Minutes from the 14 February 2000 WRF meeting

Participants

NCAR: Joe Klemp, Bill Skamarock, Jim Dudhia, Chris Davis, Dale Barker, John Michalakes, Dave Gill

OU: Kelvin Droegemeier, Ming Xue, Fred Carr, Keith Brewster, Jidong Gao

FSL: Tom Schlatter, John Brown, Stan Benjamin, Georg Grell, Paul Schultz, Yuanfu Xie, Chungu Lu, Leslie Hart

NCEP: Geoff DiMego, Jim Purser

AFWA: Shu-Hua Chen

NSSL: Lou Wicker
 

Amended list of WRF developers, with team and group associations. This list has the email and phone numbers of each of the members.
 
 

Handout/email #2 from Joe Klemp
 

                                       Tentative Milestones for WRF Project

FY00 Milestones

   Finalize numerics for model solver based on evaluation of prototypes.

   Design "plug compatible" interfaces for physics modules and begin
   porting essential physics.

   Implement standard model initialization procedures.

   Design and Implement basic 3D-Var analysis scheme for standard
   meteorological data.

   Release of a basic version of WRF with (limited physics, standard
   initialization) and associated documentation

FY01 Milestones

   Adapt ``best" physics packages from existing models to the WRF Model,
   and  begin extensive case-study testing.

   Construct tangent linear and adjoint models of the WRF forecast model.

   Work with broader research community in developing advanced model physics.

FY02 Milestones

   Test forecast verification strategies and collect a broad set of
   verification statistics.

   Implement 4D-Var assimilation system for research applications.

   Release a research quality NWP version of WRF.

FY03 Milestones

   Evaluate all aspects of the forecast system through comprehensive
   real-time testing.

   Establish a dedicated database for the archive and validation
   of data and forecasts from the WRF system.

   Adapt WRF Model to the NCEP and AFWA computing environments.

FY04 Milestones

   Implement an operational data-assimilation system for WRF.

   Release of enhanced versions of the WRF system including 4D-Var
   data-assimilation and advanced model physics.

   Begin operational use of WRF as a nested regional model.
 

Handout/email #3 from Joe Klemp

                                                                  Agenda

The main purpose of this meeting is to get our act together in moving forward
on the various aspects of the WRF development. There are a number of things
we should try to accomplish in these discussions:

  o Clarify the organizational structure for WRF development work (development
    team and working groups and how they should function)

  o Update leaders and membership of the existing working groups (I'll send out
    the most recent list I have that is admittedly out of date)

  o Refine the timeline for significant milestones (I'll send out the timeline
    I put together for the USWRP SSC Meeting last October)

We need to go over the progress and plans in each of the working group areas.
I would like each group leader to:

  o Briefly summarize the current status of development work in their area

  o Briefly describe the strategy for development needed for their part of
    milestones that we seek to achieve in the coming year (what needs to be
    done, who will do it, when it will be done)

  o Seek agreement on any issues that have reached a decision point for our
    broader planning team

  o Identify action items for specific planning/development work that should
    be accomplished and presented at our meeting at the end of March
 
 

                                                       Discussion

Working Group #1: Dynamics and Model Numerics, Bill Skamarock

There are three current development streams:

NCAR

• 3d Eulerian geometric height, with split explicit, moist prototype being tested with leap-frog and Runge-Kutta
• 2d mass-based coordinate  prototype being tested (squalll lines, etc.)

• start building the 3d mass-based coordinate to run along side the geometric height version by April
• testing by June for evaluation of coordinates
• 3rd order Runge-Kutta by April for both geometric and mass coordinate models
• come up with tests that discriminate between two coordinate systems (mountain waves are not satisfactory, perhaps diabatic heating, surface heating, latent heating to see the effects of heating with respect to constant pressure vs constant volume
• encourage the WRF members to get involved with testing to gain experience with the prototypes, though no broader community involvement is appropriate at this time

 

• 2d semi-Lagrangian, forward, conservative

• not expected that a 3d version will be ready for comparative testing by June

• 3d Eulerian, hybrid coordinate, in testing

• FSL version still in testing by June?

Working Group #2: Software Architecture, Standards & Implementation, John Michalakes

There are three basic areas of development that are ongoing: code development, coding standards, and
I/O & data format.  WG3 (Standard Initialization Procedures) will take charge of defining the metadata convention, but with input from other concerned groups.

