WORKING GROUP 1: NUMERICS AND MODEL DYNAMICS

Representatives:

Latest Update: 6 January 2004

Mission:

The mission of this working group is to formulate, test and evaluate approaches for solving the compressible nonhydrostatic equations for use in the WRF model, and to produce WRF prototype solvers for the most promising candidates.

Current Events:

Recently, the WRF-mass core (see Status (1) below) and the WRF-NMM (the  NCEP NMM (Nonhydrostatic Mesoscale Model) ported the WRF framework) have been evaluated by computing kinetic energy spectra from real-time and experimental forecasts.  Results from these tests and comments can be found at http://www.mmm.ucar.edu/individual/skamarock/spectra_discussion.html. Additionally, a detailed evaluation of the spectra produced by the WRF model can be found in the Monthly Weather Review manuscript Evaluating NWP Models Using Kinetic Energy Spectra.

At the recent SRNWP (Short-Range Numerical Weather Prediction) workshop in Bad Orb, Germany in October 2003, a proposal for A Standard Test Set for Nonhydrostatic Dynamical Cores of NWP Models was presented and strongly endorsed by the workshop participants.  At the 2004 AMS NWP-WAF conference (January, Seattle), the proposal was also presented and the wider community was invited to particiapate in the development of the test set.  A preliminary web page has been established.

Status:

There are 2 candidate solvers being examined by the working group at this time.

(1)   An Eulerian solver based on a flux formulation of the fully compressible nonhydrostatic equations with a mass (hydrostatic pressure) vertical coordinate has been constructed and is being tested within the WRF coding framework.  Prognostic variables for this solver are the mass in the column (hydrostatic surface pressure),  and coupled with the column mass - potential temperature, horizontal velocities u and v the vertical velocity w, and the geopotential.  This core is available in the current WRF release, and will be the basis for the research release due out in early 2004.

For further information:
Contact: Bill Skamarock (skamaroc@ucar.edu, 303-497-8161)

Available papers and presentations:
Equations and Discretization for the Eulerian mass-coordinate model
Time-splitting Integration Techniques: 2nd-order Runge-Kutta and  3rd Order Runge-Kutta
Presentations showing some of the results from this model:  2D prototype, 3D prototype

(2)   A semi-Lagrangian solver based on the fully compressible nonhydrostatic equations with a generalized vertical coordinate is being constructed.

The  motivating philosophy behind the development of the semi-Lagrangian WRF dynamical core is that, for some atmospheric  simulations, especially real-time forecasting (where faithful treatment  of phase information becomes vital), it is advantageous to use numerical operators possessing a high degree of formal accuracy. In the spatial  dimensions, the best differencing schemes for a given order are the so-called "compact" (spatially implicit) methods. Experience, and the corroborating evidence from the formal analysis of truncation errors, suggests that any advantages of staggering the variables of a model (such as the wind components relative to mass and tracers) is greatly  diminished once high-order spatial operators are adopted. Having all variables collocated obviates the need to interpolate among grids when access to one variable is required at the point where another is stored. But it also simplifies the implementation of semi-Lagrangian schemes, since only one family of grid-associated trajectories is ever required.

The Lagrangian framework is very often the one in which the  state evolves in the slowest and smoothest way and, in combination with a semi-implicit treatment of acoustic and deep gravity modes,  it provides the most practical escape from the costly Courant-Friedrichs- Lewy time step restrictions that Eulerian models must obey. By dimensionally splitting the grid-to-grid interpolations in a "cascade" of one-dimensional sweeps, the semi-Lagrangian algorithm is able to accommodate both high-order interpolation operators and first-order conservation at a reasonable cost. Moreover, by facilitating the use of forward trajectories, the method does not hinder the adoption of various high-order temporal discretizations for  the explicitly treated modes. Then, consistent with the philosophy expressed above, the truncation errors can be kept low even in the time dimension,  despite the relatively long time steps which the semi-Lagrangian approach encourages.

For further information:
Contact: Jim Purser (wd23jp@sun1.wwb.noaa.gov, 301-763-8000, ext 7267)

Available papers:
Radiative Upper Boundary Conditions for a Nonhydrostatic Atmosphere. R. J. Purser and S. K. Kar, May 2001
Parallel Implementation of Compact Numerical Schemes.  T. Fujita and R. J. Purser, July 2001.
Proposed Semi-Implicit Adaptations of Two Low-Storage Runge-Kutta Schemes.  Part I: Theoretical Formulation and Stability Analysis.  R. J. Purser, August 2001
 

 Information:

2001 AMS Mesoscale Processes Conference extended abstracts:

A semi-Lagrangian dynamical core for the non-hydrostatic WRF model; Purser, Fujita, Kar , and Michalakes

Prototypes for the WRF (Weather and Research Forecasting) model; Skamarock, Klemp, and Dudhia

A manuscript examining the choice of parameters for use in the turbulence parameterizations found in the mass and height coordinate cores: The effects of subgrid model mixing and numerical filtering in simulations of mesoscale cloud systems; Takemi and Rotunno

Plans:

The Eulerian mass coordinate core (WRF-mass) has been released for testing to the community.  The mass core supercedes the height  core as the community model.  The next release (in early 2004) of the mass core should constitute a research-ready model, and will include grid nesting.

The semi-Lagrangian prototype will be evaluated in the near future as the prototype becomes available.

Statement and plans for initial implementation of nesting in WRF

Interaction with other WRF Groups:

Test are also ongoing involving the software architecture working group regarding the efficiency of these codes on various parallel computers.
Numerous other working groups (physics, standard initialization, 3Dvar and 4Dvar, etc. are also involved in various aspects of the dynamical solver development.