This document describes research on developing a high-precision tsunami runup calculation method coupled with structure analysis. It discusses the need to evaluate damage from giant tsunamis considering structural destruction and debris. It proposes a 3D numerical simulator to analyze overflow, scouring, and flooding of buildings. The research aims to develop a system connecting tsunami propagation simulation with 3D structure analysis simulation. It describes a multiphysics tsunami simulator framework coupling models at different scales from the tsunami source to inundation. The framework includes STOC and CADMAS simulators connected using MPI communication. Example applications to the 2011 Tohoku tsunami demonstrate the approach.

1. © 2016, Amazon Web Services, Inc. or its Affiliates. All rights reserved.
Taro Arikawa, Chuo University
Development of High Precision Tsunami
Runup Calculation Method Coupled with
Structure Analysis

2. Direction of Research
• In order to evaluate the damage due to giant tsunamis, influence of
destruction of structures, debris, etc. is required.
• The power of the tsunami is greatly different depending on the place and the
condition
• 3 dimensional numerical simulator should be required to analyze overflow,
scour, flood into buildings and so on.
• The system which connects tsunami propagation simulator and 3-D numerical
simulator should be developed.

3. Multiscale and Multiphysics Tsunami Simulator
波源域
Inundation
Epicenter
Tsunami
Propagation
Structure
destruction
STOC-ML STOC-IC CADMAS CADMAS-2F STR
Nonlinear Long
wave equations
Three dimensional NS
equations
with continuity
surface model
Structure and geo
analysis by FEM
or DEM
Three dimensional
NS equations
with VOF model
Three dimensional
two phase NS
equations
with VOF model
STOC-ML
STOC-IC
STR
CADMAS

4. Multiscale analysis
- Coupled with wave propagation simulation -

5. The STOC-CADMAS system
STOC-ML
Tsunami source STOC-IC
3D model Calculates the free water surface with
a vertically integrated continuity equation
Computation load: moderate
Quasi-3D model (multi-level model)
Assumes hydrostatic pressures at each level
Computation load: light
CADMAS-SURF/3D
3D model Estimates the free
water surface with the VOF
method
Computation load: heavy
Coupled
with
DEM/FEM
STOC system (Tomita et. al., 2005)
CADMAS system (Arikawa et. al., 2005)

6. Connections between simulator calculations
STOC-ML
STOC-ML
STOC-IC
CADMAS-
SURF/3D-MG
STOC-ML STOC-IC
CADMAS-
SURF
/3D-MG
Communication
by MPI
Communica
tion by MPI
Must not
touch
All connections are made using MPI
communications.
Although all three models are capable
of segmenting their respective areas of
interest, when different calculation
methods are used in the same area (for
example, STOC-IC calculations are used
in an STOC-ML area), the parent area
containing the different calculation
methods is regulated to prevent
segmentation.
Consequently, when a CS3D area is
made sufficiently large, the STOC-IC
area that contains it ultimately
becomes larger as well, as a single area.

7. Image of calculation at Onagawa
Domain for CADMAS-SURF
Domain for STOC-IC

8. Domain 08, CADMAS-SURF/3D

9. Comparison of Maximum Inundation height
(Tsunami Source: Takagawa andTomita (2012)
measured

10. Comparison of Flow depth
Fujii, Satake –ver 4.0
Central Disaster Prevention Council

11. Improvement of Peak
Performance
- Performance of 3D numerical wave simulators -

12. Result of reduction in execution costs
Ration of number of node: ML:IC:CS=9:1:48
Ration of number of node: ML:IC:CS=9:1:90
With thread parallelization
With thread parallelization
Totally, more than three times reduction in
execution costs
Synchronization time becomes almost zero.
Next step, more improvement in the speed
of calculation of an individual program
should be investigated and more larger area
of CADMAS-SURF would be tried
Including
communication
time in each
program
The average of dt is around 0.005s. If 1 hour
calculation would be required, it takes
around 10 to 20 days
Total number of cores are 800

13. Number of grids around 12 billion for CS3D
Calculation time per
one step by using
900 nodes is almost
10 times as the
calculation time for
120 million grids by
using 90 nodes
The number of iteration
for convergence of
matrix analysis is
increasing
Calculation time
=（the rate of increase of grids）×（ the rate of increase of
iteration）
= proportional to the square of the rate of increase of grids
The results say that the calculation time by using 9000 nodes is almost
same as that for 120 million grids by using 90 nodes

14. The improvement of the number of Iteration
of Poisson solver

15. Improvement of Peak Performance Ratio
• ijk ordering
• SIMD
• Software pipeline
• Prefetch
• Etc..

16. Comparison of each case
Computational Time Performance
Mflops Performance
MIPS Performance
Mem throughput Performance

