1
22nd North America/Europe Data Exchange Meeting Reading December 9-11, 2009
Status report Bruno Lacroix (DPrévi/COMPAS)
With contributions from CNRM/GMAP
Outlines•Operational suite(s)
–current configurations (Computers, Models)–Use of data
•Issues under development •French data•E-suite•Future plans
2
Computing platform
NEC Configuration• NEC SX9 13 nodes of 16
processors
• 102.4 Gflops/CPU ,
• 1 To mem / node
• 2 machines:
• Operations 6 nodes since 22nd September 2009
• Research 7 nodes
• Next (and last) step 2010 Q1
• 20 nodes (2*10) :
• 32,7 Tflops
• + SX8 32 nodes / 8 processors 9.1 Tflops max
• Until February 2012, 2013 or 2014
3
Data management system: Soprano Architecture
4
Global model up to 102H at 00UTC (cut off 2H20), 72H at 06 (3H), 84H at 12UTC(1H50), 60H at 18UTC (3H)ARPEGE global spectral model TL538 C2.4 L60
60 levels, from 17m to 5Pa, horizontal resolution from 15km (over France) to 87kmLinear grid with T360 C2.4 orography (1080x540 pts)12 processors for ARPEGE forecast (10’ for 24H forecast)
4DVAR assimilation : 2 loops of minimization T107 C1 L60 (25 it.), T224 (30 it.)16 processors (1 SX9 node) for assimilation (40’ between cut-off and P0)
data used: SYNOP, SHIP, BUOY, AIREP, AMDAR, ACARS, TEMP, PILOTCMW winds GOES 11, 12 + Meteosat 7, 9 + MTSAT-1R, ModisSEVIRI radiances (Meteosat 9)AMI (ERS2), Seawind (Quickscat) and ASCAT (Metop) winds HIRS, AMSU-A, AMSU-B/MHS NOAA15, 16, 17, 18, Metop & AQUASSM/I (DMSP F13), AIRS AQUA, GPS ZTD, GPS RO, IASI (Metop)SST 1/12 degree from NCEP/NESDIS + SSM/I sea ice mask
Models configuration
5
ARPEGE horizontal resolution (km)
70km
6
Assimilation and forecast cycles
ARPEGE-Métropole, very short cut-off ( 1H05 at 00UTC )
54H run based on 3DVAR FGAT and P6 from previous short cut-off forecast P24H forecast avalaible at 0145 UTC
Guess
Analysis
Short cut-off18UTC Forecast
60H
Analysis
long cut-off18UTC
Guess
Analysis
Short cut-off00UTC
Forecast102H
Analysislong cut-off00UTC
Guess
3DVAR ARPEGE
Very Short cut-off 00UTC
Forecast 54H
8
Models configuration (follow up)
Regional model up to D2 06UTC (at 00, 06, 12 and 12UTC)
ALADIN spectral limited area model 9.5 km resolution on 2740kmx2740km
domain, 60 levels (289x289 pts) 3DVAR data assimilation: same data as
ARPEGE plus SEVIRI radiances Idem as dynamical adaptation of IFS Many coupling files
Tropical model 72H range at 00 and 12 UTC ARPEGE uniform model (TL539 C1 L60) ~37km No own data assimilation (interpolation of
stretched model analysis) To be stopped in 2010
Short Range Ensemble Prediction System 102H range 11 runs ARPEGE TL358 C2.4 L55 (23 to 133km) Based on singular vector perturbation
9
ALADIN : 24 operationnal domains
10
Coupling files for assimilation ALADIN
11
AROME-France operational since Dec 18 2008
AROME 600x512pts, Dx=2.5km, 41L, Dt=1mnAnd ALADIN-France 300x300 domain
four 30-h forecasts per day over France3-hourly 3DVar assimilation cycle including radar doppler radial winds, Meteosat radiances, synop T, Hu, wind NH model with 5-species "ICE3" microphysics, 1D TKE scheme, "EDKF" shallow convection, ECMWF radiation"SURFEX" surface model with tiles: soil/vegetation, sea, lake, town
12
AROME operational configuration the ALADIN-FRANCE operational suite provides :
– Lateral boundary conditions– Surface initial conditions : CANARI analysis (OI) at 00, 06, 12 and 18
UTC (the previous AROME forecast is used otherwise).
