Erroneous results using ETKF - issue with hybrid_opt=2?

[thomasuh]

Hello,

I am running WRFDA in cycling hybrid mode with a 10 member ensemble using ETKF. After running the ETKF, I get wind increments of more than 100 m/s and QVAPOR increments of > 20 g/kg which then leads to negative QVAPOR values and extreme wind velocities in the updated ensemble and a crash of the ensemble members.
The gen_be_etkf_nl is the following:

&gen_be_etkf_nl
num_members =10 ,
nv = 7 ,
cv = 'U' , 'V' , 'W' , 'PH' , 'T' , 'QVAPOR', 'MU' ,
naccumt1 = 1 ,
naccumt2 = 1 ,
nstartaccum1 = 0 ,
nstartaccum2 = 0 ,
nout =4 , ! e.g. 4th cycle
tainflatinput = 1 ,
rhoinput = 1 ,
infl_fac_file = 'ETKFDIR/param/inflation_factor.dat' ,
infl_let_file = 'ETKFDIR/param/inflation_letkf.dat' ,
eigen_val_file = 'ETKFDIR/param/eigen_value.dat' ,
inno2_val_file = 'ETKFDIR/param/innovation_value.dat' ,
proj2_val_file = 'ETKFDIR/param/projection_value.dat' ,
infl_fac_TRNK = .false. ,
infl_fac_WG03 = .false. ,
infl_fac_WG07 = .true.,
infl_fac_BOWL = .false. ,
letkf_flg =.false. ,
rand_filt = .false. ,
rnd_seed =2022111500 ,
rnd_nobs = 5000
etkf_erro_max = 20.
etkf_erro_min = 0.00001
etkf_inno_max = 20.
etkf_inno_min = 0.00001
etkf_erro_flg = .true.
etkf_inno_flg = .true.
etkf_wrfda =.true.
/

The obtained Eigenvalues are very small and the posterior standard deviation strongly increases.

The wrfinput files for the ETKF are completely reasonable and there are no erroneous values in the ob.etkf.* files.
Is there an issue with the ETKF when using hybrid_opt=2 and CV7 option?

The WRFDA version I am running is 4.4.2 with Ifort and the filtered observations are generated using the "QC-OBS" and "VERIFY" option of the 3DVAR as described in the WRFDA user guide section

Any help is highly appreciated.

Thomas

 

[thomasuh]

I made further tests. It seems that the inflation factor of the WG07 method become quite large as during the RUC nout, naccumt1, and naccumt2 are increasing. Thus, the inflation factor is further increasing and reaches values > 250 after 2 cycles. This leads to large U,V, etc. values in the new ensemble members.
I am not sure how to solve this. Does it make sense, to keep NOUT=1 throughout the cycle?

Thomas

 

[thomasuh]

Apparently there is a severe issue with the ETKF when the number of observations is small and/or the observations are unevenly distributed.
To me it seems that in this case the ETKF cannot be used.