hokkaido left traces of a great runup at Monai Beach on
Okushiri Island. This runup is regarded as one of the highest of the 20th cen-
tury, and its severity is attributed in part to the very complex 3-dimensional
trough-shaped topography of that beach. This study reproduced the runup
using 2-dimensional analysis that incorporates a “boundary-fitting cell”
(BFC) system. The system allows calculations to flexibly incorporate com-
plex topographic shapes.

2. Method of Numerical Analysis

2.1. Dividing the Analysis Domain
Numerical analysis was divided into two domains: the steeply undulating
Monai Beach, and other locations. For discretization of topographic shapes,
the former domain used BFCs to allow maximum incorporation of topo-
graphic shapes at the location, and the latter domain used a Cartesian
grid.
2-dimensional analysis1 using BFCs allows accurate and efficient incor-
poration of topographic shapes by being able to combine triangular and

279
August 26, 2008 17:30 WSPC/Trim Size: 9in x 6in for Review Volume 15˙yasuda

280 H. Yasuda

polygonal cells without restriction. An additional benefit of efficient dis-

cretization is the significantly reduced calculation time. This allows calcu-
lation that can be one hundred times faster than that using a uniformly
subdivided Cartesian grid. Furthermore, it allows incorporation of linear
structures, such as road or wall structures, that greatly affect the flow but
that are too narrow to be incorporated as grids.
Discretization of the two steep-sloped trough and ridge topographic
shapes at Monai Beach was performed by first defining the sides of the
cell along the trough line and ridge line. Then the sides of the cell were
defined as either parallel to or perpendicular to the elevation contour line.

2.2. Governing Equations

Two governing equations of fluid motion were applied to the Cartesian grid
domain: the equations were the shallow water equation, all terms in the
equation were integrated from the bottom to the water surface. The seabed
frictional resistance is given by Manning’s equation. One governing equation
was applied in the BFC domain: It was obtained by expanding the shallow
water equation in the 2-dimensional plane, ignoring the convection terms,
to allow calculation using polygonal cells.
1 ∂Q ∂η gn2 Q|Q|
+ gD =− , (1)
l ∂t ∂s D7/3

1X
k
∂η
− Qi = 0 , (2)
∂t A i=1
where t is the time coordinate, Q is flow rate at the sides of the cell, s is
the plane coordinate, η is the water level, D is the water depth, g is the
gravitational acceleration, n is Manning’s roughness coefficient, A is the
area of the cell where the water level is to be obtained, and k is the number
of sides of the cell.
The numerical scheme used in Eqs. (1) and (2) is based on the Staggered
Leap-Frog scheme, which is an explicit differential scheme.

2.3. Boundary and Initial Conditions

For the Cartesian grid domain, these were the boundary conditions: the
upper edge, the lower edge, and the section of the right edge with elevation
of less than 0 cm were considered a fully reflective boundary. The section
at the right edge with elevation of more than 0 cm was not considered a
August 26, 2008 17:30 WSPC/Trim Size: 9in x 6in for Review Volume 15˙yasuda

Numerical Simulation of Tsunami Runup onto a Complex Beach 281

Experiments, Calculation
Water level (cm)

Water level (cm)

Water level (cm)

5.0 5.0 5.0
Point a (x = 4.521, y = 1.196) Point b (x = 4.521, y = 1.696) Point c (x = 4.521, y = 2.196)

4.0 4.0 4.0

3.0 3.0 3.0

2.0 2.0 2.0

1.0 1.0 1.0

0.0 0.0 0.0

-1.0 -1.0 -1.0

0 5 10 15 20 25 30 0 5 10 15 20 25 30 0 5 10 15 20 25 30
time (s) time (s) time (s)

(a) point a (b) point b (c) point c

Fig. 1. Time-lapse comparison of water levels: experiments vs. numerical analysis.

reflective boundary and tsunami was able to run up onto land. The entire
surface of the left edge was considered an incidence boundary and a water
level was assigned. For the BFC domain, this was the boundary condition:
the water levels obtained from the Cartesian grid calculation were assigned
to cells at the domain boundary.
For both domains, the initial conditions were these: at 0 cm height,
static water surface was assigned; and at points over 0 cm height, the water
depth was set at 0 cm.

3. Analysis Results
3.1. Cartesian Grid Domain
The accuracy in calculating sea level variations is an indication of the valid-
ity of boundary conditions in the BFC domain. Figure 1 shows time-lapse
waveforms of water levels obtained from the hydraulic experiment and nu-
merical analysis. The three points in Fig. 1 show that the time-lapse water
levels for both methods are in close agreement.

3.2. BFC Domain

The results of 2-dimensional tsunami runup analysis at Monai Beach using
the BFCs described in §2.1 are shown in Fig. 2(b). The figure shows water
depth distributions at the time of maximum runup area. The curves in this
figure, from left to right, show 2.5, 5.0, and 7.5 cm elevation contours.
The BFC and Cartesian grid results generally agree in terms of flood-
ing area and water depth distribution. There is only a slight difference of
0.2 sec in the calculated time of maximum flood area. The maximum runup
August 26, 2008 17:30 WSPC/Trim Size: 9in x 6in for Review Volume 15˙yasuda

282 H. Yasuda

t =17.70 sec, .
M aximum height: t = 17.90(s)

2.50
Maximum runup at Monai 7.10cm (by BFC)
7.
50

0.0
(by Rectangle Grid) 5.
00

2.
50

2.5
0
BFC Area 2.50

7.5
0
5. Max. heig
ht:
00 7.30cm
z=

5.
00
y 7.
5
2. 0cm
50

0.
cm

0
x=3.00

0c
y=1.00 7.50

m
x
Water Level (cm) D epth of W ater (cm )

-2.25 -1.50 -0.75 0.00 0.75 1.50 2.25 0.0 0.5 1.0 1.5 2.0 2.5 3.0

(a) Water level distribution at Monai Beach (b) Flow regime of flood from the
at the time of maximum flood runup BFC method

Fig. 2. Analysis results for the tsunami runup at Monai Beach. (a) shows the results of
analysis using the Cartesian grid: the maximum runup height at Monai Beach is 7.1 cm.
In BFC analysis, the water level on these results are obtained.

height calculated using BFC was expected to be higher, but it remained

at 7.3 cm, just slightly higher than the 7.1 cm height calculated using the
Cartesian grid. The main reason the calculated runup height did not reach
the observed value, even with the BFC method, is thought to be that the
convection terms is neglected in the equation of BFC. Even though inertial
forces in the flow predominated, the neglect of convection terms resulted in
an underestimation of runup height.

4. Conclusion
This numerical analysis calculated tsunami runup onto a complex beach,
using the BFC system, which accurately reproduces complex topographic
features. The calculations obtained maximum runup heights that were
similar to results, without 3-dimensional calculation. This shows the im-
portance of using discretization methods that are appropriate for the flow
characteristics and topography.