ALADIN cycle
AROME cycle time
13
Op. d’obs ARPEGE
Hu2m, T2m
V10m
SEVIRI HR
ALADIN (+ SEVIRI HR, Hu2m,T2m,V10m) AROME (+ radar)
GPSRO
GPS sol
IASI, AIRS
SEVIRI CSR
Données assimilées dans les modèles
14
Number of observations (counts of bits of info.)
The number of observations depends on the assimilation time SYNOP, RADAR Doppler winds, Aircraft measurements and SEVIRI
radiances are of great interest to supply information to the data assimilation system. SYNOP / BUOY [5000-6000] / a few units
RADAR Doppler winds [0-1000]
Ground GPS [150-170]
Radiosondes (TEMP, PILOT) [700-4000]
Various Aircraft messages [500-5000]
Cloud motion winds [0-20]
Scatterometer winds [0-80]
ATOVS pixels [100-200]
SEVIRI [150-300]
total [7000-15000]
Radar data Assimilation AROMERadar data Assimilation AROME
16
24 radars , 17 Doppler bande-C giving between 2 and 11 PPIs / 15’
• BUFR (Z,Vr,statut) archived into BDM (a file /elevation, 1km res.)
•Data center Opera in January 2011 with UK Met Office (about 70 radars over 29 countries)
.. ..
.. . .
.. . .... ... ..
0 100 km
10 km Observations used as profiles
Radar products from AROME
17
Explicit observation perturbations, andimplicit (but effective) background perturbations.
Ensemble assimilation (operational with 6 members…) : simulation of the error evolution
Flow-dependent B
b = M a (+ m )
a
3DVAR FGATT359C1L60
18
SIGMAB’s
« CLIMATOLOGY »
SIGMAB’s
« OF THE DAY »
8 dec 2006 r0
19
PEARP2
PEARP2 is based on ARPEGE model Two runs : at 06TU range 72h / 18TU range 108h
35 members : 1 control member and 34 pertubated membres
Initial state Perturbation : – Singulars vectors over 4 zones > > >
– Use f 6 analyses from AEARP (Assimilation Ensemble ARPege, L. Berre & G. Deroziers)
– Amplitude limited by variance-covariance matrix coming from assimilation cycle
Mdel Errors : multi-physics (physic ARPEGE operationnal scheme+ 7 schem validated by GMAP/PROC)
Resolution PEARP2 T358L65 C2.4 / augmentation en 2010 T538L65 C2.4 or C3.6 (~15km or 10km over France)
OTI (h) résolutionEURAT 12 Tl95
HNC and HS 24 Tl44TROP 12 Tl44
20
PEARP T358L55 C2.4 (~23km over France)
21
Targeted area for singular vectors
22
Assimilation/Forecast Suites
Operational suites Atmospheric models:– Limited-Area ALADIN
• La Réunion 3100x4600km with 3DVAR assimilation, • several research, commercial and transportable dynamical adaptation
versions
Chemical Transport Model MOCAGE, Forecasts of air quality up to 96H– 3 domains:Global/Europe/France, Horz. resolutions: 4°, 0.5°, 0.1°– Observations currently only used for validation
Ocean Wave Models, Forecasts up to 102H– Global (2), Europe, France, Horz. resolutions: 1°, 1°, 0.25°, 0.1°– assimilate Jason-1 and Envisat altimeter wave height data
23
Operational Suite on SX9 (96 procs)
Hour
Nb proc
24
Changes in NWP systemChanges in NWP system
07-2008 : IASI, SSM/I F14, statistics from ensemble assimilation cycle (6 members 3DVAR with T359C1L60 forecast)04/02/2009 Arpège/Aladin: new physical parameterizations, in operation using a Prognostic Turbulent Kinetic Energy (TKE) scheme March 2009 move to SOPRANO data managment environmentApril 2009 : new ALADIN-France configuration, coupled with IFS at 00 and 12UTC, without data assimilation (dynamical adaptation).22/09/2009: move to SX9 supercomputer
