ECMWF forecast system upgrades 2004 Martin Leutbecher

1
ECMWF forecast system upgrades 2004 Martin
Leutbecher

• IFS Cycle
• Oct 7th 2003 26r3
• Mar 9th 2004 28r1
• June 29th 2004 28r2
• Sept 28th 2004 28r3

acknowledgement Martin Miller RD
2
(No Transcript)
3
(No Transcript)
4
(No Transcript)
5
Cycle 26r3 (October 2003)

• New humidity analysis (q ? normalized RH)
• First operational assimilation of AIRS data
• AMSU-B on NOAA16 and 17, AMSU-A on AQUA (4th
  AMSU-A!), Clear Sky WV radiances from Meteosat-5
  (India), GOES-9 (Japan) and 12 (replaces GOES 8)
• Added AMV from GOES-12 and Japanese wind profiler
  data
• Assimilation of MIPAS ozone retrievals
• New TL radiation scheme in the 4D-VAR
• New aerosol climatology
• Relaxation of mass-flux limiter

6
229 cases N Hem S Hem Europe
7
(No Transcript)
8
Satellite data assimilated operationally in
October 2003

• 4 xAMSUA (NOAA-15/16/17 AQUA)
• 2xAMSU-B (NOAA-16/17)
• 3 SSMI (F-13/14/15)
• 2xHIRS (NOAA-16/17)
• AIRS (AQUA)
• Radiances from 5 GEOS (Met-5/7 GOES-9/10/12)
• Winds from 4 GEOS (Met-5/7 GOES-10/12) and
  MODIS/TERRA
• SeaWinds from QuikSCAT
• 2 Altimeter (ERS-2, ENVISAT) SAR (ERS-2)
• Ozone from SBUV (NOAA 16) and MIPAS (ENVISAT)

28 satellite data sources
9
Goes-W
Goes-E
Met-7
Met-5
GMS(Goes-9)
Enhanced space-based observing system from GEO
satellites
10
Impact of ozone retrievals from MIPAS on ENVISAT
11
Comparison of CY26R3 and CY26R1 at T511 for JJA
2003
T925 Anomaly Correlations
V850 at D10 (CY26R3-CY26R1)
THE CHANGE IN ATLANTIC TROPICAL / SUBTROPICAL
WINDS REDUCES THE PREVIOUS BIAS
EXPERIMENTS AT T95 SHOW THAT THE IMPROVEMENT IS
DUE TO THE CHANGE IN AEROSOL CLIMATOLOGY
12
DJFM Z500 Model Bias
CY26R1
CY26R3
dam
1962-2001 T95
13
DJF Model Bias
THE MAJOR BENEFITS IN DJF SEASONAL RUNS COME FROM
THE CHANGE IN AEROSOL CLIMATOLOGY
THE MASS-FLUX LIMITER CHANGE WAS MOTIVATED BY,
AND HAS A BENEFICIAL IMPACT ON, MEDIUM RANGE
ERRORS
MODEL 1962-2001 T95 PRECIP OBS XIE-ARKIN
79/80-98/99 SHADED mm day-1 Z500 OBS
ERA40 CONTOUR 2 dam SIGNIFICANCE 10 (GREY
NOT SIGNIFICANT) V925 OBS ERA40 SIGNIFICANCE
10
14

•  
• Cycle 28r1
• Implemented in Operations on March 9 2004
  
• Fix to the stratospheric noise problem  
• New snow analysis
• Corrected VarQC in 3D-Var
• Assimilation of GOES BUFR winds
• Unresolved bathymetry
•  

15
(No Transcript)
16
(No Transcript)
17

Level 11 D 5hPa
H12 Z10 from 28-12-2002 at 12z
18
SETTLS with LLSI (linear least squares
interpolation) at both departure and arrival in
the vertical trajectory computation
H12 Z10 from 28-12-2002 at 12z
Level 11 D
19
ECMWF Snow Analysis Revision (uses NESDIS snow
analysis) Original
Revised
NWS National Operational Hydrologic Remote
Sensing Center
20
Early delivery system 28r2 July 2004
21
(No Transcript)
22
(No Transcript)
23
(No Transcript)
24
Early Delivery N Hem 190 cases
25
Early Delivery S Hem 190 cases
26
Early Delivery Europe 190 cases
27
EPS early delivery configuration (cy28r2)

• computation of singular vectors at a non-critical
  time by using short-range forecast from DCDA
  (first guess of DA an) instead of DA analysis as
  initial condition
• Brier skill score for pos. anomalies of Z500