References
1. H. Yasuda, M. Shirato, C. Goto and T. Yamada, Journal of Hydraulics, Coastal
and Environment Engineering No.740/II–64 (2003) (in Japanese).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

CHAPTER 16

A 1-D LATTICE BOLTZMANN MODEL APPLIED TO

TSUNAMI RUNUP ONTO A PLANE BEACH

J. B. Frandsen
School of Civil Engineering, The University of Sydney
NSW 2006, Australia
Email: [email protected]

1. The Numerical Model

We consider the finite discrete-velocity model of the Boltzmann equation
with a finite discrete-velocity set. The discrete Boltzmann equation with a
BGK collision term, referred to as the LBGK model, as originally described
by Bhatnagar et al. (1954), has been extended to simulate free-surface flows
in shallow water.
Our model approximates the 1-D depth-averaged NonLinear Shallow
Water (NLSW) equations in rotational flows,
∂h ∂(h u)
+ = 0;
∂t ∂x
(1)
∂(h u) ∂(h u2 ) ∂
h2  ∂
∂u 
+ = −g + hν +F ,
∂t ∂x ∂x 2 ∂x ∂x
where g is gravity due to acceleration, h = h0 + ζ, h0 is the still water
depth, ζ denotes the free-surface elevation measured vertically above the
still water level, u is the depth-averaged horizontal velocity and ν is the
kinematic viscosity. The force term, F = −g h ∂h b
∂x , in the present test case,
accounts for the slope of the beach where hb is the slope height, as shown
in Fig. 3(a). We have not accounted for the effects of bed friction, sediment
transport and wave breaking.
The numerical flow field is represented by particles which follow lattice
points which stream and collide when meeting. With reference to the Boltz-
mann equation of classical kinetic theory, the distribution of fluid molecules

283
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

284 J. B. Frandsen

x x + ∆x

f1 f2
t _c
+c

(1 − α )f2 (1 − α ) f1
t
α f1 α f2

f f
t + ∆t 2 1

∆x
j (x + 2 ) = c ( f2 f1 )

Fig. 1. D1 Q3 collision model.

is represented by the particle distribution function fi (x, ci , t) where i de-

notes propagation direction. The function defines the mass density ρ of
lattice particles, equivalent to the water column h, at time t which moves
with the microscopic velocity ci . The discrete-velocity set, at a lattice node,
satisfies the Maxwell distribution of equilibrium.
Herein, the numerical model is a simplification of the D2 Q9 model ap-
proach, as shown in Chapter 5 in this volume. In the present case study, it
is assumed that, at any time, the LB fluid is characterized by the popula-
tions of the three discrete microscopic velocity model, referred to as D1 Q3 ;
representing a 1-D model. Figure 1 shows a collision between two particles
of a D1 Q3 model. During the collision a portion α of the lattice particle
collide and reverses velocity. The net flux j at the boundary between lattice
cell at x and x + ∆xi /2 as
j(x + ∆xi /2) = ci fi
(x, t) + c̃i f˜i
(x + ∆xi , t)

= ci [fi (x + ∆xi , t + ∆t) + f˜i
(x + ∆xi , t + ∆t)] , (2)

where fi
denotes the post-collision particle distribution at time t, fi is
the pre-collision particle distribution at t + ∆t, and c̃i denotes a particle
propagating in the opposite direction of particles with velocity ci (c̃i = −ci ).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 285

We introduce the single phase 1-D LBGK formulation for shallow water
flows on sloped beds as

2
∂fi ∂fi (fi − fieq ) ci ∆t ∂hb 1 
+ ci =− − gh , where Ni = ci ci ,
∂t ∂x τ N i c2 ∂x c2 i=0
(3)
with Ni as a constant which depends on the lattice geometry and is defined
Ni = 2 for the D1 Q3 velocity set. The first term represents the effect of
the local change of the fluid motion in time and the second term describes
the convection. Note the convection operator of the Lattice Boltzmann
Method (LBM) is linear. This property originates from kinetic theory. The
first term of (3) on the right-hand-side (RHS) is the non-equilibrium distri-
bution function which describes the effect of the collisions (Succi (2001)).
The second term on the RHS represents the force term F . The time-scale
parameter τ = 3 ν/c2 + ∆t/2 where ν is the kinematic viscosity of wa-
ter and c = ∆x/∆t (the time step is ∆t and the grid spacing is ∆x). It
describes the collisional relaxation to the local equilibrium and 1/τ is the
collision frequency. In the present formulation τ is limited to a single value
(BGK approximation). To ensure numerical stability τ > ∆t/2 when using
the standard LBGK scheme. Further, we note that assuming a constant
value of τ is considered to be a crude approximation. The equilibrium dis-
tribution functions fieq represents the invariant function under collision (no
gradients are involved) and is dependent upon the microscopic velocity ci .
It is essential that the equilibrium distribution functions satisfy the Navier-
Stokes equation for any water wave model. In our case, the depth-averaged
shallow-water equations with rotational flow. The fieq functions can be de-
rived from the Boltzmann equation using a Chapman-Enskog expansion.
Details of the Chapman-Enskog method for the classical Boltzmann equa-
tion can be found in Gombosi (1994). The LBGK based fieq expressions
are velocity dependent, both at the micro and macroscale levels. A new set
of equilibrium distribution functions following the 1-D flow approximation
has been derived and are given as,

g h2 h u2 gh2 hu h u2
f0eq = h − − ; eq
fi=1,2 = + c i + , (4)
2c2 c2 4c2 2c 2c2

where the discrete velocity set is ci=1 = 1 and ci=2 = −1, respectively.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

286 J. B. Frandsen

The hydrodynamic moments of the equilibrium distribution functions

are conserved at every time step,
2
 2
 2
 1 2
fieq = h ; ci fieq = hu ; ci ci fieq = gh + hu2 . (5)
i=0 i=0 i=0
2

For the present application, it is noted that the pressure field exhibit near
hydrostatic behavior and therefore the dynamic pressure term (hu2 ) is neg-
ligible. The macroscopic variables of the free surface (ζ) and the depth
averaged velocities (u) are calculated as the first and second moments of
the distribution function,
2
  2
1
ζ= f i − h0 and u = ci f i . (6)
i
(h0 + ζ) i

2. LB and the Depth-Averaged Navier-Stokes Equations

Equation (3) can be viewed as a special finite-difference discretization of
the single time-relaxation approximation of the Boltzmann equation for
discrete velocities. One approach to solve the discrete Boltzmann equations
is to use a first-order forward Euler time-difference scheme and a first-order
upwind space discretization for the convection term in a uniform lattice
spacing ∆x. One can then obtain following algebraic relation,
∆t
 ∆t
fi (x + ∆x, t + ∆t) − fi (x, t) = − fi (x, t) − fieq (x, t) + ci Fi . (7)
τ N i c2
This equation is commonly referred to as the Lattice-BGK equation or
Lattice-Boltzmann Equation (LBE). The LBE is second-order accurate in
space and first-order in time (Junk et al. (2005)). See also further discussion
on this matter in Chapter 5 in this volume. Further, it should be noted that
using a first-order Euler scheme does not translate into the same lack of ac-
curacy as when used in traditional CFD schemes because of the nonlinearity
embedded in the collision term.
The macro dynamical behavior arising from the LBE can be found from
a multiple-scale analysis such as the Chapman-Enskog technique, as de-
scribed by Gombosi (1994). We shall show that for a given set of fi (which
are solutions of (3) for specific polynomials of the macroscopic quantities
fieq and moments (5)) are solutions of the depth-averaged Navier-Stokes
equations. We assume small Knudsen numbers ε = l/L (≈∆t) where l is
the scale in the mean free path and L is the characteristic macroscopic
length. The pressure is defined as the ideal gas equation p = c2s ρ in flows
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 287

of low Mach numbers where c2s = ∂p/∂ρ = c2 /3 is the speed of sound and
ρ is the fluid density.
In the following we use the Chapman-Enskog technique to show that
the present LB formulation recovers the NLSW equations in rotational flow.
The Chapman-Enskog expansion assumes that the convection and diffusion
processes operate at different time scales. The diffusion time scale being
much slower than the convective counter part. First, we perform a Taylor
expansion of (7) and retain the terms up to second-order,
∂ ∂ ∆t2 ∂ ∂ fi − fieq
∆t( + ci )fi + ( + ci )2 fi = − + Fi ∆t . (8)
∂t ∂x 2 ∂t ∂x τ
Then we expand the particle distribution function as
(1) (2)
fi = fieq + εfi + ε2 fi + ··· . (9)
Substituting (9) into (8), the equations with the first-order and second-order
of the small parameter ε are obtained as
(1)
∂fieq ∂f eq f
ε1 : + ci i = − i + Fi , (10)
∂t ∂x τ
   
eq (1) (1)
2 ∂fi ∂fieq 1 ∂fi ∂fi
ε : + ci +ε 1− · + ci
∂t ∂x 2τ ∂t ∂x

(1) (2)
+ εfi fi
=− + Fi . (11)
τ
The conservation of mass and momentum requires
 (n)  (n)
fi = 0, fi ci = 0, n = 1, 2, · · · . (12)
i i

Then summing (10) over i gives

 ∂f eq  ∂fieq f (1) 
i
+ ci =− i
+ Fi . (13)
i
∂t i
∂x i
τ i

From (6) and (12), we obtain the macro continuity equation (1). Next, we
multiply (11) by ci and sum over i giving
 ∂f eq      (1) (1)

∂fieq 1 ∂fi ∂fi
ci i
+ ci ci +ε 1− · ci + ci
i
∂t i
∂x 2τ i
∂t ∂x

 (1)
fi + εfi
(2) 
=− ci + ci Fi . (14)
i
τ i
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

288 J. B. Frandsen


Then we use (6), (12) and ci ci fieq = 12 gh2 + h u2 leading to the macro
i
momentum equation of the NLSW equation (1),
∂(hu) ∂(hu2 ) ∂ h2 ∂Fν
+ = −g ( ) + + Fi , (15)
∂t ∂x ∂x 2 ∂x

∂(h2 /2) 2
)
where Fν = ν · ∂(hu)
∂x + ∂t + 3 ∂(hu
∂t and the kinematic viscosity is
c2
ν= 3 (τ − 12 ).

