Assimilating Sounding, Surface and Profiler Observations with a WRFbased EnKF for An MCV Case during

1
Assimilating Sounding, Surface and Profiler
Observations with a WRF-based EnKF for An MCV
Case during BAMEX

• Zhiyong Meng Fuqing Zhang Texas AM
  University

2
The MCV event of 10-12 June 2003 (IOP 8 of BAMEX)
a)
c)
b)

f)
e)
d)
3
Forecast Model WRF2.1
D1
D2

• Two domains with grid sizes of 90 30 km
  one-way nesting
• Physical parameterizations The Grell cumulus
  scheme, the WSM 6-class microphysics scheme with
  graupel, and YSU PBL scheme
• Data assimilation is only performed in D2

4
Ensemble forecast in D2Simulated reflectivity
(colored) and MSLP (blue lines, every 2 hPa)
12h
24h
11/00
11/12
36h
30h
12/00
11/18
X
X
L
L
L Observed MCV position at surface
X simulated MCV position at surface
5
Data to be assimilated
Profiler (27) (thinned in vertical) 3-h interval
Sounding (31) 12-h interval
Half Surface (458) 6-h interval
X
6
A WRF-based EnKF

• A sequential filter Whitaker and Hamill
  (2002),
  Snyder and Zhang (2003)
• Covariance localization Gaspari and Cohn (1999),
  
• ROIs Vertical - 15 levels
  for all data.
• Horizontal - 300 km
  for surface data,
• 
  900 km for radiosonde profiler
• Assimilated variables u, v and T (same obs
  errors as NCEP)
• Ensemble generation perturbations sampled from
  WRF/3Dvar background error covariance (Barker et
  al. 2003)
• Ensemble size 30

7
WRF-3DVAR
Objectives in here to generate the initial
ensemble be a benchmark for the EnKF. Control
variables stream function, pseudo relative
humidity, unbalanced part of velocity potential,
temperature, and surface pressure. Background
error covariance NMC method. Minimization
Conjugate gradient method.
8
Experiment design

• Sounding assimilation
• Profiler assimilation
• Surface assimilation
• Sounding Profiler Surface assimilation
• Model error treatments
• (with Sounding Profiler Surface assimilation)
• - Covariance relaxation
• - Multi-scheme ensemble

All results are verified with sounding except for
otherwise specified
9
Sounding assimilation - cycling at 12h interval
(h)
(h)
(h)
10
Profiler EnKF assimilation - cycling at different
intervals
(h)
(h)
(h)

• The final forecast error decreases from 12-h
  interval to 3-h interval
• Further increase of obs frequency worsens the
  result in general.

11
Profiler assimilation - cycling at 3h interval
(h)
(h)
(h)
12
Surface observation assimilation - Cycling at
6-h interval - Verified with the other half
surface (upper) and soundings (lower)
(h)
(h)
(h)
(h)
(h)
(h)
13
Assimilation of SoundingSurfaceProfiler obs -
Cycling at 3-h interval
(h)
(h)
(h)
14
MCV positions at 36h (00UTC Jun.12)Observed
radar echo, simulated reflectivity (colored) and
MSLP (blue lines, every 2 hPa)
No EnKF
OBS
SND
X
X
SFC
SNDSFCProfiler
Profiler
X
X
L
L
L
X
X Simulated MCV position at surface
L Observed MCV position at surface
15
Model error treatment

• Covariance inflation through relaxation

(Zhang et al. 2004)
(h)
(h)
(h)
16
Model error treatment

• Multi-cumulus-scheme-ensemble
• (Fujita et al. 2006 Meng Zhang 2006)
• The schemes used in ensemble KF, Grell, and
  BM

(h)
(h)
(h)
17
Model error treatment

• Multi-cumulus-scheme-ensemble
• ( Fujita et al. 2006 Meng Zhang 2006)

(h)
(h)
(h)
18
Summary

• The WRF-based EnKF behaves well when assimilating
  real observations. It performs better than 3DVAR
  for this MCV event.
• Sounding and profiler assimilation can improve
  the analysis significantly. Impact of surface
  data is rather weak and short-term. The best
  performance is obtained by assimilating all three
  data sources.
• Higher temporal frequency of profiler may give
  better performance until down to 3-h intervals.
• Covariance relaxation and multi-scheme-ensemble
  can apparently improve the performance of the
  EnKF in this MCV event, consistent with Fujita et
  al. (2006) Meng and Zhang (2006).

19
What to do next?

• Assimilate surface pressure and humidity in
  addition to u, v and T.
• Dropsonde data assimilation
• Higher resolution model
• Radar data assimilation