N-Hem, 29 cases

• neutral impact on
• spread and
• probabilistic scores

28
(No Transcript)
29
Cycle 28r3 28 September 2004
1. The new TL/AD cloud and convection schemes but
by default not activated (Philippe, Marta) 2.
Postprocessing of total column liquid water
(TCLW, 78) and total column ice water (TCIW, 79)
(Adrian). 3. Revised convection numerics with
additional call to the cloud scheme before
convection (Peter). 4. Improved numerics o tile
coupling with separate library for surface
routines (Pedro, Anton, Jean-Francois) -
Revised convection scheme numerics and calling
of the cloud scheme Use of TL/AD of vertical
diffusion in the 1st minimization of 4D-Var.
Reduction of radiation frequency to 1 hour in the
high resolution forecasts. Improved numerics
of surface tile coupling New linear and adjoint
cloud and convection schemes (not activated)
Post-processing of total column liquid water and
ice (TCLW/I, 78/9) Option of tracer transport
by convection and turbulent diffusion
30
Cycle 28r3 (continued)
1. The new TL/AD cloud and convection schemes but
by default not activated (Philippe, Marta) 2.
Postprocessing of total column liquid water
(TCLW, 78) and total column ice water (TCIW, 79)
(Adrian). 3. Revised convection numerics with
additional call to the cloud scheme before
convection (Peter). 4. Improved numerics o tile
coupling with separate library for surface
routines (Pedro, Anton, Jean-Francois) -
Revision to assimilation of AIRS improved
bias correction, improved cloud
detection Assimilation of MSG water vapour
radiances (and withdrawal of Met-7
radiances) New fast radiative transfer model
RTTOV8, accompanied by a retuning of all
satellite instruments Corrected use of AMSU-B
over land Assimilation of EARS (EUMETSAT ATOVS
Retransmission Service) AMSU data
Improved use of TEMP and SYNOP humidity
observations Technical developments for rain
assimilation (inactive) Revision of
perturbations targeted on tropical cyclones in
the EPS

31
34 cases
new phys revised numerics for cloud scheme ½
cloud ? convection ? cloud
32
34 cases
T511 does also improve in the tropics
33
34 cases
20N/20S
34
First estimation of stratospheric CO2 from AIRS
a potential contribution to an expanded
capability for global monitoring (GEMS)
35
(No Transcript)
36
EPS revised initial perturbations (cy28r3)

• Gaussian sampling for extra-tropical singular
  vectors
• Revision of initial condition perturbations for
  tropical cyclones (TC)
• New algorithm to determine optimisation regions
  based on predicted TC tracks from previous EPS
  run (if available)
• Initial condition perturbations extended to
  latitude belt 40S40N
• Up to 6 regions, computation also for tropical
  depressions
• Tropical singular vectors are computed in the
  subspace orthogonal to the used extra-tropical
  singular vectors
• The Caribbean remains an optimisation region if
  no TC in the vicinity
• Ortho-normalisation applied to set of all
  tropical SVs

37
Caribbean
Fabian
tropical depression
38
(No Transcript)
39
(No Transcript)
40
20 dates in Sep, Oct 03 and Jun 04, 33 TCs
41
Error of T255 EPS control forecast
61 cases in Feb/Mar and Aug/Sep 04
42
spread of perturbed forecasts around control
48 cases in Feb/Mar and Aug 2004
43
Brier skill score (pos. anomalies of Z500)
28r2 old EPS config
28r3 new EPS config
61 cases in Feb/Mar and Aug/Sep 04
44
Principal targets for operational implementations

• Increased vertical resolution (L91/62)
• Rain-affected µ-wave radiance assimilation
• Use of AMSR and SSMIS radiances
• Regional variation to Jb covariances
• New PBL and Turb. orog.-drag schemes
• Increased horizontal resolution (T799) with
  T255 in the inner loops of 4D-Var

1Q05
2/3Q05
45
T511- L60 32 MPI Tasks and 4 OpenMP threads 28R2
hpca
hpcc
hpcc speed-up1.7 times hpca
46
T511- L60 compared with T799 L91 64 MPI Tasks
and 4 OpenMP threads
hpcd T511
hpcd T799
Extra cost for T799 L91 3.5 times
47
L91
60- and 91-level vertical resolutions
L60
48
(No Transcript)
49
Assimilation of rain-affected microwave radiances
Background
PATER retrieval
Typhoon MITAG (12UTC 5 March 2002) Surface
rainfall rates (mm hr-1) from TMI data
50
4D-VAR RESULTS FC TRACK AND MSLP for cyclone
ZOE
BT RR Control Observed
51
Low Cloud Cover JJA87 climate
CY26R1 (mean36.3)
new PBL (mean34.7)
52
T511
10m winds
T799
53
Impact of T799 on 2 metre temperatures (3 week
mean)