WORKING GROUP 10: 4D-VAR DATA ASSIMILATION SYSTEM
Representatives:
- Dale Barker (lead), NCAR
- Yuanfu Xie, FSL
- Milija Zupanski, NCEP
- Ming Xue, CAPS
- Xiaolei Zou, FSU
Status:
WRF data assimilation resources are currently being devoted to the
development of a 3DVAR system (see WG4 -
3DVAR). Development of the 4DVAR components of the variational
data assimilation system will commence when resources are available.
Documentation/links - to come.
Plans:
Initial WRF 4DVAR meeting/conference call to decide issues including.
- Do we agree that 4DVAR is the way to go? Do the
limitations of the variational approach at high resolution (e.g.
highly nonlinear processes and observation operators) justify a
concerted 4DVAR effort for the 1-10km scale? What are the alternatives?
- Starting point for 4DVAR - use legacy code (as WRF physics is doing)
or re-write from scratch (as WRF dynamical cores are doing)?
- General approach - Incremental? Adjoints for multiple dynamical cores?
Status of automatic differentiation tools (e.g. TAMC/TAF), etc.
- Reviewing procedure and code maintenance.
- Adoption of mission (draft below) and general plan including particular
tasks, effort required, estimated timescales.
- Initial allocation of tasks to resources available.
- Updating of WRF 4DVAR representatives.
- Listing of 4DVAR issues that need to be publicized to other working
groups.
Draft Mission:
The mission of this working group is to develop a robust, efficient and
flexible WRF 4-D variational assimilation system for use by both operational
and research communities. Specific 4DVAR tasks supplement those of the WRF
3DVAR system and include:
- Perform an early assessment of the possible use of automatic
differentiation tools.
- Develop a forecast model for the evolution of small yet
finite perturbations (tangent-linear/perturbation forecast model) -
including the representation of significant physics (components TBD) as
well as dynamics.
- Develop the adjoint code for the perturbation forecast model.
- Include representation of model error evolution.
- Merge the above with the WRF 3DVAR system.
- Inclusion of high-frequency/resolution observations e.g. radar,
automated surface stations, satellite.
Interaction with other WRF Groups:
- WG1 - Numerics/Dynamics:
Development of tangent-linear (TL) and adjoint
code for WRF dynamical core(s), consideration of 4DVAR needs in design of
forecast model e.g. vertical co-ordinate.
- WG2 - Software Architecture:
Parallelization and vectorization strategy,
WRF code library, WRF coding conventions.
- WG3 - Standard Initialization:
Interface between 3DVAR and tangent-linear/adjoint code.
- WG4 - 3DVAR:
Ensure common framework for 3/4DVAR e.g. grid, minimization,
observation/grid data formats, quality control, observation operators,
observation/background/model errors etc.
- WG5 - Model Physics:
Development of TL/adjoint code for physics routines.
- WG7 - Verification:
Exchange of background/forecast errors (case-study
and statistical), forecast fit to observations, estimates of analysis
error.
- WG8 - Support:
Documentation, presentations, web-pages.
- WG13 - Ensemble Prediction:
Comparison of techniques, evolution of
model error, exchange of code (e.g. observation operators).
