A. General Conclusions:

WRF represents an implementation of a conservative split-explicit numerical scheme that approximates the compressible nonhydrostatic equations governing the Earth's atmosphere in a limited domain. As such, it satisfies quite a few of the requirements we had for a new nonhydrostatic core for use at GFDL, and should be evaluated in depth for inclusion in the FMS. The principal modification that would be required at the outset is an expression of the equations in a form that fully accounts for sphericity.

B. Specific Comments:

- My guess is that, at least initially, WRF could be included into FMS in much the same way as the nonhydrostatic ZETA model, namely as a self-contained atmospheric model that can be made to provide the required fluxes to the other component models. Variable definitions, I/O, etc. would all be done within WRF. WRF would be modified on an as-needed basis to correspond to some of the standard FMS conventions, such as the selected use of the mpp modules.

- WRF's focus is on limited-area simulations at resolutions from 1-10 km, with an emphasis on the higher-resolution end of that range. A WRF prototype is scheduled to be released in 10/00.

- WRF is intended to be open source. It may be beneficial to licence it under the GNU Public License (GPL).

- There are 3 dynamical cores under consideration:

        1. Spit-explicit C-grid in height-based coordinates
        2. Spit-explicit C-grid in mass-based coordinates (i.e., a sigma coordinate based on hydrostatic pressure)
        3. Semi-implicit semi-Lagrangian on an A-grid

  The WRF prototype will almost certainly use either options 1 or 2, but only one of these will be carried forward. That decision has yet to be made.  Option 1 is simpler to work with but only when a rigid lid upper boudnary condition is employed.The dynamical cores are on a limited-area domain that employ map factors in the horizontal, although a global version of Option 3 may be developed. Full sphericity and global domains will not be included. The applicability of option 3 to cloud-resolving models has yet to be determined.

- There is no obvious reason why deep domains (i.e., stratosphere/mesosphere) can't be used. Simulations with domains of this type have not been done.

- Third-order Runga-Kutta is employed as a time-stepping scheme for its balance of efficiency and stability with higher-order spatial differencing.

- First-order quantities are conserved, but no estimates of the conservation of second-order quantities have been made. It is unknown a priori what  level of effort is required to maintain first-order conservation in WRF expressed in spherical coordinates, although we're pretty familiar with the necessary numerical techniques in other GFDL models.

- Different advection schemes can be employed for the momentum vector and scalar/tracer advection (e.g., high-order finite differences for the momentum, positive-definite for tracers). One of the goals for scalar transport is consistency, defined here as the reduction of a numerical scalar conservation law to the continuity equation for a constant scalar field.

- Specification of the WRF physics module interface should be available in 06/00. I provided them with a copy of the 1D Physics Module Specification found on the FMS home page, but warned them that it was a 2-year-old draft.

- Initialization/Data Assimilation Timeline:
 

        10/00 - A "Standard Initilization" will be available that allows input  from a variety of datasets. Idealized initialization (e.g.,   user-specified) will be supported.
        10/01 - 3D variational assimilation should be available
    > 10/01 -  4DDA, perhaps 4D variational assimilation, will eventually be available
The co-development of model adjoints is encouraged, but the number of cores and their associated adjoints will need to be limited just from the standpoint of maintanance.

C. Answers to our specific questions:

1. General

   a) Call sequence in WRF

   This is given pretty much by the routines in the WRF Model Layer depicted in the "WRF Hierarchical Software Archtecture" slide from http://wrf-model.org/PRESENTATIONS/2000_04_18_Klemp/index.htm. The goal is to have as much of WRF be run-time configurable, rather than configured at compile-time. A lot of the development headaches have been minimized by using a registry (ACSI text) of model variables that is parsed to produce the appropriate declarations within the code.

   b) Support for external users

   They're struggling with this resource issue as much as we are. Clearly some of the external users (GFDL included) may become collaborators on the WRF project. The intent is for MM5 users to develop for WRF as well.

   c) Code availability/customization

   WRF will soon be under  CVS. GFDL may be able to get a CVS module in July, but certainly we'll get the prototype to be released by October. Since it's open source, modfication shouldn't be an issue.

   d) Can WRF be internally initialized?

   Internal (idealized) initialization should be supported in the "Standard Initialization" package scheduled for release with the protype.

2. Dynamical Core

   a) Numerical methods

      i. Are tendencies explicitly calculated?

      Yes.

      ii. Is Runga-Kutta applied to the Eulerian or Lagrangian time derivative?

      Eulerian.

      iii. Why 3rd order RK?

      Balance between efficiency and stability with higher-order spatial differencing.

      iv. Is mass-based or height-based model results better?

      To the extent that there are no lenth-scale dependencies in the test cases, these systems converge to the identical result.

      v. How different is the core from ARPS?

      vi. Advection schemes
 

   A) Modularity

   B) Conservation of first and higher-order quantities

          Conservation of first-order quantities is implemented. No estimates have been made of the conservation of higher-order quantities.

   C) Treatment of positive-definite quantities vs. other model variables

Different advection schemes can be applies to different variables.
   b) Boundary conditions (Lateral/Vertical)

   Right now, radiative and probably periodic lateral, outward gravity-wave energy propogation at the top.

   c) Anticipated domains (Horizontal/Vertical)

   Limited-area 1-10 km resolution in the horizontal. No obvious limitations to the depth of the domain.

   d) Nesting capabilities (moving meshes?)

   Nesting is explicitly implemented via recursive calls, but has not been tested. If GFDL makes the conversion to spherical coordinates and a global domain, nesting should be taken care of.

3. Physics

   a) Software interface

   To be determined by 06/00.

   b) Moisture schemes

      i. When and how is saturation evaluated?

      Saturation is evaluated last.

      ii. How is feedback to potential temperature achieved?

      iii. Subgrid (fractional) condensation?

      Being considered.

      iv. Conservation properties of q_r flux?

      In flux form, but under CFL constraints.

      v. Ice Schemes

      Effectively the Goddard ice microphysics

   c) Boundary layer schemes

   MRF PBL as a place-holder

   d) Gravity wave drag schemes

   Probably unecessary at the targeted resolutions.

4. Tests

   a) What's WRF been tested on so far?

   Mountain waves, moist convection, standard gravity current tests. Baroclinic wave simulations will be evaluated soon. One of the guiding principals is for WRF to acheive the hydrostatic limit at the appropriate scales.

   b) Properties of numerics in long integrations (e.g., channel Rossby wave)

   Nondissipative test cases so far have been limited by the development of singularities (e.g., fronts) in the flow.