25
Evolution RMSE Z500 EuropeRegular improvment over 23 years
26
Telecom and data received (files)
link to Toulouse relevant to US/Europe data exchange:
Daily volume of satellite data files received from Exeter :HIRS NOAA16, 17, 19AMSU-A NOAA15, 16, 18, 19, AQUAAMSU-B/MHS from NOAA15, 16, 18, 19: 700 MbytesSSM/I and IS from DMSP F13->F15, F16, F17: 280 MbytesSeawind from Quikscat 240 Mbytes?AIRS from Aqua 500 Mbytes
RMDCN = ECMWF + GTS
8Mb/s
Lannion 4Mb/s
Recent advances in the use of observations in the French NWP models
September 2006:
20 stratospheric AIRS channels, SSM/I F13 and F15, Ground-based GPS data over Europe
September 2007:
GPS radio-occultation, ATOVS on MetOp (AMSU-A, MHS), ERS scatterometer,
February 2008:
Variational Bias Correction for radiances, ASCAT assimilation
June 2008:
HIRS on Metop, SSM/I F14, Emissivity parametrisation over land for micro-wave, CSR Meteosat, IASI
28
Evolution in obs number
H. Bénichou
Since July 2008, more than 2 million data per day
29
Radiances: ATOVSreceived with long cut-off
– NOAA15 (AMSU-A)– NOAA16 (AMSU-A, AMSU-B)– NOAA17 (HIRS, AMSU-B)– NOAA18 (AMSU-A, MHS)– Aqua (AMSU-A)– Metop (HIRS, AMSU-A, MHS)
H. Bénichou
30
Radiances: ATOVSused with long cut-off
– NOAA15 (AMSU-A)– NOAA16 (AMSU-A, AMSU-B)– NOAA17 (HIRS, AMSU-B)– NOAA18 (AMSU-A, MHS)– Aqua (AMSU-A)– Metop (HIRS, AMSU-A, MHS)
H. Bénichou
31
–SSMI (DMSP-F13): 7 channels
– AIRS (Aqua) : 54 channels over 324
– IASI (Metop): 51 channels over 314
Radiances:SSMI, AIRS and IASI
32
Winds: CMW (all in BUFR), MODIS, Seawind, AMI, ASCAT
33
Assimilation of MSG SEVIRI Clear Sky Radiances
10.8
m
cha
nnel
Associated percentage of cloud
free
CSR product from Meteosat-8/-9 (MSG/MSG-2) Hourly product Assimilation of
– 2 WV channels in 4DVar
250 km thinning
CSR
34
Ground-based GPS: Station selection
35
Radio Occultation GPS
Before screening
After screening10% data used
36
Evolution in managed/used data ratio
75% satellite data / 25% in situ data used
37
Land surface emissivity at microwave frequencies
Developments to assimilate surface sensitive satellite channels over land (Karbou et al., 2009)
Use of a dynamically retrieved emissivity to better assimilate AMSUA/B sounding channels over land in operations since July 2008
38
Assimilation of AMSUB over land
surface
upper atmosphere
observation structure function
Assimilation of AMSUB surface sensitive channels over land
channels 2 (150 GHz) and 5 (183+/-7 GHz) where orog > 1000 m
Emissivity dynamically derived from 89 GHz channel assigned to those channels
AMSUB channels already assimilated over land
channel 3 (183+/-1 GHz, where orog > 1500 m) channel 4 (183+/-3 GHz, where orog > 1000 m)
39
Impact on total column water vapour (TCWV) Average over the period 1 Aug-14 Sep’06
EXP = CTR + additional AMSUB channels over land
EXP-CTR
CTR
TCWV diurnal cycle at TOMB
40
Assimilation of SSM/I over land
surface
upper atmosphere
observation structure function
Assimilation of SSM/I channels 3 to 7 over land– 22V / 37V / 37H / 85V / 85H
Emissivity– dynamically retrieved from 19V/19H channels– assigned to channels of same polarization with a
frequency parameterization Quality control
