Development of Coastal Ocean Modeling Infrastructure (COMI…nsgl.gso.uri.edu/lsu/lsuwr09003.pdf ·...

32
Development of Development of C C oastal oastal O O cean cean M M odeling odeling I I nfrastructure nfrastructure (COMI) at LSU (COMI) at LSU Q. Jim Chen Department of Civil and Environmental Engineering & Center for Computation and Technology Louisiana State University

Transcript of Development of Coastal Ocean Modeling Infrastructure (COMI…nsgl.gso.uri.edu/lsu/lsuwr09003.pdf ·...

  • Development of Development of CCoastal oastal OOcean cean MModeling odeling IInfrastructure nfrastructure

    (COMI) at LSU(COMI) at LSU

    Q. Jim Chen

    Department of Civil and Environmental Engineering& Center for Computation and Technology

    Louisiana State University

  • Acknowledgements

    COMI Team:Drs. Gabrielle Allen, Mayank Tyagi (CCT)Drs. Claes Eskilsson, Kelin Hu (CEE/CCT)Yaakoub El-Khamra, Lei Jiang (CS)Qi Fan, Ranjit Jadhav, Qian Zhang, Haihong Zhao (CEE)

    Funding:NSF, ONR, NOAA

    Gulf Shores, Alabama

  • Motivation

    Combined waves and surge

    Depth-integrated coastal models

    Higher-order finite element methods

    Modeling framework and results

    Conclusion and future work

    Outline

  • Courtesy of Lake Pontchartrain Basin Foundation

    Animation of Wave-Structure-Seabed Interaction

    Louisiana Coastal Protection Assessment of Multiple-lines of Defense

    Chen et al. (2006)

  • Wave overtopping

    Bridge damaged by storm waves Combined

    Waves & Surge

    news.bbc.co.uk/.../uk_enl_1194609737/img/1.jpg

    Kevork Djansezian/ AP

    Gulfport, Mississippi during Gustav

    US -90 bridge over Biloxi Bay, MS

    Impact of Katrina on WetlandsUK

  • Challenges

    Hurricane protection and coastal restoration require a suite of numerical models for different physical processes.

    Different models use different numerical methods to solve different governing equations, e.g. FDM, FEM, FVM, SEM

    There is a need for developing a common modeling framework to enable the modeling of multi-physics, multi-scale coastal processes using HPC.

  • CFD Toolkit

    ComposingFluid-flow solver

    Solvers

    SuperLU

    Trilinos

    PETSc

    Numerics

    Finite difference

    Finite Volume

    Spectral Element

    Infrastructure

    Cartesian

    M-B curvilinear

    Unstructured

    Physics

    Navier-Stokes

    RANS

    Boussinesq

    Other coastal models

    To develop the capability of modeling coastal circulation and nearshoresurface waves in deltaic environments using high-order numerical methods

    To integrate application-oriented coastal modeling systems with massive-processor, high-performance computing facilities and technologies available at LSU

    Objectives

  • Depth-Integrated Coastal Models

    3-DimensionalEuler Equations

    Boussinesq-typeEquations

    Nonlinear ShallowWater Equations

    Linear Long Wave Equations

    Mild SlopeEquations

    Wind Stress

    Storm Surge

    Wave Growth Storm Surge

  • Modeling Surge/Wave Attenuation by Porous Media or Vegetation

    Porous Media

    Cruz and Chen (2007)

  • Why higher-ordernumerical methods?

  • Unstructured Spectral/hpDG Methods

    Solution approximated inside an elementby a pth order polynomial expansion

    (note: can be non-uniform).

    Elements coupled throughso-called numerical fluxes

    (computed by approximate Riemann solvers)

    Supports unstructured meshes (conforming or non-conforming) consistingof elements of size h. Note that the solution

    is allowed to be discontinuous over theelement boundaries.

