SimCenter | Modeling of 500-year Cascadia Subduction Zone Tsunami Inundation, November 1, 2017


Welcome. Today is Wednesday,
November 1st, 2017 and this is the NHERI Early Career Researcher Forum. This forum is intended
to foster the exchange of ideas and be a platform of best practices for
successful research. It is intended to highlight compelling research, build up
the natural hazards engineering student community and provide presentation
opportunities. For more information, visit the NHERI website at designsafe-ci.org
where you can find links to the SimCenter and the NHERI Learning Center.
Today’s webinar is coordinated by the Natural Hazards Engineering Research
Infrastructures Simulation and Computational Modeling Center. This
webinar is supported by the National Science Foundation under Awards
1612843 and 1520817. Any statements in this webinar are those of the presenter and do not necessarily
represent the views of the National Science Foundation. Today’s presentation
is by Shawn Qin. He is a fourth year PhD student at the University of Washington.
He received a Bachelor’s degree in Ocean Engineering from Shanghai Jiao Tong
University in China in 2014. His primary areas of research are numerical modeling
of tsunami inundation and the impact on coastal structures using large-scale
supercomputers. Recently, he has also been working on advancing — sorry — accelerating
tsunami analyses with graphical processor units. Shawn is co-advised by Michael Motley, Randall LeVeque and Frank Gonzalez. The
title of his presentation is “Multi-scale Modeling of a 500-year Cascadia
Subduction Zone Tsunami Inundation Including the Constructed Environment.”
And Shawn, I invite you to begin. Hello everybody. Thank you for being here.
So, my name is Xinsheng Qin, and I’m 4th year PhD student at the University of Washington,
Seattle, and I’m co-advised by Michael Motely, Randall LeVeque and Frank Gonzalez.
And today I’m gonna present my work on comparing, actually comparing a 2D and 3D
tsunami inundation model with the constructed environment, and specifically
it’s a 500-year Cascadia subduction zone tsunami. And as you might
still remember, in the past several decades, there are several here which
took tens of thousands of people’s lives. And the most recent one is
the Japan 2011 tsunami and the dark strip in this figure, shows all areas in the
world that are prone to a tsunami hazard. And note that the West Coast of the United
States is also in this region. And so the figure on the left shows what a
subduction zone is — where are the subduction zones in the world by the red
solid line, and specifically there is a subduction zone on the West Coast of the
United States of the west of North American plates. And here, we show
what a Cascadia Subduction Zone is. So, it’s where the Pacific plates
interact with the North American plates and you can see on this figure on the
left the Pacific plate moving toward the North American plate and there’s a
rupture between the two plates denoted by that solid line in this figure. And if
such a subduction zone — if an earthquake happens in such as a
subduction zone, we are expected to have a very huge tsunami. And it is reported
that magnitude nine tsunami earthquake is very likely to happen in the next
several decades, and the Washington State, Oregon State, and California State are
all possible to be affected by such an earthquake and the tsunami generated by
such an earthquake. So, that’s why we would like to study
this Cascadia Subduction Zone. And here, these figures show some very severe
damage from tsunami events. And you can see how powerful a wave generated by
tsunami can be. And here the figure on the left shows a bridge damaged,
destroyed by a tsunami. And a bridge is very critical after a tsunami,
because like you want to transport water and food into the region. But if the
bridge is damaged, then it’s hard to get those those things into
the region. And the figure on the top right also shows a building that actually
moved away by several meters, move away several meters by the tsunami
wave. And you can also see how powerful a tsunami wave can be. And so, we would like to
be able to predict, to model the tsunami as well as, in addition, or to be able to
predict the impact from a tsunami to those coastal structures such that
when we design, say a bridge, we would like to design such that it can survive
a tsunami. Then we can we can transport water and food into the region
after a tsunami, and we also would like to design some critical buildings like
hospitals, such that it is strong enough to survive a tsunami because people in
a hospital are not very likely to be able to run away from the region after a tsunami, when a tsunami comes. And here, a tsunami
actually consists of several steps and I here I divided it into four steps. So
the first step is tsunami generation. And so, a tsunami, and so here
we were talking about tsunami generated by an earthquake under the sea, and after that it will propagate toward the shoreline,
and then after it arrives at the shoreline, we we have inland inundation of a
tsunami. And then after that, the tsunami will interact with coastal structures
like bridges and buildings inland. And the scale of these steps
actually decrease from very large which is in the Tsunami Generation step, to
quite small and the Interaction with Structure step. And usually, when we
model the first two steps, people would use, would solve to the Shallow
