Overview of LACE data assimilation systems (with a certain regard) Contributions from many DA colleagues

Transcription

1 Overview of LACE data assimilation systems (with a certain regard) Contributions from many DA colleagues

2 Outline Data Assimilation in the ALADIN system LACE DA history The key elements of successful implementations OPLACE LACE DA networking LACE DA systems Questions

3 Data Assimilation Data assimilation is achieved by solving the BLUE x a =x b +K ( y H ( x b ) ) where the initial state (xa), background (xb), observations (y), Kalman gain (K) and observation operator (H) are denoted. A local data assimilation is substantially more expensive to implement than spin-up initialization (both computationally and regarding manpower), but in turn it enables a more accurate representation of small scale phenomena in the initial state. 3

4 Data Assimilation Comparison of a spin-up and a 3DVAR analyses 4

5 Data Assimilation RMSE and BIAS verification of AROME forecasts for T2m DYNA vs 3DVAR 5

6 Data Assimilation In ALADIN system there are two main data assimilation configurations which provide analysis tool for all CMCs. Variational method: Optimal interpolation: (analysis (xa), background (xf), observations (y), Kalman gain (K), observation operator (H), linearized observation operator (H), background error covariance matrix (Pf) and observation error covariance matrix (Po) Variational method is mostly applied for upper-air, optimal interpolation is used for surface analysis, but in principle both methods can be employed for either purposes. 6

7 Data Assimilation More details of DA can be found in literatures. Without aiming to give an exaustive list: Data Assimilation concepts and methods (ECMWF Training Course, F. Bouttier and P. Courtier) Analysis methods for numerical weather prediction (A.C. Lorenc, 1986, Quart. J. R. Meteorol. Soc. Atmospheric Data Analysis (R. Daley, 1991, Cambridge University Press, 457 pp.) Evensen, 2009: Data Assimilation, The Ensemble Kalman Filter, Springer, pp279 Atmospheric Modeling, Data Assimilation and Predictability (E. Kalnay, 2003, Cambridge University Press, 341 pp.) Bölöni, G., 2006: Development of a variational data assimilation system for a limited area model at the Hungarian Meteorological Service. Időjárás, 110, Fischer, C., Montmerle, T., Berre, L., Auger, L., Stefanescu, S. 2005: An overview of the variational assimilation in the Aladin/France numerical weather prediction system, Quart. J. Roy. Meteor. Soc., 131, Brozkova, R., Klaric, D., Ivatek-Sahdan, S., Geleyn, J.-F., Casse, V., Siroka, M., Radnoti, D., Janousek, M., Stadlbacher, K., and Seidl, H.: DFI Blending, an alternative tool for preparation of the initial conditions for LAM, PWRP Report Series No. 31, WMO-TD, No. 1064, 2001 The technical details will be provided in the frame of a DA kit. 7

8 Data Assimilation (OI method) T2mobs Tsoil,Tdeepsoil analysis RH2mobs Wsoil, Wdeepsoil analysis Illustration of technical steps: OBS in ODB Background SST anal 8 CANARI Namelist Climate files ISBA polynomes Increment file Const. files ANAL

9 Data Assimilation (OI method) T2mobs Tsoil,Tdeepsoil analysis RH2mobs Wsoil, Wdeepsoil analysis Illustration of technical steps: Background SURFEX OBS in ODB Background SST anal 9 CANARI Namelist Climate files ISBA polynomes Increment file Const. files ANAL OI_main (SURFEX) Namelist Climate files ISBA polynomes PGD ECOCLIMAP ANAL (SURFEX)

10 Data Assimilation (3DVAR method) vor,div,t,q,ps as control variables Non-linear H allows to use e.g. SATS, RADAR, GNSS Illustration of technical steps: OBS in ODB Background 10 OBS in ODB + QF, OMG SCREEN MINIM Namelist Climate files Const. files Namelist B matrix Const. files ANAL

