Time Integration Schemes Bill Skamarock NCAR/MMM 1.Canonical equations for scheme analyses. 2.Time-integration schemes used in NWP models.

Slides:

Advertisements

Similar presentations

Stable Fluids A paper by Jos Stam.

Advertisements

Finite Difference Discretization of Hyperbolic Equations: Linear Problems Lectures 8, 9 and 10.

ECMWF Governing Equations 3 Slide 1 Numerical methods IV (time stepping) by Nils Wedi (room 007; ext. 2657) In part based on previous material by Mariano.

Introductory Control Theory I400/B659: Intelligent robotics Kris Hauser.

2ª aula Evolution Equation. The Finite Volume Method.

Reduction of Temporal Discretization Error in an Atmospheric General Circulation Model (AGCM) Daisuke Hotta Advisor: Prof. Eugenia Kalnay.

Finite Difference Methods

(c) MSc Module MTMW14 : Numerical modelling of atmospheres and oceans Staggered schemes 3.1 Staggered time schemes.

Ecosystem Modeling I Kate Hedstrom, ARSC November 2008.

Günther Zängl, DWD1 Improvements for idealized simulations with the COSMO model Günther Zängl Deutscher Wetterdienst, Offenbach, Germany.

Ordinary Differential Equations

Numerical Integration of Partial Differential Equations (PDEs)

Page 1© Crown copyright 2007 High-resolution modelling in support of T-REX observations Simon Vosper and Peter Sheridan Met Office, UK T-REX Workshop,

Total Recall Math, Part 2 Ordinary diff. equations First order ODE, one boundary/initial condition: Second order ODE.

CSE245: Computer-Aided Circuit Simulation and Verification Lecture Note 5 Numerical Integration Spring 2010 Prof. Chung-Kuan Cheng 1.

Interactive Animation of Structured Deformable Objects Mathieu Desbrun Peter Schroder Alan Barr.

Atms 4320 Lab 2 Anthony R. Lupo. Lab 2 -Methodologies for evaluating the total derviatives in the fundamental equations of hydrodynamics  Recall that.

ECIV 301 Programming & Graphics Numerical Methods for Engineers Lecture 31 Ordinary Differential Equations.

Modeling, Simulating and Rendering Fluids Thanks to Ron Fediw et al, Jos Stam, Henrik Jensen, Ryan.

Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 2 Instructor: Tim Warburton.

Numerical Methods for Partial Differential Equations CAAM 452 Spring 2005 Lecture 6 Various Finite Difference Discretizations for the Advection Equations.

1 The Spectral Method. 2 Definition where (e m,e n )=δ m,n e n = basis of a Hilbert space (.,.): scalar product in this space In L 2 space where f * :

Finite Differences Finite Difference Approximations  Simple geophysical partial differential equations  Finite differences - definitions  Finite-difference.

M. Baldauf, DWD Numerical contributions to the Priority Project ‘Runge-Kutta’ COSMO General Meeting, Working Group 2 (Numerics) Bukarest,

A Look at High-Order Finite- Volume Schemes for Simulating Atmospheric Flows Paul Ullrich University of Michigan.

Matthew Fischels Aerospace Engineering Department Major Professor : Dr. R. Ganesh Rajagopalan REDUCING RUNTIME OF WIND TURBINE SIMULATION Los Alamos National.

– Equations / variables – Vertical coordinate – Terrain representation – Grid staggering – Time integration scheme – Advection scheme – Boundary conditions.

Aktionsprogramm 2003 AP 2003: LMK Properties of the dynamical core and the metric terms of the 3D turbulence in LMK COSMO- General Meeting.

Most physically significant large-scale atmospheric circulations have time scales on the order of Rossby waves but much larger than the time scales.

– Equations / variables – Vertical coordinate – Terrain representation – Grid staggering – Time integration scheme – Advection scheme – Boundary conditions.

An introduction to semi-Lagrangian methods III Luca Bonaventura Dipartimento di Matematica “F. Brioschi”, Politecnico di Milano MOX – Modeling and Scientific.