Water Equation which is much less computational expensive than what is
typically used in modeling the last step that is the 3D Navier-Stokes
Equation. Here, our focus is actually on the offshore propagation inland
inundation and interaction with structures. So since the case we
are considering actually, consists of three steps out of those four. It’s not very clear whether we
should use a traditional water equation to model this, or the 3D Navier- Stokes
Equation. So, in this study, i’m going to present my result of comparing, of using
both the 2D Shallow Water Equation to model the three steps and the 3D Navier-
Stokes Equation to model these three steps. Okay, and specifically we built a
model on a case of Seaside, Oregon. So the two figures on the left are the top view of
the city Seaside in Oregon. And you can see it’s right next to the
Pacific Ocean and the figure on the right shows that your graphic location of Seaside, and it’s very close to a Portland and
Seattle. Okay, and the reason why we choose, why we chose this case study
is that we have some experimental data by some researchers at Oregon State University who did a series of experiments several
years ago. And they built a physical model of part of the city Seaside in the
numerical wave basin. And here is a snapshot of the experiment, from the
experiment. And you can see a single tsunami wave is coming toward us,
and it’s going to hit the buildings categorize by three different colors: red,
blue and yellow. Okay, so the figure on the left shows
— the yellow rectangle on the figure on the left — shows the region where we are going to model, and this is also the
region that the physical experiment was modeled. And a wave of ten meters
high is coming, is going to enter the region from its left boundary and
hit. And it’s going to hit the buildings in the small yellow rectangle on the
right. And the length of the rectangle is about two kilometers and
it’s about 1.5 kilometers wide and it’s a 1.50 scale model. So, the wave in
the experiment is actually only twenty centimeters, which correspond to a ten
meter wave in prototype. And when we used our 2D and 3D models to
to model this scenario, and ideally we would like to have our
numerical model, model the same region or the same domain as in the physical
experiment. This is true for the 2D model, but it turns out, if we try to put
everything, if we try to put the entire rectangle into the three 3D model, it’s
just too expensive. And I was gonna, I will show you how expensive it is in the
next several slides. So we actually came up with a different strategy when we used
the 3D model to model this domain. And here is — so in the experiment, they
measured velocity and wave height at several locations. So, here you can see, on
the right, you can see A2, A3, A1, A2, A3 until A9 and B1, B2 until B9, C1 to C9,
D1 to D4. So, those 31 locations where they measured those flow parameters. So,
those are the places where we would like our numerical model to be able to
predict the same flow parameters. So, now let’s say if we would like to
predict flow parameters at A1 and A9, we carefully chose a subsection of the
rectangular region. And here, I denoted by the blue dash line, and so we use
our 3D model to only model this subsection to get flow parameters at A1
to A9. So, in this way, we don’t have to, don’t have to model the entire
rectangular region. And you can see we repeat this process for other groups of
gauges or other group of places where we want to predict flow parameters. So,
this, the green dashed line here shows a long thin subsection, by model, which we
model in which we predict flow parameters at gauge B1, B2 until B9.
And we repeat the same strategy for prediction of flow parameters from C1 to C9
and the last is D1 until D4. So this is a summary of the four subsections
that we carefully chose to model in order to get a prediction of flow
parameters at all 31 places. By doing this, we don’t have to
model the entire rectangular region at once which is impractical with the
computational resources we have at hand. And here is a summary or a comparison of
the two models that we used. And so for the 2D model, we solved the Shallow Water
Equations and and we used the package called GeoClaw which is a package of
Clawpack developed by Randall LeVeque at the University of Washington. And
specifically, we used a technique, mesh technical Adaptive Mesh Refinement
which we find only places where we need higher resolution, and it actually
automatically refined the grid. Or we generate the grid at every several time
steps. So, you can imagine as the front of the wave or the front of the tsunami
pour, or moves, it automatically we find where near those places where you
you know we should have finite weight. And that’s the good thing
about using Adaptive Mesh Refinement. And it turns out, you
actually save a lot of computational efforts, and so in this specifically
for this case, we monitored the entire rectangular region, and it takes only
five to six hours on my own laptop. And the 3D model is, on the other hand, is
much much more expensive, and it’s not because we are solving the Navier-Stokes
equation with the last model. It’s also because we are not using Adaptive
Mesh Refinement. So, the mesh we are using here is just regular unstructured mesh
in which you refine wherever you think you need, you need finer mesh
during the simulation and the mesh is just static during the whole simulation.