11 Data Assimilation (3DVAR method) vor,div,t,q,ps as control variables Non-linear H allows to use e.g. radiance, RADAR, GNSS Illustration of technical steps: OBS in ODB OBS in ODB + QF, OMG Missing NH, HYD fields Background SCREEN MINIM Missing Surface fields Namelist Climate files Const. files Namelist B matrix Const. files 11 ANAL

12 Data Assimilation (Diffs between CMCs) However, the same DA methods, configs can be used for both AROME and ALARO (for different CMCs), but few settings have to be carefully set. When SURFEX is used (i.e. atmospheric and surface models are completely separated) some of the surface variables have to be inserted into first-guess in order to pass the setup level of either Screening or Canari. 12 CDPREFM(1) ='SURF', CDSUFM(1) ='Z0.FOIS.G', CDPREFM(2) ='SURF', CDSUFM(2) ='ALBEDO', CDPREFM(3) ='SURF', CDSUFM(3) ='ALBEDO.SOLNU', CDPREFM(4) ='SURF', CDSUFM(4) ='ALBEDO.VEG', CDPREFM(5) ='SURF', CDSUFM(5) ='EMISSIVITE', CDPREFM(6) ='SURF', CDSUFM(6) ='IND.VEG.DOMI', CDPREFM(7) ='SURF', CDSUFM(7) ='RESI.STO.MIN', CDPREFM(8) ='SURF', CDSUFM(8) ='IND.FOLIAIRE', CDPREFM(9) ='SURF', CDSUFM(9) ='RES.EVAPOTRA', CDPREFM(10)='SURF', CDSUFM(10)='GZ0.THERM', CDPREFM(11)='SURF', CDSUFM(11)='ET.GEOPOTENT', CDPREFM(12)='SURF', CDSUFM(12)='PROP.VEGETAT', CDPREFM(13)='SURF', CDSUFM(13)='PROP.ARGILE', CDPREFM(14)='SURF', CDSUFM(14)='PROP.SABLE', CDPREFM(15)='SURF', CDSUFM(15)='VAR.GEOP.ANI', CDPREFM(16)='SURF', CDSUFM(16)='VAR.GEOP.DIR', CDPREFM(17)='SURF', CDSUFM(17)='EPAIS.SOL', CDPREFM(18)='PROF',CDSUFM(1)='TEMPERATURE', CDPREFM19) ='SURF', CDSUFM(2) ='RESERV.EAU', CDPREFM(20) ='PROF', CDSUFM(3) ='RESERV.EAU', CDPREFM(21) ='SURF', CDSUFM(4) ='RESERV.GLACE', CDPREFM(22) ='PROF', CDSUFM(5) ='RESERV.GLACE', CDPREFM(23) ='SURF', CDSUFM(6) ='RESERV.INTER',

13 Data Assimilation (Diffs between CMCs) Furthermore DA configurations are not going to write out NH and Hydrometeor variables, therefore these should be added to analysis before model integration (from first-guess). CDPREFM(1)='S',CDSUFM(1)='CLOUD_WATER ',NBNIV(1)=60,LOPC(1)=.T., CDPREFM(2)='S',CDSUFM(2)='ICE_CRYSTAL CDPREFM(3)='S',CDSUFM(3)='SNOW CDPREFM(4)='S',CDSUFM(4)='RAIN ',NBNIV(2)=60,LOPC(2)=.T., ',NBNIV(3)=60,LOPC(3)=.T., ',NBNIV(4)=60,LOPC(4)=.T., CDPREFM(5)='S',CDSUFM(5)='GRAUPEL CDPREFM(6)='S',CDSUFM(6)='TKE ',NBNIV(5)=60,LOPC(5)=.T., ',NBNIV(6)=60,LOPC(6)=.T., CDPREFM(7)='S',CDSUFM(7)='PRESS.DEPART CDPREFM(8)='S',CDSUFM(8)='VERTIC.DIVER ',NBNIV(7)=60,LOPC(7)=.T., ',NBNIV(8)=60,LOPC(8)=.T., CDPREFM(9)='S',CDSUFM(9)='CLOUD_FRACTI ',NBNIV(9)=60,LOPC(9)=.T., 13