– no coastal point, no land point with | lat | > 60° Variational bias correction (VarBC)
– “Ts” instead of “Ts” as one of the predictors – Emissivity dynamically retrieved from 19V channel
Only used over sea for the moment
Water vapour (TCWV & specific humidity profile) Average over the period 15 Jul-13 Sep’06
Control TCWV increments Mean= 0.027 kg.m-2 (0.1%)
Experiment TCWV increments Mean= 0.041 kg.m-2 (0.2%)
EXP-CTR TCWV analysis difference Mean= 0.165 kg.m-2 (0.6%)
EXP-CTR q analysis difference iso = 0.05 g.kg-1
500 hPa
20°N
more humidity in EXP
42
Impact of advanced infrared sounder radiances in the french global NWP ARPEGE model
1. Overview Current operational configuration
2. Use of IASI data Channels selection + Impact on forecasts Increase of IASI density Extension to Water Vapour channels
3. Cloud-affected Radiances Method Impacts from AIRS (analysis + forecasts
43
1. Current operational configuration1. Current operational configurationIASI operationally assimilated in :
- ''long wave'' temperature channels are assimilated, - clear condition (1 flag/channel, McNally & Watts, 2003):
AIRS operationally assimilated in :- ''long wave'' temperature channels are assimilated, - Clear and cloudy conditions- Over open sea
Sept, 06 → Jul, 08 1 Jul. 08 → 4 Feb. 09 Since 4 Feb. 09
Clear 19 channels (stratos) 54 channels (+35 tropos) 54 channels
Cloudy Ø Ø 54 channels
Sept, 06 → Jul, 08 1 Jul. 08 → 4 Feb. 09 Since 4 Feb. 09
Open sea Ø 50 channels 64 channels
Land Ø Ø 50 channels
Sea ice Ø Ø 32 channels
44
IASI assimilation: general features
Level 1C radiances are received via EumetCast in Toulouse(whole BUFR including 8461 channels)
A subset of 314 channels is retained in the Operational Observational DataBase (commonly chosen with other NWP centres)
Radiances are bias corrected using VarBC
45
2.a. Use of IASI data2.a. Use of IASI dataChannels selectionChannels selection
Sea 64 channels Land 50 channels sea-ice32 channels
Weighting functions
46
Geopotential: RMSE(noIASI wrt ECMWF) –RMSE(OPER wrt ECMWF)
Positive impact in mid-latitude and polar region in the troposphere
2.a. Use of IASI data2.a. Use of IASI dataImpact of IASI on forecastImpact of IASI on forecast 100
50
40
30
20
10
-10
-20
-30
-100
96h forecast range
NH SH
SH
72h forecast range
NH
NH
SH
SH
47
In operational configuration:– Pre-selection:
• Only data from detector #1• 1 fov AMSUA over 2• 1 scanline over 2
– Selection during screening:1 profile per 250km box
In order to increase density– Pre-selection:
• Only data from detector #1• More complex pattern
– Selection during screening: 1 profile per 125km box– Between 3.5 and 4 more profiles are assimilated
2.a. Increase IASI density
48
2.a. Increase IASI densityTypical data coverage over a 6-hour assimilation window(# of used channels / profile)
example for 4th March 2009,00UTC analysis time
1 profile / 125km box
1 profile / 250km box
49
2.b. 2.b. Impact of IASI density increaseImpact of IASI density increase
250 km 125 km Positive impact mainly for
southern hemisphere72h forecast range
NH SH
96h forecast range
NH SH
Geopotential:
RMSE(noIASI wrt ECMWF) – RMSE(OPER wrt ECMWF)
100
50
40
30
20
10
-10
-20
-30
-100
50
Add 9 WV channels (1320, 1349.5 and between 1392.5 and 1401.5 cm-1)
•Everywhere (sea, land, sea ice).
