1. Recent Developments at the Deutscher Wetterdienst (DWD)
WGNE-meeting
Oct. 2010, Tokyo
Michael Baldauf (DWD)

2. Supercomputing environment at DWD (Sept. 2010)
One production and one research computer NEC SX-9, both with:
14 nodes with16 processors / node = 224 vector processors
Peak Vector Performance / CPU: 100 GFlops Peak Vector Performance / node: 1.6 TFlops Peak Vector Performance total: 23 TFlops
main memory / node: 512 GByte main memory total: 7.1 TByte
Internode crossbar switch (NEC IXS): 128 GB/s bidirectional
Login nodes: SUN X4600
15 nodes with 8 processors (AMD Opteron QuadCore)/node = 120 processors (2 nodes for interactive login)
main memory / node: 128 GB
NEC SX-9 was first machine to break the limit of 100 GFlops
Database server: two SGI Altix 4700 2

3. The operational Model Chain of DWD: GME, COSMO-EU and -DE
(LME)
COSMO-DE
(LMK)
hydrostatic
parameterised convection
x  30 km
* 60 GP
t = 100 sec., T = 7 days
non-hydrostatic
parameterised convection
x = 7 km
665 * 657 * 40 GP
t = 40 sec., T = 78 h
non-hydrostatic
convection-permitting
x = 2.8 km
421 * 461 * 50 GP
t = 25 sec., T = 21 h
(since 16. April 2007) 3
GME: 39 Mio GPe (seit März 2010)
GME bis Feb. 2010: x  40 km, * 40 GP, t = 133 sec.
COSMO-EU: 17.5 Mio GPe
COSMO-DE: 10 Mio GPe 3 3

4. GME 40 km / L40  GME 30 km / L60 since 02. Feb. 2010
reduction of mean grid box size 1384 km²  778 km²
increase of number of vertical levels, increase of resolution in troposphere/tropopause
prognostic rain and snow  additional output fields  boundary values for COSMO-EU
Mainly improvement on northern hemisphere
but decrease in skill over southern hemisphere
(H. Frank, K. Fröhlich, T. Hanisch, DWD)

5. GME 30 km / L60 31 forecasts from 01.-31.01.2010 ANOC pmsl NH
BIAS pmsl NH
GME 30km / L60
GME 40km / L40
GME 40km / L40
GME 30km / L60

6. COSMO-EU with boundary values for QR, QS from GME30L60
COSMO-EU Parallelsuite
COSMO-EU Routine

7. Use of GPS - radio occultation (bending angles) in the 3DVar-Assimilation of GME (since 03. Aug. 2010)
Advantages of GPS radio occultations (bending angles)
high vertical resolution  even vertical thinning of data required!
globally accessible, approximately equally spaced
not influenced by clouds
measurement of the bending angle is almost bias free, temporally stable, independent from the instrument
number of profiles is proportional to the product of the sending GNSS-satellites (GPS, Galileo, GLONASS) and receiving LEOs:
CHAMP, GRACE-A (research satellites)
FORMOSAT-3 / COSMIC ( 6 research satellites)
GRAS (Metop-A)
 ~ 2000/d (May 2010)
(H. Anlauf, DWD)

8. Use of GPS - radio occultation in the 3DVar-Assimilation of GME
geopotential in 500 hPa: anomaly correlation of southern hemisphere for July 2010
with GPS
without GPS
Advance in predictability for about 7-8 hours!
(A. Rhodin)

9. COSMO-EU (7 km): (since 29. June 2010)
Replacement of the dynamical core ('Leapfrog-scheme', Klemp, Wilhelmson (1978) MWR) by the 'Runge-Kutta-scheme' (Wicker, Skamarock (2002) MWR, Baldauf (2010) MWR)
Goal:
'convergence' of the dynamical cores of COSMO-EU and COSMO-DE
Motivation:
higher accuracy of the RK-scheme towards leapfrog (in particular better horizontal advection for the dynamic variables); additionally better transport schemes for humidity variables
maintenance: only to foster one dynamical core
future developments are easier to do with a 2-timelevel scheme instead of a 3-timelevel scheme, e.g. physics-dynamics-coupling
(G. Zängl, M. Baldauf, A. Seifert, J.-P. Schulz, DWD)

10. Measurements to reduce a pressure bias
COSMO-EU / RK
Measurements to reduce a pressure bias
more accurate discretization of metrical terms (in pressure gradient) for the stretched vertical coordinate (Gal-Chen-coord.) ( definition: main levels geometrically are situated in the middle of the half levels)
improved lower (slip-) boundary condition for w: upwind 3rd order + extrapolation of vh to the bottom surface
Introduction of a subgrid scale orography (SSO)-scheme (Lott, Miller (1997) QJRMS)
use of a new reference atmosphere (allows z )
consistent calculation of base state pressure p0(z) on the main levels (i.e. not by interpolation but by analytic calculation)
SSO-Schema verbesserte vor allem den RMSE des Drucks (nicht so sehr den Bias)
p0(z): bisher Mittelung aus den Nebenflächenwerten
Diskr. metr. Terme: damit verschwand der Unterschied zw. Gal-Chen und sigma-Koordinate

11. SYNOP-Verification, 03.02.-06.03.2010, 0 UTC runs
COSMO-EU / RK
SYNOP-Verification, , 0 UTC runs
COSMO-EU RK (new)
COSMO-EU Leapfrog (old)
(U. Damrath)

12. COSMO-EU / RK
Measurements to improve the precipitation forecast
'checkerboard' pattern in precipitation can be eliminated by an increase in the calling frequency of the convection scheme (nincconv=10  4!) (remark: different time steps: Leapfrog dt=40 sec.; RK dt=66 sec.)
precipitation underestimation during summer was caused by a bug in the physics-dynamics-coupling: qi-detrainment tendencies of the improved Tiedtke convection scheme were lost.

