See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325640636

Performance Comparison of NAMI DANCE and FLOW-3D® Models in Tsunami

Propagation, Inundation and Currents using NTHMP Benchmark Problems

Article  in  Pure and Applied Geophysics · June 2018

DOI: 10.1007/s00024-018-1907-9

CITATIONS READS

2 326

2 authors:

Deniz Velioglu Sogut Ahmet Cevdet Yalciner

Stony Brook University Middle East Technical University
30 PUBLICATIONS   71 CITATIONS    204 PUBLICATIONS   2,256 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

NEAREST_project View project

SEAHELLARC View project

All content following this page was uploaded by Deniz Velioglu Sogut on 06 September 2018.

The user has requested enhancement of the downloaded file.

Pure Appl. Geophys.
Ó 2018 Springer International Publishing AG, part of Springer Nature
https://doi.org/10.1007/s00024-018-1907-9
Performance Comparison of NAMI DANCE and FLOW-3DÒ Models in Tsunami
Propagation, Inundation and Currents using NTHMP Benchmark Problems
DENIZ VELIOGLU SOGUT1,2
Abstract—Field observations provide valuable data regarding
nearshore tsunami impact, yet only in inundation areas where
tsunami waves have already flooded. Therefore, tsunami modeling
is essential to understand tsunami behavior and prepare for tsunami
inundation. It is necessary that all numerical models used in tsu-
nami emergency planning be subject to benchmark tests for
validation and verification. This study focuses on two numerical
codes, NAMI DANCE and FLOW-3DÒ, for validation and per-
formance comparison. NAMI DANCE is an in-house tsunami
numerical model developed by the Ocean Engineering Research
Center of Middle East Technical University, Turkey and Labora-
tory of Special Research Bureau for Automation of Marine
Research, Russia. FLOW-3DÒ is a general purpose computational
fluid dynamics software, which was developed by scientists who
pioneered in the design of the Volume-of-Fluid technique. The
codes are validated and their performances are compared via ana-
lytical, experimental and field benchmark problems, which are
documented in the ‘‘Proceedings and Results of the 2011 National
Tsunami Hazard Mitigation Program (NTHMP) Model Bench-
marking Workshop’’ and the ‘‘Proceedings and Results of the
NTHMP 2015 Tsunami Current Modeling Workshop’’. The vari-
ations between the numerical solutions of these two models are
evaluated through statistical error analysis.
Key words: Tsunami, depth-averaged shallow water,
Reynolds-averaged Navier–Stokes, benchmarking, NAMI
DANCE, FLOW-3DÒ.
1. Introduction
Coastal regions have always been attractive set-
tling grounds for human populations since they have
provided a great deal of marine resources and
opportunities for transportation and trade. Today, a
1
Middle East Technical University, 06800 Ankara, Turkey.
E-mail: deniz.velioglusogut@stonybrook.edu; yalciner@metu.
edu.tr
2
Present Address: 1212 Computer Science, Department of
Civil Engineering, Stony Brook University, Stony Brook, NY
11794, USA.
Pure and Applied Geophysics
and AHMET CEVDET YALCINER1
large part of the world’s population inhabits coastal
areas, and the population density on these regions is
still growing rapidly. However, coastal environments
are dynamic natural systems and they pose a great
challenge to human habitation since they are prone to
water-related hazards such as floods, storms and
tsunamis, tsunamis being one of the most devastating
hazards related to these dynamics.
Theoretical approaches taken towards the near-
shore tsunami motion are difficult to apply due to
strong nonlinearities of equations that govern tsunami
behaviors, the three dimensionality of the flow, and
the turbulence that develops due to shoaling effects.
Numerical modeling has proved to be an accurate and
useful method of modeling tsunami inundation on a
coastline with the rapid development of computing
technology. Over the past several decades, many
numerical models have been developed to compute
wave transformation from deep water to shallow
water (Liu 1994; Wu 2001). The computational cost
of three-dimensional (3D) numerical models, which
require the numerical solution of the full Navier–
Stokes equations, is higher than that of two-dimen-
sional (2D) numerical models, which are based on the
shallow water approximation. Determining which
model would be suitable for a particular problem
requires knowledge and experience.
It is necessary that all numerical models used in
tsunami emergency planning be validated and veri-
fied. Validation/verification of a numerical code must
never cease; even proven models must be tested
continuously as new knowledge and data are acquired
(Synolakis et al. 2008). Careful and explicit valida-
tion/verification of tsunami models through the
comparison of their predictions with benchmark
analytical solutions, laboratory experiments and field
 D. V. Sogut and A. C. Yalciner
measurements has helped them evolve in the last two
decades. A numerical code that has performed well in
all benchmark tests will not necessarily make realistic
inundation predictions in every instance. However, in
the results of validated/verified codes, the level of
uncertainty is largely reduced to the uncertainty in the
geophysical initial conditions.
This study focuses on two numerical tools; NAMI
DANCE and FLOW-3DÒ. The codes are applied to
analytical, experimental and field benchmark prob-
lems for validation and performance comparison.
These problems are capable of identifying the short-
comings of numerical models in terms of tsunami
motion and inundation parameters such as water
surface elevation, current velocity and run-up. A total
of six benchmark problems are considered in this
study. The first three benchmark tests were proposed
in the ‘‘Proceedings and Results of the 2011 NTHMP
Model Benchmarking Workshop’’ (NTHMP 2012).
The rest of the benchmark problems was documented
in the ‘‘Proceedings and Results of the NTHMP 2015
Tsunami Current Modeling Workshop’’ (NTHMP
2015). The first benchmark problem (BMP 1) covers
analytical and experimental investigations of the
evolution, propagation and run-up of a single solitary
wave climbing up a plane beach. Benchmark problem
2 (BMP 2) is composed of physical model studies
which were carried out to investigate the run-up of a
solitary wave on a conically shaped island and to
provide a better understanding of three-dimensional
tsunami run-up. An extreme run-up height of 31.7 m
was measured near the village of Monai, in Okushiri
Island after the 1993 Okushiri tsunami. Benchmark
problem 3 (BMP 3) provides the data of a 1:400 scale
laboratory experiment of the Monai run-up. Bench-
mark problem 4 (BMP 4) describes a series of
experiments that analyze the transformation of a
single solitary wave as it propagates up a triangular-
shaped shelf with an island feature located at the
offshore point of the shelf. Benchmark problem 5
(BMP 5) covers a series of experiments having a
single long period wave that propagates up a piece-
wise linear slope and onto a small-scale model of the
town of Seaside, Oregon. The last benchmark prob-
lem, Benchmark problem 6 (BMP 6), is a case study,
consisting of a field data set of the Japan 2011 tsu-
nami recorded in Hilo Harbor, Hawaii. This study is
Pure Appl. Geophys.
also the first in-depth study to employ such a large
number of benchmark problems to validate the
commercial numerical model FLOW-3DÒ in tsunami
wave evolution, propagation and inundation.
2. Recent Studies in the Literature
This section summarizes the most prominent and
recent studies which are conducted using NAMI
DANCE and FLOW-3DÒ. A more detailed literature
review can be found in Velioglu (2017).
Yalciner et al. (2005) issued a report about the
field survey results concerning run-up, arrival time
and damages of the 2004 Indian Ocean tsunami. In
addition, they simulated the 2004 Indian Ocean tsu-
nami using NAMI DANCE and compared the
numerical results with the observed field data of other
tsunami survey teams at Sumatra, Thailand, Sri
Lanka, India and Maldives. Zaitsev et al. (2009)
presented the data of the instrumental registration of
the tsunami on Sakhalin and Hokkaidō islands as well
as performing numerical simulations of the tsunami
propagation in the Tatar Strait using NAMI DANCE
software. They found that the numerical results were
in satisfactory agreement with the observed data.
Yalciner et al. (2014) conducted a numerical mod-
eling study, whose main focus was the western part of
Peloponnese in Greece, in the framework of Euro-
pean Union funded project, SEAHELLARC. In the
study, simulations of possible tsunami source sce-
narios for the Pylos, Kyparissia, Filiatra and
Zakynthos regions, located at the western part of
Greece, were carried out using NAMI DANCE to
estimate extreme but possible tsunami wave effects in
these regions. Necmioglu and Ozel (2014) studied the
sensitivity of earthquake source parameters in tsu-
nami generation and propagation, focusing on the
Eastern Mediterranean, using NAMI DANCE. Patel
et al. (2016) modeled the Makran 1945 tsunami using
NAMI DANCE for the coasts of Pakistan, Iran, India
and Oman. The results were consistent with the
available reports and published data.
Choi et al. (2007) employed FLOW-3DÒ to sim-
ulate wave run-up on a conical island, based on the
unexpectedly large tsunami run-up elevations which
had been observed in the circular Babi Island in the
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models
1992 Flores tsunami and in the pear-shaped Okushiri
Island in the 1993 Okushiri tsunami. They maintained
that the results obtained using two- and three-di-
mensional models were in good agreement with each
other as well as the laboratory data of the physical
model studies conducted earlier by Briggs et al. in
1995. Basu et al. (2009) assessed the impulse waves
generated by the subaerial landslide at Lituya Bay,
Alaska by conducting numerical simulations. FLOW-
3DÒ software was used to track the free surface and
shoreline movements. The simulation results were
compared with the experimental ones obtained in a
physical model of a real case and they were found to
be in good agreement with the measured data. It was
concluded that the model captured the basic flow
properties which were associated with the time-de-
pendent evolution of the flowfield as the landslide
interacts with the bay. Pelinovsky et al. (2013) cited
both the highest wave run-up which was recorded at
Lhok Nga Twin Peaks after the 2004 Sumatra–An-
daman earthquake and the 2011 earthquake, which
generated extremely destructive tsunami waves up to
37.9 m in height at the tiny fishery port Koborinai
situated north of Miyako City in the Iwate Prefecture.
The researchers reproduced these extremely large
run-ups of tsunami waves in both cases via FLOW-
3DÒ and discovered that the water flow that climbed
the coast featured a strong vertical velocity compo-
nent. Kim et al. (2015) reproduced the dynamics of
extreme run-up height, which was measured as
31.7 m at Monai Valley as a result of the 1993
Hokkaidō Nansei-Oki tsunami. They used FLOW-
3DÒ to investigate the role of diffraction in the
extreme run-up. The results revealed that the shape
and layout of the coast line as well as the diffraction
of tsunami waves by the two islands located in front
of the valley had led tsunami waves to converge,
leading to the extreme run-up height. Wu et al. (2016)
focused on the generation of solitary waves in a wave
flume using a piston-type wavemaker. They made
experimental observations to evaluate the stability of
the generated solitary waves, defining stability as the
generated solitary wave travelling a longer distance
without an apparent decay. They carried out numer-
ical simulations using FLOW-3DÒ to support their
experimental results.
At this point, other studies on NTHMP tsunami
benchmark problems should also be referred to. A
number of researchers have performed similar work
on NTHMP benchmark problems with their particular
models. A case in point is Apotsos et al. (2011),
Nicolsky et al. (2011), Tolkova (2014), Horrillo et al.
(2015), Lynett et al. (2017) and Macias et al. (2017).
Furthermore, the proceedings of NTHMP (2012) and
NTHMP (2015) are also good resources that describe
the detailed work of other authors.
3. Theoretical Framework of the Numerical Models
NAMI DANCE is an in-house tsunami numerical
model developed by the Ocean Engineering Research
Center (OERC) of Middle East Technical University
(METU), Turkey and Laboratory of Special Research
Bureau for Automation of Marine Research, Far
Eastern Branch of Russian Academy of Sciences,
Russia. NAMI DANCE gained importance and pop-
ularity in a short span of time in the area of coastal
modeling and was recommended by the United
Nations Educational, Scientific and Cultural Organi-
zation (UNESCO) to be used in the simulations of
possible tsunami scenarios.
NAMI DANCE solves nonlinear forms of non-
dispersive two-dimensional depth-averaged shallow
water (2D-NSW) equations. Navier–Stokes equa-
tions, which are the governing equations in the
prediction of fluid motion, form the basis of long
wave/shallow water theory (Shuto 1991), in which
the vertical motion of water particles is not taken
into consideration because of its negligible effect
on the pressure distribution. Therefore, the pressure
distribution is accepted to be hydrostatic. Given
this approximation, necessary dynamic and kine-
matic conditions are used and the bottom friction
terms (for nonlinearity) are included so that the
fundamental equations of NAMI DANCE are
obtained and are discretized by means of the
staggered leapfrog scheme. The equations of con-
tinuity (conservation of mass) and momentum are
given by the following set of equations (Velioglu
et al. 2016):
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

