Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
1
Current status and plans at DWD
D. Majewski
Deutscher Wetterdienst
• Global 3D-Var assimilation
• SSO-Scheme for regional COSMO-EU model
• Prognostic precipitation
• New headquarters, new supercomputer
• GME 20 km / L60
• Global soil moisture assimilation
Numerical Weather Prediction at DWD
COSMO-EUGrid spacing: 7 km
Layers: 40
Forecast range:
78 h at 00 and 12 UTC
48 h at 06 and 18 UTC
1 grid element: 49 km2
COSMO-DEGrid spacing: 2.8 km
Layers: 50
Forecast range:
21 h at 00, 03, 06, 09,
12, 15, 18, 21 UTC
1 grid element: 8 km2
Global model GMEGrid spacing: 40 km
Layers: 40
Forecast range:
174 h at 00 and 12 UTC
48 h at 06 and 18 UTC
1 grid element: 1384 km2
2
Global 3D-Var data assimilation
• Replacement of the OI in the global (GME) data assimilaton system• Operational since 17 September 2008• Technical characteristics:
• PSAS (Physical Space Assimilation System)• 3-h assimilation window• Wavelet representation of the B-Matrix
• Observation usage currently similar to OI, but:• Direct assimilation of radiances (currently AMSU-A)• Temperature (TEMP, AMDAR) instead of geopotential (TEMP)• No use of “Pseudo-TEMPS” (profiles derived from IFS analysis)
• Scores still very similar to OI (which includes Pseudo-Temps)• Current work:
• Migration to NEC SX-8R (11 / 2008) and NEC SX-9 (01 / 2009)• Extended use of remote sensing data• B-Matrix without separability constraint
PSAS 3D-Var Data Assimilation for GME, Part I
3
PSAS 3D-Var Data Assimilation for GME, Part II
Anomaly correlation MSL pressure 00 and 12 UTC, northern hemisphere
• Observations used– Conventional (in situ) observations (3h interval) : 80 000– Radiances (AMSU-A) : 120 000
• Usage of remote sensing data will be extended:– Scatterometer 2008/2009– Radio occultations 2008/2009– AMSU-B 2009/2010– AIRS/IASI 2009/2010
PSAS 3D-Var Data Assimilation for GME, Part III
4
DWD 3D-Var – B-Matrix Representation
• Sparse matrix representation of wavelet transformed covariances– Representation of arbitrary (NMC or ensemble-derived) B-Matrices– No constraints regarding isotropy, symmetry, separability
• Accuracy– 1% (due to sparse matrix representation)– Noise due to limited ensemble size is an issue
• Status– Separability assumption still used
• Current work– Full 3-D non-separable formulation– Localization approach
DWD 3D-Var – Plans in the long term
• Hybrid 3D-Var / EnKF system for GME / ICON• Goals:
– Improve global assimilation system– Provide boundaries for local ensemble forecast system (COSMO-DE)– Share Resources with local ensemble forecast system
• Schedule:– 2008 1st version of LETKF for GME– 2009 3D wavelet representation in 3D-Var– 2009 Hybrid EnKF & 3D-Var– 2010 3D-Var for ICON– 2011 Operational EnKF for ICON
5
SSO scheme for 7-km regional model COSMO-EU
The problem …
Verification at DWD shows that the surface pressure in forecasts of the COSMO-EU model is systematically biased. In particular, in wintertime high pressure systems tend to develop a positive pressure bias, by 1-2 hPa after 48h, low pressure systems a negative bias ("highs too high, lows too low").
At the same time the wind speed tends to be overestimated by up to 1 m/s throughout the entire troposphere. The wind direction near the surface shows a positive bias of some degrees.
SSO Scheme for the 7-km COSMO-EU Model
The solution …
Implement SSO scheme of GME (Lott and Miller, 1997) in COSMO-EU.
6
Bias of wind speed [m/s]COSMO-EU experiment period: 26 Feb. – 17 Mar. 2008, 00 UTC
REF
SSO
Bias of wind direction [deg]COSMO-EU experiment period: 26 Feb. – 17 Mar. 2008, 00 UTC
REF
SSO
7
RMSE of vector wind [m/s]COSMO-EU experiment period: 26 Feb. – 17 Mar. 2008, 00 UTC
REFSSO
RMSE of MSL pressure [hPa]COSMO-EU experiment period: 26 Feb. – 17 Mar. 2008, 00 UTC
REFSSO
8
Prognostic precipitation in GME
Processes considered in the grid-scale precipitation scheme.
Currently, rain and snow are treated diagnostically.
Prognostic treatmentof rain and snow(including advection)is now being tested.
Prognostic precipitation
9
Cloud cover oflow clouds July 2008
96 h fct.
20 km
40 km
OLR intercomparison, July 14, 2007
10
New DWD Headquarters by June 2008
Offenbach, Frankfurter Strasse 135
New DWD Supercomputer
• NEC SX-9, Sun Login nodes and SGI Altix data base server
• Installation phase 0 in June 2008 (because of delay of SX 9): 2 x 0.3 TFlop/s sustained (SX-8R)
• Installation phase I in October 2008 until February 2009: 2 x 4.5 TFlop/s sustained (SX-9)
• Installation phase II in 2010: Upgrade by a factor of 3
• Main cost driver: Ensembles (data assimilation, forecasting)
11
GME 20 km / L60
Pre-operational trial of GME 20 kmTopography in the Alpine region
40 km, ni=192, Max: 2551 m 20 km, ni=384, max: 2918m
12
GME L40, GME L60, ECMWF L91 and ECMWF L60
GME scales well with the number of CPUs
13
First results based on ECMWF analysis
Schedule of introduction of GME 20 km / L60
Since summer 2007
Test runs of GME 20 km / L60 based on ECMWF analyses
February 2008
Daily test runs of GME 20 km / L60 based on ECMWF analyses
October 2008 until February 2009
New NEC SX-9 supercomputer (2 x 4.5 TFlop/s sustained) at DWD
March 2009
Pre-operational test runs of GME 20 km / L60 with 3D-Var at DWD
End of Q2 2009
GME 20 km / L60 with 3D-Var and soil moisture assimilation operational at DWD
14
Global soil moisture analysis
T0=0:00 LT, vv=12, 15 hrs
T0=3:00 LT, vv=9, 12 hrs
Analyses in different time zones need background fieldwith different forecast lead time
Data coverage active synops over land, 2007/10/07,12:00 UTC
Analysis for main run at 0:00 UTC, Observations at 12:00, 15:00 LT
15
2D-Var (z,t) soil moisture analysis (SMA)
)()()()( 221
221 obs
mmTobs
mmbT
b TTOTTwwBwwJ −−+−−= −−
))(()( 221
211
21
2 bmmT
mTmTT
mTbana wTTOBOww obs−
−−−− Γ+ΓΓ+=
Cost function penalizes deviations from observations and initial soil moisture content
Analysed soil moisture depends on T2m forecast error and sensitivity ∂T2m/∂w
0=∇J
)00:0,()00:15,00:12(2
kwT m
∂∂
Soil moisture increments compensate for T_2M forecast errors
Bias T_2M (K) Soil moisture increment (mm)
16
T_2M forecast errors are effectively reduced
Bias T_2M (K) with SMA Bias T_2M (K) without SMA