1 A Framework for The Analysis of Physics-Dynamics Coupling Strategies Andrew Staniforth, Nigel Wood (Met Office Dynamics Research Group) and Jean Côté.

Implementation of an Advection Scheme based on Piecewise Parabolic Method (PPM) in the MesoNH The main goal of this study was to estimate if the RGSW model.

Applied NWP [1.2] “Once the initialization problem was resolved in the 1960s, models based on the primitive equations gradually supplanted those based.

Comparison of convective boundary layer velocity spectra calculated from large eddy simulation and WRF model data Jeremy A. Gibbs and Evgeni Fedorovich.

1 Computational Methods Dr. Farzad Ismail School of Aerospace and Mechanical Engineering Universiti Sains Malaysia Nibong Tebal Pulau Pinang Week.

A cell-integrated semi-Lagrangian dynamical scheme based on a step-function representation Eigil Kaas, Bennert Machenhauer and Peter Hjort Lauritzen Danish.

© Crown copyright Met Office A Framework for The Analysis of Physics-Dynamics Coupling Strategies Andrew Staniforth, Nigel Wood (Met O Dynamics Research)

Point Source in 2D Jet: Radiation and refraction of sound waves through a 2D shear layer Model Gallery #16685 © 2014 COMSOL. All rights reserved.

3.3.3: Semi-Lagrangian schemes AOSC614 class Hong Li.

Properties of Filter (Transfer Function) Real part of a Transfer Function for different order periodic filters.

7. Introduction to the numerical integration of PDE. As an example, we consider the following PDE with one variable; Finite difference method is one of.

Model Task 0B: Implementing different schemes ATM 562 Fall 2015 Fovell 1.

The Case for Using High-Order Numerical Methods in Mesoscale and Cloudscale NWP Bill Skamarock NCAR/MMM.

24-28 Sept. 2012Baldauf, Reinert, Zängl (DWD)1 Michael Baldauf, Daniel Reinert, Günther Zängl (DWD, Germany) PDEs on the sphere, Cambridge, Sept.

CIS888.11V/EE894R/ME894V A Case Study in Computational Science & Engineering We will apply several numerical methods to find a steady state solution of.

Large Timestep Issues Lecture 12 Alessandra Nardi Thanks to Prof. Sangiovanni, Prof. Newton, Prof. White, Deepak Ramaswamy, Michal Rewienski, and Karen.

Deutscher Wetterdienst 1FE 13 – An Improved Third Order Vertical Advection Scheme for the Runge-Kutta Dynamical Core Michael Baldauf (DWD) Bill.

5. Integration method for Hamiltonian system. In many of formulas (e.g. the classical RK4), the errors in conserved quantities (energy, angular momentum)

Instability in Leapfrog and Forward-Backward Schemes by Wen-Yih Sun Department of Earth and Atmospheric Sciences Purdue University West Lafayette, IN.

Requirements of High Accuracy Computing 1) Adopted numerical scheme must resolve all physical time and length scales. 2) Numerical scheme must be neutrally.

Standardized Test Set for Nonhydrostatic Dynamical Cores of NWP Models

CHANGSHENG CHEN, HEDONG LIU, And ROBERT C. BEARDSLEY

Mass Coordinate WRF Dynamical Core - Eulerian geometric height coordinate (z) core (in framework, parallel, tested in idealized, NWP applications) - Eulerian.

Scales of Motion, Reynolds averaging September 22.

1 Reformulation of the LM fast- waves equation part including a radiative upper boundary condition Almut Gassmann and Hans-Joachim Herzog (Meteorological.

ECE 576 – Power System Dynamics and Stability Prof. Tom Overbye Dept. of Electrical and Computer Engineering University of Illinois at Urbana-Champaign.

Numerics of Parametrization ECMWF training course May 2006 by Nils Wedi  ECMWF.

The story so far… ATM 562 Fovell Fall, Convergence We will deploy finite difference (FD) approximations to our model partial differential equations.

Animating smoke with dynamic balance Jin-Kyung Hong Chang-Hun Kim 발표 윤종철.