And we used an open source, quite famous open source, CFD package called Open FOAM. And here you can see the computational time is 8 to 10 days with 128 computer cores
in parallel. So a CPU on a supercomputer typically will have 16
cores, so 128 computer cores means a a supercomputer. And in the last column, I actually evaluated, roughly evaluated, the
floating operations, number of floating-point operations, in both models.
And you can see in terms of floating operation, floating-point operations, and
the 3D model can be 10,000 times more expensive than the 2D model. Okay, so
on these slides, I’m showing the results, the prediction of flow parameters at B2.
And the first row is the water depth and the second row is velocity, and the third
row is momentum flux which is computed from multiplying water depth with
velocity square. And here you see — so let’s have a look at the first row. Here
you see, so the storyline is the experimental measurement, and you can see in both the
3D model and the 2D model actually agree quite well with the experimental data.
And for the velocity, for the prediction of velocity here you, so the
both this blue solid line and the red dashed lines are from experimental
measurement. And here you can see in terms of the peak, peak value, the 3D
model which is, which is the cross. Actually it is much closer than the 2D model, the 2D
GeoClaw model. And since the momentum
flux is computed from water depth and velocity, you can also, so it is also
expected that the 3D model actually match better with the
experimental data in terms of the peak value. So, here we are showing a
prediction of flow parameters around, at two other locations. And also,
you can see at B3 the 3D model is obviously much better than the 2D model
in terms of water level and velocity. And at at location D2,
especially for the prediction of velocity, you can see the 2D model
actually is quite bad at predicting and velocities at D2. This is because
A1 to A9, B1 to B9 and C1 to C9 are all along like a street, but D1 and D2
and D2 is behind a building where the flow can be much more complex
and the 2D model actually has some difficulty at predicting flow parameters there. There’s a very interesting, so we know
that there’s a very interesting discrepancy in velocity. So, here we show
the prediction of velocity at gauge A3 and here you can see, so the
peak value from the 3D model actually is much higher than the peak value from
the experiment, and we actually have a quite interesting finding about this.
So here you should know you should know that the experimental data,
experimental measurement are represented by both the story line and red dashed line,
and here’s how they got the red dash line from the experiment. And so in the
experiment, when they get data represented by a black dashed line, it is the
measurement from the velocimeter. But the data here represented by red dashed line is not
from the velocimeter because in the experiment, so because in the experiment
there are so many error in the edge of the bore so the
velocimeter actually failed to record a consistent, the results
consistently between different trials. So, here is how they get the the data
represented by red dashed line instead. So, here, if you have a look at the
figure on the right, so let’s say the red — sorry, the green solid line represents
the position of the bore edge at time T1, and after a second it moves to the
position represented by the purple dashed, the purple solid line. So let’s
call that time T2, and the way they, so they get the position of the edge
of the bore at every at every moment by taking photos or videotaping it. And the way they
estimate the velocity here is they compare the distance between two
different, between two different edges and then divide that by and divide that
by the difference in time. So, the expression they use is “u” equals the position of
leading edge at time T2 and minus the position of leading edge at time T1 and then
divide that by how long it takes. So that’s how, how they get the
peak value of velocity here represented by, so I used a red dot
to represent the 2.2 meters in the figure. And however in
the CFD model, in the 3D model, we actually, so that the result
you see here, the peak value from OpenFOAM, that value is not from this
approach, is actually just from direct measurement which we are able to do
in numerical simulation. And so here we actually, if we use the same approach as
they used in the experiment where, so here you can see if we cut a slice
along gauge A1 to A9 and then so the figure on the left shows the profile of
the wave if we cut the slice through through A1 until A9. And the figure on
the left snapshot of the simulation at two different time steps. And here again,
we can, using a similar idea, we can get the location of leading
edge at T1 and the location of the leading edge at T2. And using a similar idea, if
we compare, if we estimate the peak velocity by doing the same algebra
we actually get exactly the same results which is 2.2 meters per second.
