Initial Perturbations for NCEP EPS

1
Initial Perturbations for NCEP EPS

• Breeding Analysis error variances at 500mb are
  specified in breeding mask,
• not flow dependent.
• ETKF Analysis perts are based on ensemble
  Kalman filter real-time
• observations can be used. But for real obs,
  large variation of obs number
• is hard to deal with.
• ET Operational mask in breeding used to
  specify the initial analysis error variances
• computationally cheaper than ETKF less
  influence from the variation of obs number
• high dimension of ensemble subspace testing
  for possible implementation.
• Future Development the analysis covariance
  matrix will be extracted from
• operational data assimilation system GSI.
  The observational data impact will be
• automatically reflected in the analysis
  covariance matrix.

2
Data Exchange with NRL

• NCEP
• 80 ensembles, 6-hour forecasts
• 
  
• ET on 160 ensemble members
• 80 (NCEP) 80 (NRL)
• Produce 16-day forecasts for 20 members, short
  forecasts for 60 members. Get 80 members from NRL
• ET on 160 ensemble members
• 80 (NCEP) 80 (NRL)
• 
  
  
  

• NRL
• 80 ensembles, 6-hour forecasts
• 
  
  
• ET on 160 ensemble members
• 80 (NRL) 80 (NCEP)
• Produce 16-day forecasts for 20
• members, short forecasts for 60
• members Get 80 members from
• NCEP
• ET on 160 ensemble members
• 80 (NRL) 80 (NCEP)
• 
  
  
  

3
Anormaly Correlation and RMS Error
4
Amplification factor and Dimension
5
PECA and Variances