14 Data Assimilation (Diffs between CMCs) Few other important parameters which have to be specified differently during assimilation LDIRCLSMOD: L to take 2m model equivalent directly from input file LAEICS :.T. surface fields initialisation LBLVAR : KEY FOR CALLING SPECIFIC 2M OBS OPERATOR FROM LAST LEVEL LSPRT :.T.: if R*T/Rd "virtual temperature" as spectral variable During the preparation of B matrix, few more parameters have to be set differently for AROME and for ALARO. Specific humidity in grid-point space in AROME (FEMARS!) LSPRT 14 :.T.: if R*T/Rd "virtual temperature" as spectral variable

15 Data Assimilation (Monitoring and diagnosing) Observations can add useful information only when background and observation errors are correctly specified. O-B and O-A statistics can provide relevant information about it O-B bias should be negligible O-B distribution should be gaussian sigma(o-a)/sigma(o-b) should be smaller than 1 Check Jo/n in diagnostics which should be order of 1 Check the increments (e.g. with independent observations) Do long-term experiments and make monthly statistics 15

16 LACE DA history The DF Blending of global ARPEGE analyses and LAM first-guesses was started in 2001 in Prague and later it was combined with CANARI surface assimilation around In DVAR became operational in Hungary (and in MeteoFrance as well). Only two Members had operational DA in LACE project ( ) triggered implementations of operational DA systems at all Members, taking into account the local infrastructure & staffing and considering the opportunities for operational task sharing. After 2010 the number of operational DA systems started to grow and nowadays there are ~10 in LACE countries. 16

17 LACE DA history DF Blending +surface DA 2006 (CZ) DF Blending 2001 (CZ) 3DVAR 2005 (HU) 3DVAR +surface DA 2008 (HU) DF Blending 2006 (SK) DF Blending +surface DA 2012 (SK) 3DVAR+surface DA 2011 (SI,CR) only surface DA (AT) DA project 2009 (OPLACE) 2010 (DA works technically) 17 BlendVAR +surface DA 2015 (CZ) AROME 3DVAR 2013 (HU) AROME 3DVAR+surface DA 2014 (AT)

18 The key elements of successful implementations During the LACE project of : 18 scientific & technical support of experienced partner (OMSZ,MF) operational task sharing (OPLACE) close cooperation and exchange of know-how (Forum, DAWD) on various aspects (obs handing, bias correction, tuning, ) exchange of the tools (observation monitoring and QC, tools for computing background error structure functions, ) manpower (1-2 FTE)

19 The OPLACE system The common LACE observation pre-processing system (OPLACE) was build in order to provide observations in appropriate format for DA. The main aims were to support DA implementation, avoid duplication of work on observation pre-processing, share maintenance and to let people to focus on DA. Observation pre-processing comprises mainly decoding, conversion to the local databases, simple QC, conversions to suitable format for ODB conversion. More information will come in Alena's presentation tomorrow. 19 SYNOP local DB SEVIRI NWC SAF OULAN OBSOUL Tb calibration, conversion to GRIB BATOR BATOR ODB ODB

20 LACE DA Networking LACE Data Assimilation Working Days (DAWD) are organized yearly in Autumn for 2,5 days. Program consists of status presentations of local DA systems and research activities, discussions on issues, coordination and planning, but so far almost zero practical exercises or code developments. Non-LACE DA colleagues are also invited. In previous years colleagues were also participating from HIRLAM (Norway, Sweden, Netherlands), ALADIN (Tunisia, Portugal, Belgium) and remotely from MF. Another valuable floor for spreading information, direct interaction and sharing tools is LACE Forum ( 20