3. Second-Order Finite Difference LB Model

We have compared the standard LB solution (7) with a second-order Finite
Difference (FD) LB Model in an attempt to improve the accuracy and sta-
bility of the predictions as well as the CPU efficiency. The numerical FD
scheme proposed consists of a spatial discretization of convection/collision
operator, force terms and a time integration of (3). For the spatial dis-
cretization we use the second-order accurate Lax-Wendroff approximation
of the convection and force terms,
∂fi (x, t) fi (x + ∆x, t) − fi (x − ∆x, t)
=
∂x 2∆x
∆t[fi (x + ∆x, t) − 2fi (x, t) + fi (x − ∆x, t)]
− ,
2∆x2
(16)
dhb hb (x + ∆x, t) − hb (x − ∆x, t)
=
dx 2∆x
∆t[hb (x + ∆x, t) − 2hb (x, t) + hb (x − ∆x, t)]
− ,
2∆x2
where the force term takes the averaged value of the two val-
ues at the lattice point and its neighboring lattice point Fi =
1
2 [Fi (x, t) + Fi (x + ∆x, t + ∆t)]. The force term representing the beach
slope is included in the streaming step. The discrete formulation for the
convection operation is
∂fi
Ci = ci · ∇fi = ci . (17)
∂x
There currently exists different suggestions to the discretization of the
collision operator, e.g. Mei and Shyy (1998); Guo and Zhao (2003). For
the argument of simplicity and preliminary testing, we used the sugges-
tion as described by Mei and Shyy (1998). They proposed a simple linear
extrapolation approach to calculate the fieq at a new time level t + ∆t ,
fieq,t+∆t = 2fieq,t − fieq,t−∆t . (18)
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 289

The collision operator Ωt+∆t

i can be expressed as
∆t t+∆t
Ωt+∆t
i =− [f − (2fieq,t − fieq,t−∆t )] . (19)
τ i
The mixed numerical scheme utilizes a second-order Runge-Kutta time
integrator. The time step update of the LBGK (3) reads
t+∆t/2 t+∆t/2 t+∆t/2
fit+∆t = fit + ∆t(−Ci + Ωi + Fi ), (20)
where
t+∆t/2 t+∆t/2 t+∆t/2 ∆t
Ci = ci · ∇fi ; fi = fit + (−Cit + Ωti + Fit )
2
(21)
t+∆t/2 ∆t t+∆t/2 eq,t+∆t/2
Ωi =− (fi − fi ).
τ
Substituting (17) and (19) into (20) yields the finite-difference LB equations
applied herein.

4. The Free-Surface Dynamics Treatment

It is notable that the present 1-D LBGK model does not include the con-
ventional dynamic and kinematic boundary conditions at the free-surface.
Instead, the physics of the free-surface dynamics are accounted for through
the non-equilibrium particle distribution function (f neq = fi − fieq ). So no
additional surface boundary conditions are necessary to be prescribed in
the present non-overturning numerical NLSW wave model. Furthermore, it
is essential to satisfy the constraints of the hydrodynamics moments (5) at
all times; the LB solvers equivalent means of conserving mass and momen-
tum. We should stress that the present model can only expect to work well
for a non-breaking surface in shallow water depths. Accounting for wave
breaking and/or other water depths would involve a different LB formula-
tion than the model presented herein. For example, an LB formulation in
arbitrary water depth could include a free-surface treatment as known from
traditional CFD solvers, e.g. a volume of fluid algorithm, a level set, etc. to
account for breaking. Therefore a host of models could potentially be devel-
oped to bridge the gap between micro and macroscopic fluid flow behavior
and thus potentially advance the way we currently model free-surface flows
and wave runup.

5. The Wet-Dry Interface and Other Boundary Treatments

The computational domain is defined over x ∈ [−500, 50500] m, discretized
with N lattices. Beyond the beach slope, at the shore boundary end of the
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

290 J. B. Frandsen

computational domain (x = −500 m), we have not prescribed any boundary

condition as the water level does not reach this level in the benchmark test
case 1 (see problem set-up in §7). A numerical damping zone is required
to avoid non-physical reflections at the open boundary (x = L). In the
LB method, the boundary conditions are described through distribution
functions fi . At x = L, we are required to calculate fi when i = N . To
handle reflections, we have extended the flow domain with a stretch of flat
bed from N = 50 km to Nd = 500 m to mimic some form of sponge layer.
We have adopted the simple extrapolation equation fi (Nd , t) = 2fi (Nd −
1, t) − fi (Nd − 2, t) of Yu et al. (2005) at x = 50, 500 m. It should be noted
that by imposing these constraints, the reflections are not suppressed. They
are only delayed so that the present solutions in the time duration of interest
are not affected. It should also be mentioned that this has not been dealt
with previously, that is, in the LB free-surface framework.
Regarding the wet-dry interface treatment, the “slot method” is one
technique to use in wave runup predictions, e.g. Kennedy et al. (2000).
The method assumes a porous beach allowing for the water level to be
below the beach level. Several other techniques were reviewed and evalu-
ated by, e.g. Balzano (1998); Prasad and Svendsen (2003). We have tested
two methods in simulating runup. The first method prescribes a thin film
(h = hs = 10−5 m) in dry areas (h = 0) along the slope of the beach.
The shoreline location is identified when ζ → hs . The second method in-
volves the use of a shoreline algorithm. We have chosen to apply the al-
gorithm of Lynett et al. (2002). Their moving boundary technique utilizes
linear extrapolation of the surface properties near the wet-dry boundary
in which the shoreline is traced on a fixed grid. In the following, we shall
refer to our results as predictions with or without a shoreline algorithm,
respectively.

6. Solution Procedure
Initially we prescribe the waveform of Fig. 3(b) together with the non-
physical values of the velocity u = 0 and the distribution functions fi = fieq .
We solve (7) using a non-splitting operator scheme for both the standard
LBGK and the FD scheme. The solution procedure is outlined  in Fig. 2.
The links relevant to the LB schemes herein are labeled: standard LBGK

and FD LB
, respectively. The hydrodynamic moments of the equilibrium
distribution functions of (5) are conserved at every time step. It is assumed
that the Mach number Ma = u/cs  1, where cs is the speed of sound. The
 August 26, 2008
17:43
WSPC/Trim Size: 9in x 6in for Review Volume
A 1-D Lattice Boltzmann Model
standard LBGK
1.17
FD scheme

1.7
1.20

16˙frandsen
291
Fig. 2. Solution procedure. Non-splitting operator scheme applies to the standard LBGK and FD LB solvers.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

292 J. B. Frandsen

Peclet number Pe = u ∆x/ν < 2 and the Courant number Cr = u ∆t/∆x <
1 are also satisfied.