Code Development

• current area of interest is the index ordering: i,j,k vs k,i,j.   Vector vs cache  performance
• tentatively and in a nutshell: cache based machines tend to be less sensitive to the array index ordering, while vector machines tend to be quite sensitive
• a decision was made to continue to consider the ordering as an open issue

• a presentation will be made with more data points to propose a final decision on the index ordering issue
• a plan will be presented for version control
• it was agreed that the DA system performance (parallelization, FFTs, inhomogeneous obs network, etc.) is within the scope of the WRF architecture working group

• no definitive effort exists for the WRF coding standard
• development on the code has proceeded through the implicit use of certain ideas
• concerns were voiced about the perceived restrictiveness of the (not yet even written) proposed coding standards
• the primary intent should be to specify interfaces, as most developers will want to insert physics routines; the desire should not be to subject developers to mandatory stylistic guidelines

• prior to the meeting, forward a draft of the coding standards within the working group and to the development team managers for comments and review
• present a proposal for the WRF coding standards

• need an architecture document, possibly in the form of a white paper
• define the metadata convention

At this point, a significant discussion with the Working Group #3 (Standard Initialization Procedures) began.  The definition of the metadata convention will be handled by WG3.
• once metadata convention is defined, then the implementation can begin: GRIB, netCDF, HDF, etc

• architecture document (outline already exists) for I/O, to include issues such as:
• types of data to consider: gridded, observation, Fourier transforms
• required format for each data type
• how to insure parallel and scalable I/O
• WRF I/O plans:
• comprehensive design for WRF I/O and data that encompasses program modules, interfaces, transport and storage layers
• hierarchical architecture (similar to the software architecture that exists for the model itself) that encapsulates details at each layer
• package independence

Working Group #3: Standard Initialization Procedures, Paul Schultz (for John McGinley)

Handout/email from John McGinley
 

Our task is one of the first items in the WRF critical path.  I would
like to start an e-mail dialog of how to develop a consensus on
defining, locating, and acquiring the necessary background fields to
best run the WRF. I'm sure each of our institutions has such
procedures. In our FSL/LAPS project we have a flexible system that has
had to adapt to grids and background data sets around the globe.  Below
is a list of fields we might want to consider as a first cut in
defining what may be needed for WRF model setup.

Land Characteristics
 1. Terrain height
 2. Soil type
 3. Surface temperature
 4. Soil Moisture
 5. Snow Cover
 6. Land vegetation, natural
 7. Land/water ratio
 8. Surface Albedo
 9. Surface roughness
 10. Surface emissivity(IR and MW)

Ocean Characteristics
 1. Sea Surface Temperature
 2. Below Surface Temperatures
 3. Ice Coverage
 4. Ocean currents

Human Impact
 1. Smoke/pollutant sources
 2. Land use (urban/agricultural areas, crop distribution)

Atmospheric initialization/backgrounds
 1. Coarse Grid Model Backgrounds
 2. Atmospheric Climatological Fields
 3. Ozone Climatology

I believe we should conduct a survey to identify the latest
and best of these data sources, and look into the optimum methods
for acquisition for any selected grid location.

Plans for March:

• the first use of the utility must to to provide initial and boundary condition data to WRF, prior to the availability of the 3dvar code
• possibly namelist oriented, similar to LAPS
• could be used to drive any of the constituent models (eta, MM5, arps, etc.)
• will be able to handle real data cases as well as idealized studies
• survey to identify data sources and variables, presentation at March meeting of findings
• responsible for metadata convention to be used by WRF, proposal presented at March meeting; this group expects to have close ties with WG4 (Data Assimilation) and WG6 (Post Processing) in detailing and defining the metadata convention

The current status from the November meeting.

NCEP

• expect to have 3dvar anisotropic covariance in a few months
• after that, begin to verify the statistical parameters
• developing analysis scheme, isotropic but inhomogeneous, shoud be available soon

• working on obs operators for radar data

• working on obs operators for GPS
• several groups have obs operators for precipitable water, rate of precipitation
• a  web survey  was constructed to help define the difference between essential and desirable (essential should go into the first release of the WRD 3d-var system, desirable should go into a later release); this is presented in the form of a table detailing a comparison of each center's DA system