•sigma_o(WV) = 4 K
•(sigma_o(LW) = 0.5 – 1 K)
2.c. Extension to WV channels:2.c. Extension to WV channels:(Settings + impact on the analysis) (Settings + impact on the analysis)
Slight improvement of the innovation (obs- first guess) for other satellite humidity observations (MHS, HIRS 11 & 12)
51
2.c. Extension to WV channels:2.c. Extension to WV channels:impact on forecasts impact on forecasts
Positive impact on forecast wrt ECMWF analysis for large domains
Statistically significant
• for geopotential in the upper-troposhere for 72-96 hour for NH
Geopotential at 96h forecast range
IASIWV
REF
rmseBias
52
Relative humidity (12h forecast) wrt ECMWF analysis
Statistically significant in the whole troposphere until 24h forecast range for NH
IASIWV
REF
2.c. Extension to WV channels:2.c. Extension to WV channels:impact on forecasts impact on forecasts
rmseBias
53
3.a. 3.a. Cloud-affected radiances:Cloud-affected radiances:Method (Pangaud et al, 2009, MWR)Method (Pangaud et al, 2009, MWR)
CO2-SlicingCO2-Slicing
Cloud parameters retrieval (CTP et Ne)
Use of CTP and Ne into RTTOV
Simulation of cloudy radiance
Cloud-DetectCloud-Detect
Flag cloudy channels
Assimilation of cloudy channels• 600hPa<CTP<950hPa• AIRS: sigma_o(cloudy) = sigma_o(clear) = 1
54
3.b. 3.b. Cloud-affected radiances:Cloud-affected radiances:Impact on AIRS analysisImpact on AIRS analysis
EXP: assim clear + cloudy observations REF: assim clear observations only
Cloudy obs assimilated
Clear obs assimilated
More observations are assimilated, particularly for tropospheric channels (potentially more contaminated by clouds).
Geographical coverage of assimilated observations for the channel 239 (478 hPa:mid-troposphere). 01/09/06 à 00UTC
Bath 21-25 september 2009, EUMETSAT Meteorological Satellite Conference
55
3.b. 3.b. Cloud-affected radiances:Cloud-affected radiances:Impact on forecasts from AIRSImpact on forecasts from AIRS
blue:positive = reduction of RMSE red :negative = increase of RMSE
Statistics accumulated from 01/09/06 to 04/10/06RMSE difference with respect to radiosonde data
Altitude(hPa)
Forecast range (h)
GEOPOTENTIAL TEMPERATURE
Forecast range (h)
Significant up to 72h forecast range
56
Data monitoring
•http://www.meteo.fr/special/minisites/monitoring/menu.htmlUser/password available upon request [email protected]
57
News on upper-air observations
58
TEMP/TEMPSHIP
Nancy (12UTC) : stop end 2010 Lyon : 06UTC only Rapa : 18UTC only Takaroa: Stopped in August 2009 Tubuai and Amsterdam: impact study
Nimes :Robotsonde (MODEM) in 2010– Autosonde (Vaissala) at Bordeaux
One more ASAP end 2009 A fourth one in early 2010
59
Windprofilers network
La Ferté Vidame available on GTS
Marignane, Clermont Ferrand and Lannemezan Available on bilateral basis
End of Nice profiler
60
GPS surface network
IGN : RGP about 170 stations in January 2009
61
E-suite in research environnement
• Cycle 35T2_op1 (including RTTOV9)• New resolution : T798 C2.4 L70 (10km over France) Dt=600s (first version : 720s)• 2 loops of minimization in 4DVAR:
T107 C=1 L70 Dt=1800s 25iter T323 C=1 L70 Dt=1350s 30 iter
• Use of a stratiform precipitation scheme in second minimization• Use of a 6-member assimilation ensemble with 4D-VAR T399 C1 L70 , use of background error variances depending on the flux for all parameters (only vorticity in oper version) with a tuned spatial filter
• Evolution of turbulence scheme
•ALADIN-France:• Change of resolution: 7.5 km, 70 levels• Switch off Aladin-France as intermediate coupling model between global and
convective scale systems beg•AROME L60 direct coupling with ARPEGE
62
E-suite observation part
Reduction to 125 km of box sizes used to select satellite data instead of 250 km in oper version. (P. Moll …)
9 additional chanels Water Vapor IASI (land + sea) and 4 surface IASI channels de surface (sea)
Assimilation of humidity observations in low troposphere with AMSU-B over land
New RTTOVS coefficients for AIRS
Use of clear sky MODIS CMW
improved sea-ice mask
63
(A. Joly)
64
Change of horizontal thinning for radiances in ARPEGE
Operational horizontal thinning presently is 250 km
In E-suite, horizontal thinning is decreased to 125 km => ~ 3.5 times more radiances are assimilated
More impact in Southern Hemis.because this area has less conventional data& because we assimilate more data over sea than over land
Geopotential height1 isoline = 1 m
Wind 1 isoline = 0.2 m/s
Example: increased density only for IASI
Scores with respect to ECMWF analysesover a 3-week periodRMS(250km) – RMS(125km)
65
125km (instead of 250km) thinning (P. Moll)
• Multiplication by 4 of data incoming the screening more expensive !
