Transcript Recent [Global DA] Developments at the Met Office Dale Barker, Weather Science, Met Office THORPEX/DAOS Meeting, 28 June 2011 © Crown copyright Met Office.

Recent [Global DA] Developments at
the Met Office
Dale Barker, Weather Science, Met Office
THORPEX/DAOS Meeting, 28 June 2011
© Crown copyright Met Office
Outline of Presentation
• Where Are We Now?
• What’s New:
• Observation Sensitivities.
• Resolution/climatological covariance upgrade (Nov 2010).
• Hybrid 4D-Var, moist control Variable (July 2011).
• What’s next?
© Crown copyright Met Office
Where Are We Now?
© Crown copyright Met Office
Operational NWP Models: Jun 2011
Global
25km 70L
4DVAR – 60km
60h forecast twice/day
144h forecast twice/day
+24member EPS at 60km 2x/day
NAE
12km 70L
4DVAR – 24km
60h forecast
 4 times per day
 +24member EPS at 18km 2x/day
UK-V (& UK-4)
1.5km 70L
3DVAR (3 hourly)
36h forecast
Crown copyright
Met Office
 4©times
per day
What’s new?
© Crown copyright Met Office
Observation impacts: Model Comparison
Richard Marriott
Recent Operational Upgrades
Parallel Suite 25 : Nov 2010
Global Data Assimilation - 4DVAR to 60km; CovStats from EC Ensemble
Seasonal Forecast Model to L85 (from L38)
Parallel Suite 26 : Mar 2011
Global Model – GA3.1 – Removal of Spurious Light Problem (since PS24)
UK models - Improvements to Drizzle/Fog in (DrFog) + JULES
Seasonal Forecast System - more members for 30day forecast
Post-processing - Best Data via Blending/Lagging - 5000 sites
UK4 run as Global Model Downscaler
Parallel Suite 27 : Jul 2011
Global DA– Hybrid Data Assimilation, Moisture Control Variable, New Obs
Global Model – Non-interactive Prognostic Dust
UK DA – Doppler Radar Winds
UK Model – More complete DrFog package
© Crown copyright Met Office
Climatological Covariances
Training data
Old (NMC method)
New (ECStats)
UM forecast
differences
T+30 – T+6 lagged
UM T+6 ensemble(10) from
EC analysis perturbations
Resolution
N216L70
N320L70
Period
Jan 2007
1-17 Oct 2006
Cov model
current
unchanged
© Crown copyright Met Office
Impact Of N216 4D-Var + ECStats
Verif Vs Obs (Above), Vs Anal(below)
Res impact Vs Obs = +0.37
ECStats impact Vs Obs = -0.02
Joint impact Vs Obs = +0.17
Res impact Vs Anal = +1.50
ECStats impact Vs Anal = +4.11
Joint impact Vs Obs = +5.52
© Crown copyright Met Office
Towards ‘Quasi-Continuous’ 4D-Var
OPS
(QG12)
UM
(QU06)
UM
model
background
analysis
increment
VAR
N108
vguess
(QG12)
VAR
N216
Hessian
eigenvectors
1445
1503
1623 GMT
Preconditioning N108 4D-Var reduces final N216 4D-Var cost from 21mins to 13mins
Rick Rawlins
SSMIS UAS + GPSRO to 60 km :
November 2009
Solid: Control
Dashed: GPSRO+SSMIS
• DA covariances introduces slow temperature drift near model model.
• Solutions:
• Analysis increment ramping (reduce increments near top).
• Additional data: GPS RO + SSMIS (Channels 21 and 22).
© Crown copyright Met Office
PS27: Global Data Assimilation
Upgrade (July 2011)
• Assimilation method
• Hybrid 4D-Var algorithm: Coupling DA and MOGREPS.
• Moisture control variable: Replacing RH with scaled
humidity variable
• Observation changes
• Introduce METARS
• GOES/Msat-7 clear-sky radiances, extra IASI (land)
• Revisions to MSG clear-sky processing and GPSRO
• Reduced spatial thinning
(ATOVS/SSMIS/IASI/AIRS/aircraft)
© Crown copyright Met Office
PS27 Hybrid data-assimilation
• Basic idea: Use data from MOGREPS-G to improve the representation of
background error covariances in global 4D-Var:
u response to single
u observation:
Climatological COV
MOGREPS COV
• MOGREPS is sensitive to the position of the front, and gives covariances that
stretch the increment along the temperature contours.
