Quasi-3D Multiscale Modeling Framework as a Physics Option in CAM-SE

Transcription

1 Quasi-3D Multiscale Modeling Framework as a Physics Option in CAM-SE Joon-Hee Jung, Celal S. Konor, Don Dazlich, David A. Randall, Department of Atmospheric Sciences, Colorado State University, USA Peter Lauritzen, and Steve Goldhaber National Center for Atmospheric Research, Boulder, Colorado, USA This research has been supported by NSF AGS (and partly by DOE ACME DE-SC and DOE CMDV DE-SC ). CESM: Atmosphere Model Working Group Meeting NCAR, Mesa Lab, Boulder, CO, February 12-14, 2018

2 Multiscale Modeling Framework (aka Superparameterization) GCM Cloud Resolving Model (CRM) Traditional subgrid-scale parameterizations in the GCMs are replaced with explicit simulations by a 2D CRM embedded in each GCM grid column. MMF community: CSU, State Univ. of New York at Stony Brook, Univ. of Washington, Univ. of Chicago, UC at Irvine, Univ. of Oxford, Harvard Univ., George Mason Univ., NCAR, NASA GSFC & LRC, ECMWF, LBNL, SIO, ESRL, PNNL, MIT, IITM, DOE (E3SM)

3 Q3D MMF Jung and Arakawa (2010, 2014), Jung (2016) CRMs in GCM grid columns are seamlessly connected; Two perpendicular sets of CRMs are used; CRMs are three-dimensional, although covering only channel-like domains. Q3D MMF MMF y y z z x x GCM grid cell GCM grid CRM grid CRM ghost grid

4 Q3D MMF Jung and Arakawa (2010, 2014), Jung (2016) CRMs in GCM grid columns are seamlessly connected; Two perpendicular sets of CRMs are used; CRMs are three-dimensional, although covering only channel-like domains. Convections are allowed to propagate across the GCM cell boundaries; The surface orography can be resolved by the CRMs; Vertical momentum transport by convection and waves can be simulated. Q3D MMF MMF y y z z x x GCM grid cell GCM grid CRM grid CRM ghost grid

5 Q3D MMF inherits the structure of the conventional GCMs in Coupling the Dynamical Core and Physics FORCING GCM Effects on CRM GCM Dynamical Core FEEDBACK CRM Effects on GCM CRM Parameterized Physics This is like the first-generation MMF.

6 Why does the Q3D MMF still need a separate GCM dynamical core?

7 Why does the Q3D MMF still need a separate GCM dynamical core? The network of CRM channels cannot replace a GCM. The CRM channels do not and cannot directly interact with each other; Averaging the solutions of the CRM channels does not represent fully interacting 3D large-scale dynamical processes; To simulate well-behaving large-scale features, a reasonable 3D GCM dynamics is necessary; The computational cost of a GCM dynamics is not a burden comparing to that of CRM channels; It is good to maintain compatibility with conventional GCMs.

8 There is no double counting or spurious competition between the GCM and CRM solutions.

9 There is no double counting or spurious competition between the GCM and CRM solutions. GCM solution at t=t0 + dtgcm CRM-mean within a GCM cell Eddy transport and diabatic effects added to GCM s dynamical effects t = t0 + dtgcm t ep s GCM solution at t=t0 t0 x GCM grid GCM solution interpolated to CRM grids R M st dtgcm C G C M st ep t t = t0 Decomposition of a CRM variable: background-field CRM grid x (GCM grid-scale) + eddy GCM gathers only eddy transport and diabatic effects (i.e., subgrid-scale effects) from the CRMs. CRM maintains its GCM grid-scale solution close to GCM s solution through a relaxation.

10 Scale-awareness of the Q3D MMF All subgrid-scale processes are represented by the CRMs with the same grid size, regardless of the resolution of the GCM; CRMs can figure out the GCM s resolution; For example, with a fine-grid, the vertical velocities of the GCM can be much larger (e.g., in a mesoscale system) and also the horizontal gradients can be much larger (e.g., near a front). The feedbacks of CRMs are sensitive to the GCM s resolution, because the collective effects of subgrid-scale processes will correspond to the smallest scale (i.e., grid size) of the GCM.

11 Development of a Global Q3D MMF A North-South CRM channel CAM-SE dynamical core + Vector Vorticity Model (CRM developed at CSU/UCLA) An East-West CRM channel

12 Development of a Global Q3D MMF A North-South CRM channel CAM-SE dynamical core + Vector Vorticity Model (CRM developed at CSU/UCLA) An East-West CRM channel The VVM has been developed and applied as a limited-area model. To be applied globally (i.e., wrapped around the planet), it has been extended to the curvilinear coordinates. To couple the VVM with the CAM-SE dynamical core, a computational infrastructure has been designed and is now under construction.