  • CACTUS Framework

    CACTUS is a freely available, modular, portable and manageable environment

    for collaboratively developing parallel, high-performance multi-dimensional simulations

    http://www.cactuscode.org/Successful Applications: Astrophysics,Petroleum Engineering,

    http://www.cactuscode.org/

  • COMI Architecture

    UMDriver provides the underlying parallel layer. At present it uses the Zoltan library to provide mesh partitioning, load balancing and mesh migration

    LocalToGlobal provides local re-indexing of elements, edges and vertices The Nektar++ thorn initializes and populates the data structures of the

    Nektar++ library MeshReader provides a simple ASCII mesh file reader, and also allows users to

    register their own mesh readers (e.g. the mesh reader from Nektar++) The core thorn for the coastal modeling toolkit in Cactus is CoastalWave. This thorn

    defines the generic variables, parameters, and methods for coastal models Thorn SWE contains the actual SWE solver

  • Nektar++ library

    Nektar++ is an open-source spectral/hp element library presently in the last stages of development (www.nektar.info)

    As the name suggests Nektar++ is written in C++

    Nektar++ is developed and maintained by Prof. Sherwin (Imperial College London) and Prof. Kirby (University of Utah)

    COMI's shallow water codes are based upon Nektar++ and will be distributed as part of the Nektar++ solver library

    http://www.nektar.info/

  • Discontinuous GalerkinSpectral Element Model

    Flood Wall

    Flood wal

    Breach

  • Domain distributed over 8 cores

    Sixth order triangular elements

    COMI Result Eskilsson et al. (ICCS 2009)

  • Parallel DG Model

  • Weak Scaling 900 quadrilateral elements per core 100 time steps Times without I/O, initializing and partitioning of the mesh 128 cores, p = 8 roughly 28 million DoF

  • Hurricane Gustav Animation of the combined H*wind and background winds (NCEP/NCAR Reanalysis)

  • Wind Comparison

    0

    10

    20

    30

    40Buoy 42040

    Win

    d sp

    eed

    (m/s

    )

    0

    10

    20

    30

    40Buoy 42007

    Win

    d sp

    eed

    (m/s

    )

    0

    10

    20

    30

    40Dauphin Island

    Win

    d sp

    eed

    (m/s

    )

    Aug.31 00 12 Sep.1 00 12 Sep.2 000

    10

    20

    30

    40PSTL1 SW Pass

    Win

    d sp

    eed

    (m/s

    )

    Time (days, Aug.30 ~ Sep.2, 2008)

    0

    90

    180

    270

    360Buoy 42040

    Win

    d di

    rect

    ion

    (deg

    rees

    )

    0

    90

    180

    270

    360Buoy 42007

    Win

    d di

    rect

    ion

    (deg

    rees

    )0

    90

    180

    270

    360Dauphin Island

    Win

    d di

    rect

    ion

    (deg

    rees

    )

    Aug.31 00 12 Sep.1 00 12 Sep.2 000

    90

    180

    270

    360PSTL1 SW Pass

    Win

    d di

    rect

    ion

    (deg

    rees

    )

    Time (days, Aug.30 ~ Sep.2, 2008)

    90.5 90.0 89.5 89.0 88.5 88.0 87.5 87.0 86.5 86.0 85.5 85.028.5

    29.0

    29.5

    30.0

    30.5

    31.0

    Longitude (degrees)

    Latit

    ude

    (deg

    rees

    )

    Buoy 42040

    Buoy 42039

    Buoy 42007

    SWP BURL1

    CSBF1

    Grand Isle

    Dauphin Island Perdido PassFort Walton

    Panama Beach

    Apalachicola

    PSTL1 SW Pass

    State Docks

    Mobile Airport Pensacola AirportDauphin Island Sea Lab StationMeteorology Stations

    Tide Gages

    C-MAN StationsBuoy Stations

    42007

    42040

  • Total triangle elements: 63,757 Total nodes: 32,544. Time step: 4 s

    The basinThe basin--scale mesh provides the scale mesh provides the open boundary conditions.open boundary conditions.