Priority project CDC Task 1.4: Choice of the anelastic equation system and Milestone 3.2: Suitability of fundamental approximations PP CDC-Meeting, ,

Regional Weather and Climate Modeling:

Reporter: Prudence Chien

CHAPTER 2 - EXPLICIT TRANSIENT DYNAMIC ANALYSYS

Lecture 4: Numerical Stability

Introduction to Numerical Methods I

CSE 245: Computer Aided Circuit Simulation and Verification

Objective Numerical methods Finite volume.

Topic 8 Pressure Correction

Presentation transcript:

Time Integration Schemes Bill Skamarock NCAR/MMM 1.Canonical equations for scheme analyses. 2.Time-integration schemes used in NWP models. 3.Leapfrog scheme. 4.Semi-implicit schemes. 5.Runge-Kutta schemes. 6.Combining the schemes 7.Summary Outline

Canonical Equations for Scheme Analyses Navier-Stokes equations: Transport Linear (e.g. scalar advection with uniform flow) Nonlinear (e.g. momentum) Wave Propagation Dissipation Energy and Enstrophy

Canonical Equations for Scheme Analyses Possible canonical equations: Oscillation equation Exponential decay Exponential growth Nonlinear ODEs

Canonical Equations for Scheme Analyses Navier-Stokes equations: Transport Linear Nonlinear Wave Propagation Dissipation Energy and Enstrophy

Canonical Equations for Scheme Analyses Linear oscillation equation Exponential decay equation

Canonical Equations for Scheme Analyses is a frequency (T -1 ) (phase speed/wavelength) Atmosphere - consider the shortest resolvable wavelengths Transport velocities: maximum O(100 ms -1 ) Gravity wave phase speeds, internal waves: < 100 ms -1 Deep gravity wave phase speeds: > ms -1 (e.g. external modes in pressure coordinate models) Sound (acoustic) waves: 300 ms -1

Canonical Equations for Scheme Analyses is a frequency (T -1 ) (phase speed/wavelength)

Canonical Equations for Scheme Analyses is a decay rate (T -1 ) (viscosity/wavelength 2 ) Atmosphere - Decay rates proportional to eddy turnover times: Large scales: 1/days PBL: 1/minutes

Time Integration Schemes Used in NWP Models. ECMWF IFS, JMA GSM, DWD GME, NCEP GFS All these models use some form of Leapfrog semi- implicit (semi-Lagrangian) time integration. Leapfrog Semi-implicit wave propagation

Time Integration Schemes Used in NWP Models. ECMWF IFS, JMA GSM, DWD GME, NCEP GFS Dissipation is either handled implicitly or using forward Euler (in the leapfrog context). Forward Euler Semi-implicit dissipation

Time Integration Schemes Used in NWP Models. UKMO Unified model, GRAPES Predictor-Corrector Semi-implicit Forward Euler Semi-implicit wave propagation dissipation

Time Integration Schemes Used in NWP Models. MM5, ARPS, COAMPS Leapfrog (advection, gravity waves) Forward-backward (acoustic modes) wave propagation

Time Integration Schemes Used in NWP Models. MM5, ARPS, COAMPS Dissipation is either handled implicitly or using forward Euler (in the leapfrog context). Forward Euler Semi-implicit

Time Integration Schemes Used in NWP Models. WRF (ARW), COSMO, NICAM Runge-Kutta (advection) Forward-backward (acoustic modes) wave propagation RK3 RK2

Time Integration Schemes Used in NWP Models. WRF (ARW), COSMO, NICAM Dissipation is either handled implicitly or using forward Euler or using Runge-Kutta. Forward Euler Semi-implicit

Leapfrog Time Integration continuousdiscrete assume solutions of the form Stability: continuous solutions do not grow

Leapfrog Time Integration Two roots: A+ physical mode A- computational mode (parasitic mode or root) Stability:

Leapfrog Time Integration Relevance of the computational mode? Consider T = 0 2dt Odd and even timestep are decoupled. The amplitude of the computational mode depends on starting procedure. In practice, nonlinearities will put energy into the computation mode during an integration.