og oM oN conversion coefficients, RDIF is a turbulent diffusion

þ þ ¼ 0; ð1Þ
ot ox oy term and RSOR is a mass source. The equations of
    motion are the Navier–Stokes equations with some
oM o M 2 o MN og additional terms (Flow Science 2002):
þ þ þ gD
ot ox D oy D ox  
2 p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi ð2Þ ou 1 ou ou ou Ay v 2
n þ uAx þ vAy R þ wAz n
þ M M 2 þ N 2 ¼ 0; ot VF ox oy oz xVF
7=3
D 1 oP RSOR
    ¼ þ Gx þ fx  bx  ðu  uw  Dus Þ;
oN o MN o N2 q ox qVF
þ þ ð5Þ
ot ox D oy D
ð3Þ  
og n2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi ov 1 ov ov ov
þ gD þ N M 2 þ N 2 ¼ 0; þ uAx þ vAy R þ wAz
oy 7=3 ot VF ox oy oz
D  
Ay uv 1 oP
where x and y are the horizontal x and y axes in the þn ¼ R þ Gy ð6Þ
xVF q oy
Cartesian coordinate system, z is the vertical axis in
RSOR
the Cartesian coordinate system, t is the time, g is the þ f y  by  ðv  vw  Dvs Þ;
gravitational acceleration, n is the Manning’s qVF
 
roughness coefficient, g is the free surface elevation, ow 1 ow ow ow
M and N are the discharge fluxes on horizontal plane þ uAx þ vAy R þ wAz
ot VF ox oy oz
along x and y axes, sij is the normal or tangential 1 oP RSOR
¼ þ Gz þ f z  bz  ðw  ww  Dws Þ;
shear stress in i direction on the j normal plane, and D q oz qVF
is the total water depth given by D = g þ d. ð7Þ
FLOW-3DÒ, which was developed and com-
mercialized by Flow Sciences, Inc., is a general where P is the pressure, Gx , Gy and Gz are the body
purpose CFD software. Flow Sciences, Inc. was accelerations, f x , f y and f z are the viscous accelera-
founded in Santa Fe, New Mexico U.S. in 1980 by tions, bx , by and bz are the flow losses across porous
Dr. C. W. (Tony) Hirt. Dr. Hirt was one of the media, uw , vw and ww represent the velocity of the
scientists who pioneered in the design of the Vol- source component and us , vs and ws represent the
ume-of-Fluid (VOF) technique, which was first fluid velocities at the surface of the source relative to
reported in Nichols and Hirt (1975), and more the source itself.
completely in Hirt and Nichols (1981). FLOW-3DÒ FLOW-3DÒ uses the quadratic law for bottom
is a direct descendant of this work. The model shear stress calculation for turbulent flow (Flow
solves the fully three-dimensional mass continuity Science 2002):
* *
and momentum (Navier–Stokes) equations via the sb ¼ qCD u u; ð8Þ
finite difference or finite volume method in the *
Cartesian coordinate system (x, y, z) to describe where u is the depth-averaged velocity and CD is the
fluid motion. The general mass continuity equation drag coefficient.
is given by (Flow Science 2002): The default value of CD is 0.0026, which is a
typical value in hydraulic and coastal engineering
oq o o 
applications. However, it is also possible to define
VF þ ðquAx Þ þ R qvAy
ot ox oy component specific CD or let the code to calculate CD
ð4Þ
o quAx for a component based on its surface roughness. For
þ ðqwAz Þ þ n ¼ RDIF þ RSOR ;
oz x the latter option, the surface roughness, flow depth,
where q is the fluid density, u, v and w are the water and drag coefficient are related by (Flow Science
particle velocities in x, y and z directions, VF is the 2002):
fractional volume, Ax , Ay , Az are the fractional areas
in x, y, z directions, R and n are the coordinate
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 1
A sketch of the 1:19.85 sloping beach connected to a constant depth region (distorted scale) (Horrillo et al. 2015)

