An Efficient Ensemble Data Assimilation Approach and Tests
with Doppler Radar Data
Jidong Gao Ming Xue
Center for Analysis and Prediction of Storms,
University of Oklahoma, Norman
Research Goals
• To develop an efficient ensemble Kalman filter (EnKF) method for high-resolution NWP, by using a dual resolution approach.
• To evaluate the efficiency and accuracy of the method through OSSEs, with simulated radar radial velocity data for a supercell storm.
Introduction• EnKF was first introduced by Evensen (1994)
and has become very popular in recent years
• Recently, the EnKF method has been successfully applied to the radar data assimilation problem (e.g., Snyder and Zhang 2003; Zhang et al. 2004; Dowell et al. 2004; Tong and Xue 2005).
• Effective assimilation of radar data is essential for initializing convective-scale NWP models
Radar Data Assimilation• The EnKF data assimilation method is especially
suitable for radar data assimilation because
– Radar only observes Vr and Z, and data coverage is usually incomplete
– All other variables have to be ‘retrieved’– EnKF ‘retrieves’ the unobserved variables via background
error covariance obtained through a forecast ensemble
• But, EnKF is expensive, because of the need for running a usually rather large ensemble of forecasts and analyses
• In this work, we propose a dual-resolution (DR) hybrid ensemble DA strategy, with the goal of improving the EnKF efficiency
• With the method, an ensemble of forecasts and analyses is run at a lower resolution (LR), while a single system of analysis and forecast is performed at a higher resolution (HR)
• The LR forecast ensemble provides estimated background error covariance for the HR analysis
• The HR forecast is used to replace or partially adjust the mean of the LR analysis ensemble
The Methodology
LR
EnK
F A
nalysis
LR E
nKF
Analysis
LR E
nKF
Analysis
HR EnKF
Single higher-resolution analysis and forecast
Lower-resolution analysis and forecast ensemble
covarian
ce rep
lace m
ean
covarian
ce co
varianc
e
replace
mean
replace
mean
HR EnKFHR EnKF
OSSEs with a Simulated Supercell Storm
• A truth simulation is created using ARPS with the Del City supercell sounding, at x = 2 km
• The model domain: 92 x 92 x 16 km3.
• LR has x=4 km, HR has x=2 km
• z = 500 m.
•Vr data collected at grid point locations are assimilated, at 5 min intervals
•20 ensemble members are used
List of EnKF OSSEs
Experiment Description
EXP1 Single-reslution EnKF at HR (2 km)
EXP2 Single-resolution EnKF at LR (4 km)
EXP3 Dual-resolution hybrid EnKF (2 & 4 km)
RMS Errors of the Analyses for the Three Experiments
HR EnKF (EXP1)
LR EnKF (EXP2)
DR EnKF (EXP3)
’(contours), Z(color shades) and Vh (vectors) at Surface
Truth
EXP2
LR-EnKF
EXP1
HR-EnKF
EXP3
DR-EnKF
’, Z and Vh at Surface after 80 min assimilation
Truth EXP1
HR-EnKF
EXP2
LR-EnKF
EXP3
DR-EnKF
W at 6 km AGL after 80 min assimilation
Truth
EXP2
LR-EnKF
EXP1
HR-EnKF
EXP3
DR-EnKF
2-h Forecasts of ’, Z and Vh at surface
Truth
EXP2
LR
EXP1
HR
EXP3
DR
2-h Forecasts of w at 6 km AGL
Truth
EXP2
LR
EXP1
HR
EXP3
DR
Summary and Discussion• A new efficient dual-resolution (DR) approach for
EnKF is proposed and tested with simulated radar data for a supercell storm.
• It is shown that the EnKF analysis using DR is almost as good as the HR analysis, but is much better than the LR analysis.
• For this case, we save CPU 3-4 times. However, depending on the resolution one choose, the method have the potential to save CPU 10-50 times more than Original EnKF methods.
Summary and Discussion
• My new experiments: using Dx =Dy= 4km with model EnKF run, to provide error structure for Dx =Dy= 1km, single model run. The result is also very positive.