7. Benchmark Problem 1 Set-up

In the following test case, we are concerned with testing the solver’s ability
to handle wave runup on beaches. The predictions of the shoreline trajec-
tory represent a classical bench mark test of numerical models, especially
because of the challenge of accurately predicting the wave motion when
the depths are vanishing into dry-states. The present test case represents
a tsunami wave generated runup/run-down study on a beach. The initial
shoreline location is located 50,000 m from starting point of the slope and
z0 = 5000 m. The slope of the beach is constant ∂hb /∂x = 1/10 and uniform
and spans the whole computational domain, as shown in Fig. 3(a).

4
3

z
ζ [m]

ζ
x
11111111111111111
00000000000000000
00000000000000000
11111111111111111 z0
00000000000000000
11111111111111111
h0
00000000000000000
11111111111111111
00000000000000000
11111111111111111
hb
00000000000000000
11111111111111111
00000000000000000
11111111111111111
−8.81
−10
00000000000000000
11111111111111111 0 8.25 20.55 35 50
L (a) x [km] (b)
Fig. 3. (a) Definition sketch of wave runup study. (b) Initial wave profile.

Initially the velocity in the flow domain is zero and the free-surface de-
scribed by the form of a leading depression N-wave shape, typically caused
by an offshore submarine landslide,
2 2
ζ = a1 e(−k1 (x−x1 ) ) − a2 e(−k2 (x−x2 ) )
and u = 0 at t=0 , (22)
where a1 = 13 a2 = 0.006, k1 = 19 k2 = 0.4444, x1 = 4.1209 and x2 = 1.6384,
after Carrier et al. (2003). The initial wave profile is also shown in Fig. 3(b).

8. Results
We shall, in the following, present 1-D wave runup results based on our stan-
dard LBE and second-order accurate FD LBM solvers. First we undertook
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 293

the standard LBE simulations without any shoreline algorithm. In these

analysis, a thin film is prescribed to treat the wet-dry interface, as men-
tioned. We observed relative good comparisons regarding the inundation
length (x) except for the maximum run-down location which was underes-
timated. The accuracy of the shoreline velocity (us ) was poor. The solver
breaks near the maximum run-down occurrence. The second-order FD LB
showed similar results although the accuracy of the maximum run-down
inundation length improved (and fewer nodes were required). To overcome
these difficulties, other options for the wet-dry interface treatment were ex-
plored, and, as mentioned, the shoreline algorithm of Lynett et al. (2002)
was implemented. The above simulations were repeated and the second-
order FD LB solutions were found to yield similar results as the semi-
analytical solutions of Carrier et al. (2003), as shown in Fig. 4. Using the
FD solver, we need about 5,000 nodes and about 1/2 hour on a standard
PC of today of 3 GHz to solve the wave runup problem of benchmark 1.
This is not as fast as a standard NLSW solver but still acceptable when
it comes to tsunami predictions. To support these findings, we display the
behavior of the macroscopic variables in Fig. 8. First, we show the grid sen-
sitivity study of the inundation length and shoreline velocity (Fig. 5). It is
the velocity component which is the governing variable of the required grid
resolution. Since the 5,000 and 6,000 node solutions yielded similar shore-
line motion behavior, the following results will be based on the 5,000 node
FD LB solver. Figures 6(a–d) to 7(a–d) show the FD LB time dependent
trajectories for the free-surface elevation and corresponding velocities. We
observe the highly nonlinear wave transformation patterns from the (a) off-
shore to (d) shoreline motion location. For example, Figs. 7(a–d) illustrate
further the highly nonlinear trajectories of the shoreline variables. The ones
offshore exhibit non-circular trajectories which are transformed drastically
to near disappearing ones, as the wave approaches the maximum runup
location. Figures 8(a–c) show the FD LB solutions with the second-order
NLSW and Boussinesq solutions produced by Pedersen (Chapter 17 in this
volume). Pedersen models are based on a time step of 0.09 s and about
3171 nodes uniformly distributed. The three selected time instances rep-
resent maximum runup (≈220 s), maximum run-down (≈175 s) and near
maximum run-down with maximum offshore velocity (≈160 s). Key data
from the models are outline in Table 1.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

294 J. B. Frandsen

200
x [m]

−200

100 160 175 220 280

t [s] (a)

10

0
us [m/s]

−10

−20
100 160 175 220 280
t [s] (b)

Fig. 4. Comparison of standard LBGK and FD LBM solutions. − · −, Standard LBGK;

−, FD LB solutions; − −, Carrier et al. (2003).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 295

230
200
x [m]

−166
−200
100 170 215 280
t [s] (a)

10
7.4

0
us [m/s]

−15.2

−20
100 161 181 215 280
t [s] (b)
Fig. 5. FD LB solutions. Grids: − −, 1000 nodes; −, 5000 nodes; − · −, 6000 nodes.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

296 J. B. Frandsen

15 15

10 10

5
5

0
0
ζ [m]

ζ [m]
−5
−5
−10

−10
−15

−15 −20

−20 −25
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] t [s]
(a) (b)
15 18

16
10

14
5
ζ [m]

ζ [m]
12

0
10

−5
8

−10 6
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] t [s]
(c) (d)
10 20

8
15
6
10
4
5
2
u [m/s]

u [m/s]

0 0

−2
−5
−4
−10
−6
−15
−8

−10 −20
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] t [s]
(e) (f)
25 30

20
20
15

10 10

5
u [m/s]

u [m/s]

0
0

−5 −10

−10
−20
−15

−20
−30
−25
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] t [s]
(g) (h)
Fig. 6. FD LB free-surface (a–d) elevation and (e–h) velocity solutions of wave form
transformation x ∈ [−166, 500] m where x = −166 m and x = +230 m represent maxi-
mum runup and run-down locations. (a) Offshore, (b–c) Approaching shoreline, (d) The
shoreline motion behavior towards maximum runup. (a,e) −, 500 m; − −, 400 m; − · −,
300 m. (b,f) −, 230 m; − −, 200 m; − · −, 150 m. (c,g) −, 100 m; − −, 50 m; − · −, 0 m.
(d,h) −, −50 m; − −, −100 m; − · −, −166 m.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 297

10 20

8
15
6
10
4
5
2
u [m/s]

u [m/s]
0 0

−2
−5
−4
−10
−6
−15
−8

−10 −20
−20 −10 0 10 20 −30 −20 −10 0 10 20 30
ζ [m] ζ [m]
(a) (b)
20 20

15 15

10 10

5 5
u [m/s]

u [m/s]
0 0

−5 −5

−10 −10

−15 −15

−20 −20
−30 −20 −10 0 10 20 30 −30 −20 −10 0 10 20 30
ζ [m] ζ [m]
(c) (d)

20 20

10 10
u [m/s]

u [m/s]

0 0

−10 −10

−20 −20

30 30
15 15
300 300
0 0
200 200
−15 100 −15 100
−30 0 −30 0
ζ [m] t [s] ζ [m] t [s]
(e) (f)

20 20

10 10
u [m/s]

u [m/s]

0 0

−10 −10

−20 −20

30 30
15 15
300 300
0 0
200 200
−15 100 −15 100
−30 0 −30 0
ζ [m] t [s] ζ [m] t [s]
(g) (h)
Fig. 7. FD LB velocity and free-surface trajectories of wave form transformation x ∈
[−166, 500] m where x = −166 m and x = +230 m represent maximum runup and
run-down locations. (a) Offshore, (b–c) Approaching shoreline, (d) The shoreline motion
behavior towards maximum runup. (a,e) −, 500 m; − −, 400 m; − · −, 300 m. (b,f) −,
230 m; − −, 200 m; − · −, 150 m. (c,g) −, 100 m; − −, 50 m; − · −, 0 m. (d,h) −, −50
m; − −, −100 m; − · −, −166 m.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

298 J. B. Frandsen

200

x [m]
0

−200

100 160 175 220 280

t [s] (a)
10

0
us [m/s]

−10

−20
100 160 175 220 280
t [s] (b)

20

t=220s
10
ζ [m]

0
t=175s
−10
t=160s

−20

−200 0 1000
x [m] (c)

Fig. 8. Snapshots of shoreline location movement. −, FD LB solutions; − −, Carrier

et al. (2003); · · ·, NLSW: Pedersen (Chapter 17); − · −, Boussinesq: Pedersen (Chapter
17).
 August 26, 2008
17:43
WSPC/Trim Size: 9in x 6in for Review Volume
Table 1. Model comparison of key data. LB solutions include shoreline algorithm of Lynett et al. (2002).