• Ensemble currently too small to
provide the full covariance, so we
blend the MOGREPS covariances
with the current climatological
covariances; i.e., we use a hybrid
system:
Hybrid COV
© Crown copyright Met Office
Pre-operational hybrid trials
Verification vs. obs
Better/neutral/worse
NH
TR
SH
Dec uncoupled (29 days)
29/94/0
6/117/0
12/109/2
Jun coupled
34/89/0
9/114/0
46/74/3
(28 days)
Dec uncoupled: +1.2
Skill:
RMSE:
© Crown copyright Met Office
Jun coupled: +1.6
Pre-operational hybrid trials
Verification vs. own analyses
Better/neutral/worse
NH
TR
SH
Dec uncoupled (29 days)
16/91/16
7/69/47
3/106/14
Jun coupled
49/63/11
9/86/28
18/82/23
(28 days)
Dec uncoupled: -4.0
Skill:
RMSE:
© Crown copyright Met Office
Jun coupled: -0.2
Pre-operational hybrid trials
Verification vs. ECMWF analyses
Better/neutral/worse
NH
TR
SH
Dec uncoupled (29 days)
35/79/0
39/75/0
14/100/0
Jun coupled (23/28 days)
51/63/0
27/87/0
46/66/2
Dec uncoupled: 1.721 (1.338%)
Skill:
RMSE:
© Crown copyright Met Office
Jun coupled: 1.337 (1.298%)
PS27 Moisture control variable
Limits on q/RH skew distribution
• Plot shows O vs B B vs A
similar
• Near 0% or 100% RH, (A-B)
is very skewed
• Transform to a function of
(A+B)/2 (Holm) – distribution is
much more symmetric
• This makes the analysis
nonlinear
• Much better fit of humiditysensitive satellite obs to
background
• Reduced spin-down of
precipitation
© Crown copyright Met Office
Moisture control variable
- improved fit of observations to background forecast
© Crown copyright Met Office
PS27: Impact of Package Components
Combined Winter/Summer Results
Hybrid 4D-Var
NWP index NWP index
vs obs
vs anl
+1.4
-1.9
Moisture control variable, replacing
RH with scaled humidity variable
0
+1.2
Introduce METARS
+0.8
+0.4
GOES/Msat-7 clear-sky radiances,
extra IASI (land)
+0.1
+0.1
Revisions to MSG clear-sky
processing and GPSRO
+0.2
-0.2
Reduced spatial thinning
(ATOVS/SSMIS/IASI/AIRS/aircraft)
+0.2
+0.4
© Crown copyright Met Office
Global Modelling Centre Comparison
Northern Hem. NWP Index (v. Anl) basket
% diff relative to Met Office
CAWC
NCMRWF
Met
Office
© Crown copyright Met Office
KMA
R
What’s next?
© Crown copyright Met Office
Operational NWP Configs:
Spring 2013 (Tentative)
Global
16-20km 85L (85km top)
Hybrid 4DVAR (50km inner-loop)
60 hour forecast twice/day
144 hour forecast twice/day
48/12member 40km MOGREPS-G
4*/day
MOGREPS-EU
Common NWP/reanalysis domain.
12Km 70L (40km top)
3D-Var (or NoDA)
48 hour forecast
 12 members ; 4 times per day
UKV
1.5km 70L (40km top)
3DVAR (hourly)
36 hour forecast
 4 times per day
12 member 2.2km MOGREPS-UK
Nowcasting Demonstration Project
•
1.5 km NWP-based nowcasting system
•
Southern UK only
•
Hourly cycling, ~12 hour forecasts
•
To be run in experimental mode during
London’s summer Olympics in 2012
•
3/4DVAR with Doppler winds/reflectivities.
Result below fro 4 cases/10-20cycles per
case:
David Simonin
DAE Techniques: Strategy Going
Forward
•
Continue to optimize 4D-Var: SE + algorithmic changes.
•
Continue to develop hybrid for short/medium-term (1997-2015):
•
Increase ensemble size, more sophisticated localization, etc.
•
Consider replacing ETKF as ensemble perturbation generator.
•
Develop UKV DA hybrid 3/4D-Var (2012-2015).
•
Develop 4D-Ensemble-Var:
•
Code and test within current VAR framework (2011-2013).
•
Extend to an ‘Ensemble of 4D-Ensemble-Vars’ (2013-2015).
•
Retire PF model if/when 4D-Ensemble-Var beats 4D-PF-Var.
•
Coupled DA: Monitor DA developments in ESM components and
increase modularity/sharing of algorithms where appropriate: ‘seamless
DA’.
Thank You
Questions?
© Crown copyright Met Office
PS28 (late 2011): Prepare for 6 hour
EPS Production Cycle
24 members 12
hourly
Still MOGREPS-R
domain at PS28
Not until PS30
© Crown copyright Met Office
12 members 6
hourly
Fraction Of Beneficial Observations
UKMO UM Fraction of Beneficial Obs
Ship
Satwind
SSMIS
Raob
QSCAT
Land
Aircraft
AMSUA
0.46
0.48
0.50
0.52
Conclusion: On average, 51-52% of observations are beneficial!
0.54
0.56
Richard Marriott