    Both ADCIRC and SWAN are coupled Both ADCIRC and SWAN are coupled using the same regional mesh.using the same regional mesh.

    Fine spatial resolution is needed to Fine spatial resolution is needed to resolve topographic and hydrodynamic resolve topographic and hydrodynamic features of flooding and coastal waves.features of flooding and coastal waves.

    Nested Domain for Wave-Surge Coupling

    Lake Pontchartrain

    New Orleans

    ADCIRC Mesh

  • Animation of Modeled Significant Wave Heights (m) during Gustav

  • Modeled Maximum Significant Wave Heights (m) during Gustav

  • Wave Comparison

    90.5 90.0 89.5 89.0 88.5 88.0 87.5 87.0 86.5 86.0 85.5 85.028.5

    29.0

    29.5

    30.0

    30.5

    31.0

    Longitude (degrees)

    Latit

    ude

    (deg

    rees

    )

    Buoy 42040

    Buoy 42039

    Buoy 42007

    SWP BURL1

    CSBF1

    Grand Isle

    Dauphin Island Perdido PassFort Walton

    Panama Beach

    Apalachicola

    PSTL1 SW Pass

    State Docks

    Mobile Airport Pensacola AirportDauphin Island Sea Lab StationMeteorology Stations

    Tide Gages

    C-MAN StationsBuoy Stations

    25 30 35 40 45 50 55 60 65 700

    2

    4

    6

    8

    10

    12Buoy 42040 (Red for the observed;Blue for the unstructured modeled;

    Hs

    (m)

    25 30 35 40 45 50 55 60 65 700

    2

    4

    6

    8

    10

    12Black for the rectangular modeled; Green for the unstructured modeled with no elevation change)

    Wav

    e pe

    riod

    (s)

    25 30 35 40 45 50 55 60 65 700

    2

    4

    6

    8

    10

    12Buoy 42007

    Hs

    (m)

    Hours from 2008-08-30 00h25 30 35 40 45 50 55 60 65 70

    0

    2

    4

    6

    8

    10

    12

    Wav

    e pe

    riod

    (s)

    Hours from 2008-08-30 00h

    42007

    42040

  • Contribution of Storm Surge to Maximum Significant Wave Heights during Gustav

  • Conclusion and Future Work

    A coastal modeling framework has been developed at LSU using spectral/hp DG methods and HPC technologies.

    Effects of vegetation and fluid mud are being incorporated into the solvers.

    Recent hurricanes provide an excellent testbed for skill assessment of coupled, unstructured surge, wave, salinity and sediment transport models

  • Surge Comparison

    20 30 40 50 60 70 800

    0.5

    1

    1.5

    2PSTL1 (red for the observed; blue for the calculated)

    Sur

    ge (m

    )

    20 30 40 50 60 70 800

    0.5

    1

    1.5

    2

    Sur

    ge (m

    )

    Dauphin Island (red for the observed; blue for the calculated)

    20 30 40 50 60 70 800

    0.5

    1

    1.5

    2

    Sur

    ge (m

    )

    Panama City

    20 30 40 50 60 70 800

    0.5

    1

    1.5

    2

    Sur

    ge (m

    )

    Pensacola

    Hours from 08-30-2008 00:00

    20 30 40 50 60 70 800

    0.5

    1

    1.5

    2

    Sur

    ge (m

    )

    Mobile State Docks

    Hours from 08-30-2008 00:0020 30 40 50 60 70 80

    0

    0.5

    1

    1.5

    2S

    urge

    (m)

    GDIL1

    Hours from 08-30-2008 00:00

    90.5 90.0 89.5 89.0 88.5 88.0 87.5 87.0 86.5 86.0 85.5 85.028.5

    29.0

    29.5

    30.0

    30.5

    31.0

    Longitude (degrees)

    Latit

    ude

    (deg

    rees

    )

    Buoy 42040

    Buoy 42039

    Buoy 42007

    SWP BURL1

    CSBF1

    Grand Isle

    Dauphin Island Perdido PassFort Walton

    Panama Beach

    Apalachicola

    PSTL1 SW Pass

    State Docks

    Mobile Airport Pensacola AirportDauphin Island Sea Lab StationMeteorology Stations

    Tide Gages

    C-MAN StationsBuoy Stations

  • Modeled Maximum SurgeHeights (m, MSL) during Gustav

  • What is the cause of Katrinas record high surge?

