Applications of Parallel and Hybrid Simulation for Nearshore Tsunami Evolution Patrick Lynett Assist

1
Applications of Parallel and Hybrid Simulation
for Nearshore Tsunami EvolutionPatrick
LynettAssistant ProfessorCoastal Ocean
DivisionDepartment of Civil EngineeringTexas
AM UniversityCollege Station, Texas
2
Goal

• Develop a coupled, hybrid hydrodynamic
  computational model for the simulation of wave
  processes from deep water to the beach
• Source
• Propagation
• Nearshore

3
Available Models

• Shallow water (2HD, lgt25h, 1CU)
• Earthquake source
• Deep ocean propagation
• Large-scale (O(1 km)) runup patterns
• Boussinesq (2HD, lgt2h, 50CU)
• Many landslide sources
• Dispersive (short) wave propagation
• but if we want to model dispersion, we have to be
  able to resolve dispersive waves, DxO(h)
• Nearshore, nonlinear evolution
• Empirical, but calibrated breaking models
• Navier-Stokes (3D, 500CU)
• Anything
• have to resolve the scale of interest

4
The Scale Problem

• Hypothetical We want the hydrodynamic force on a
  overtopped vertical coastal structure due to a
  specific source, and we want it quickly
• Best to model source with LSW
• Expect nearshore to be nonlinear and dispersive,
  would like to use Boussinesq
• Ideally, we want the force estimate from N-S
• N-S is inclusive of all physics, but wholly
  impossible to use for entire domain (decades of
  CPU time)
• Solution mix n match

5
NLSW or Boussinesq in refined nested grid for
nearshore detail O(1-100m)
(N)LSW for oceanic propagation O(1 km)
3D N-S model with turbulence closure O(lt1m)
Couple/nest the models, parallelize the
individual components, and then figure out how to
balance the total load
6
The whole domain is divided into several
sub-domains, each is processed in a single
processor.
Boussinesq Work
7
Model Performance
Boussinesq Work
Global grid size 2000x2000
Global grid size 500x500
Global grid size 1000x1000
8
(No Transcript)
9
2D(V) RANS-1HD Boussinesq Coupling
WAVE
BOUSSINESQ (COULWAVE)
RANS w/ k-e turbulence closure (COBRAS)

• Boussinesq model does good job at intermediate
  water depth, computationally inexpensive but not
  good near breaking zone and cannot deal with
  structures.
• RANS model does good job from deep water to
  breaking zone, can deal with structures but
  computationally is very expensive.
• Solution use HYBRID wave model which couples
  Boussinesq and RANS models to simulate wave
  propagation from deep water to swash zone and its
  interaction with structures.

10
RANS-Boussinesq Coupling
Domains overlap, adjacent domain acts as a
ghost-cell boundary condition

• Data passed from Boussinesq to RANS
• Vertical profiles of u,w, constructed based on
  Boussinesq theory
• Free surface elevation

• Data passed from RANS to Boussinesq
• u at z-0.53h
• Free surface elevation

• All data exchanged at the beginning of each time
  step
• Time step of the two models are forced to be the
  same
• Developing a adaptive time step scheme for
  Boussinesq
• Horizontal grid spacing between the models can be
  different

11
RANS-Boussinesq Coupling

• Challenges
• Different numerical schemes
• Iterative high-order predictor-corrector vs
  low-order, two-step projection
• Massage the interface
• Increase the length of the overlapping region,
  expands the transition
• Different physics
• Weakly dispersive/inviscid vs fully
  dispersive/turbulent
• Carefully choose interface location
• Working on dynamic interface position
• Moves based on the physics near the interface

12
Parallel Hybrid Wave Model
DISTRIBUTED MEMORY NETWORK

• The serial hybrid wave model is parallelized
  based on a distributed data paradigm.
• RANS domain is divided into smaller sub-domains
  and distributed to nodes in a distributed-memory
  cluster. The Boussinesq domain occupies only 1
  node.
• All loops in RANS are parallelized and data
  communication takes place between adjacent nodes.
• Use distributed (Incomplete Cholesky
  factorization) preconditioned Conjugate Gradient
  iterative solver to solve the penta-diagonal
  linear system of equations (for pressure) arising
  from the Poisson equation in RANS.

13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
Conclusions/Future Work

• Depth-integrated 2D(V) RANS hybrid allows for a
  range of nearshore tsunami problems to be
  investigated accurately and efficiently
• 100 km domain scales, with sub-meter vertical
  and horizontal resolution
• Wave-structure
• Turbulent tsunami hydrodynamics
• Extension to LES (3D) 2HD Boussinesq hybrid,
  incorporation of sediment transport