13 The VVM predicts the Vorticity Vector. Large-scale motions are controlled by the vertical component of vorticity Small-scale motions are controlled by the horizontal component of vorticity

14 The VVM predicts the Vorticity Vector. Large-scale motions are controlled by the vertical component of vorticity Small-scale motions are controlled by the horizontal component of vorticity 3D nonhydrostatic anelastic model; Pressure gradient force is eliminated; Vertical velocity is a solution of a 3D elliptic equation; (Lower boundary condition over steep topography is easily formulated.) CRM-type physics parameterizations are included. Suitable for the Q3D-type application

15 Governing Equations of the VVM in the Curvilinear Coordinates Vorticity Equation ξ 1 = t G x η 1 = t G x ( G uξ + y ) ( u u 1 θ u G vξ ( wξ ) + ξ + η + (ζ + f ) + g y x z G y θ 0 z ) ( G uη + y ) ( v v 1 θ v G vη ( wη ) + ξ + η + (ζ + f ) g y x z G x θ 0 z ) w w ζ 1 w = Gu ζ + f + Gv ζ + f w ζ + f + ξ + η + ζ + f ( ) ( ) ( ) ( ) z t x y z x y G where x = q1 2 y = q u = u1 2 v = u u! = u1 v! = u2 G : Jacobian of transformation 1 ξ = (ω H ) = 2 η = (ω H ) = ζ = ω z = 1 w v! y z G 1 u! w z x G 1 v! u! x y G

16 Governing Equations of the VVM (Cont.) Nondivergence of 3D vorticity ζ = z zt 1 G x ( ) Gξ + y ( ) Gη dz + (ζ )z=z T Continuity Equation 1 G x ( ) Gρ0u + y ( (ρ0 w ) Gρ0 v + =0 z ) w-equation 1 G 11 w 21 w 12 w 22 w +g + G g +g G g x y y x y x! ξ! 1 (ρ0 w ) 1 η + = z ρ0 z G x y

17 Latitude (degrees) EQ - VVM: Barotropic Instability Test SPHERICAL LOW RESOLUTION t = 96 hr dx=dy ~ 100 km EQ - (dx=dy ~ 100 km) CUBED-SPHERE dx max =dy max ~ 100 km dx min =dy min ~ 84 km t = 96 hr Latitude (degrees) EQ - t = 144 hr Longitude (degrees) EQ - t = 144 hr Longitude (degrees) (s -1 )

18 Latitude (degrees) EQ - VVM: Barotropic Instability Test SPHERICAL LOW RESOLUTION t = 96 hr dx=dy ~ 100 km EQ - (dx=dy ~ 100 km) CUBED-SPHERE dx max =dy max ~ 100 km dx min =dy min ~ 84 km t = 96 hr Latitude (degrees) Latitude (degrees) Latitude (degrees) EQ Longitude (degrees) EQ - EQ - SPHERICAL Longitude (degrees) t = 144 hr EQ - HIGH RESOLUTION t = 96 hr t = 144 hr dx=dy ~ 5 km t = 144 hr Longitude (degrees) EQ EQ - (dx=dy ~ 5 km) CUBED-SPHERE Longitude (degrees) dx max =dy max ~ 5 km dx min =dy min ~ 4 km t = 96 hr t = 144 hr (s -1 )

19 VVM: Baroclinic Instability Test LOW RESOLUTION (dx = dy ~ 100 km) Zonally Uniform Initial State ( ) Idealized setting: f = 2Ωsin ϕ + π 4 Height (km) Latitude (deg) Initial Random Perturbation Longitude (deg) Latitude (deg) (K) Latitude (deg) (m/s) Potential Temp. at the lowest model layer (Day 12) SPHERICAL CUBED-SPHERE Longitude (deg) Latitude (deg) Longitude (deg)

20 Computational Infrastructure for coupling the VVM with CAM-SE Dycore Collaboration with Peter Lauritzen and Steve NCAR

21 Computational Infrastructure for coupling the VVM with CAM-SE Dycore Collaboration with Peter Lauritzen and Steve NCAR Using the finite-volume physics grid for CAM-SE, including the coupling to the spectral element dynamics grid (CAM-SE-physgrid). This physics grid forms the virtual GCM (vgcm) grid for the Q3D system. dynamics grid y physics grid y y x vgcm grid x x CRM grid CRM ghost grid

22 Computational Infrastructure (Cont.) Computing the (cube) face-edge ghost cells in the dycore, which have never existed in CAM before (these ghost cells are not part of the original physics grid). Background Field (GCM grid scale) of the VVM is interpolated from the vgcm fields. y x Moving the vgcm data from the dynamics decomposition to the physics decomposition.

23 Computational Infrastructure (Cont.) (In the physics decomposition) F5 Organizing all vgcm data (including the ghost cells) in a way that they can be properly assigned to the CRM channels. F4 F4 F1 F2 F6 Group 1 Group 2 Group 3 F5 F5 F5 F1 F6 F2 F3 F4 F1 F6 F2 F3 F3 F4 F1 F6 F2 F3

24 Computational Infrastructure (Cont.) Moving the vgcm data from the physics decomposition to the channel decomposition: Each vgcm s data is placed to the correct location of CRM channel. (Also, moving the mean CRM data from the channel decomposition to the physics decomposition.) channel decomposition is where each task has all the data for one or more CRM channels. (No MPI communications are necessary for the CRM calculations.)

25 POST-CRM Procedures DYCORE COMPUTE Distribution of CRM feedbacks to Dycore points vgcm communications (moving data to physics decomposition) CRM PREDICTION Calculation of mean CRM feedbacks (physics tendencies) CRM COMPUTE PRE-CRM Procedures CRM COMPUTE CRM COMPUTE CRM COMPUTE Interpolation of vgcm data to CRM points vgcm communications (moving data to channel decomposition) Generation of data on physics grids Q3D MMF Computation Algorithm Similar to the coupling approach of CAM-SE Q3D CAM-SE PREDICTION (Dycore + CRM effects) physics t0 t

26 Summary Our goal is to make global versions of the Q3D MMF: Q3D MMF as a Physics Option in CAM-SE (also in E3SM) Finished the preparation of the VVM with the standard cubedsphere grid, which will be coupled with the CAM-SE Dycore to form a global Q3D MMF. Created the Q3D physics package in a CAM branch. Making progress in designing and constructing a computational infrastructure (started implementing the codes and the communication layout).