Participants
Following members attended the meeting.
Tom Auligne
Jordan Powers
Kevin Manning
Monika Krysta
Michel Yann
Zhiquan Liu
Syed Rizvi
Agenda
10:00: Introduction
- 5-minute introduction about the B matrix, how it is calculated in GEN_BE and used in WRF-Var (Tom).
10:05: AMPS developments
- Short informal presentation on GEN_BE developments in AMPS domain (Yann).
- Discussion on the strategy to test new Background Errors (BE) in AMPS.
Jordan and Kevin are free to leave (or stay...) at this stage.
10:50: Humidity developments
- Short informal presentation on the coupling of humidity in GEN_BE (Monika).
- Discussion on the work calendar and how to coordinate efforts on GEN_BE.
11:30: Adjourn
Expected Outcome
- Jordan and Kevin being aware of developments to the B matrix (and potential benefits)
- A strategy to test current and future developments on AMPS
- Collective solutions to the current issues
- A coordinated working plan on BE.
Presentations
Variations on BE according to Berre (Monika Krysta and Syed Rizvi)
Background Error Modeling: Climatological Flow-Dependence (Yann Michel)
Minutes (thanks to Rizvi)
Tom talked briefly about the role of background error (BE) in WRF-Var and how it gets computed using "gen_be". Jordan wanted to know exactly what are the contents of "be.dat" file, which goes as input to WRF-Var.
Yann gave presentation highlighting some week points of WRF-Var in handing BE like, the use of homogenous horizontal lengthscale, variance etc,.Following NCEP's logic, for AMPS domain Yann has successfully computed lenthscale for different WRF_Var control variables. His results indicated that the lengthscales for stream function and unbalanced velocity potential fields were smaller whereas, for unbalanced temperature and moisture they are larger compared to the existing BE which are computed using WRF-Var "gen_be" code. Yann plans to work on implementing latitudinal variation of lengthscales in WRF-Var. However, he did notice that these lengthscales are highly inhomogeneous and thus needs proper smoothing before using it in WRF-Var. Rizvi pointed out that before doing so, it needs to be checked that whether the horizontal lengthscales computed using NCEP method is consistent with WRF-Var horizontal transforms.
Currently WRF-Var do have a provision for utilizing latitudinally dependent regression coefficients. However, past experience with T8 domain indicated that latitudinal regression coefficients lead to vary noisy analysis increments. Yann also wants to try latitudinal variation of BE variance in WRF-Var with homogeneous EOF.
Monika discussed single observation results with new approach (Berre, 2000) of generating BE. This method introduces additional regression coefficients between different control variables. The results indicated that Temperature and Moisture had higher lengthscales compared to the existing BE.