A 1-D Lattice Boltzmann Model

Lattice Boltzmann Lattice Boltzmann NLSW Semi-Analytical
Maximum Standard LBGK 2nd-Order FD (Pedersen (Chapter 17)) (Carrier et al. (2003))

Runup [m] −147 (t = 226 s) −166 (t = 215 s) −164 (t = 216 s) −164 (t = 217 s)
Run-down [m] +226 (t = 170 s) +230 (t = 170 s) +239 (t = 172 s) +242 (t = 173 s)
Shoreward velocity [m/s] −11.29 (x = 130 m) −15.22 (x, t) = (181 m,180 s) −17 (x = 176 m) −15.76 (x, t) = (174 m,177 s)
Offshore velocity [m/s] +7.09 (x = 179 m) +7.44 (x, t) = (161 m,162 s) +7.28 (x = 164 m) +7.28 (x, t) = (164 m,162 s)
No. of nodes 100,000 5,000 3,171
∆x [m] 0.5 10 15.78
∆t [s] 0.002 0.00015 0.09
τ [−] 0.732 0.5 —
CPU time [s] 6702 1975 ≈30

16˙frandsen
299
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

300 J. B. Frandsen

70 40

35
60
30
50
25
f [m]

f0 [m]
40 20
0

15
30
10
20
5

10 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (a) t [s] (b)
25 9

8
20 7

6
15
5
f0 [m]

f0 [m]
4
10
3

5 2

0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (c) t [s] (d)
−3 −4
x 10 x 10
2
5
1.5

1
0
0.5
f1 [m]

f1 [m]

−0.5 −5

−1

−1.5 −10

−2
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (e) t [s] (f)
−4 −4
x 10 x 10
4 1

2 0.5

0 0
f [m]

f [m]

−2 −0.5
1

−4 −1

−6 −1.5

−8 −2
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (g) t [s] (h)
Fig. 9. The temporal variations of the distribution function (a–d) f0 and (e–h) f1 for
x ∈ [−166, 500] m where x = −166 m and x = +230 m represent maximum runup and
run-down locations. (a,e) Offshore, (b–c, f–g) Approaching shoreline, (d,h) The shoreline
motion behavior towards maximum runup. (a,e) −, 500 m; − −, 400 m; − · −, 300 m.
(b,f) −, 230 m; − −, 200 m; − · −, 150 m. (c,g) −, 100 m; − −, 50 m; − · −, 0 m. (d,h)
−, −50 m; − −, −100 m; − · −, −166 m.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 301

0
(a) (b)
0

(c) (d)
−4 −4
x 10
5

0
[m]

feq
1
[m]

−5

−10

0 50 100 150 200 250 300 350

(e) t [s] (f)
−4 −4
x 10 x 10
4 1.5

2 1

0.5
0
0
feq [m]

feq [m]

−2
1

−0.5
−4
−1

−6 −1.5

−8 −2
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (g) t [s] (h)
Fig. 10. The temporal variations of the equilibrium distribution function (a–d) and f0eq
(e–h) f1eq . (a,e) Offshore, (b–c, f–g) Approaching shoreline, (d,h) The shoreline motion
behavior towards maximum runup. (a,e) −, 500 m; − −, 400 m; − · −, 300 m. (b,f) −,
230 m; − −, 200 m; − · −, 150 m. (c,g) −, 100 m; − −, 50 m; − · −, 0 m. (d,h) −, −50 m;
− −, −100 m; − · −, −166 m.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

302 J. B. Frandsen

−4 −4
x 10 x 10
1.5
4
1
2
0.5
0
f −feq [m]

f −feq [m]
0
0

0
−2
−0.5
0

0
−1 −4

−1.5 −6

−2 −8
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (a) t [s] (b)
−4 −4
x 10 x 10
1

1
0.5

0 0
f −feq [m]

[m]
−0.5
f0−feq
0

−1
0

−1

−2
−1.5

−2
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (c) t [s] (d)
−5 −4
x 10 x 10
10

8 2

6
1
4
f −feq [m]

[m]

2
f1−feq
1

0
1

−2
−1
−4

−6
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (e) t [s] (f)
−4 −5
x 10 x 10
2 10

8
1.5
6
1
4
f −feq [m]

f −feq [m]

0.5 2
1

1
1

0
0
−2
−0.5
−4

−1 −6
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
t [s] (g) t [s] (h)
Fig. 11. The temporal variations of the collision term (a–d) and (e–h) f1 − f1eq . f0 − f0eq
(a,e) Offshore, (b–c, f–g) Approaching shoreline, (d,h) The shoreline motion behavior
towards maximum runup. (a,e) −, 500 m; − −, 400 m; − · −, 300 m. (b,f) −, 230 m;
− −, 200 m; − · −, 150 m. (c,g) −, 100 m; − −, 50 m; − · −, 0 m. (d,h) −, −50 m; − −,
−100 m; − · −, −166 m.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 303

The FD LB and NLSW solution of Pedersen (Chapter 17) compares well

with the semi-analytical solution of Carrier et al. (2003) (as expected, since
they are all NLSW models) whereas the Boussinesq model seems more
inaccurate for the same grid refinement. Any discrepancy and difficulty
for all models were mainly related to the large accelerations at maximum
run-down. We looked further into the variation of the collision term, as
this would be a means of examining free-surface and shoreline nonlinearity.
Figures 9(a–d) to 11(a–d) show the time series of fi , fieq and fi − fieq
for the discrete microscopic velocities c0 and c1 . We can observe that f0
and f0eq follow similar patterns as the free-surface elevation ζ whereas f1
and f1eq tend to follow the macroscopic behavior of the velocity u. The

−4 −4
x 10 x 10
4 4

3 3

2 2

1 1
f −feq [m]

[m]

0 0
f1−feq
0

1
0

−1 −1

−2 −2

−3 −3

−4 −4
−200 0 200 400 600 800 −200 0 200 400 600 800
x [m] (a) x [m] (b)
Fig. 12. Snapshots of collision term variations for the discrete particle velocities (a) c0
and (b) c1 . −, t = 160 s; − · −, t = 175 s; − −, t = 220 s.

(a) (b)
Fig. 13. Distribution functions at the maximum runup location (t = 175 s), for (a) c0
and (b) c1 . −, fi ; − −, fieq ; − · −, (fi − fieq ).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

304 J. B. Frandsen

collision term fi − fieq does not follow any macroscopic patterns. Due to
the coarse grid, we can unfortunately not make any further interpretation in
terms of bridging the behavior of the mesoscopic and macroscopic variables.
Figure 12 shows the variations of the collision term (fi −fieq ) for the discrete
microscopic velocities ci at t = 160 s, 175 s, and 220 s. We can observe that
the deviations from equilibrium are largest in the run-down region near
the occurrence at maximum velocity. This is further confirmed in curves
of Fig. 13 and thus agrees with the observations of the macroscopic vari-
ables. Finally, we should note that the LB maximum velocity, maximum
runup and run-down occurred at slightly different instances than the semi-
analytical solution of Carrier et al. and thus the LB extreme values would
deviate from the solutions shown.