" #2 the maximum and minimum values of the observed

j
CD ¼ k  ; ð9Þ
0:71 þ ln 30d
s
where j is the von Kármán constant, which is equal
to 0.40 and ks is the Nikuradse equivalent surface
roughness, which can be related to Manning’s
roughness coefficient by (Yen 1991):
n 6 MAX ¼
ks ¼ : ð10Þ
0:0389
4. Benchmarking
Relating the numerical results and benchmark
data within a statistical framework is a reasonable
way to assess the capability of numerical models.
National Oceanic and Atmospheric Administration
(NOAA) introduces different types of statistical
errors to determine the correlation between the cal-
culated and observed values. The most commonly
used statistical errors are: (1) the normalized root
mean square error (NRMSE) and (2) the error of the
maximum value (MAX). NRMSE is a tool used to
assess the overall model performance and it is valid
when applied within a space segment or time period
to all observed data points.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
1 ðf ðx i Þ  y i Þ2
NRMSE ¼
f ðxi Þmax f ðxi Þmin n
ð11Þ
where f ðxi Þ and yi represent the observed and pre-
dicted data, respectively, and the difference between
data is used for normalization.
MAX error, which is a relative error based on the
maximum magnitude of the observed values, is
employed to quantify the predictive accuracy of a
numerical model for the maximum value regardless
of time or location:
f ðxi Þ yimax
max
: ð12Þ
f ðxi Þmax
The most recent allowable error limits can be
found in the NOAA Technical Memorandum OAR-
PMEL-135 standard (Synolakis et al. 2007).
This section contains the NAMI DANCE and
FLOW-3DÒ results for each of the six benchmark
problems that are proposed in the scope of NTHMP
Tsunami Inundation activities. Detailed descriptions
of all benchmarks, as well as topography data when
required and laboratory or field data for comparison
when applicable, can be found in the NCTR reposi-
tory, https://nctr.pmel.noaa.gov/benchmark/, and in
the repository of University of Southern California
(USC), http://coastal.usc.edu/currents_workshop/.
For the sake of completeness, a brief description of
each benchmark problem is provided. The computa-
tional domains, input parameters and applied
boundary conditions are identical in both models for
each BMP.
;
4.1. Benchmark Problem 1
In this section, the numerical results for solitary
wave shoaling on a plane beach having an angle of
a ¼ arccot (19.85) (Fig. 1) are compared with the
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 2 pffiffiffiffiffiffiffiffi
Non-dimensional free surface profile for frictionless bottom and H=d = 0.0185 at t ¼ 60 d=g—the time of maximum run-up. The circles
and dashed black line represent the analytical solution; the purple line: Dx ¼ 0.125; the orange line: Dx ¼ 0.25; the green line: Dx ¼ 0.5; the
red line: Dx ¼ 1.0; the blue line: Dx ¼ 2.0; the pink line: Dx ¼ 5.0

Figure 3
BMP 1 friction analysis. The dashed black line and circles represent the analytical solution and laboratory data (Synolakis 1986), the red
markers represent the results when n ¼ 0; the blue markers represent the results when n ¼ 0.01; the green markers represent the results when
n ¼ 0.03. Dx ¼ 1.0

analytic solution for frictionless bottom based on the is compared with the laboratory data and the solution
shallow water equations. Moreover, the maximum of the following expression, derived by Synolakis
run-up of solitary waves, R, on the same plane beach (1986):
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 4
Non-dimensional free surface profiles for frictionless bottom and H=d ¼ 0.0185. The circles represent the experimental data; the dashed black
line represents the analytical solution; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results.
Dx ¼ 1.0

  the left side. Wall boundary condition is applied for

R pffiffiffiffiffiffiffiffiffi H 5=4
¼ 2:831 cota ; ð13Þ both models, where the normal velocity is taken as
d d
zero and the tangential velocity is set to any value for
where d is the constant offshore depth and H is the a no-slip type of wall. The incident wave at time
solitary wave height. t ¼ 0 is given by (Synolakis 1986)
All variables used in this problem are non- "rffiffiffiffiffiffiffi #
dimensional and the computations are performed H 3H
g ðx; 0Þ ¼ sech2 ðx  X1 Þ ; ð14Þ
accordingly. The computational domain along x is d 4d
between x = 10 and x = 70. The slope is located on
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 5
Non-dimensional free surface profiles for H=d ¼ 0.3. The circles represent the experimental data; the red line represents the NAMI DANCE
results; the green line represents the FLOW-3DÒ results. Dx ¼ 1.0 and n ¼ 0:01

where x is the dimensionless horizontal coordinate
and increases seaward and X1 is the dimensionless x
location where the offshore height of a solitary wave
is defined. In addition, x ¼ 0 and x ¼ X0 ¼ dcotðaÞ
define the positions of the initial shoreline and the toe
of the beach, respectively. The non-dimensional
effective wavelength, L, is estimated as (Synolakis
1986)
rffiffiffiffiffiffiffiffiffi!
2 1
L ¼ qffiffiffiffiffi cosh1 : ð15Þ
3H 0:05
4d
The benchmark data were obtained from Syno-
lakis (1986). Frictionless bottom was used for
comparison of model results with the analytical
solution. On the other hand, for the maximum run-
up comparison, a friction analysis was made. The
water surface elevation time series used at the
upstream end of the computational domain (i.e.,
minimum x boundary) was located L=2 distance away
from the toe of the slope.
The non-dimensional free surface profile at t ¼
60 was predicted using grid resolutions of Dx ¼
0.125, Dx ¼ 0.25, Dx ¼ 0.5, Dx ¼ 1.0, Dx ¼ 2.0 and
Dx ¼ 5.0 (Fig. 2). The results of the grid analysis
show that the finer the grid size, the closer are the
predicted results to the exact solution. No consider-
able change was observed in model results for
resolutions finer than Dx ¼ 1.0. The effect of the
vertical grid size, Dz, on the FLOW-3DÒ results was
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 6
Water level time series at x=d ¼ 0:25 (upper panel) and x=d ¼ 9:95 (lower panel). Dx ¼ 1.0 and n ¼ 0

also analyzed. The predictions got closer to the

Figure 7
Maximum run-up values on 1:19.85 slope when n ¼ 0.01: The
dashed black line represents the Synolakis’s (1986) run-up law; the
circles represent the Synolakis’s (1986) experimental data; the red
markers represent the NAMI DANCE results; the green markers
represent the FLOW-3DÒ results. Dx ¼ 1.0 and n ¼ 0:01
analytical solution for finer vertical grid sizes with a
constant Dx. No considerable change was observed
for vertical resolutions finer than Dz ¼ 1.0. Thus, a
uniform grid size of 1.0 unit was used in the rest of
the simulations. Variable time stepping based on a
CFL condition was used for both models.
The friction analysis of BMP 1 reveals that the
predicted maximum run-up values are not dependent
on n for small amplitude waves, H=d  0:01. The
computed run-up heights are slightly affected by the
bottom friction when 0:01\H=d\0:044. In the case
of breaking solitary waves, H=d  0:044, the maxi-
mum run-up strongly depends on n (Fig. 3). Similar
results were reported by Lynett et al. (2002). For
large wave heights, it is obvious that an appropriate
choice of n improves the FLOW-3DÒ results,
whereas the NAMI DANCE results, which have
already been deteriorated by the excessive dissipation
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Table 1
Ò
NAMI DANCE and FLOW-3D model free surface profile errors with respect to the analytical solution for frictionless bottom, where H=d ¼
0.0185
pffiffiffiffiffiffiffiffi
t g=d NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

30 0.3 0.2 2.3 0.3

40 1.2 0.8 1.7 0.5
50 0.8 5.0 1.5 4.0
60 0.8 1.7 1.7 0.1
70 2.0 4.2 6.5 2.3

Table 2
Ò
NAMI DANCE and FLOW-3D model free surface profile errors with respect to laboratory data for n ¼ 0:01, where H=d ¼ 0.3
pffiffiffiffiffiffiffiffi
t g=d NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

15 11.0 19.9 1.5 5.0

20 9.5 20.0 5.0 3.5
25 6.3 4.5 3.5 1.0
30 4.0 17.0 4.4 4.2

Table 3
Ò
NAMI DANCE and FLOW-3D model water level dynamics errors with respect to the analytical solution for frictionless bottom at
x=d ¼ 0:25 and x=d ¼ 9:95, where H=d ¼ 0.0185

x=d NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

0.25 0.4 1.2 0.5 1.1

9.95 3.3 5.4 1.1 0.6

during wave breaking, got worse with the addition of 3. Numerically and analytically computed water
the friction term. level dynamics at locations x=d ¼ 0:25 and x=d ¼
To accomplish BMP 1: 9:95 were compared for frictionless bottom
(Fig. 6).
1. Numerically
pffiffiffiffiffiffiffiffi computed water
pffiffiffiffiffiffiffiffi level pprofiles
ffiffiffiffiffiffiffiffi at
4. Maximum run-up of the solitary wave was
t ¼p30ffiffiffiffiffiffiffiffi d=g , t ¼ 40 d=g
pffiffiffiffiffiffiffiffi , t ¼ 50 d=g , t ¼
numerically computed and R=d vs. H=d graph
60 d=g and t ¼ 70 d=g were compared with
was constructed for n ¼ 0:01 (Fig. 7).
both analytical solutions and laboratory data for
frictionless bottom, where H=d = 0.0185 (Fig. 4). The existence of strong wave breaking hinders the
2. Numerically
pffiffiffiffiffiffiffiffi computed water
pffiffiffiffiffiffiffi
ffi level profiles
pffiffiffiffiffiffiffi
ffi at agreement of the predicted values with the laboratory
t ¼p15ffiffiffiffiffiffiffiffi d=g, t ¼ 20 d=g, t ¼ 25 d=g and t ¼ measurements for the two-dimensional depth-aver-
30 d=g were compared with the laboratory data, aged model, NAMI DANCE. On the other hand, the
where H=d = 0.3 and n ¼ 0:01 (Fig. 5). predicted wave profiles obtained via the three-
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 8
Basin geometry, coordinate system and location of gauges (not to scale) (NTHMP 2012)

