NCAR
Mesoscale and
Microscale Meteorology
Data Assimilation in AMPS
Dale Barker
S. Rizvi, and M. Duda
MMM Division, NCAR
Email: dmbarker@ucar.edu
http://www.mmm.ucar.edu/3dvar
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
Talk Overview
1. The MM5/WRF 3DVAR system.
2. MM5 3DVAR in Alaska.
3. AMPS observation study.
4. 3DVAR performance in AMPS.
5. Ensemble Kalman Filter in AMPS.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
1.The MM5/WRF 3DVar System
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
MM5/WRF 3DVAR Algorithm
•
Define analysis increments: xa = xb + I x’
•
Solve model space, incremental cost function:
1 T 1 1
J  x'   x' B x'  ( y'  y' o )T (O  F )  1( y'  y' o )
2
2
where y’ = Hx’, yo’ = yo - y.
•
Preconditioned control variable v analysis space (B = UUT):
x'  Uv  UpUvUhv
•
Choice of background error covariance model (NCAR, NCEP).
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR in the MM5/WRF Modeling Systems
MM5/WRF
Background
Preprocessing
Observation
Preprocessor
Background
Error
Calculation
xb
yo
3DVAR
xa
B
Cold-Start Mode
April 1st 2003
Update
Boundary
Conditions
MM5/WRF
Forecast
NCAR
Mesoscale and
Microscale Meteorology
3DVAR in the MM5/WRF Modeling Systems
xb
Observation
Preprocessor
Background
Error
Calculation
yo
3DVAR
xa
B
Cycling Mode
April 1st 2003
Update
Boundary
Conditions
MM5/WRF
Forecast
NCAR
Mesoscale and
Microscale Meteorology
2. MM5 3D-Var in
Alaska
Courtesy of AFWA
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR/MM5 AFWA Alaska “T1” Theater
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
MM5 3D-Var Comparisons: Alaska Theater (T1)
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
MM5 3D-Var Comparisons: Alaska Theater (T1)
• MM5 production (MVOI) compared to 3D-Var
initialized MM5 over the Alaska Theater
• Two cycles were run: 6Z and 18Z
• Data from both cycles is averaged together
• Model runs occurred between 9/7/02 and 9/15/02
• All verification is compared to observations
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
MM5 3D-Var Comparisons: Alaska Theater (T1)
April 1st 2003
AFWA Europe “T3A” 45km Verification: June 4-July 10 2002.
•3DVAR 
 vs. MVOI ( ).
•Verification against radiosondes: 00hr, 12hr, 24hr.
Height
Relative Humidity
NCAR
Mesoscale and
Microscale Meteorology
3. AMPS Observation
Study
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
Observations Available (September 2003)
 In-Situ:
- Surface (SYNOP, METAR, SHIP, BUOY).
- Upper air (TEMP, PIBAL, AIREP, ACARS).
 Remotely sensed retrievals:
- Wind profiler.
- Atmospheric Motion Vectors (SATOBS).
- ATOVS thicknesses (SATEMs).
- GPS total precipitable water.
- GPS refractivity.
- SSM/I oceanic surface wind speed and TPW.
- SSM/T1 temperature.
- SSM/T2 relative humidity.
- Scatterometer (Quikscat) oceanic surface
winds.
- Radar radial velocity.
 Radiances:
- SSM/I brightness temperatures.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
December 2002 AMPS Observation Statistics
a) Present statistics for
30km AMPS domain 2.
b) 3DVAR performed at 00
and 12 UTC.
c) First Guess = NCEP
“final” analysis.
d) Total 62 analyses.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
December 2002 AMPS 30km Temperature Statistics
T
Good
Reject
% Reject
Mean O-B
(K)
St. Dv. O-B
(K)
Synop
1919
118
6
0.0134
3.1521
Metar
979
15
2
0.1101
2.6628
Ships
452
30
6
-0.0865
1.4380
Sound
11554
95
1
0.0029
1.3585
Aircraft
16
19
54
0.7754
2.3836
Current setup: 12 hourly “cold starts” from NCEP global
analysis
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
December 2002 AMPS 30km Wind Statistics
WIND
Good
Reject
% Reject
Mean O-B
(m/s)
St. Dv. O-B
(m/s)
Synop U
1378
272
16
0.1064
2.5642
Synop V
1275
375
23
0.3804
2.5565
Metar U
816
92
10
0.3396
2.4143
Metar V
712
196
22
0.2779
3.0824
Ship U
167
18
10
0.3396
2.4143
Ship V
170
15
8
0.1552
1.9525
Sound U
10378
131
1
-0.0578
1.9560
Sound V
10371
138
1
-0.0289
2.0858
Pilot U
46
0
0
0.2503
2.0571
Pilot V
46
0
0
0.0436
2.2041
Aircraft U
39
8
17
-0.2501
5.6123
Aircraft V
42
5
11
-0.0332
5.5312
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
December 2002 Synop T, p variation by station.