13. COSMO-EU / RK
checker-board pattern in precipitation

14. Model climatology: monthly average of precipitation 12/2009
COSMO-EU / RK
Model climatology: monthly average of precipitation 12/2009
observation
COSMO-EU Leapfrog
COSMO-EU RK
(A. Seifert)

15. Main changes in the COSMO-DE
use of the extended radar composit for the Latent Heat Nudging (LHN): 16 additional radar stations from Netherlands , Belgium, France, and Switzerland (since ) up to now only crude quality control by clutter filtering and 'gross error detection' (K. Stephan)
vertically implicit TKE diffusion (instead of an explicit scheme; stability)
Baldauf, Seifert, Majewski, et al.: "Operational convective-scale numerical weather prediction with the COSMO model", submitted to MWR
current developments:
COSMO PP KENDA: km-scale ensemble data assimilation use LETKF methods project leader: Chr. Schraff (DWD)
COSMO PP UTCS: Unified turbulence - shallow convection scheme project leader: D. Mironov (DWD)
COSMO PP CDC: Conservative dynamical core project leader: M. Baldauf (DWD)

16. Current Status of COSMO-DE-EPS
Susanne Theis, Christoph Gebhardt, Michael Buchhold,
Zied Ben Bouallègue, Roland Ohl, Marcus Paulat, Carlos Peralta
with support by: Helmut Frank, Thomas Hanisch, Ulrich Schättler, etc
Gebhardt et al., 2010, Atmospheric Research, in revision
Start of pre-operational phase: Oct (20 members)
operational: ~2012 (40 members)

17. Generation of Ensemble Members
COSMO-DE-EPS
Generation of Ensemble Members
Variations in Forecast System
for the Representation of Forecast Uncertainty
Initial Conditions Boundaries Model Physics

18. Generation of Ensemble Members
COSMO-DE-EPS
Generation of Ensemble Members
COSMO 7km
Variation of boundary conditions
By COSMO 7km runs driven by different global models
Which computers are used?
at ECMWF: „7 km Ensemble“
at DWD: COSMO-DE-EPS
GME
IFS
GFS
…etc…
transfer of data

19. Generation of Ensemble Members
COSMO-DE-EPS
Generation of Ensemble Members
variation of „model physics“
Selection of Configurations
subjective, based on experts, verification
Selection Criteria:
1. large effect on forecasts
2. no „inferior“ configuration
different configurations
of COSMO-DE 2.8 km:
entr_sc
rlam_heat
q_crit
tur_len 1 2 3 5 4

20. Talagrand Diagram COSMO-DE-EPS Oct 7 – Nov 24 2009 15 days selected
(15 ensemble members)
grid point verification compared to
Radar observations
man erkennt deutlich die 3*5- Struktur: d.h. die 3 verschiedenen Randantriebe zerlegen das Ensmble in 3 Bündel.
die jeweils 5 Physikstörungen bringen dagegen nur wenig spread.
man sollte dazusagen, daß short range Modelle und Verifkation von Bodenvar. generell Unterdispersivität erzeugt.
1hr-precipitation
24 hrs lead time

21. Probabilistic Verification of Ensemble
COSMO-DE-EPS
Probabilistic Verification of Ensemble
Oct 7 – Nov
15 days selected
(15 ensemble members)
Brier Skill Score
0.4
0.3
0.2
0.1
0.0
grid point verification compared to
Radar observations
Ensemble clearly (?) better than deterministic run
reference:
deterministic COSMO-DE
1hr-precipitation
6-24 hrs lead time
threshold in mm/h

22. Requirements to a next generation global model
ICON (Icosahedral Nonhydrostatic)
Common project DWD - Max-Planck-Inst. f. Meteorology, Hamburg
Requirements to a next generation global model
Applicability on a wide range of scales in space and time → „seamless prediction“
(Static) mesh refinement and limited area model (LAM) option
Scale adaptive physical parameterizations
Conservation of mass (chemistry, convection resolving), energy?
Scalability and efficiency on massively parallel computer systems with more than 10,000 cores
Operators of at least 2nd order accuracy
G. Zängl, D. Majewski + ICON-Team 22

23. Baroclinic wave test with moisture
Modified baroclinic wave case of Jablonowski, Williamson (2008) test suite with moisture and Seifert, Beheng (2001) cloud microphysics parameterization (one-moment version; QC, QI, QR, QS)
Initial moisture field: RH=70% below 700 hPa, 60% between 500 and 700 hPa, 25% above 500 hPa; QV max g/kg to limit convective instability in tropics
Transport schemes for moisture variables:
Horizontal: Miura (2007) 2nd order with flux limiter
Vertical: 3rd-order PPM with slope limiter
Grid resolutions 70 km and 35 km, 35 vertical levels
Results are shown after 14 days

24. Temperature at lowest model level on day 14
Mesh refinement in ICON (G. Zängl, DWD)
Temperature at lowest model level on day 14
70 km
35 km
70 km, nested
nest, 35 km