And so if we do this, the numerical prediction actually agrees quite well
with the experimental measurement in terms of the peak value. Okay. And in
fact the results I’m showing here has a peak value of 2.8 meters per second. So
why is that, and where is it coming from? So, if you have a look at the figure on
the left again and so the two snapshots are all colored
by velocity and you can see there’s a dark region, quite dark region, after the
leading edge of the bore which means if you put a velocimeter
there and after the leading edge of the bore has already moved, has
already passed, you will get a 2.8 meters per second sometime after that. So
that’s why we get 2.8 meters per second in the numerical model
because we are consistently using pseudo velocimeter in the numerical experiment
to get, to measure velocity at the place we want. Okay. So, as you have seen the 3D model is
so expensive. It’s ten thousand, it’s ten thousand times more expensive than the
2D model and you also see that at some of the locations, the 2D model is
actually not that bad. So, why do we still want to use the 3D model? So one
important reason is that the 3D model actually can predict tsunami forces on
structures directly because we are solving for pressure field and if you
integrate pressure on those structures you get you get tsunami forces. But for the 2D model, we can’t get tsunami forces on those structures directly. And
here I’m showing time history of tsunami forces on some buildings in
this region. And here you can see we can see at the peak value of forces
on building number one and building number two is quite high compared to other
buildings. This is obvious because it’s in front of other buildings. And here we,
if we compare the peak forces on this building number one and
convert it to prototype scale, we get fifty-five thousand kN. And if
we compare this to seismic load, due to maximum considered earthquake for
this building. And it turns out, if you design this building such that it can survive an earthquake, and you only have to make it withstand three
thousand, about three thousand kN tsunami loads which means you
have to strengthen the building by around fifteen times if you want
this building to stand there after a tsunami wave like this. And here, we are also able to predict with the 3D
model, we also are able to predict localized forces on structures. So here,
the figure on the left, shows tsunami forces on each wall of building of
building number two, and especially you can see, so the black dashed line
corresponds to tsunami forces on the south wall of this building
which points up, and the orange story line is forced on the
north wall of this building which is pointing down. And you can see the two
forces actually cancel each other each other out. And but this figure also shows
it is important to compute forces on individual components of a structure
because if you only have a look at the total force, the two, as in this
example, two forces cancel each other out and you didn’t know, like forces
on individual of them can be large and they can be destroyed by the tsunami and
which can affect the property of the intact structure and cause some critical damage. And also, obviously the 3D model
can give more details. So this is a comparison of a snapshot from an experiment
and a 3D model. And you can see they look pretty similar and the 3D model actually
gives you a lot of details. And here, I’m showing the
animation from both the 2D and 3D models. And for the 3D model, it’s only a
subsection of the entire domain. And we have four subsections like this.
And for the 2D model, you can see it’s actually able to model the entire rectangular domain at one time. Okay,
so here are the conclusions from this study. And so first of all, so when you use a 3D
model to model a tsunami inundation region, and if the 3D model is too expensive
to model the entire region yourself, how we can think about using modeling
only sub sections. And we have shown that if you choose to subsection carefully
enough, you can do it actually with much less computational resources. You
can do it much faster without loss of accuracy and where you want to predict
in areas where you want to take flow parameters, or where you want information
from the model. And the second is that such a 500 years cascadia tsunami
can actually cause very severe damage to buildings as you
have already seen. And for the building number one and number two, it can be
destroyed, it’s very obvious that it will be destroyed by a building
according to the first predictions. And the third, so the 3D
model, one good thing about the 3D model over the 2D model is that you can
give direct prediction of tsunami loads on those structures which can give
very useful advice when designing those buildings, the coastal
structures. And the 2D model, although it’s less accurate, you can see it’s
not that bad. And since it can run so fast, and when you need to run the
model for many many times, for example, if you want to do
some probabilistic tsunami hazard assessment, where you need multiple runs
the 2D model can be a very good choice. And I think that’s all I have
today. And thanks everybody. Well, thank you, Shawn, very much. I want to
thank you for your presentation and at this time, we’ll begin with the question
and answer session. Attendees, you are reminded that questions should be
submitted through the chat panel and sent to the moderator. Shawn, we have
a couple questions. The first is: could you tell me about the grid size you used
for both simulations? Okay, let me go to Okay. So for the 3D model, the
degree is actually, we actually refined the grid allowed near the buildings and
as I remember, if I remember correctly, the size of each, the smallest mesh
can be like in the order of centimeters, and for the 2D model we used Adaptive Mesh
Refinement and so it’s structural mesh and but we still have a very fine
mesh inland, near the buildings. And I think it’s in the order of, I would
say, in the order of 0.1 meters. Yeah. So, it’s 10 centimeters. So, the mesh
in the 2D model is about 10 times closer than the mesh in the 3D
model. I’m talking about the finest mesh. And another question: is there any
experimental data to validate the forces you reported on the structures?
