1 A Framework for The Analysis of Physics-Dynamics Coupling Strategies Andrew Staniforth, Nigel Wood (Met Office Dynamics Research Group) and Jean Côté (Meteorological Service of Canada)
2 Outline What do we mean by physics? Extending the framework of Caya et al (1998) Some coupling strategies Analysis of the coupling strategies Summary
3 What do we mean by physics? Molecular effects Radiation Microphysics (conversion terms, latent heating) Underpinning fluid dynamical equations: Physics = Boundary Conditions+RHS:
4 Filtered equation set Added sub-filter physics (neglect molecular effects): Modified Boundary Conditions Turbulence Convection Gravity-wave stress
5 Physics Physics split into two parts: –True physics (radiation, microphysics) (albeit filtered) –Sub-filter physics = function of filter scale (NOT a function of grid scale though often assumed to be! Convergence issues Mason and Callen ‘86) »CRMs, LES etc = turbulence »NWP/GCMs = turbulence + convection + GWD »Mesoscale = turbulence + ? Boundary conditions also!
6 What do we mean by physics-dynamics coupling? For explicit models with small t no issue - all terms (physics/dynamics) handled in the same way (ie most CRMs, LES etc) and even if not then at converged limit Real issue only when t large compared to time scale of processes, then have to decide how to discretize terms - but in principle no different to issues of dynamical terms (split is arbitrary - historical?) BUT many large scale models have completely separated physics from dynamics inviscid predictor + viscid physics corrector (boundary conditions corrupted)
7 Holy grail of coupling Large scale modelling ( t large): SISL schemes allow increased t and therefore a balancing of the spatial and temporal errors whilst retaining stability and accuracy (for dynamics at least) If physics is not handled properly then the coupling will introduce O( t ) errors and the advantage of SISL will be negated Aim: Couple with O( t 2 ) accuracy + stability
8 Framework for analysing coupling strategies Numerical analysis of dynamics well established Some particular physics aspects well understood (eg diffusion) but largely in isolation Caya, Laprise and Zwack (1998) simple model of coupling: Regard as either a simple paradigm or F(t) is amplitude of linear normal mode (Daley 1991)
9 CLZ98’s model represents either a damping term (if real and > 0) or oscillatory term (dynamics) if imaginary G = const. forcing (diabatic forcing in CLZ98) Model useful (CLZ98 diagnosed problem in their model) but: –neglects advection (and therefore cannot analyse eg SL orographic resonance) –neglects spatio-temporal forcing terms
10 Extending CLZ98’s model These effects can be easily included whilst retaining the analytical tractability of CLZ98: Since Eq is linear can Fourier decompose where is the k th Fourier mode of Assumed only one frequency of oscillation
11 Exact Regular Solution Consider only 1 dynamics oscillatory process, 1 (damping) physics process: Solution = sum of free and forced solution:
12 Exact Resonant Solution Resonance occurs when denominator of forced solution vanishes, when: Solution = sum of free and resonant forced solution: which, as all terms are real, reduces to:
13 Application to Coupling Discretizations Assume a semi-Lagrangian advection scheme Apply a semi-implicit scheme to the dynamical terms (e.g. gravity modes) Consider 4 different coupling schemes for the physics: –Fully Explicit/Implicit –Split-implicit –Symmetrized split-implicit Apply analysis to each
14 Fully Explicit/Implicit Time-weights: dynamics, physics, forcing =0 Explicit physics - simple but stability limited =1 Implicit physics - stable but expensive
15 Split-Implicit Two step predictor corrector approach: First = Dynamics only predictor (advection + GW) Second = Physics only corrector
16 Symmetrized Split-Implicit Three step predictor-corrector approach: First = Explicit Physics only predictor Second = Semi-implicit Dynamics only corrector Third = Implicit Physics only corrector
17 Analysis Each of the 4 schemes has been analysed in terms of its: –Stability –Accuracy –Steady state forced response –Occurrence of spurious resonance
18 Stability Stability can be examined by solving for the free mode by seeking solutions of the form: and requiring the response function to have modulus 1
19 Accuracy Accuracy of the free mode is determined by expanding E in powers of t and comparing with the expansion of the analytical result:
20 Forced Regular Response As for the free mode, the forced response can be determined by seeking solutions of the form: Accuracy of the forced response can again be determined by comparison with the exact analytical result.
21 Steady State Response of the Forced Solution A key aspect of a parametrization scheme (and often the only fully understood aspect) is its steady state response when k =0 and >0. Accuracy of the steady-state forced response can again be determined by comparison with the exact analytical result:
22 Forced Resonant Solution Resonance occurs when the denominator of the Forced Response vanishes For a semi-Lagrangian, semi-implicit scheme there can occur spurious resonances in addition to the physical (analytical) one
23 Results I Stability: –Centring or overweighting the Dynamics and Physics ensures the Implicit, Split-Implicit and Symmetrized Split-Implicit schemes are unconditionally stable Accuracy of response: –All schemes are O( t) accurate –By centring the Dynamics and Physics the Implicit and Symmetrized Split-Implicit schemes alone, are O( t 2 )
24 Results II Steady state response: –Implicit/Explicit give exact response independent of centring –Split-implicit spuriously amplifies/decays steady-state –Symmetrized Split-Implicit exact only if centred Spurious resonance: –All schemes have same conditions for resonance –Resonance can be avoided by: »applying some diffusion ( >0) or »overweighting the dynamics (at the expense of removing physical resonance)
25 Summary Numerics of Physics-Dynamics coupling key to continued improvement of numerical accuracy of models Caya et al (1998) has been extended to include: –Advection (and therefore spurious resonance) –Spatio-temporal forcing Four (idealised) coupling strategies analysed in terms of: –Stability, Accuracy, Steady-state Forced Response, Spurious Resonance