9. Other Beaches and Initial Wave Forms

We extended the benchmark test 1 to study the wave runup/run-down
effects when varying the (1) beach slope, and (2) initial wave form. The
steepest slope case (1/5) shows the smallest runup/run-down where the
opposite was the case for the flattest slope (1/20), as expected (Fig. 14).
We were somewhat surprised by the similar form of the inundation length

400
1/20
1/10
200
1/5
x [m]

−200

−400
0 100 200 300
t [s]

Fig. 14. Slope sensitivity. −, FD LB solutions; · · ·, NLSW: Pedersen (2007); − · −,

Boussinesq: Pedersen (2007).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 305

for all three slopes of the FD LB model. For example, one would expect
that the long wave models would be less accurate for relatively steep slopes,
however, the steepest slope of 1/5 we tested showed good agreement be-
tween the LB and the NLSW solutions. The largest discrepancy occurred
at maximum run-down. The grid convergence study of Pedersen (Chapter
17) showed no runup, i.e. breaking around maximum withdrawal where pre-
dicted with minor differences between the NLSW and Boussinesq solutions
when ∂hb /∂x = 1/20 whereas the LB model did predict shoreline motion
profile similar to the 1/10 case (just enlarged in shape). It is unclear why

10 20

10
us [m/s]
ζ [m]

0
−10

−4 −20
−0.05
0 5 7 100 152 210 280
x [m] 4
x 10 (a) t [s] (b)
4 20

10
0
us [m/s]
ζ [m]

−10

−10 −20
−0.05
0 5 7 100 166 182 280
x [m] x 10
4
(c) t [s] (d)
10 20

10
us [m/s]
ζ [m]

0
−10

−4 −20
−0.05
0 5 7 100 142 207 280
x [m] x 10
4
(e) t [s] (f)
Fig. 15. Shoreline velocity time history (b,d,f) based on initial wave forms (a,c,e). −,
FD LB solutions; −−, Carrier et al. (2003).
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

306 J. B. Frandsen

20 20

10 10
us [m/s]

us [m/s]
0 0

−10 −10

−20 −20
−400 −200 0 200 400 −400 −200 0 200 400
x [m] (a) x [m] (b)
20 20

10 10
u [m/s]

us [m/s]
0 0
s

−10 −10

−20 −20
−400 −200 0 200 400 −400 −200 0 200 400
x [m] (c) x [m] (d)
Fig. 16. Shoreline velocity versus shoreline location. (a) Initial Gaussian wave form,
case a, (b) negative Gaussian shape, case b, (c) leading depression N-wave, case c,
(d) waveform caused by submarine landslide, case d (bench mark test 1). −, FD LB
solutions; −−, Carrier et al. (2003).

Pedersen’s model solutions did not predict runup for the slope of 1/20. The
other beach slope cases showed similar trajectory effects on the shoreline
motion when comparing the LB with the NLSW and Boussinesq solutions
(Pedersen (Chapter 17)).
Finally, we tested the LB solver for other initial wave forms, after Car-
rier et al. (2003). We found fairly good agreement for the 5000 LB node
solutions, as shown in Figs. 15 and 16.

10. Concluding Remarks and Future LBM Endevours

We have tested two LB schemes which include a standard LBE formulation
and a second-order accurate FD LB model. We investigated two approaches
to treat the wet-dry interface: (1) a thin film layer on the beach, and (2) the
shoreline algorithm of Lynett et al. (2002). Our model is a 1-D idealization.
The FD LB scheme with the algorithm of Lynett et al. agreed well with
results of other investigators. This includes all four case studies, as described
by Carrier et al. (2003). We found that the FD LB solver (5000 nodes) is
competitive for tsunami predictions, as the CPU is about 30 min on a
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 307

single processor 3 GHz on a standard PC of today. The CPU of the LB

model is, however, not as fast as the NLSW/Boussinesq models presented
by Pedersen (Chapter 17). The LB model required higher resolution and
the additional velocity set in the luggage. The standard LBE solver with the
shoreline algorithm of Lynett et al. (2002) also performed reasonably well.
But is was less accurate and less efficient than the FD LB solver. About
100,000 nodes was required and the CPU was about 1 hour 50 min. None of
the LB schemes produced accurate velocity without a shoreline algorithm,
i.e. the thin film approach did not work well. We should note, however, that
the inundation length was agreeing well with other investigators’ solutions.
The present Boltzmann modeling approach may be a new competitive
candidate for predictions of free-surface water waves and nearshore pro-
cesses, especially where high density resolution would be required. Obvi-
ously unstructured lattices are highly desirable for the runup/run-down
problems and could be a simple explanation for not achieving better agree-
ment for the LB schemes without a shoreline algorithm. From “a big
picture” perspective, the LB solutions herein may be viewed as a special
finite-difference discretization of the depth-averaged transport equations.
The LB solutions herein do not offer anything new on the details of physi-
cal processes during runup/run-down on the beach, compared to traditional
NLSW models, due to the coarse and simple 1-D model. Like any other
numerical model the LB solutions should most likely be obtained on un-
structured fine resolution lattices in the near shore region. Reaching this
level of the LB development, we would be in a better position to conclude
whether or not this model approach would offer further insight into the
physics more than other current methods. But at least for now, one has
hopefully convinced the reader that a method with roots in gas dynamics
could be applied to water wave problems of this kind. Regarding future re-
search, we hope that it can work for breaking waves; the original idea/goal
of using the LB approach.
There are several other fundamental issues which need further investiga-
tion. First, the approximation of the collision operator should be improved
through a Multi-Relaxation-Time (MRT) scheme in which the relaxation
times for different kinetic modes are separated (τ is no longer constant).
This would offer realistic modes between particles and perhaps reveal the
deeper physics we are looking for. Second, grid efficiency in terms of ir-
regular denser resolution in the nearshore region could be achieved using
multi-block algorithm similar to that of Yu et al. (2002). Furthermore, there
exists several unresolved issues in the solutions presented both at the near
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

308 J. B. Frandsen

shore and at the open boundary. The near shore physics could be repre-
sented better, as the bed friction and sediment transport effects are not
included in our results. Nor is the effect of wave breaking considered. To
delay wave reflections at the open boundary, we adopted a simple extrapo-
lation equation at the open boundary and extended the flow domain with
500 m to mimic some form of sponge layer. Other algorithms could be
explored to improve accuracy and efficiency.
In a final remark, we should emphasize that our intention with the
LB modeling approach is to use it in local areas where physical details
are needed such as the nearshore regions. Therefore a realistic goal and
potential improvement to the current tsunami model literature would be
to propose a LB-model coupled with a macro-scale level model for large
domain tsunami model predictions.
We attempted to write-up the LB solutions of benchmark test case 1
as self-contained as possible. We realize, however, the shortcomings of ex-
planations, and would like to refer the reader to further discussions, details
and references about the LB method which are given in Chapter 5 in this
volume.

References
Balzano, A. (1998) Evaluation of methods for numerical simulation of wetting
and drying in shallow water flow, Coastal Engineering 34, pp. 83–107.
Bhatnagar, P., Gross, E. and Krook, M. A. (1954) A model for collision processes
in gases i: small amplitude process in charged and neutral one component
systems, Physics Review A 94, pp. 511–526.
Carrier, G. F., Wu, T. T. and Yeh, H. (2003) Tsunami run-up and draw-down on
a plane beach, Journal of Fluid Mechanics 475, pp. 79–99.
Frandsen, J. B. (2008) “Free-surface Lattice Boltzmann modeling in single phase
flows”, in Advanced Numerical Models for Simulating Tsunami Waves and
Runup, Advances in Coastal and Ocean Engineering, Vol. 10, eds. Liu, P.
L.-F., Yeh, H. and Synolakis, C. (World Scientific Publishing Co.).
Gombosi, T. I. (1994) Gaskinetic Theory (Cambridge University Press).
Guo, Z. and Zhao, T. S. (2003) Explicit finite-difference lattice Boltzmann method
for curvilinear coordinates, Physical Review E 67, 066709.
Junk, M., Klar, A. and Luo, L.-S. (2005) Asymptotic analysis of the Lattice
Boltzmann equation, Journal of Computational Physics 210, pp. 676–704.
Kennedy, A. B., Chen, Q., Kirby, J. T. and Dalrymple, R. A. (2000) Boussinesq
modeling of wave transformation. breaking and runup. I:1D, Journal of
Waterway, Port, Coastal, and Ocean Engineering 126(1), pp. 39–47.
Lynett, P. J., Wu, T.-R. and Liu, P. L.-F. (2002) Modeling wave runup with
depth-integrated equations, Coastal Engineering 46, pp. 89–107.
August 26, 2008 17:43 WSPC/Trim Size: 9in x 6in for Review Volume 16˙frandsen