Table 4
BMP 2 gauge locations

Gauge ID x (m) y (m) z (m) Transect

6 9.36 13.80 31.7 2700

9 10.36 13.80 8.2 2700
16 12.96 11.22 7.9 00
22 15.56 13.80 8.3 900

dimensional model, FLOW-3DÒ, are in good agree- 0.0185 case, it is found that there is good agreement
ment with the laboratory data (Tables 1, 2). between the two solutions during the propagation and
When the analytical and numerical solutions at reflection of the wave (Table 3).
x=d ¼ 0:25 and x=d ¼ 9:95 are compared for H=d ¼
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 9
Run-up predictions according to different n values. The dashed blue line represents the initial shoreline; the red line represents the results
when n ¼ 0; the blue line represents the results when n ¼ 0.01; the green line represents the results when n ¼ 0.03, Dx ¼ Dz ¼ 0:1 m
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 10
Comparison between the computed and measured water level at Gauges 6, 9, 16, and 22 for Case A (H=d ¼ 0.045). The circles represent the
laboratory measurements; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

The inundation of the 1:19.85 sloped beach by
solitary waves with different H=d ratios is simulated
in several series of numerical model runs. In each
series, the bottom friction is parameterized by
Manning’s roughness coefficient, n ¼ 0.01 (Fig. 7).
NAMI DANCE predicted the maximum run-up
values with an NRMSE of 13%, whereas FLOW-3DÒ
results have an NRMSE of 4%. Due to the fact that
the experimental parameters violate the shallow
water wave assumption, the predicted wavefront is
smaller and steeper than that of the measured wave
amplitude. Consequently, NAMI DANCE underesti-
mates the run-up distances of large amplitude solitary
waves by approximately 20%.
4.2. Benchmark Problem 2
The eastern region of Flores Island was struck by
an earthquake with a magnitude of 7.8 Mw on
December 12, 1992, which was followed by 25-m-
high tsunami waves. Babi Island, which is about
5 km offshore from Flores, was also severely affected
by the tsunamis. The fact that an extremely high
inundation was observed in the back (lee) side of
island suggested that there was a need for a better
understanding of the important physical parameters
involved in a three-dimensional tsunami run-up.
Therefore, a study was initiated in 1992 by the
National Science Foundation to analyze the param-
eters involved in three-dimensional tsunami run-up
(Briggs et al. 1995) (Fig. 8) (Table 4).
To perform BMP 2, it is necessary to demonstrate
that the modeled wavefronts split in front of the
island and collide behind it producing an extreme
localized run-up. Moreover, the computed water
levels at the four gauges around the island; namely,
Gauges 6—in the front of the island at the toe, 9, 16
and 22—closest to the shoreline located at the 00, 900,
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 11
Comparison between the computed and measured water level at Gauges 6, 9, 16, and 22 for Case B (H=d ¼ 0.091). The circles represent the
laboratory measurements; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

and 1800 radial lines, are replicated and the angular
distribution of run-up is compared with the laboratory
data for the various cases of incident waves. Three
different cases are selected from the laboratory
experiments, namely Case A, Case B and Case C,
where the initial H=d ratios are equal to 0.045, 0.091
and 0.181 at 32 cm water depth, respectively.
The computational domain is 23.5 m along x
direction and 25 m along y direction. The maximum x
boundary has a rigid wall boundary condition, which
defines the downstream end of the wave tank. The
minimum and maximum y boundaries are also
defined as rigid walls to prevent outflow from the
computational domain. The initial wave is a solitary
wave whose profile and length at time t ¼ 0 are given
by Eqs. (14) and (15), respectively. It is extended to
two dimensions with a constant wave amplitude and
zero velocity in y-direction.
The convergence of the numerical models is
tested by comparing the predicted run-up values at
Gauge 9, for Cases A, B and C, respectively. The
match between the predicted data and laboratory
measurements increases as the computational grid
resolution is decreased from 1.0 to 0.01 m. The most
obvious manifestation of this convergence is the
improved form of the first wave. No considerable
change is observed for resolutions finer than Dx ¼
0:1 m. The predicted run-up values differ by less than
0.5% for Dx  0:1 m. When different vertical grid
sizes are used, no improvement was observed for
Dz  0:1 m. Thus, a uniform grid size of 0.1 m is
used in the simulations. Variable time stepping based
on a CFL condition is used for both models.
The effect of friction varies spatially over the
computational domain since the friction term is a
function of water depth. Introduction of the friction
coefficient improves the numerical predictions. It is
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 12
Comparison between the computed and measured water level at Gauges 6, 9, 16, and 22 for Case C (H/d = 0.181). The circles represent the
laboratory measurements; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

Table 5
Ò
NAMI DANCE and FLOW-3D model water level time series error with respect to laboratory data for Case A (H=d ¼ 0.045)

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

6 7.3 7.1 3.4 7.2

9 7.9 7.8 4.1 5.8
16 8.3 10.0 3.5 5.6
22 7.0 7.3 4.7 7.1

seen that, around the perimeter of the conical island,
the predicted and measured run-up values match
when n ¼ 0.01, including the extreme run-up behind
the conical island. Accordingly, the predicted inun-
dation level on the frictionless surface is higher than
the measured one. These situations are most spotted
in Cases B and C, where H=d ¼ 0.091 and H=d ¼
0.181, respectively (Fig. 9). In this study, to reflect
the real laboratory conditions as much as possible, the
run-up predictions are made for a Manning’s rough-
ness coefficient n ¼ 0:01, which is used for smooth
concrete.
The wave transformations around the conical
island that are simulated by both models are given in
Figs. 10, 11 and 12 along with the statistical errors
(Tables 5, 6, 7).
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Table 6
Ò
NAMI DANCE and FLOW-3D model water level time series error with respect to laboratory data for Case B (H=d ¼ 0.091)

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

6 8.2 10.6 3.0 2.2

9 8.7 7.6 3.8 4.2
16 11.5 10.7 5.5 3.3
22 8.7 10.7 6.1 0.9

Table 7
Ò
NAMI DANCE and FLOW-3D model water level time series error with respect to laboratory data for Case C (H=d ¼ 0.181)

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

6 7.0 9.1 3.9 5.9

9 9.3 7.8 5.3 9. 8
16 12.2 4.3 5.3 8.4
22 10.5 17.9 5.9 6.1

The inundation around the perimeter of the island
for the three cases is given in Fig. 13. The predicted
inundation levels are fairly uniform around the
perimeter of the island for Case A.
Good agreement is seen overall, and in particular,
between predicted and measured time series for the
first wave. The agreement for the later wave details
becomes progressively worse, as multiple reflections
and refraction occur at the basin boundaries. In Case
C, it is obvious that 3D model shows an appreciable
improvement over the 2D one. Case C is an extreme
case, in which a distinctive pattern of run-up emerges
because of the edge waves that propagate around the
island. The run-up on the back side of the island is
almost as large as that on the front side. The results
reveal that NAMI DANCE predicts the run-up on the
back of the island approximately 15% less than the
measured value (Table 8).
4.3. Benchmark Problem 3
A 7.7 Mw earthquake occurred west of Hokkaidō,
Japan on July 12, 1993 at 13:17 GMT. The damage
caused by the earthquake was not very severe; yet,
the tsunami destroyed an entire coastal community on
Okushiri Island before it hit the west coast of
Hokkaidō. What made this tsunami remarkable was
the maximum run-up height of 31.7 m measured at
the bottom of the Monai Valley, which opens onto a
small pocket beach. Matsuyama and Tanaka (2001)
carried out a laboratory study at the Central Research
Institute for Electric Power Industry (CRIEPI) to
investigate the 31.7 m run-up in Monai zone during
the 1993 Hokkaidō Nansei-Oki earthquake tsunami.
The experimental setup was a 1:400-scale model of
the bathymetry and topography around Monai. The
computational domain recommended for the simula-
tion of BMP 3 is a 5.488 m by 3.402 m portion of the
wave tank near the shore, which is divided into
0.014 m 9 0.014 m grid cells (Fig. 14). The recom-
mended time step size, Dt, is 0.05 s (NOAA Center
for Tsunami Research, n.d). It is worth to mention
that variable time stepping based on a CFL condition
was also used for both models. However, no signif-
icant change was observed in the numerical
predictions.
The free surface elevation time series of the
incident wave at the water depth 13.5 cm is known.
The initial wave, which is a LDN wave, has a
leading-depression height of 2.5 mm and a crest of
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 13
Inundation around the perimeter of the conical island for Cases A, B and C. The circles represent the laboratory measurements; the dashed
blue line represents the initial shoreline; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

Table 8
Ò
NAMI DANCE and FLOW-3D model run-up errors with respect to laboratory data