• In screening output, observations really used :
Numbre in million Total Sat % sat obs
Oper 1,3 0,95 73%
New 3,8 3,48 92%
3H background errors statistics,given by new assimilation ensemble Arpege
assim. d’ens. 4D-Var
assim. d’ens. 3D-Var Fgat
Klaus storm, maxim error variances better forecasted
(position+amplitude) with 4D-Var version
24/01/2009 à 00h/03h
67
Preliminary Results …
scores (70 cases) with respect to radiosondes (TP) and IFS analysis (AC): Geopotential Temperature Vent
TP TP TPAC AC AC
68
New vertical resolution
0
1000
2000
3000
4000
5000
6000
7000
0 20 40 60 80 100
120
140
160
180
200
220
240
260
280
300
320
340
360
380
400
épaisseur de la couche (en m)
altit
ude
(en
m)
60 niveaux
41 niveaux
AROME From L41 to L60 (+ 37% CPU) :Increased vertical resolution mainly in the boundary layer:-1st level from 17mto 10m-27 level below 3000m (instead of 15)Spectral coupling above 100hPa
–Vorticité, divergence et temperature–20 first wave numbers (scale > 100 km)
-
alt L41 ARO L60 ARO L70 ARP/ALA (m) oper dbl dbl
)()()1()( tXtXtX coupleuraromearome
69
24 radars: 16 in C band (yellow circles) + 8 in S band (green circles). Volumes reflectivity (from 2 to 13 elevations).
22 Doppler radars (red circles), 2 planned (dashed red circles)
Radar data assimilation : French network
70
Radar data assimilation : Inversion method of reflectivity profiles
Caumont, 2006: use of model profiles in the vicinity of the observation as representative database
Consistency between the retrieved profile and clouds/precipitations that the model is able to create Possibility of wrong solution if the model is too far from reality… needs check
2
0
20
21exp
21exp
||xyy||
||xyy||x=xE
jsj
js
jj
71
Données utilisées
RADAR AROME Guess AROME ANALYSEZpseudo-anZobs
• Important for increments alance in convective situations• High departure to first guess allowed• Thinning:1 obs. on 15 kms boxes to avoid correlated observations and representaivity
erros • Sigma Obs increasing linearly up to 160 kms
dbZthresNielev
listenoirespatiale
thresthres ZZZZsimZZZZobs
0,1
#/
)()(
Retrieved profiles when
72
Future plansPEARP with 35 members at 06 and 18 UTC
Spring 2010:
Nec phase 2 (2* 10 nodes)
ARPEGE 10km L70, AROME 2.5km L60 (direct coupling with ARPEGE), ALADIN 7.5km L70,
high density radiances, more IASI and AIRS channels, extended condition of use, NOAA-19, radar reflectivity (AROME only)
Late 2010
ALADIN 3D-VAR Outre-Mer (Polynesia, New Caledonia, Antilles-Guyana) Configuration coupled with IFS using LBC project
New data : SSM/IS F16 and F17, AVHRR winds, GRAS on Metop, Iscat
Thanks for
your attention