21 LACE DA systems 7 LACE countries (Austria, Croatia, Czech Republic, Hungary, Romania, Slovenia, Slovakia) 9 operational systems (+3 pre-operational ones)

22 LACE DA systems Summary table of operational LACE DA systems LACE DA progress report

23 LACE DA systems Summary table of operational LACE DA systems LACE DA progress report

24 ALARO DA in Hungary ALARO-1 physics, ISBA surface 8km horizontal, 49L vertical cy38t1_bf03 SMS environment 4 forecasts/day up to 60 hours 6 hours assimilation cycling 3 hourly coupling IFS global CANARI+3DVAR Observations: SYNOP, AIREP, TEMP, Geowind AMV, SEVIRI from MSG-10, ATOVS from NOAA18, passively ATOVS from NOAA-19, METOP-A,B and Windprof Validation: 24 Configurations: Bator, 701, 002, 131

25 ALARO DA in Hungary Monthly monitoring is prepared and sent to all LACE partners Regular information about the DA system and the quality of the observations! 25

26 ALARO DA in Czech Republic ALARO-1 physics, ISBA surface 4.7km horizontal, 87L vertical cy38t1_op5 4 forecasts/day up to max. 72 hours 6 hours assimilation cycling 3 hourly coupling ARPEGE global DF Blending E87x69 CANARI+BLENDVAR Observations: SYNOP, AIREP, TEMP, Geowind AMV, SEVIRI from MSG-10 Validation: 26 Configurations: DFBlending, Bator, 701, 002, 131

27 ALARO DA in Czech Republic BlendVAR: DF Blending Bucanek et al., 2015) LAM better representation of small scale component of the analysis, but have difficulties to correctly specify large scales. The BlendVar scheme is inspired by a variational assimilation in global spectral models that use the multi-incremental approach for faster convergence (Veerse & Thepaut 1998). The DF Blending is able to transfer long wave part of analysis of driving model (ARPEGE) to LAM ALADIN while keeping short wave part of the first-guess. DF Blending (without 3DVAR) does not require local use of observations, but it is constrained to the time schedule of the global DA system DVAR (see

28 AROME DA in Austria AROME-MesoNH, SURFEX 2.5km horizontal, 90L vertical cy40t1 export (3DVAR: cy36t1 export) 8 forecasts/day up to 48 hours 3 hours assimilation cycling 3 hourly coupling IFS global OI_main+3DVAR Observations: SYNOP, AIREP, TEMP, PILOT, Geowind AMV, SEVIRI from MSG-10, ATOVS, HIRS, IASI from NOAA and METOP sats Lake Temperature replaced from measurements Snow initialization with Modis data and SNOWGRID model Validation: 28 Configurations: Bator, OI_main, 701, 002, 131, local tools

29 Challenges (just one example) 12 operational and pre-operational DA systems Mostly 6 hours assimilation cycle used, but there are 3 hourly cycled and 1 pre-operational hourly cycled assimilation system as well. During last years the efforts of the LACE DA colleagues are moved towards maintenances, validations and operations duties instead of doing research activities. Oper Core developments Research ETKF Around DVAR/HU 3RUC 4DVAR 3DVAR/CR BLEND/SK 3DVAR/HU BLENDVAR/CZ 3D-FGAT OI/EKF BLEND 1RUC OI/EKF 3RUC/SL Around RUC/HU 3DVAR/RO BLEND/SK 3DVAR/CR OI/AT 3RUC/AT 29

30 Conclusions DA can significantly improve NWP forecasts. On the other hand, DA is manpower and time demanding. DA requires continuous monitoring and maintenance Collaborations in common areas are highly recommended Exchange of experiences and know-how can help to reduce duplicated work It is also important to explore local observations and local specialities (e.g. LACE has limited experiences with observations over sea!) 30