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

Case A 7.2 9.3 4.1 1.4

Case B 11.9 6.0 5.0 1.0
Case C 15.3 13.4 8.2 1.1

1.6 cm, which is produced in the model using FFT
analysis and the WK TIME SERIES wavemaker
option (NTHMP 2012) (Fig. 15). The input data only
covers 22.5 s and it is defined as the left boundary
condition.
The primary theme of this benchmark problem is
to model the propagation of the incident and reflec-
tive wave, to compare the numerical and
experimental water level dynamics at Gauges 5, 7
and 9 and to show snapshots of the numerically
computed water level at time synchronous with those
of the video frames.
The convergence of both models are tested by
comparing the run-up measurements along the tran-
sects at y ¼ 1.88 m—where the maximum run-up
was recorded, y ¼ 2.2062 m and y ¼ 2.32 m for
Dx ¼ 0.007 m, Dx ¼ 0.014 m, Dx ¼ 0.028 m and
Dx ¼ 0.056 m. The results reveal that the grid size
has a negligible effect on the run-up predictions up to
Dx ¼ 0:028 m and that a much coarser grid size (i.e.,
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 14
The computational domain—1:400 scale model around Monai

that the effect of varying friction is only felt during

the inundation, as expected. The predicted run-up
values are slightly larger on a frictionless bottom,
whereas when n ¼ 0:03, lower run-up values are
computed, which are approximately 10–15% less
than the laboratory measurements. These results
Figure 15 indicate that friction plays an important role in the
Incident wave profile defined from t ¼ 0 to t ¼ 22.5 s
predicted run-up values and n ¼ 0:012 is a good
choice for the simulations.
Dx  0:028 m) may lead to unreasonable run-up
The NAMI DANCE and FLOW-3DÒ results
values. The same values are tested for the vertical
regarding the wave transformation in the shallow
grid size and almost identical results are obtained. As
zone, at Gauges 5, 7 and 9, where ðx; yÞ ¼ (4.521,
a result, the recommended grid size, which is 1.4 cm,
1.196 m), (4.521, 1.696 m) and (4.521, 2.196 m), are
is used in the simulations (NTHMP 2012). The time
given in Fig. 16 and Table 9.
step size is 0.05 s, as recommended.
Figure 17 compares the movie frames that are
To assess the effect of friction on the maximum
extracted from the overhead camera used during the
run-up values, different Manning’s roughness coeffi-
laboratory experiments with the results of numerical
cients are considered: (1) n ¼ 0 (i.e., frictionless
simulations. The frames focused on the narrow gully
bottom), (2) n ¼ 0:012 (i.e., painted plywood), (3)
where the highest run-up is observed. The actual
n ¼ 0:03 (i.e., very rough surface). The results are
shoreline location is ambiguous; therefore, the dashed
compared along the transects at y ¼ 1.88 m, y ¼
yellow lines on the frames are provided as part of the
2.2062 m and y ¼ 2.32 m. The analysis results reveal
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 16
The predicted and measured free surface profiles at: a gauge 5; b gauge 7; c gauge 9. The dashed black line represents the laboratory
measurements; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

Table 9
Ò
NAMI DANCE and FLOW-3D model errors for water level time series with respect to laboratory data

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

5 7.1 4.2 6.5 2.7

7 7.5 4.3 5.8 1.3
9 7.7 0.7 6.0 0.5

benchmark specification and show the approximate
shoreline (NTHMP 2012).
The run-up and inundation comparisons of the
NAMI DANCE and FLOW-3DÒ results are shown in
Fig. 18.
Both NAMI DANCE and FLOW-3DÒ reproduce
the overall flow and inundation patterns in fully
hydrostatic and non-breaking events considerably
well and produce similar results to the previous
studies (NTHMP 2012). Both models are able to
capture the rapid sequence of run-up and rundown
and reproduce the extreme characteristics of the
tsunami in the Monai Valley, including the observed
extreme run-up value.
4.4. Benchmark Problem 4
BMP 4 is a test case whose experiments were
carried out at the O.H. Hinsdale Wave Research
Laboratory (HWRL), Oregon State University (OSU)
in 2009. This benchmark problem serves to investi-
gate the three-dimensional kinematic properties
associated with a breaking solitary wave as it
propagates, evolves and breaks over an irregular
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 17
a Frames 10, 25, 40, 55, and 70 from the overhead movie of the laboratory experiment (Liu et al. 2008; NTHMP 2012); b snapshots of the
NAMI DANCE simulation at the time intervals corresponding to the movie frames; c snapshots of the FLOW-3DÒ simulation at the time
intervals corresponding to the movie frames

shallow water bathymetry. It also deals with currents NAMI DANCE results for BMP 4 were published
induced by breaking and overtopping of a solitary previously (Velioglu et al. 2016). The improved
wave on a circular island and a complex shelf. The results are presented here.
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 18
Run-up and inundation comparisons: a run-up; b inundation. The circles represent the laboratory data of Matsuyama and Tanaka (2001); the
red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

The walls and underlying bathymetry of the basin
were made of finished concrete. The bathymetry used
in BMP 4 is given in Fig. 19.
The free surface elevation time series recorded at
Gauge 1 is defined as the input wave. It propagates
over a uniform water depth of 0.78 m before
transforming over the shelf system and inundating
the backshore area (Fig. 20).
The grid analysis reveals that both the horizontal
and vertical numerical resolutions have no evident
influence on the free surface profiles, even in the
areas with high vorticity concentration (i.e., Gauges 3
and 7) (Fig. 21). A uniform grid size of 0.1 m is used
in the rest of the simulations. Variable time stepping
based on a CFL condition is used for both models.
The varying friction is mostly effective at the
gauges located farther behind the obstacle, where
smaller friction produces results closer to the mea-
sured data. It is observed that, in general, frictionless
bottom or n\0:01 leads to an overestimation of flow
speed, and in turn, to a small mismatch in the wave
amplitude and arrival times at the wave gauges
(Fig. 22).
The NAMI DANCE and FLOW-3DÒ results
regarding the wave transformation in the shallow
zone and in the vicinity of the conical island are given
in Fig. 23. Moreover, the predicted velocity compo-
nents are given in Figs. 24, 25 and 26.
The computations are terminated after the first
reflection of the wave from the island. FLOW-3DÒ
accurately describes the evolution and transformation
of the wave, as well as the complex free surface
profiles occurring behind the island. On the other
hand, NAMI DANCE underestimates the propagation
speed of the wave and somewhat misrepresents the
waveform, especially behind the obstacle, on the
centerline. The results reveal that the depth-averaged
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 19
BMP 4: a schematic sketch of the experimental setup; b gauge locations (Velioglu et al. 2016; Velioglu 2017)

Figure 20
The free surface elevation time series recorded at Gauge 1 (NTHMP 2015)

model predicts the wave amplitude nearly 10% less
prior to wave breaking (Tables 10, 11).
NAMI DANCE model results show a noticeable
delay in the wave arrival times at Gauges 3 and 7.
The wave overtopping the island is properly modeled
with FLOW-3DÒ, hence giving a direct path to the
gauges located behind the island. However, NAMI
DANCE cannot simulate the wave overtopping
properly, leading to a longer path around the island
and delaying the arrival time.
The discrepancy between the 3D model result and
the measured data regarding the estimate of vertical
velocity, w, at Gauge 2 may be attributed to an
instrumental and/or human error (probably related to
the current meter). This can be further investigated
further if needed.
4.5. Benchmark Problem 5
Several studies have been conducted into macro-
roughness and tsunami velocity variation. Cox et al.
(2008) carried out physical model tests of Seaside,
Oregon, which showed that the macro-roughness
reduced the tsunami inundation velocity by 40%
(Rueben et al. 2011). BMP 5, which investigates the
effect of macro-roughness elements on inundation
flow depth, cross-shore velocity and momentum flux,
is based on these tests. The model, constructed at
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 21
Convergence test at Gauges 3 and 7. The circles represent the laboratory measurements; the blue, red, orange and green lines represent model
results for Dx ¼ 0.01 m, Dx ¼ 0.05 m, Dx ¼ 0.1 m and Dx ¼ 0.2 m; respectively