Leapfrog Time Integration Controlling the computational mode T = 0 2dt Asselin-filter leapfrog Asselin filter

Leapfrog Time Integration Controlling the computational mode T = 0 2dt Drawbacks: damping - first order in time (error O(dt), not O(dt 2 )) Asselin filter damps all modes! unfiltered filtered

Semi-Implicit Time Integration - The Implicit Component - continuous Stability: continuous solutions do not grow Implicit Stable for all timesteps with no damping!

Semi-Implicit Time Integration - The Implicit Component - Drawbacks of centered implicit integration (1)Phase Errors stability gained by reducing the frequency. (2) Need to solve 3D elliptic equation

Semi-Implicit Time Integration - The Implicit Component - Offcentering is used to stabilize the integration Replace with Amplitude

Semi-Implicit Time Integration - The Implicit Component - Amplitude Consequences: In semi-Lagrangian semi-implicit (SLSI) models that use large timesteps, physically relevent modes may be strongly damped.

Runge-Kutta Time Integration RK3RK2 (1) (2) Stable for the oscillation eqn (1) and decay eqn (2). Unstable for the oscillation eqn (1). Stable for the decay eqn (2). RK2 transport must be dissipative!

Runge-Kutta Time Integration From the ARW tutorial.

Combining Time-Integration Schemes Canonical equation Example: Linearized acoustic-mode equations Assume solutions of the form e ikx

t t+dtt+dt/3 t t+dtt+dt/2 t t+dt U t = L fast (U) + L slow (U) L s (U t )U*U* L s (U * )U ** L s (U ** ) U t+dt 3rd order Runge-Kutta, 3 steps Split-Explicit time integration ARW RK3 and Leapfrog tt+dt L s (U t )U t+dt Leapfrog - 1 step U t-dt t-dt Acoustic mode integration: forward-backward

Split-Explicit time integration ARW RK3 and Leapfrog Leapfrog, n_s = 6 (shaded regions unstable; A > 1) Perfect advection: Unstable modes exist in Leapfrog and FB stable Courant numbers. LF Asselin time filter removes these instabilities as does 3D divergence mode damping (acoustic mode damping).

Split-Explicit time integration ARW RK3 and Leapfrog RK3, n_s = 6 (shaded regions unstable; A > 1) Perfect advection: Divergence damping helps stabilize the split scheme

U t = L fast (U) + L slow (U) Split-Explicit Time Integration Forward-In-Time Transport FIT - 1 step t t+dt L s (U t ) U t+dt 1st order Upwind, n_s = 6 (shaded regions unstable; A > 1) Upwinding with FB scheme is unstable, divergence damping does not provide stability needed for applications.

Semi-Lagrangian Semi-Implicit Time Integration (UKMO Unified Model) Continuous linear acoustic equations Discrete linear acoustic equations Grid-point value Departure point following fluid trajectory - quantities needed here must be interpolated.

Semi-Lagrangian Semi-Implicit Time Integration (UKMO Unified Model) Continuous linear acoustic equations Discrete linear acoustic equations For  = 1/2, the scheme is absolutely stable for all  t for the linear problem However, for the full nonlinear equations, truncation errors in the trajectory calculations, and nonlinear terms not included in the implicit formulation lead to the need for offcentering the implicit calculation. Typically 0.6 <  < 0.8

(courtesy of Tim Palmer, 2004) Implicit-Scheme Damping - Does It Matter? ECMWF Model

Implicit-Scheme Damping - Does It Matter? ECMWF Model Correct mesoscale spectrum for the wrong reason?

Summary Linear analyses of the oscillation and decay equations reveal stability properties of the individual time integration schemes. When schemes are combined to integrate slow and fast modes, the stability of the combination is not necessarily indicated by the stability of the individual schemes. Most combined schemes need some form of filtering for stability. References: For atmospheric models, see the textbook Numerical Methods for Wave Equations in Geophysical Fluid Dynamics by Dale Durran (Springer, 1998) and references therein.