A 1-D Lattice Boltzmann Model 309

Mei, R. and Shyy, W. (1998) On the finite-difference-based Lattice Boltzmann

method in curvilinear coordinates, Journal of Computational Physics 143,
pp. 426–448.
Pedersen, G. (2008) “A Lagrangian model applied to runup problems”, in
Advanced Numerical Models for Simulating Tsunami Waves and Runup,
Advances in Coastal and Ocean Engineering, Vol. 10, eds. Liu, P. L.-F.,
Yeh, H. and Synolakis, C. (World Scientific Publishing Co.).
Prasad, R. S. and Svendsen, I. A. (2003) Moving shoreline boundary condition
for nearshore models, Coastal Engineering 49, pp. 239–261.
Succi, S. (2001) The Lattice Boltzmann Equation for Fluid Dynamics and Beyond
(Oxford University Press).
Yu, D., Mei, R. and Shyy, W. (2002) A multi-block Lattice Boltzmann method for
viscous fluid flows, International Journal for Numerical Methods in Fluids
39, pp. 99–120.
Yu, D., Mei, R. and Shyy, W. (2005) Improved treatment of the open boundary in
the method of the Lattice Boltzmann equation, Progress in Computational
Fluid Dynamics 5(1/2), pp. 3–12.
This page intentionally left blank
August 26, 2008 17:46 WSPC/Trim Size: 9in x 6in for Review Volume 17˙Geir

CHAPTER 17

A LAGRANGIAN MODEL APPLIED TO

RUNUP PROBLEMS

G. Pedersen
Mechanics Division, Department of Mathematics, University of Oslo
E-mail: [email protected]

1. The Models
The main runup model of this section is based on Lagrangian Boussinesq
equations that are fully nonlinear and possess standard dispersion proper-
ties. It is run in both hydrostatic (NLSW) and dispersive mode. For com-
parison we employ also an Eulerian FDM code for the standard Boussinesq
equations and a Boundary Integral Method (BIM) for full potential the-
ory. The first is used for linear computations and is without any particular
runup feature; runup heights are found from the vertical elevation at the
shore. The BIM is based on Cauchy’s formula, Lagrangian surface nodes
and spline interpolation. All models allow for a variable spatial resolution,
but this feature is employed only for the BIM and the Eulerian models.
Systematic grid refinement is employed for all applications reported.
Expressed in terms of the depth averaged velocity, u, and the total
depth, H the Lagrangian long wave equations read
∂x ∂u H ∂H ∂h
H = H0 (a), (1−r1 )= −g +g −r2 +S1 +S2 (b),
∂a ∂t H0 ∂a ∂x
3
∂ h ∂ h ∂u ∂ 2 h ∂H
2 2
∂3h
 
1 ∂ h 1
S1 = H 2 , S2 = − 2u + 2 − +Hu ,
2 ∂t ∂x g ∂t∂x ∂t ∂t ∂t∂x ∂t ∂t∂x2
(1)
where a is the Lagrangian coordinate, h is the equilibrium depth and
H0 = H(a, 0). The position x relates to the velocity according to ∂x/∂t = u.
Only the shallow water (hydrostatic) terms are spelled out in (1b). A more
complete description of the equations in absence of a slide, including the

311
August 26, 2008 17:46 WSPC/Trim Size: 9in x 6in for Review Volume 17˙Geir

312 G. Pedersen

dispersion terms (r1 and r2 ), is found in Jensen et al. (2003)2 and refer-
ences therein. The principal forcing due to a time dependent bottom is
the modified source of horizontal momentum from the second term on the
right hand side of (1b). Higher order source terms, S1 and S2 , have been
included particularly for benchmark 3. Replacing H by h in S1 we obtain
the linear part that is employed in the Eulerian Boussinesq equation, while
S2 is exclusively nonlinear.
A moving shoreline is associated with a fixed a, where the condition
H = 0 is employed. The set (1) is solved by finite differences on a grid
that is staggered in space and time. The hydrostatic (NLSW) version is
explicit, while dispersion requires implicitness. No smoothing or filtering is
employed with the long wave models.

Bouss(NL) ref.
ref. 220s NLSW
NLSW
Bouss(L)
SW(L)
BIM(NL)
175s
160s

t/sec x/km
Fig. 1. Left: withdrawal (m). Right: surface (m) at selected times.

2. Benchmark Problem 1
The reference solution1 exists in a semi-infinite domain, but have been
made available until 50 km from the coast. We add a deep water region
(x > 50 km) to avoid that reflections from the seaward boundary reach the
shore region for t < 280 s. The alternatives are open boundary conditions
or sponge layers that may introduce additional errors and uncertainties.
The inundation lengths shown in Fig. 1 are all close to convergence
(increment ∆a = 16 m for the nonlinear long wave models). The NLSW
solution agrees very well with the reference solution (ref.). Dispersion has
some effect and reduces the draw-down and increases the runup. The maxi-
mum runup and drawdown are nearly equal for linear and nonlinear models.
August 26, 2008 17:46 WSPC/Trim Size: 9in x 6in for Review Volume 17˙Geir

A Lagrangian Model Applied to Runup Problems 313

Since the initial condition inherits very small elevation/depth ratios this is
to be expected in the hydrostatic approximation4 . That also the linear
(L) and nonlinear (NL) Boussinesq models yield nearly identical extrema
indicates that dispersion and nonlinearity are important in the deep and
shallow regions, respectively, with no or little overlap.
The Boussinesq and full potential (BIM) models yield nearly identi-
cal solutions near shore, while there are small discrepancies in the waves
propagating into deeper water (not shown).
From a modeling point of view the NLSW solution is the most inter-
esting (challenging) one. At maximum drawdown there is a (near) cusp on
the inundation curve corresponding to huge accelerations, a solution that
is close to being “weak” and slow convergence of the numerical model. In
Fig. 2 we observe that the accuracy for the velocity is much poorer than
for the inundation length. The convergence of the numerical solution is
particularly slow close to t = 173 s where the reference solution is slightly
multi-valued. These features have not been pursued further. Moreover, for
the maximum withdrawal the NLSW model displays a linear convergence
except for extremely fine resolutions. For the smooth Boussinesq solution
we obtain quadratic convergence (dashes in right panel).

Bouss(NL)
15.8 NLSW
7.9 ref.
3.9
2.5
252
63 ref.
16
8
ref.

t/sec ∆a (m)
Fig. 2. Left and mid panels: shoreline velocities (m/s). NLSW computations, marked by
∆a (m), are compared to the reference solution. Mid panel: blow-up of region (172, 176)×
(−18, 5). Right: Convergence of maximum withdrawal (m).