1:50 scale, was an idealization of the town of Seaside,
Oregon and featured several of the characteristic
structures, including a 2-m (prototype) high seawall,
large hotels along the water front, and smaller
commercial and residential structures (Fig. 27). The
measurement locations are divided into four lines: A
to D. Lines A, B and C are located on a city street
parallel to the primary inundation flow direction and
numbered from 1 to 9. Line D is located mostly
behind the buildings and only has four measurement
locations (Park et al. 2013).
A synthetic time series of incident wave elevation
at x ¼ 5 m is used to force the numerical model at
x ¼ 5 m (Fig. 28). The overland flow depth, cross-
shore velocity and cross-shore specific momentum
flux at four locations, B1, B4, B6 and B9 are
compared to efficiently tun BMP 5.
To simulate BMP 5 efficiently, it is necessary to
model the macro-roughness elements properly. To
achieve this, the FAVORTM module of FLOW-3DÒ
(Flow Science, 2002) is used to decide whether the
elements are modeled accurately enough. It is found
that the horizontal and vertical grid sizes have to be
smaller than 0.1 m, i.e., Dx  0:1 m and Dz  0:1,
since for larger grid resolutions, macro-roughness
elements can not be modeled accurately. A finer grid
size of 0.05 m is preferred for the simulations. For the
sake of consistency, the same grid size is used in
NAMI DANCE simulations. Variable time stepping
is preferred in the simulations.
The results reveal that the effect of varying
friction is stronger in more inland areas (Fig. 29).
When the bottom surface conditions are very rough,
in the furthest areas inland, there is a decrease in the
overland flow depth by 20%, in the cross-shore
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 22
Frictional analysis at Gauges 3 and 7. The circles represent the laboratory measurements; the red line represents the results when n = 0; the
blue line represents the results when n = 0.01; the green line represents the results when n = 0.03

velocity by 60%, and in the specific momentum flux
by 80%. This fact reveals that the predicted velocity
and momentum flux terms are highly sensitive to the
bottom friction factor further inland. The arrival time
of the inundation wave is earliest when no friction is
set in the model.
The NAMI DANCE and FLOW-3DÒ model
predictions regarding the overland flow depth,
cross-shore velocity and cross-shore specific momen-
tum flux are compared at B1, B4, B6 and B9
locations (Figs. 30, 31, 32, 33). It is a fact that
inundation parameters are very sensitive to bottom
friction for this specific case; therefore, it is important
to calibrate the models according to the most
appropriate roughness coefficient. The numerical
simulations are conducted using n ¼ 0.01 since it is
found that the results most similar to the physical
model are produced by this value, which is then used
for all subsequent analyses with both models.
The overland flow depth computed by NAMI
DANCE fits better to the measurements compared to
FLOW 3D results at location B1 (Fig. 30). In terms
of cross-shore momentum flux, both models compute
lower values. The agreement between measured and
computed results is in acceptable limits at locations
B4 and B6, for both models (Figs. 31, 32). However,
the results for location B9 show delayed arrival of
predicted wave by both models compared to the
measured data (Fig. 33).
The general tendencies and magnitudes of pre-
dicted overland flow depth and cross-shore velocity
are well matched with the physical model results for
the locations that are close to the shoreline. In both
models, it is observed that the maximum overland
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 23
The predicted and measured free surface profiles at Gauges 1–9. The circles represent the laboratory data; the red line represents the NAMI
DANCE results; the green line represents the FLOW-3DÒ results

flow depth and cross-shore momentum flux rapidly from B1 to B9, the maximum overland flow depth
decrease as the wave moves landward, while the decreases by 77%, the maximum cross-shore velocity
cross-shore velocity slowly decreases. Specifically, decreases by 30% and the maximum momentum flux
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 24
The predicted and measured velocity components at Gauge 2. The circles represent the laboratory measurements; the red line represents the
NAMI DANCE results; the green line represents the FLOW-3DÒ results

Figure 25
The predicted and measured velocity components at Gauge 3. The circles represent the laboratory measurements; the red line represents the
NAMI DANCE results; the green line represents the FLOW-3DÒ results

decreases by 97% in the physical model. According approximately 20% less than the 3D model FLOW-
to the measurements of the cross-shore velocities at 3DÒ.
B1, B4, B6 and B9, the 2D depth-averaged model The NRMSE of the NAMI DANCE and FLOW-
NAMI DANCE predicts the cross-shore velocity 3DÒ predictions for the overland flow depth, cross-
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 26
The predicted and measured velocity components at Gauge 10. The circles represent the laboratory measurements; the red line represents the
NAMI DANCE results; the green line represents the FLOW-3DÒ results

Table 10
Ò
NAMI DANCE and FLOW-3D model errors for water level time series with respect to laboratory data

Gauge NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

1 0.6 0.0 0.9 0.0

2 8.5 9.7 6.2 6.2
3 19.2 17.7 7.5 5.0
4 4.4 7.6 2.8 7.4
5 10.7 13.2 9.1 5.3
6 13.8 9.2 9.1 4.7
7 14.2 9.2 8.6 4.9
8 15.2 11.9 5.1 3.6
9 15.9 12.6 9.8 8.4

Table 11
Ò
NAMI DANCE and FLOW-3D model errors for velocity components with respect to laboratory data

Gauge Velocity component NAMI DANCE FLOW-3DÒ

% NRMSE % MAX % NRMSE % MAX

2 u 12.0 18.4 9.8 3.3

v 19.9 34.3 9.7 6.8
w NA NA 11.2 9.6
3 u 19.5 11.9 9.8 7.6
v 21.3 30.1 10.7 10.2
w NA NA 8.3 10.1
10 u 8.5 16.1 11.0 9.3
v 14.1 23.5 10.7 11.2
w NA NA 10.5 9.2
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 27
BMP 5 computational domain (left panel) (Velioglu 2017) and detailed plan view of macro-roughness elements with measurement locations
(right panel) (Park et al. 2013)

Figure 28
Incident wave time series at x ¼ 5 m—simulated data, which can be used to drive a stationary input wave boundary condition at x ¼ 5 m
(NTHMP 2015)

shore velocity and cross-shore momentum flux at
lines A, B, C and D are given in Fig. 34. The NRMSE
for line D is relatively large. This anomaly may be
attributed to the difference in the measurement
locations. In the experimental model, the gauges
along lines A, B, and C are located on the road, with
no obstructions between the locations and the ocean,
while along line D the gauges are situated mostly
behind the buildings.
4.6. Benchmark Problem 6
During the 11 March 2011 Tōhoku tsunami, the
328 current meters that had been placed around the
islands of Hawaii recorded the time series of the fluid
velocity at varying depths of water (Lynett et al.
2014). BMP 6 introduces a part of this data set from
the recordings at Hilo Harbor. To accomplish BMP 6,
the free surface elevation time series from the Hilo
tide station and the velocity information from the two
ADCPs, namely HA1125 and HA1126 are to be
compared and reliable agreements with the measured
data are to be achieved. The first ADCP, HA1125, is
located at the Hilo Harbor entrance, at (lat, long) =
(19.745, 204.918). The second ADCP, HA1126, is
located inside the Hilo Harbor, at (lat, long) =
(19.742, 204.930). Here, it is important to note that
the data recorded at the ADCPs are averaged over
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 29
Comparison of predicted and measured cross-shore velocity for locations B1 and B6. The dashed black line represents the measured data; the
red, green, purple, blue and orange lines represent the model predictions when n ¼ 0, n ¼ 0:005, n ¼ 0:01, n ¼ 0:02 and n ¼ 0:03,
respectively

depth and filtered to remove the long period tidal
components.
The computational domain is the full geographic
footprint of the 1/3 arc second benchmark bathymetry
at longitude 204.901° B x B 204.965° and latitude
19.710° B y B 19.774°. The x axis of the Cartesian
coordinate system is parallel to the Hilo breakwater
and the y axis is perpendicular to the wave propaga-
tion direction (Fig. 35).
In BMP 6, the simulations are driven with an
offshore simulated free surface elevation time series
which is supposed to be located at (lat, long) =
(19.758, 204.930) (i.e., Control Point) (Fig. 36).
The computed and measured time series of the
water surface elevation at the Hilo tide station
(Hilo_TS) and the average flow speeds recorded at
the two ADCP locations, HA1125 and HA1126 are
compared using different spatial grid resolutions to
identify the grid resolution effect. The numerical
modeling of BMP 6 is carried out at 5, 10 and 20 m
grid resolutions with a Manning’s roughness coeffi-
cient of 0.025. The study shows reasonable overall
convergence of the results, going down from 20 m
resolution to 10 m and then a 5 m (Fig. 37).
The effect of friction is investigated through the
application of different roughness coefficients to the
BMP 6 numerical model. The results indicate that the
effect of varying friction is very limited and only
slightly felt in the maximum and minimum values of
the time series in the current speed and surface
elevation (Fig. 38).
The incident wave propagates southward and
enters the harbor, in the middle of which ADCP
HA1125 is located. The waves refract and diffract
around the breakwater before reaching HA1126 and
the Hilo tide station. The reflected waves interact
with the subsequent arrivals to produce a complex
wave field in Hilo Bay (Fig. 39). The predicted u and
v components of current velocities in N–S and E–W
directions are favorably compared with their
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 30
Comparison of predicted and measured overland flow depth, cross-shore velocity and cross-shore momentum flux for location B1 (33.721,
- 0.588). The black line represents experimental data; the red and green lines represent the NAMI DANCE and FLOW-3DÒ results,
respectively

Figure 31
Comparison of predicted and measured overland flow depth, cross-shore velocity and cross-shore momentum flux for location B4 (35.176,
- 0.406). The black line represents experimental data; the red and green lines represent the NAMI DANCE and FLOW-3DÒ results,
respectively

observed counterparts. It is seen that the predicted These values are not observed in the recorded data.
velocity components have lots of fluctuation and thus Because of the coarse time sampling of the tsunami
several maxima/minima after t ¼ 9 h (Fig. 40). currents during the Tōhoku 2011 event, some peak
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 32
Comparison of predicted and measured overland flow depth, cross-shore velocity and cross-shore momentum flux for location B6 (36.635,
- 0.229). The black line represents experimental data; the red and green lines represent the NAMI DANCE and FLOW-3DÒ results,
respectively