17. How about AWS
0
2
4
6
8
10
12
14
16
0 50 100 150 200 250 300
RatioofCalculationtime
(1.0isthetimebyusing9threadsin1instance)
Threads
1instance 2instance
4instance 8instance
Total number of grids is 10 millions

18. 0
0.5
1
1.5
2
2.5
3
3.5
4
0 100 200 300 400 500 600
RatioofCalculationtime
(1.0isthetimebyusing9threadsin4instances)
Threads
4instance 8instance
16instance
Total number of grids is 100 millions
The memory overflow was occurred by using 1 or 2 instances

19. 0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0 50 100 150 200 250 300
Cost[$]
For100steps
Threads
1instance 2instance
4instance 8instance
1instance $2.016/hour
Total number of grids is 10 millions
100 steps is almost 0.1s as integrated times. So, the cost of 1s is almost 5$ when the
total number of grids is 10 millions.

20. K Computer
0
5
10
15
20
25
0 50 100 150 200 250 300
RatioofCalculationtime
(1.0isthetimebyusing9threadsin9nodes)
Nodes
1 nodes has 8 cores
Total number of grids is 10 millions
The total calculation time is almost 3 times as that of AWS

21. How to use AWS
• If we uses 1 or 2 instances, then calculation time may be faster than
the super computer. Because it depends on the performance of CPU
more than the speed of Network.
• By using our numerical simulators, around 10 millions grids should be
efficient to calculate by using AWS. It is almost enough sizes for the
practical use, such as the coastal design works.
• The relation between the accuracy and the grid size has been verified.
So, the cost for some accuracy will be shown by using AWS near
future.

22. Multiphysics Simulation
- Coupled with Structure Analysis -

23. Cross section of damage of North Breakwater
DL=0.00
DL=-50.00
港 外 側 港 内 側
DL=0.00
DL=-50.00
港 外 側 港 内 側
– h” g’ çŒv‰æ– @• ü
DL=0.00DL=0.00
DL=-50.00DL=-50.00
港 外 側 港 内 側港 外 側 港 内 側
DL=0.00DL=0.00
DL=-50.00DL=-50.00
港 外 側 港 内 側港 外 側 港 内 側
DL=0.00
DL=-50.00
港 外 側 港 内 側
DL=0.00DL=0.00DL=0.00
DL=-50.00DL=-50.00DL=-50.00
港 外 側 港 内 側港 外 側 港 内 側港 外 側 港 内 側
DL=0.00
DL=-50.00
港 外 側 港 内 側Sea Side Harbor Side

24. Numerical Simulations
24
(2) CADMAS-STR PCT-girder
System of CADMAS-SURF/3D-STR (Arikawa et al., 2009)
Main program
CADMAS-SURF/3D
(VOF method)
STR3D
(FEM)
calculation step calculation step
Data on pressure
Data on displacement
call subroutine call subroutine
Images of CADCADMAS-SURF/3D

25. Experimental Video under tsunami overflow

26. Numerical Conditions
26
CADMAS dx=dy=dz=0.10 m
STR
Young's modulus ： 2.35e11
Poisson's ratio ： 0.333
Density
test body ： 2135
dummy caisson ： 2349
Coefficient of friction
static ： 0.6
dynamic ： 0.2
dummy

27. Overflow Animation
27

28. Cross section

29. Kamaishi Area
Outer areas are STOC-ML
CS-STR

30. Breakwaters in Kamaishi Bay

31. Numerical Simulations (2) CADMAS-STR PCT-girder
31
OBST
Bridge
POROUS
Slope 1:1
Girder
1.31
0.5
4.0
10.0
1.85
1.35
0.7
0.5
2.2
OBST
POROUS
Girder
2.4
0.2
0.1
0.2
0.1
Calculation Conditions

32. Numerical Simulations (2) CADMAS-STR PCT-girder
32
Physical Property
Young's modulus ：2.0E11
Poisson's ratio ：0.333
Density ：2450
Coefficient of static friction ： 0.6
Coefficient of dynamic friction： 0.2
Calculation Conditions（STR)

33. Numerical Simulations (2) CADMAS-STR PCT-girder
33
Animation
(Arikawa, 2016)

34. (a) Time = 4s
(b) Time = 8s
(c) Time = 12s
(d) Time = 24s
(e) Time = 84s
40s
80s
120s
240s
840s
Scour due to overflow calculated by using iSPH

