Numerical methods III (Advection: the semi-Lagrangian technique)

Slides:



Advertisements
Similar presentations
Chapter 11 Differentiation.
Advertisements

Chapter 17 Multivariable Calculus.
Formulation of linear hydrodynamic stability problems
Lectures D25-D26 : 3D Rigid Body Dynamics
Finite Difference Discretization of Hyperbolic Equations: Linear Problems Lectures 8, 9 and 10.
An Introduction to Numerical Methods
ECMWF Governing Equations 2 Slide 1 Governing Equations II: classical approximations and other systems of equations Thanks to Clive Temperton, Mike Cullen.
by Sylvie Malardel (room 10a; ext. 2414)
Numerics of physical parametrizations
ECMWF Governing Equations 1 Slide 1 Governing Equations I by Clive Temperton (room 124) and Nils Wedi (room 128)
Governing Equations IV
Finite differences Finite differences.
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.
Visualization Techniques -
Non-Central Coordinate System
Motion in Two and Three Dimensions
Splines I – Curves and Properties
2ª aula Evolution Equation. The Finite Volume Method.
Maria Ugryumova Direct Solution Techniques in Spectral Methods CASA Seminar, 13 December 2007.
P. Venkataraman Mechanical Engineering P. Venkataraman Rochester Institute of Technology DETC2009 – Solving Inverse ODE using Bezier Functions 29.
State Variables.
Chapter 8 Elliptic Equation.
P. Venkataraman Mechanical Engineering P. Venkataraman Rochester Institute of Technology DETC2013 – 12269: Continuous Solution for Boundary Value Problems.
Remark: foils with „black background“ could be skipped, they are aimed to the more advanced courses Rudolf Žitný, Ústav procesní a zpracovatelské techniky.
By S Ziaei-Rad Mechanical Engineering Department, IUT.
CE33500 – Computational Methods in Civil Engineering Differentiation Provided by : Shahab Afshari
Semi-Lagrangian Dynamics in GFS Sajal K. Kar. Introduction Over the years, the accuracy of medium-range forecasts has steadily improved with increasing.
Coupled Fluid-Structural Solver CFD incompressible flow solver has been coupled with a FEA code to analyze dynamic fluid-structure coupling phenomena CFD.
Atms 4320 Lab 2 Anthony R. Lupo. Lab 2 -Methodologies for evaluating the total derviatives in the fundamental equations of hydrodynamics  Recall that.
Motion Analysis (contd.) Slides are from RPI Registration Class.
Introduction to Numerical Methods I
ON MULTIVARIATE POLYNOMIAL INTERPOLATION
Chapter 6 Numerical Interpolation
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.
A Look at High-Order Finite- Volume Schemes for Simulating Atmospheric Flows Paul Ullrich University of Michigan.
Most physically significant large-scale atmospheric circulations have time scales on the order of Rossby waves but much larger than the time scales.
A cell-integrated semi-Lagrangian dynamical scheme based on a step-function representation Eigil Kaas, Bennert Machenhauer and Peter Hjort Lauritzen Danish.
Integration of 3-body encounter. Figure taken from
3.3.3: Semi-Lagrangian schemes AOSC614 class Hong Li.
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.
+ Numerical Integration Techniques A Brief Introduction By Kai Zhao January, 2011.
1 Finite Elements. 2 Finite elements Formally identical to spectral technique Dependent variables expanded as linear combinations of basis local, non.
The status and development of the ECMWF forecast model M. Hortal, M. Miller, C. Temperton, A. Untch, N. Wedi ECMWF.
Optical Flow. Distribution of apparent velocities of movement of brightness pattern in an image.
Numerical methods 1 An Introduction to Numerical Methods For Weather Prediction by Mariano Hortal office 122.
Discretization Methods Chapter 2. Training Manual May 15, 2001 Inventory # Discretization Methods Topics Equations and The Goal Brief overview.
1 Solving the algebraic equations A x = B =. 2 Direct solution x = A -1 B = = Applicable only to small problems For the vertical in the spectral technique.
Optimal Path Planning Using the Minimum-Time Criterion by James Bobrow Guha Jayachandran April 29, 2002.
Deutscher Wetterdienst Flux form semi-Lagrangian transport in ICON: construction and results of idealised test cases Daniel Reinert Deutscher Wetterdienst.
The story so far… ATM 562 Fovell Fall, Convergence We will deploy finite difference (FD) approximations to our model partial differential equations.
Lecture 3 & 4 : Newtonian Numerical Hydrodynamics Contents 1. The Euler equation 2. Properties of the Euler equation 3. Shock tube problem 4. The Roe scheme.
Date of download: 7/9/2016 Copyright © ASME. All rights reserved. From: Convective Motion and Heat Transfer in a Slowly Rotating Fluid Quasi-Sphere With.
ECMWF Semi-Lagrangian semi-implicit technique in IFS Slide 1 The semi-Lagrangian semi-implicit technique in the ECMWF model by Michail Diamantakis (room.
The semi-Lagrangian semi-implicit technique in the ECMWF model
Lecture 4: Numerical Stability
Introduction to Numerical Methods I
Modeling of Traffic Flow Problems
Nodal Methods for Core Neutron Diffusion Calculations
Linearized Block Implicit (LBI) Method applied to Quasi-1D Flow
3D Image Segmentation and Multigrid Method
Objective Numerical methods Finite volume.
Introduction to Fluid Dynamics & Applications
topic4: Implicit method, Stability, ADI method
topic4: Implicit method, Stability, ADI method
topic4: Implicit method, Stability, ADI method
topic4: Implicit method, Stability, ADI method
6th Lecture : Numerical Methods
Physics 319 Classical Mechanics
Topic 3 Discretization of PDE
Presentation transcript:

Numerical methods III (Advection: the semi-Lagrangian technique) by Nils Wedi (room 007; ext. 2657) Based on previous material by Mariano Hortal and Agathe Untch

Advection: The semi-Lagrangian technique material time derivative or time evolution along a trajectory thus avoiding quadratic terms; disadvantage: not flux-form! x x x x x x x From a regular array of points we end up after Δt with a non-regular distribution x x x Semi-Lagrangian: (usually) tracking back Solution of the one-dimensional advection equation: computing the origin point via trajectory calculation origin point interpolation

Stability in one dimension e.g. McDonald (1987) Linear advection equation without r.h.s. p Origin of parcel at j: X* =Xj-U0Δt x α j “multiply-upstream” p: integer Linear interpolation α is not the CFL number except when p=0, then=> upwind Von Neumann: |λ|≤1 if 0 ≤α ≤1 (interpolation from two nearest points) Damping!

Cubic spline interpolation S(x) is a cubic polynomial - S(xj)=φj at the neighbouring grid points - ∂S(x)/ ∂x is continuous - ∫d2S/dx2 dx is minimal Then: S(x)=Dj-1(xj-x)2(x-xj-1)/(Δx)2-Dj(x-xj-1)2(xj-x) /(Δx)2 + +φj-1(xj-x)2[2(x-xj-1)+ Δx] /(Δx)3+ φj(x-xj-1)2[2(xj-x)+ Δx] /(Δx)3 where (Dj-1+4Dj+Dj+1)/6=(φj+1- φj-1)/2 Δx

Shape-preserving interpolation • Creation of artificial maxima /minima x: grid points x x x x: interpolation point x x • Shape-preserving and quasi-monotone interpolation - Spline or Hermite interpolation derivatives modified derivatives x x interpolation φmax φmin x x - Quasi-monotone interpolation

Interpolation in the IFS semi-Lagrangian scheme ECMWF model uses quasi-monotone quasi-cubic Lagrange interpolation Cubic Lagrange interpolation: with the weights Number of 1D cubic interpolations in two dimensions is 5, in three dimensions 21! x y To save on computations: cubic interpolation only for nearest neighbour rows, linear interpolation for rest => “quasi-cubic interpolation” => 7 cubic + 10 linear in 3 dimensions.