Figure 33
Comparison of predicted and measured overland flow depth, cross-shore velocity and cross-shore momentum flux for location B9 (40.668,
- 0.269). The black line represents experimental data; the red and green lines represent the NAMI DANCE and FLOW-3DÒ results,
respectively
 D. V. Sogut and A. C. Yalciner
Figure 34
%NRMSE of NAMI DANCE and FLOW-3DÒ predictions along lines A, B, C, and D. The circles, triangles, squares and diamonds represent
the predictions along lines A, B, C and D, respectively
Figure 35
Computational domain for BMP 6 (top view)
velocities might be missing in the measurements. The
sampling of the predicted current velocity compo-
nents is 3 s for both models, while the sampling
interval of the recorded velocity data is 6 min. This
may be the main reason why there is a mismatch
between the predicted and recorded velocity values
after 9 h.
The NRMSE of the NAMI DANCE and FLOW-
3DÒ results for the free surface dynamics recorded at
Pure Appl. Geophys.
the Hilo tide station is calculated as 24 and 21%,
respectively, and the MAX errors are 33 and 27%,
respectively.
While time series comparisons are useful for
understanding how errors and variations evolve
temporally, they do not help to understand how
quickly flow properties change spatially. To examine
spatial variability, maps for the predicted maximum
current distribution are constructed for both models
(Fig. 41). The greatest speeds are found in the area
around breakwater head and along the coast where
the depths are shallow. In these areas, the predicted
current speeds are in an excess of 4 m/s. These results
are consistent with the findings of the previous
studies (Allan et al. 2012; Cheung et al. 2013;
Barberopoulou et al. 2015; Lynett et al. 2017).
Figure 41 also reveals that the spatial variability
results of the two models differ considerably. This
may be attributed to the effect of the vertical velocity
component, which provides a more detailed spatial
distribution of the maximum current speed for the 3D
model results. Furthermore, it is seen that the eddies
take different paths in the models in accordance with
their strength; therefore, in areas that are affected by
eddies, the models are expected to show significant
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models
Figure 36
The incident wave of BMP 6: time series of ocean surface elevation at Control Point (NTHMP 2015)
differences. Similar results are reported in Lynett
et al. (2017).
5. Results and Discussion
The NAMI DANCE and FLOW-3DÒ numerical
models are validated and verified using some of the
NOAA benchmarks for tsunami propagation, inun-
dation and currents. The numerical solutions are
tested against analytical solutions, laboratory mea-
surements and field observations.
In general, the numerical results are found to be
below the established thresholds given in the OAR
PMEL-135 standard for both models. However, the
2D model gave relatively large errors when non-hy-
drostatic effects were present. BMP 1 analyses reveal
that the 2D model cannot accurately describe the
wave form of the large amplitude solitary wave in
very shallow areas. The wavefront steepens, which is
typical of any hydrostatic model, and the predicted
wave heights are approximately 30% smaller than the
laboratory measurements. This leads to the underes-
timation of the run-up height of large amplitude
solitary waves, by approximately 20%. When the
results of BMP 2 are assessed, it is seen that the 3D
model shows an appreciable improvement over the
2D one in the case of the large amplitude solitary
wave. BMP 3 test case shows that both models are
able to reproduce the rapid sequence of run-up and
rundown as well as the maximum run-up in the nar-
row gully near the village of Monai. BMP 4 involves
energetic breaking waves with overtopping, vortex
formation and run-up. NAMI DANCE underesti-
mates the propagation speed of the wave and
somewhat misrepresents the waveform at the gauge
located behind the obstacle, on the centerline, where
strong turbulence and nonlinearities occur. On the
other hand, FLOW-3DÒ predicts the cross-shore and
long-shore velocity components reasonably well with
no mismatch in the phase at all gauges. In BMP 5, the
time series and maximum values of overland flow
depth, cross-shore velocity and cross-shore momen-
tum flux are considered. Compared to the
measurements along line B, NAMI DANCE com-
putes the cross-shore velocity * 20% and cross-
shore momentum flux * 50% less than FLOW-3DÒ
at the farthest landward locations. This is a huge
deviation in terms of force calculations. BMP 6 is a
case study since it provides a comprehensive tsunami
current data set recorded after the 2011 Tōhoku tsu-
nami at Hilo Harbor, Hawaii. Particularly, at both
ADCPs, there is good agreement between the FLOW-
3DÒ results and the recorded data for the first and
second peaks of the current velocities in N–S and E–
W directions.
The mesh properties and process time for all
BMPs are given in Table 12.
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 37
Convergence test for BMP 6 at Hilo tide station, HA1125 and HA1126 for n ¼ 0:025. The dashed black line represents the field data; the solid
blue, red and green lines represent the numerical predictions for Dx ¼ 20 m, Dx ¼ 10 m, and Dx ¼ 5 m, respectively
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 38
Model results according to different n at Hilo tide station, HA1125 and HA1126, where Dx ¼ 10 m. The dashed black line represents the field
data; the solid blue, red and green lines represent the numerical predictions when n ¼ 0, n ¼ 0:015 and n ¼ 0:025, respectively
 D. V. Sogut and A. C. Yalciner Pure Appl. Geophys.

Figure 39
Comparison of the free surface elevation time series at Hilo tide station where Dx ¼ 10 m and n ¼ 0:025. The dashed black line represents the
field data; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results

Figure 40
Comparison of the velocity components in E–W and N–S directions at HA1125 and HA1126 where Dx ¼ 10 m and n ¼ 0:025. The dashed
black line represents the field data; the red line represents the NAMI DANCE results; the green line represents the FLOW-3DÒ results
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Figure 41
The distribution of the computed maximum current speed during the entire duration of the NAMI DANCE and FLOW-3DÒ simulations. The
resolution of computational domain is 10 m

Table 12
Mesh properties and process time for all BMPs

BM Category Tested Dx Simulation NAMI DANCE FLOW-3DÒ

test quantity time (sec) Process time Process time
IntelÒ XeonÒ IntelÒ XeonÒ E5-1650 v3 3.5
E5-1650 v3 3.5 GHz GHz processor
processor

BMP1 Analytical solution and laboratory Run-up/ 0.01 60 10 min 4h

measurements amplitude
BMP2 Laboratory measurements Run-up/ 0.05 40 40 min 6h
amplitude
BMP3 Laboratory measurements Amplitude/ 0.05 20 15 h 2 days
velocity
BMP4 Laboratory measurements Run-up/ 0.014 50 3h 3 days
amplitude
BMP5 Laboratory measurements Run-up/ 0.05 40 21 h 3 days
velocity
BMP6 Field measurements Amplitude/ 10 48 600 2 days 15 days
velocity