Unfortunately, for this case, no. I think when they did this experiment,
they didn’t measure forces on these buildings. But we have
validated our model, our 3D model with a simpler case where you have a template
problem and you have a, you have a column downstream and we have
experimental data on that. So, we have experimental measurements of forces on
that single column and we actually modeled that case and our 3D model
actually can predict that force pretty well. Okay. But it’s a simpler case where
you have only one essentially you have only one column, or you have only one building there. But in this case, we have
hundreds of buildings. Okay. We have a comment from one of the attendees,
Pedro from OSU, who says that they have executed experiments on forces of
structures in their wave tank… I’m sorry. I think I can’t hear you.
I’m sorry. I was simply saying that Pedro from Oregon State has said that they
have executed experiments on forces on structures of forces on structures
previously. Those were led by Motley, Arduino and Cox. So… Ok is that on
like, is that from a case where you have only one or two buildings or… I’m not
sure. We’ll have to relay that question… Okay, okay. So, another question, Shawn. During
a real tsunami, some of the buildings or things in the vicinity of the region
are destroyed or flushed downstream, would impact buildings. How would
those sorts of interactions with the structures where damage might
affect the results of this study? Actually, so in this study, we actually assume, so
for example…let me find a good slide. Yeah, for example, so here in this
study we actually assume all buildings are not, won’t be destroyed by this
tsunami. And if, say building number one and building number two are destroyed by
this study and it’s carried by the tsunami wave and and hits the red
building behind it, and that’s gonna cause a, I would say, that’s gonna cause a huge impact on the, you can
expect that, right? That’s gonna cause a huge impact on the red buildings after
the two blue buildings. And unfortunately, currently we are not able
to model this scenario. Okay, The boundary conditions you used in the
video show that you have reflective boundaries. It seemed to show that you
have reflective boundaries. Is that the case? Yeah, so the boundary on the left
is just inlet boundary and all three other boundaries are just wall. So you can
expect the wave will get reflected when they hit the wall. Does that impact any of
your analyses? So all the time history of forces and wave and water
depth and velocity, I show here are before the reflected wave affects
the gauges or affect that measurement of forces. Okay. When you talked about
the time it takes for you to do some of the analyses, it looked like
there were, you know, several hours to compute the 2D model and you said
that was for one core. Is that one core on your laptop, or is that one core of a
supercomputer? Where are you running these? Oh, so for the 2D model, so
that’s one core actually on my laptop. I actually want the 2D model, the
result I got I show here is actually from several runs on
my own laptop. So it’s pretty cheap. Well, if there aren’t any other questions,
I think we’ve reached the conclusion of today’s Early Career Researcher Forum.
On behalf of the attendees, thank you, Shawn, for taking the time to share
your research. To the attendees, thank you for your participation and
your questions. Please check the SimCenter’s website at
simcenter.designsafe-ci.org and check your inbox for emails from
[email protected] There, we’ll have registration links for additional
upcoming webinars in the Early Career Researcher Forum and the Natural
Hazards Engineering 101 webinar series. Thank you for attending today’s forum.

Leave a Reply

Your email address will not be published. Required fields are marked *