RegionalScale En KF in Comparison with WRF3 Dvar
- Slides: 10
Regional-Scale En. KF in Comparison with WRF/3 Dvar Fuqing Zhang and Ellie Meng Texas A&M University
WRF En. KF vs. 3 DVar over the Month of June 2003 (90 km) (30 km) Verification area Observations: standard soundings every 12 h QC’d by WRF/3 Dvar in D 2 Verifications: against soundings both before and after assimilation Boundary conditions: D 1 updated by 12 hourly FNL analysis Reference forecast: a single forecast for the period w/o assimilation
Month-long Performance of En. KF vs. 3 Dvar with WRF Reference forecast En. KF (day of the month) 3 DVar (prior, solid; posterior, dotted) (day of the month) • All in terms of domain-averaged root-mean square error verifying against soundings • Both the En. KF and 3 DVar have smaller error than reference deterministic forecast • En. KF is consistently better than 3 Dvar in terms of 12 -h forecast and posterior analysis
Month-long Performance of En. KF vs. 3 Dvar with WRF ---- Reference forecast En. KF 3 DVar (prior, solid; posterior, dotted) posterior Prior Better performance of En. KF than 3 DVar also seen in both 12 -h forecast and posterior analysis in terms of root-mean square difference averaged over the entire month
12 h Fcst from En. KF vs. WRF/3 Dvar vs. NCEP/FNL ICs DF_En. KF : 12 h deterministic forecast initiated from En. KF analysis DF_NCEP: 12 h deterministic forecast initiated from FNL analysis • DF_En. KF has smaller (larger) forecast error than 3 DVar (En. KF), suggesting En. KF does benefit from both using the ensemble mean for state estimation and using flow-dependent background error covariance. • The forecast started from the NCEP FNL analysis has error larger than that of En. KF but smaller than that of 3 DVar
En. KF 4 DVar: Coupling En. KF with 4 DVar in Lorenz’ 95 Fuqing Zhang, Meng Zhang and Jim Hansen
Comparison in Perfect model Scenario (F=8. 0) 80 -variable Lorenz 95 model with 20, 40 (above) vs. 10 (below) members
Comparison in Imperfect model Scenario (F=8. 5 but truth is 8. 0) 80 -variable Lorenz 95 model with 20, 40 (above) vs. 10 (below) members
Imperfect model with moderate model error: F=8. 5 but truth is 8. 0 80 -variable Lorenz 95 model with 20, 40 (above) vs. 10 (below) members
Imperfect model with strong model error: F=9. 0 but truth is 8. 0 80 -variable Lorenz 95 model with 20, 40 (above) vs. 10 (below) members