Acknowledgements not receive any specific grant from funding agencies

The authors wish to thank Dr. Andrey Zaytsev due to
his undeniable contributions to the development of
in-house numerical model, NAMI DANCE. The
Turkish branch of Flow Science, Inc. is also
acknowledged. Finally, the National Tsunami Hazard
Mitigation Program (NTHMP), who provided most of
the benchmark data, is appreciated. This research did
in the public, commercial, or not-for-profit sectors.
Conflicts of Interest The authors declare no conflict of
interest.
REFERENCES
Allan, J. C., Komar, P. D., Ruggiero, P., & Witter, R. (2012). The
March 2011 Tohoku tsunami and its impacts along the U.S. West
 D. V. Sogut and A. C. Yalciner
Coast. Journal of Coastal Research, 28(5), 1142–1153. https://
doi.org/10.2112/jcoastres-d-11-00115.1.
Apotsos, A., Buckley, M., Gelfenbaum, G., Jafe, B., & Vatvani, D.
(2011). Nearshore tsunami inundation and sediment transport
modeling: towards model validation and application. Pure and
Applied Geophysics, 168(11), 2097–2119. https://doi.org/10.
1007/s00024-011-0291-5.
Barberopoulou, A., Legg, M. R., & Gica, E. (2015). Time evolution
of man-made harbor modifications in San Diego: effects on Tsu-
namis. Journal of Marine Science and Engineering, 3, 1382–1403.
Basu, D., Green, S., Das, K., Janetzke, R. and Stamatakos, J.
(2009). Numerical Simulation of Surface Waves Generated by a
Subaerial Landslide at Lituya Bay, Alaska. Proceedings of 28th
International Conference on Ocean, Offshore and Arctic Engi-
neering. Honolulu, Hawaii, USA.
Briggs, M. J., Synolakis, C. E., Harkins, G. S., & Green, D. R.
(1995). Laboratory experiments of tsunami run-up on a circular
island. Pure and Applied Geophysics, 144(3/4), 569–593.
Cheung, K. F., Bai, Y., & Yamazaki, Y. (2013). Surges around the
Hawaiian Islands from the 2011 Tohoku Tsunami. Journal of
Geophysical Research: Oceans, 118, 5703–5719. https://doi.org/
10.1002/jgrc.20413.
Choi, B. H., Dong, C. K., Pelinovsky, E., & Woo, S. B. (2007).
Three-dimensional Simulation of Tsunami Run-up Around
Conical Island. Coastal Engineering, 54, 618–629.
Cox, D., T. Tomita, P. Lynett, R.A., Holman. (2008). Tsunami
Inundation with Macroroughness in the Constructed Environ-
ment. Proceedings of 31st International Conference on Coastal
Engineering, ASCE, pp. 1421–1432.
Flow Science. (2002). FLOW-3D User’s Manual.
Hirt, C. W., & Nichols, B. D. (1981). Volume of fluid (VOF)
method for the dynamics of free boundaries. Journal of Com-
putational Physics, 39, 201–225.
Horrillo, J., Grilli, S. T., Nicolsky, D., Roeber, V., & Zang, J.
(2015). Performance benchmarking Tsunami models for
NTHMP’s inundation mapping activities. Pure and Applied
Geophysics, 172, 869–884.
http://nws.weather.gov/nthmp/documents/nthmpWorkshopProcMer
ged.pdf.
http://nws.weather.gov/nthmp/documents/NTHMP_Currents_Work
shop_Report.pdf.
Kim, K. O., Kim, D. C., Choi, B.-H., Jung, T. K., Yuk, J. H., &
Pelinovsky, E. (2015). The role of diffraction effects in extreme
run-up inundation at Okushiri Island due to 1993 Tsunami.
Natural Hazards and Earth Systems Sciences, 15, 747–755.
Liu, P. L.-F. (1994). Model equations for wave propagations from
deep to shallow water. (P.-F. Liu, Ed.) Advances in Coastal and
Ocean Engineering, 1, 125–158.
Liu, P. L.-F., Yeh, H., & Synolakis, C. E. (2008). Advanced
numerical models for simulating Tsunami waves and run-up.
Advances in Coastal and Ocean Engineering, 10, 344.
Lynett, P. J., Borrero, J., Son, S., Wilson, R., & Miller, K. (2014).
Assessment of the tsunami-induced current hazard. Geophysical
Research Letters, 41, 2048–2055. https://doi.org/10.1002/
2013GL058680.
Lynett, P. J., Gately, K., Wilson, R., Montoya, L., Arcas, D.,
Aytore, B., et al. (2017). Inter-model analysis of Tsunami-in-
duced coastal currents. Ocean Modelling, 114, 14–32.
Lynett, P. J., Wu, T.-R., & Liu, P. L.-F. (2002). Modeling wave
run-up with depth-integrated equations. Coastal Engineering,
46(2), 89–107.
Pure Appl. Geophys.
Macias, J., Castro, M. J., Ortega, S., Escalante, C., & Gonzalez-
Vida, J. M. (2017). Performance benchmarking of Tsunami-
HySEA model for nthmp’s inundation mapping activities. Pure
and Applied Geophysics, 174, 3147–3183.
Matsuyama, M., & Tanaka, H. (2001). An experimental study of
the highest run-up height in the 1993 Hokkaidō Nansei-Oki
Earthquake Tsunami. Proceedings of ITS, 2001, 879–889.
National Tsunami Hazard Mitigation Program. 2012. Proceedings
and Results of the 2011 NTHMP Model Benchmarking Work-
shop. Boulder: U.S. Department of Commerce/NOAA/NTHMP;
(NOAA Special Report). p. 436.
National Tsunami Hazard Mitigation Program. (2017). Proceedings
and Results of the National Tsunami Hazard Mitigation Program
2015 Tsunami Current Modeling Workshop, February 9-10,
2015, Portland, Oregon: compiled by Patrick Lynett and Rick
Wilson, p 194.
Necmioglu, O., & Ozel, N. M. (2014). An earthquake source
sensitivity analysis for Tsunami propagation in the Eastern
Mediterranean. Oceanography, 27(2), 76–85.
Nichols, B.D. and Hirt, C.W. (1975). Methods for Calculating
Multi-Dimensional, Transient Free Surface Flows Past Bodies.
Proceedings of 1st International Conference Num. Ship Hydro-
dynamics. Gaithersburg.
Nicolsky, D. J., Suleimani, E. N., & Hansen, R. A. (2011). Vali-
dation and verification of a numerical model for Tsunami
propagation and run-up. Pure and Applied Geophysics, 168(6),
1199–1222.
NOAA Center for Tsunami Research: Tsunami Run-up onto a
Complex Three-dimensional Beach; Monai Valley. (n.d).
Retrieved from: https://nctr.pmel.noaa.gov/benchmark/
Laboratory/Laboratory_MonaiValley/.
Park, H., Cox, D. T., Lynett, P. J., Wiebe, D. M., & Shin, S.
(2013). Tsunami inundation modeling in constructed environ-
ments: a physical and numerical comparison of free-surface
elevation, velocity, and momentum flux. Coastal Engineering,
79, 9–21.
Patel, V. M., Dholakia, M. B., & Singh, A. P. (2016). Emergency
preparedness in the case of Makran Tsunami: a case study on
Tsunami risk visualization for the Western Parts of Gujarat,
India. Geomatics Natural Hazard and Risk, 7(2), 826–842.
Pelinovsky, E., Kim, D.-C., Kim, K.-O., & Choi, B.-H. (2013).
Three-dimensional simulation of extreme run-up heights during
the 2004 Indonesian and 2011 Japanese Tsunamis. Vienna: EGU
General Assembly.
Rueben, M., Holman, R., Cox, D., Shin, S., Killian, J., & Stanley, J.
(2011). Optical measurements of Tsunami inundation through an
urban waterfront modeled in a large-scale laboratory basin.
Coastal Engineering, 58, 229–238.
Shuto, N. (1991). Numerical simulation of Tsunamis—its present
and near future. Natural Hazards, 4, 171–191.
Synolakis, C. E. (1986). The run-up of long waves. Ph.D. Thesis.
California Institute of Technology, Pasadena, California.
Synolakis, C. E., Bernard, E. N., Titov, V. V., Kanoglu, U. &
Gonzalez, F. (2007). Standards, criteria, and procedures for
NOAA evaluation of Tsunami Numerical Models. 55 p. Seattle,
Washington: NOAA OAR Special Report, Contribution No
3053, NOAA/OAR/PMEL.
Synolakis, C. E., Bernard, E. N., Titov, V. V., Kanoglu, U., &
Gonzalez, F. I. (2008). Validation and verification of Tsunami
numerical models. Pure and Applied Geophysics, 165,
2197–2228.
 Performance Comparison of NAMI DANCE and FLOW-3DÒ Models

Tolkova, E. (2014). Land-water boundary treatment for a tsunami
model with dimensional splitting. Pure and Applied Geophysics,
171(9), 2289–2314.
Velioglu, D. (2017). Advanced two- and three-dimensional Tsu-
nami models: benchmarking and validation. Ph.D. Thesis.
Middle East Technical University, Ankara.
Velioglu, D., Kian, R., Yalciner, A.C. and Zaytsev, A. (2016).
Performance assessment of NAMI DANCE in Tsunami evolution
and currents using a benchmark problem. (R. Signell, Ed.) J.
Mar. Sci. Eng., 4(3), 49.
Wu, T. (2001). A unified theory for modeling water waves. Ad-
vances in Applied Mechanics, 37, 1–88.
Wu, N.-J., Hsiao, S.-C., Chen, H.-H., & Yang, R.-Y. (2016). The
study on solitary waves generated by a piston-type wave maker.
Ocean Engineering, 117, 114–129.
Yalciner, A. C., Dogan, P. and Sukru. E. (2005). December 26
2004, Indian Ocean Tsunami Field Survey, North of Sumatra
Island. UNESCO.
Yalciner, A. C., Gülkan, P., Dilmen, I., Aytore, B., Ayca, A., Insel,
I., et al. (2014). Evaluation of Tsunami scenarios For Western
Peloponnese, Greece. Bollettino di Geofisica Teorica ed Appli-
cata, 55, 485–500.
Yen, B. C. (1991). Hydraulic resistance in open channels. In B.
C. Yen (Ed.), Channel flow resistance: centennial of manning’s
formula (pp. 1–135). Highlands Ranch: Water Resource
Publications.
Zaitsev, A. I., Kovalev, D. P., Kurkin, A. A., Levin, B. V., Peli-
novskii, E. N., Chernov, A. G., et al. (2009). The Tsunami on
Sakhalin on August 2, 2007: mareograph evidence and numerical
simulation. Tikhookeanskaya Geologiya, 28, 30–35.

(Received December 22, 2017, revised May 16, 2018, accepted May 24, 2018)

View publication stats