10
8
a) Variety of diagnostic
utilities developed.
6
4
O-B (K)
2
0
88900
-2
89100
89300
89500
89700
89900
-4
b) Some stations indicate
bias w.r.t. model.
-6
-8
-10
5
4
3
c) Need to update station
elevations?
O-B (hPa)
2
1
0
88900
-1
89100
89300
89500
-2
-3
-4
-5
Station Number
April 1st 2003
89700
89900
NCAR
Mesoscale and
Microscale Meteorology
Preliminary Testing of MODIS (TERRA) data
Data time: 12 UTC 12/01/2002
Time Window: +/- 90 minutes.
QC: Reject if O-B>5 sigma_o
Observation Error: 4.5m/s
3060 obs after QC in 45km area.
O-B mean/std.dev = 0.53, 5.26m/s
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
Preliminary Testing of MODIS (TERRA) data
MODIS O-B vs Latitude
April 1st 2003
MODIS O-B vs Pressure
NCAR
Mesoscale and
Microscale Meteorology
Preliminary Testing of MODIS (TERRA) data
55 obs after QC in 15km area.
O-B u mean/rms = 4.32/6.88m/s
O-A u mean/rms = 0.77/3.72m/s
J / num_obs = 0.468
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
4. 3DVAR performance
in AMPS
Work performed by Syed Rizvi, Mike Duda
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Background Error – Vertical Eigenvectors
Streamfunction
Velocity Potential
Conclusion:
• Minor differences in
streamfunction.
•Very different
dominant mode for
velocity potential.
Old (global) – above, New (AMPS) – below.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Background Error – Horizontal Lengthscales
Streamfunction
Velocity Potential
Conclusion:
• “Local” lengthscales
significantly shorter.
•Should result in closer
fit to observations.
Old (global) – above, New (AMPS) – below.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Single Observation Test
U-Componenet RMS analysis error (AO) with new and Old Background
AMPS (Domain 2)
SFC = 1
Date: 2002120100
U-Componenet RMS analysis error (AO) with new and Old Background
AMPS (Domain 2)
SFC = 1
Date: 2003012712
5
2.5
4.5
4
2
3
New
Old
2.5
2
RMSE (m/s)
RMSE (m/s)
3.5
1.5
New
Old
1
1.5
1
0.5
0.5
0
0
sound(158)
synop(24)
metar(20)
sound(138)
airep(3)
4
1.5
3.5
1.3
3
1.1
2.5
New
Old
2
1.5
ships(3)
metar(12)
Temperature RMS analysis error (AO) with new and Old Background
AMPS (Domain 2)
SFC = 1 Date: 2003012712
RMSE (Kelvin)
RMSE (Kelvin)
Temperature RMS analysis error (AO) with new and Old Background
AMPS (Domain 2)
SFC = 1
Date: 2002120100
synop(20)
0.9
New
Old
0.7
0.5
1
0.3
0.5
0.1
0
sound(199)
April 1st 2003
synop(33)
metar(24)
airep(3)
-0.1
sound(186)
synop(29)
ships(3)
metar(13)
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Single Observation Test
po- pb = 1mb, observation error = 1mb.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Single Observation Test
po- pb = 1mb, observation error = 1mb.
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Analysis – 00 UTC 17 June 2003
Sea Level P, Surface Wind
April 1st 2003
T, p increment
NCAR
Mesoscale and
Microscale Meteorology
3DVAR Analysis – 00 UTC 17 June 2003
Sea Level P, Surface Wind
April 1st 2003
u, v increment
NCAR
Mesoscale and
Microscale Meteorology
AMPS Domain 1 Real-Time Verification: T+00
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
AMPS Domain 1 Real-Time Verification: T+12
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
AMPS Domain 1 Real-Time Verification: T+24
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
AMPS Domain 1 Real-Time Verification: T+36
April 1st 2003
NCAR
Mesoscale and
Microscale Meteorology
Conclusions and Future Work
•
“Basic” 3DVAR operational in AFWA Alaskan domain.
•
Antarctic December 2002-January 2003 data collection
underway (GPS, MODIS, includes tuning of ob errors).
•
AMPS real-time data ingest issues isolated, working on.
•
Initial performance of 3DVAR in AMPS is satisfactory.
•
3DVAR/Ensemble Kalman Filter comparison begun.
April 1st 2003