3-t-l Semi-Lagrangian schemes in 2-D L: linear operator N: non-linear function • Interpolating x x x x G I x o x Two interpolations needed • Ritchie scheme U=U*+U’ V=V*+V’ x x x x G I’ 2V*Δt o’ o x 2V’Δt The three of them are second-order accurate in space-time • Non-interpolating Average the non-linear terms between points G and o’

Stability of 2-D schemes In the linear advection equation the interpolating scheme is stable provided the interpolations use the nearest points In the linear shallow-water equations (with rotation), treating the linear terms implicitly, the stability limit is In the two non-interpolating schemes the stability is given by Δt f ≤1 Coriolis term (kU’+lV’) Δt ≤1 Which is always true due to the definition of U’ and V’

Treatment of the Coriolis term In three-time-level semi-Lagrangian: • treated explicitly with the rest of the rhs In two-time-level semi-Lagrangian: Extrapolation in time to the middle of the trajectory leads to instability (Temperton (1997)) Two stable options: • Advective treatment: (Vector R here is the position vector.) • Implicit treatment : Helmholtz eqs partially coupled for individual spectral components => tri-diagonal system to be solved.

Semi-Lagrangian advection on the sphere Momentum eq. is discretized in vector form (because a vector is continuous across the poles, components are not!) Interpolations at departure point are done for components u & v of the velocity vec- tor relative to the system of reference local at D. Interpolated values are to be used at A, so the change of reference system from D to A needs to be taken into account. Trajectories are arcs of great circles if constant (angular) velocity is assumed for the duration of a time step. X Y Z A V x D M i j A D Tangent plane projection Trajectory calculation

Iterative trajectory calculation V1Δt V0Δt x x x x x r0 r1 can be taken as trajectory straight line or great circle assumed constant during 2Δt or implicit

Iterative trajectory computation (1 dimension) r(n+1)=g-V0(n)Δt Where, for simplicity, we have taken a 2-time-level scheme and taken the velocity at the departure point of the trajectory Assume that V varies linearly between grid-points V=a+b.r  b = r (n+1) = g - aΔt - Δt b r(n) For this procedure to converge, it must have a solution of the form r = λn + K; (| λ| < 1) Substituting, we get K=(g - a Δt)/(1 + b Δt) and λ = -b Δt  |b|Δt < 1 more generally for three dimensions this translates to the determinant of a matrix: Lipschitz condition (Smolarkiewicz and Pudykiewicz, 1992) Physical meaning: To prevent trajectory intersections !!! It is in general less restrictive than the CFL condition and it does not depend on the mesh size.

Mass conservation Semi-Lagrangian formulations are based on the advective form of the equations but can be made mass conserving (e.g. Zerroukat 2007; Kaas 2008) 2 fundamental issues: The iterative trajectory calculation should (but normally does not) involve the continuity equation. The conservation properties of the interpolation operator are important.

Semi-Lagrangian advection with rhs • Three-time-level schemes - centered (second-order accurate) scheme - split in time (first-order accurate) - R at the departure point (first-order accurate) the r.h.s. R can be evaluated by interpolation to the middle of the trajectory or averaged along the trajectory: RM(t)={RD(t)+RA(t)}/2

Example Z = Re( Ae-ikx eωt) Let us apply each of the above schemes to the equation whose analytical solution (with appropriate initial and boundary conditions) is: Z = Re( Ae-ikx eωt) with ω=ikU0-k2K WARNING: the three-time-level scheme applied to the diffusion eq. has an absolutely unstable numerical solution! With the values A=1, k=2π/100, K=10-2, the r.m.s. error with respect to the analytical solution (before the unstable numerical solution grows too much) grows linearly with time. After 200 sec of integration, the error is: 5×10-4 Δt for the split treatment 5×10-4 Δt for r.h.s. at departure point 5×10-8 (Δt)2 for the centered scheme

Semi-Lagrangian advection with rhs Two-time-level (second-order accurate) schemes : with Extrapolation in time to middle of time interval Unstable! => noisy forecasts Forecast of temperature at 200 hPa (from 1997/01/04)

Stable extrapolating two-time-level semi-Lagrangian (SETTLS): Taylor expansion to second order With and Forecast 200 hPa T from 1997/01/04 using SETTLS

Trajectory computation with SETTLS Mean velocity during Δt The Taylor expansion from which we started is: which represents a uniformly accelerated movement Note: The middle of the trajectory is not the average between the departure and the arrival points.