• should WRF 3dvar be used  as a tool for the broader community
• if so, is this just an interpolation issue solved by the reformatter from WG3
• designing for generality is problematic
• designing too specifically requires significant work when model is modified
• the 4dvar working group needs to be started, representatives are:
• NCAR: Dale Barker    303-497-8148       dmbarker@ucar.edu
• FSU:  Xiaolei Zou    850-644-6025       zou@met.fsu.edu
• NCEP: Yuanfu Xie     303-497-6846       xie@fsl.noaa.gov
• CAPS: Ming Xue       405-325-6037       mxue@ou.edu
• FSL:  Yuanfu Xie     303-497-6846       xie@fsl.noaa.gov

• finalize an agreement of what the components of the 3dvar system should be
• propose system with detail at a subroutine level, so that the functionality of each piece is known
• present options for 3dvar initial phase, list of which will be implemented and by whom

Working Group #5: Model Physics, John Brown

There are many areas into which the group is looking.

• implementing a single package for each of the following types of physics options (the big 5): cumulus, microphysics, PBL, surface and radiation; the specific choices are simply going in as place holders to make sure that the various coding design assumptions are viable and not with the assumption that these options define the first release of the physics packages
• the interface between the physics routines and the rest of the model (driver and mediation layers) needs to be well defined for the user community; this is what will permit a wider group of users to plug their own physics packages into WRF; a paper describing the design of the separation of these model layers is available
• an important charge of WG5 is the winnowing down the possible options to include and presenting the proposed list to the WRF development team; this implies some sort of organized evaluation and testing procedures exist to objectively permit the decision process
• a concern is the interaction between parameterization schemes, possibly a future session of a WRF workshop could address this issue by presenting results of various combinations of packages
• the subgrid moist processes at the 5-10 km scale were mentioned as a specific research topic
• a suggestion was made that contributors of physics packages be required to also supply adjoints of their packages
• a suggestion was made that the group investigate using 3d radiation schemes, not just 1d, for higher resolution applications
• a question of exactly what criteria will be used to decide inclusion was left unanswered and will be addressed in the proposed work

• present a framework for writing/including physics modules in WRF
• one example of each of the big 5 incorporated into WRF height-based, (K,I,J) prototype, how each interfaces with the height based coordinate
• a proposal for which packages should be included in the first version in FY01; based on the reception of the 3dvar survey, a similar web-based collaboration may be employed to determine essential and desired attributes

Working Group #6: Post Processing, Lou Wicker

• plotting capability is necessary ASAP to allow testing and development to continue
• overall plan:
• leverage existing technology
• be able to deal with an interrogation system (SQL database) for use with archives and ensembles
• conduct a web survey for a list of functional requests that can be prioritized as essential and desirable
• comments:
• keep it simple, standardized data formats will allow users to use their traditional post-processing utilities
• development of interfaces from WRF to IDL, GRADS, Vis5D
• want the ability to integrate model and observations
• vertical interpolation is a sore point

• specific plan for meeting short range plans
• proposal for those essential features to include for initial release

Working Group #7: Model Testing and Verification, Kelvin Droegemeier and Chris Davis

The groups seemed to be in general agreement that a suite of tests that were able to be automatically run on WRF would be desirable.  The tests would be run on new releases to provide a level of assurance of the new code validity.  The tests would cover a spectrum of situations:

• symmetry and property conservation
• linear solutions
• nonlinear with converged/known solutions
• idealized cases: neutral atmosphere with a cold bubble along a permeable boundary, 3d baroclinic channel simulation
• real data cases: raw data, analyzed grids

The philosophy of the tests:

• should be available to the entire user community
• test data sets will be archived at NCAR
• supply users with readers for all obs and gridded data sets
• select cases where particular observation type is of 1st order importance
• recognized that while objective measures are required, skill scores, RMS errors are not sufficient
• maintain backward compatibility for later releases of WRF
• concern for testing may not include direct evaluation of physics packages right away, but this is an area of interest

• identify tests and objective measures for evaluation
• design an overall framework of how testing and evaluation fits into WRF picture
• work with WG6 (post processing) to view model behavior
• strawman proposal for testing numerics, dynamics, physics
• identify testing interface and driver
• model only, not data assimilation yet

Working Group #8: Web Site, Workshops, Model Support
 

• a new web page is currently under development, contact either Leslie Hart or Dave Gill
• a 1 day WRF workshop is scheduled for 23 June 2000, in Boulder CO at the NCAR Mesa site
• inviting the science advisory board and the oversight board
• anticipate a number of presentations at workshop
• WRF presentation will be part of a COMET CD-ROM
• WRF management planning meeting scheduled for week of 19 June 2000