3. Benchmark Problem 3
In this problem3 a slide with time dependent shape and acceleration pen-
etrates the water. All results for this benchmark is given in meters and
August 26, 2008 17:46 WSPC/Trim Size: 9in x 6in for Review Volume 17˙Geir

314 G. Pedersen

p
dimensionless time units, δ/gµ2 , as defined in the reference. There is a
singularity at the rear of the slide body (x = 0), while it is undefined for
x < 0. Hence, if the fluid reaches x = 0 before the height of the slide body
becomes negligible at this point, conceptual and practical problems arise in
the numerical solution. For case A this occur before the first runup max-
imum. This is not so for case B for which the NLSW and Boussinesq(∗ )
equations yield:

∆a 0.16∗ 0.16 0.31 0.63 2.51 5.04 10.17

Runup 0.495 0.495 0.495 0.496 0.491 0.458 0.468

Drawdown −0.235 −0.233 −0.231 −0.226 −0.207 −0.221 −0.318

The solutions for the different models are shown in Fig. 3. For case A and
t = 1.5 the reference solution agree closely with the nonlinear solutions ex-
cept for a small discrepancy very close to the shore. The differences between
the linear and nonlinear solutions are significant for case B at t = 2.5, while
effects of dispersion are visible only at the front of the offshore propagating
wave. The Eulerian model SW(L) is close to the reference solution for both
cases (difference up to 0.0025 m with shoreline resolution ∆x = 15 m for
case A and t = 1.5). For case B inclusion of the dispersive source term, S1 ,

Case A, t = 1.5 Case B, t = 2.5

ref. ref.
Bouss(NL) Bouss(NL)
NLSW NLSW
SW(L) SW(L)

x/m x/m
Fig. 3. η (m); comparison of models.

reduces the maximum height of the outgoing wave from 1.42 m to 1.33 m at
t = 4.5. S2 then increases the height by 0.01 m. For case A and t = 1.5 the
term S1 reduces the wave height from 0.698 m to 0.692 m, while the effect
of S2 is negligible (less than 10−5 m).
August 26, 2008 17:46 WSPC/Trim Size: 9in x 6in for Review Volume 17˙Geir

A Lagrangian Model Applied to Runup Problems 315

4. Remarks
Lagrangian (and ALE) models are undoubtedly useful for idealized stud-
ies, as benchmarks 1 and 3, where they produce very accurate results.
However, such models have poorer prospects for complex tsunami studies.
Further discussion and references on related models are given in Chapter 1
in this volume (Pedersen)4 .

References
1. G. F. Carrier, T. T. Wu and H. Yeh, J. Fluid Mech. 475, 79 (2003).
2. A. Jensen, G. Pedersen and D. J. Wood, J. Fluid. Mech. 486, 161 (2003).
3. P. L.-F. Liu, P. Lynett and C. E. Synolakis, J. Fluid Mech. 478, 101 (2003).
4. G. Pedersen, Modeling runup with depth integrated equation models, in
Advanced Numerical Models for Simulating Tsunami Waves and Runup, eds.
P. L.-F. Liu, H. Yeh and C. Synolakis (World Scientific Publishing Co., 2008),
p. 3–41.
This page intentionally left blank
April 3, 2008 3:12 WSPC/Trim Size: 9in x 6in for Review Volume divider

Appendix
The following article was omitted from Volume 9 of the review series,
entitled “PIV and Water Waves” (eds. J. Grue, P. L.-F. Liu and G. K.
Pedersen).
This page intentionally left blank
 APPENDIX

PHASE-AVERAGED TOWED PIV MEASUREMENTS FOR

REGULAR HEAD WAVES IN A MODEL SHIP
TOWING TANK

Joe Longo, Jun Shao, Marty Irvine, Lichuan Gui and Fred Stern
IIHR Hydroscience and Engineering
300 S. Riverside Drive
The University of Iowa, Iowa City, IA, USA 52242
E-mail: [email protected]

Phase-averaged towed PIV measurements are made for regular head

waves in a model ship towing tank using both fixed and towed PIV
systems in preparation for tests with a model ship. Data reduction and
uncertainty assessment software are developed and tested for calm water
and long-wave, low steepness conditions used in previous forces,
moment, and wave pattern tests with surface combatant. The results are
validated through comparisons with progressive wave theory.

1. Introduction
Focus of experimental ship hydrodynamics research is moving into
unsteady viscous flows in support of unsteady Reynolds-averaged Navier-
Stokes (RANS) code development for simulation-based design. The
forward-speed diffraction problem1, i.e. restrained body advancing in
regular headwaves, is a building block problem towards ultimate goal of
physical understanding and simulation of viscous nonlinear seakeeping
and 6DOF maneuvering. The present study is precursory for measurement
of the unsteady nominal wake of a naval combatant and provides
documentation of the data-acquisition and reduction procedures along
with detailed data of the incident headwave and comparisons with 2D
progressive-wave theory.

319
320 J. Longo et al.

2. Test Design
The tests are conducted in the 100 × 3 × 3 m IIHR tank, which is equipped
with a drive carriage and trailer, wavemaker, and moveable wave
dampeners.

(a) (b)
Fig. 1. Experimental setup for towed, unsteady PIV measurements.

The DANTEC, towed, 2D PIV measurement system2 is linked to a

servo wave gage positioned either directly over or upstream of the
measurement area to phase lock the PIV measurements (Fig. 1(a)). The
system is configurable to measure in vertical (xz; Fig. 1(b)) and horizontal
(xy) planes. A separate measurement system is used for carriage speed.
All tests are performed with Uc = 1.53 m/s and without Uc = 0 m/s forward
speed for cases with and without waves. Headwave parameters are
wavelength λ = 4.572 m, frequency fw = 0.584 Hz, and steepness Ak =
0.025 and frequency of encounter fe = 0.922 Hz, which are all based on
previous studies2,3,4. Measurement area dimensions are 192 × 1018 pixels
or 14.3 × 74.9 mm. PIV image pairs are taken at 133 ms intervals with ∆t =
490 µs between images. Data is acquired at z = −25.0, −53.34, −110.45 mm
and z = −25.0 mm for xz and xy configurations, respectively. Datasets are
acquired at each elevation and configuration through repeated carriage
runs, stockpiling 1200 and 2000 vector maps for steady and unsteady
cases, respectively, to obtain convergence. Data-reduction software was
developed to perform harmonic data analysis and includes procedures for
phase-sorting a dataset, two-stage filtering for spurious vectors, 5th order
least-squares curvefitting to filtered data, and 2nd order Fourier series
analysis of least squares. Uncertainty assessment (UA) follows standard
procedures5.
 Phase-Averaged Towed PIV Measurements 321

0.000 0.000 0.000 0.000

-0.010 -0.010 -0.010 -0.010

z

z
-0.020 -0.020 -0.020 -0.020

-0.030 -0.030 -0.030 -0.030

0.930 0.940 0.930 0.940 0.930 0.940 0.930 0.940

x x x x
t/T=0 t/T=0.25 t/T=0.50 t/T=0.75

Fig. 2. Vector field for four time instances in one encounter period.

3. Results and Comparison Theory

Figure 2 illustrates experimental wave elevation and vector fields for
z = −25.0 mm at four instances in one encounter period (T ). Expected
wave-induced velocities are displayed and correlate with free surface
elevation. Quantitative validation is performed through comparisons
of harmonic variables of 2D progressive-wave theory and experiment.
For z = −25.0 mm, variables are averaged through the measurement area
and experiment is subtracted from theory and expressed as a percentage.
Average level of agreement for 0th and 1st harmonic amplitude is 0.3%
and 0.8% for U, respectively, and 0.1% and 0.8% for W, respectively. For
1st harmonic phase, agreement is 1–2% for both U and W which is
roughly 3–7°. Average turbulence is 0.01% (i.e. 0th harmonic uu0, ww0,
uw0), however, 1st harmonic amplitudes of turbulence are small but
significant (0.005%), near the location of the incident wave trough.

4. Future Work
Future work consists of steady and unsteady PIV measurements and UA of
the nominal wake plane of model 55126 for the same conditions presented
herein. The data and UA will be archived at www.iihr.uiowa.edu/~towtank.
322 J. Longo et al.

Acknowledgments
This research was sponsored by the Office of Naval Research under Grant
N00014-01-1-0073 under the administration of Dr. Pat Purtell.

References
1. H. Rhee and F. Stern, Int. J. Num. Meth. Fluids, 37 (2001) 445.
2. L. Gui, J. Longo and F. Stern, Exp. Fluids, 31 (2001) 336.
3. L. Gui, J. Longo, B. Metcalf, J. Shao and F. Stern, Exp. Fluids, 31 (2001) 674.
4. L. Gui, J. Longo, B. Metcalf, J. Shao and F. Stern, Exp. Fluids, 32 (2002) 27.
5. Test uncertainty: Instruments and apparatus, ASME PTC (Performance Test Code)
19.1–1998 (1998), 112.
6. J. Longo, J. Shao, M. Irvine and F. Stern, 24th Symposium on Naval Hydrodynamics
(2002).