35. Comparison of the maximum depth of scour
(1) Simulated results of eddies
(2) Effect of overflow depth on maximum scour depth
z (m)
x (m)
case
Zf
(cm)
η
(cm)
v(m/s)
Dmax-
exp. (m)
Dmax-
cal. (m)
Xs-exp.
(m)
Xs-cal.
(m)
1
24
1.0 0.17 0.13 0.14 0.09 0.08
2 3.3 0.80 0.39 0.40 0.30 0.36
3 4.7 1.09 0.62 0.62 0.48 0.50
4 6.0 1.31 0.65 0.78 0.50 0.72
5 6.8 1.37 0.87 0.85 0.80 0.81

36. The Problem of speed of scour
(1) Inflow boundary conditions
(2) Scour process

37. For evacuation
- Coupled with Multiagent simulator -

38. Achievement of Resilient Society to Natural Disaster
Support of disaster response and decision making
of Local community, corporations, citizens.
・Sharing the real-time disaster information
・Information dissemination technique under disasters
・Application for the local community
・Robotics
Real-time Disaster Information
Sharing system (Cabinet Office,
Government of Japan)
Response
Capability
Prevention
Capability
Improve the seismic and
tsunami resistant capacity
and Recognize the danger of
the structure in advance
Acquire disaster information more
rapidly with higher precision.
Earthquake, tsunami, heavy rain,
tornado, volcano and fire
E-Defense
Full-scale three-dimensional vibration
destruction experimental facility
Regional
Alliance
Disaster
Response
System
Outline of the system
Improve resilient society system for protection of life, property and
industry and aim for the safe and reliable society
•Develop the resilient infrastructures to improve the prevention capability and develop the
real-time disaster information sharing system to reduce damages, which connect the
response capability with the prevention capability and usage of the prediction capability for
understanding the actual state of disasters.
Prediction
Capability
Inundation Conditions
(Example of Society 5.0)
State of damage
of buildings
State of damage
of infrastructures

39. Coupling with Evacuation Simulation
Coupling STOC, CADMAS-SURF and PARI-AGENT
STOC
CADMAS-SURF
Arikawa andTomita
(2015)
PARI-AGENT

40. 40
Tsunami Evacuation Simulator (1)
(PARI-AGENT)
CADMAS-SURF/3D
3D model Estimates
the free water surface
with the VOF method
Determinate Moving Route
is determinated by superposed two potentials.
② Crowd Potential
Follow the direction in which there are evacuees.
① Evacuation Route Potential
Evacuees select the shortest distance
to evacuation place.

41. 41
Tsunami Evacuation Simulator (2)
(PARI-AGENT)
Death Judgement
When the inundation height reach the 1.0m,
evacuees are dead.
Walking Speed is corrected by the evacuation route slope
and the inundation depth.
If evacuee is flooded, walking speed become slow.

42. Reproduction at the earthquake on 1st of April in 2014
42
 Iquique city in Chile
 Setting the evacuation places at which more than 5 the evacuees were gathering.
Initial positions and Routes Initial Positions and Evacuation places

43. Calculation conditions of PARI-AGENT
43
Item Detail Note
Number of People 285
From the results of
Questionnaire
Start time to
evacuate
0sec
Time steps for
calculations
0.1sec
End time 7,200sec
Moving velocity for
agents
Initial velocity 1.0 to
3.0 m/s
With hiking function
Depth to die 1.0m
Number of
evacuation places
22カ所
From the results of
Questionnaire
Evacuation Sign No

44. Calculation Conditions for STOC-ML
44
項目 詳細 備考
Calculation Domains See the slide after
Topography From Dr. Okumura in Kyoto University Without buildings
Grid size
Domain 1 ：810m
～
Domain 5 ：10m
1:3 all domains
Time steps 0.5sec
End time 7,200sec
Tidal level 0.0m
Tsunami Source From Dr. Okumura

45. Animation
45
Tsunami Source; Mw8.8，Virtual Fault in Iquique city

46. Comparison of Evacuation distance and places
 Almost good agreement with actual evacuation distance and places（concordance
rate 66.3%）．
 So many people chosen the minimum evacuation distance.
 One of the main reason of this phenomena is the simple route to the evacuation
places from the shore lines
46
0
5
10
15
20
25
0 5 10 15 20 25
計算による
避難箇所
実際の避難箇所
0
500
1000
1500
2000
2500
3000
3500
4000
0 500 1000 1500 2000 2500 3000 3500 4000
計算による
避難距離 (m)
実際の避難距離 (m)
Evacuation
distance
Cal
Actual
Cal
Actual
Evacuation
places

47. For inhabitants Courtesy of Nakatosa town in Kochi Prefecture
47
Begin to evacuate
10 minutes after
the seismic motion
Begin to
evacuate 20
minutes after the
seismic motion

48. Thank you for your attention!!
Let’s study Tsunami together!!
Please come to Japan!!

49. Thank you!
함께 해주셔서 감사합니다!