    250 200 150 100 50-500

    -400

    -300

    -200

    -100

    0

    Ave

    rage

    dep

    th b

    elow

    MS

    L (m

    )

    Distance from shoreline along the track of hurricane (km)

    KatrinaIvanFrederic

    50 40 30 20 10 00

    2

    4

    6

    8

    10

    Sur

    ge H

    eigh

    t (m

    ,MS

    L)

    Distance from shoreline (km)

    Sloping BottomFlat Bottom

    Chen et al. (2008)

  • Structure of Cactus

    Core Flesh

    Plug-In Thorns(modules)

    driverdriver

    input/outputinput/output

    interpolationinterpolation

    SOR solverSOR solver

    coordinatescoordinates

    boundaryboundary conditionsconditions

    black holesblack holes

    equations of stateequations of state

    remote steeringremote steering

    wave evolverswave evolvers multigridmultigrid

    parametersparameters

    gridgrid variablesvariables

    errorerror handlinghandling

    schedulingscheduling

    extensibleextensible APIsAPIs

    makemake systemsystem

    ANSI CANSI C

    Fortran/C/C++Fortran/C/C++

    Reservoir simulatorsReservoir simulators

    Coastal modellingCoastal modelling

    Molecular dynamicsMolecular dynamics

  • A Fully Nonlinear Boussinesq Model for Waves and Currents

    Continuity Equation

    Momentum Equation

    )( 04 ghOM

    t =+

    )(

    )()(

    0

    04

    3212

    lghORRR

    VVVguut

    u

    wfi

    =+

    +++++

    Wei et al. (1995) and Chen et al. (2003, 2004, 2006)

    = free surface elevationM = volume rate of flow

    Boussinesq dispersive terms

    Momentummixing term

    Wind forcing

    Conservation ofpotential vorticity

  • Sea-Dependent Wind Drag Coefficients

    A new formulation

    Wu (1980)

    10213 |)|(10 bUaaC xd ++=

    103 065.08.010 UCd +=

    a1= 0.2a2 = 18b = 0.065

    Surface slope

    (Sea-independent)

    0 5 10 15 20 25 300

    0.5

    1

    1.5

    2

    2.5

    3

    U10(m/s)

    Cd

    x 10

    3

    ka0.2

    0.15

    0.1

    0.05

    Large and Pond (81)Smith (80)Smith and Banke (75)Geernaert et al.(87)Geernaert et al.(86)Sheppard et al. (72)Donelan (82)Denman and Miyake (73)Pond et al. (71)Graf et al. (84)

    Ka=2

    (Chen et al, 2004)

    Wind Stress = Skin Friction + Form Drag

    Challenges Depth-Integrated Coastal Models Modeling Surge/Wave Attenuation by Porous Media or VegetationWhy higher-ordernumerical methods? Unstructured Spectral/hp DG MethodsCOMI ArchitectureNektar++ library Discontinuous Galerkin Spectral Element Model Parallel DG Model Weak ScalingHurricane GustavWind ComparisonNested Domain for Wave-Surge CouplingAnimation of Modeled Significant Wave Heights (m) during GustavModeled Maximum Significant Wave Heights (m) during GustavWave ComparisonContribution of Storm Surge to Maximum Significant Wave Heights during GustavConclusion and Future WorkSurge ComparisonModeled Maximum SurgeHeights (m, MSL) during Gustav What is the cause of Katrinas record high surge?Structure of CactusA Fully Nonlinear Boussinesq Model for Waves and CurrentsSea-Dependent Wind Drag Coefficients