DATA ASSIMILATION activities @SHMU
description
Transcript of DATA ASSIMILATION activities @SHMU
DATA ASSIMILATION activities @SHMU
M. Derkova, M. Bellus, M. Nestiak
ALADIN/SHMU: model characteristics
ALADIN @SHMU
horizontal resolution 9.0km
spectral truncation 106 x 95
blending spectral truncation
53 x 48
number of grid points 320 x 288
vertical levels 37
operational time step 400s
coupling frequency 3h (ARPEGE LBC)
forecast length 72h (60h at 18UTC)
model version36_t1.10 (ALARO+3MT)
data assimilation frequency
6h
HPCIBM:10 p755 computing nodes (~320CPUs) + 2 p750 management nodes
ALADIN/SHMU DA: methods (1)
upper air• spectral blending by DFI (surface fields
copied from ARPEGE analysis)• operational since 19/09/2007surface• data assimilation using CANARI (NEW !)• standard setting, no special tuning • operational since 03/04/2012 (difficult
mental step)
ALADIN/SHMU DA: methods (2)
Data assimilation scheme: 6h frequency based on long cut-off observations and ARPEGE long cut-off LBC
Production 4x/day based on short cut-off observations and short cut-off ARPEGE LBC
get_oplace_longcanari_assimblend_assim
run_assim start end
18 UTC 02:20 02:30
00 UTC 08:55 09:10
06 UTC 13:45 14:00
12 UTC 20:55 21:10
run_prod start end
00 UTC 02:55 04:00
06 UTC 09:45 10:50
12 UTC 14:35 15:35
18 UTC 21:45 22:40
get_oplace_shortcanari_prodblend_prod
ALADIN/SHMU DA: methods (3)
ALADIN/SHMU DA: methods (4)DA step
GUESS ANALYSISSURF ANALYSISBLEND
BLENDING
GUESS
(no initialization)
6h FORECAST
ANALYSISSST
CANARIcopy of SST
NEW
BLENDING
ALADIN/SHMU DA: data (1)
• data assimilated: SYNOP 2m temperature and 2m relative humidity
• data sources: OPLACE + local database
• SST copied from ARPEGE analysis
• data processing: upgraded obsoul_merge script solves problem of corrupted OPLACE files
OK
PB!
New obsoul_merge script offers full control of observations: it reads the records and their headers one-by-one and checks many things. Duplicated records, records with wrong date or wrong observation type are excluded, lat/lon discrepancy of duplicated records is checked and so on.
courtesy of M. Bellus, available upon request
ALADIN/SHMU DA: data (3)
MERGING OBSOULS:
=> READING FILE (0): /data/nwp/products/oplace_long/2012-03-03/obsoul_1_xxxxxx_xx_2012030306 NT(orig): 06 NT(conv): 060000
=> READING FILE (1): /data/nwp/products/oplace_long/2012-03-03/obsoul_5_xxxxxx_xx_2012030306 NT(orig): 06 NT(conv): 060000
=> READING FILE (2): /data/nwp/products/obsoul/2012-03-03/obsoul_1_xxxxxx_xx_2012030306 NT(orig): 6 NT(conv): 060000
DATA PROCESSING:
record number: 1 total data: 42 station ID: 13704
record number: 2 total data: 37 station ID: HU12805
record number: 3 total data: 42 station ID: 12812...
ALADIN/SHMU DA: data (4)listing with detailed debug info
... record number: 3090 total data: 732 station ID: 01415 (!) 01415 => has wrong observation date/time... record number: 3205 total data: 37 station ID: 17601 (!) 17601 => duplicated observation...record number: 4583 total data: 32 station ID: 02095 (!) 02095 => has wrong lat/lon (saved:0.67/0.23 record:0.56/0.16) (file:20120303-060000 record:-533133308--5330800)
MERGING OBSOULS:
=> READING FILE (0): /data/nwp/products/oplace_long/2012-03-03/obsoul_1_xxxxxx_xx_2012030306
=> READING FILE (1): /data/nwp/products/oplace_long/2012-03-03/obsoul_5_xxxxxx_xx_2012030306
=> READING FILE (2): /data/nwp/products/obsoul/2012-03-03/obsoul_1_xxxxxx_xx_2012030306
=> TOTAL RECORDS WRITTEN: 5349
(!) Number of skipped records due to inconsistent date/time: 104(!) Number of skipped records due to inconsistent lat/lon: 82(!) Number of skipped records due to duplicity: 2389
=> FINISHED IN: 1 secs
ALADIN/SHMU DA: data (5)listing final info
ALADIN/SHMU DA: validation (1)
• 6months of e-suite (01/08/2011-30/01/2012)• reference = operational forecast (DFIblending)
– veral– point verification– special diagnostics
CANARI
OPER
2mT analysis, 00UTC
ALADIN/SHMU DA: validation (2)
BIAS (left) and STDEV (right) of 2m temperature of the guess (blue) and of the CANARI analysis (red) computed over whole domain for few randomly selected days.
ALADIN/SHMU DA: validation (3)“basic school” example
2mT analysis scores: top: over SK (OK), bottom: over whole domain (pb!)
OPE
RCA
NAR
I
ALADIN/SHMU DA: validation (4)What happens if SST is not correctly treated? Diff between CANARIanalysis and ARPEGE(?) analysis with SST cycled (left) and copied (right)
ALADIN/SHMU DA: validation (5)
2mT analysis scores over whole domain after correction
OPERCANARI cycled SSTCANARI copied SST
Generally there was positive impact found on the analysis and subsequent forecasts; on the surface and also in lower levels; namely for temperature and humidity. The impact is more pronounced in summer period.
Worsening of the daytime scores (in the summer): a problem in the forecasts for 12 and 18h day time, for any starting analysis time and any forecast length.
A cold temperature BIAS (winter).
ALADIN/SHMU DA: validation (6)
ALADIN/SHMU DA: validation (7)
2mT analysis 00UTC
CAN
ARI
OPE
R
2mT analysis 12UTC
OPE
RO
PER
OPE
RCA
NAR
IO
PER
CAN
ARI
OPE
RCA
NAR
I
ALADIN/SHMU DA: validation (8)
2mT +72h forecast
1000hPa T +12h forecast
OPE
RCA
NAR
IO
PER
CAN
ARI
ALADIN/SHMU DA: validation (9)
2mT +24h forecast from 12UTC
2mT +36h forecast from 00UTC
OPE
RCA
NAR
I
ALADIN/SHMU DA: validation (10)
2mRH RMSE +24h forecast from 12UTC
OPER CANARI2mRH RMSE +36h forecast from 00UTC
ALADIN/SHMU DA: validation (11)
OPER CANARI
ALADIN/SHMU DA: validation (12)2mT RMSE diurnal cycle (summer?) pb
00UTC
12UTC
CANARI
OPER
ALADIN/SHMU DA: validation (13)
difference of surface soil wetness in analyses between operational and parallel run
ALADIN/SHMU DA: validation (14)cold 2mT BIAS (winter?) pb
negative temperature BIAS in general for whole integration period near surfacethe temperature BIAS is negative mainly in winter season, but the fact, that it is generally worse for forecasts based on CANARI analyses comes from the warmer part of the testing period
01/08/2011-30/01/2012 01/08/2011-31/10/2011 1/11/2011-30/01/2012
OPER CANARI
RADAR assim in AROME/HU (1)
Technical development for 3D-VAR assimilation of radial Doppler winds (using 3 HU radars). Preliminary results - analysis increments of U wind component for 3 model levels are shown.
RADAR assim in AROME/HU (2)
EXP0426 → 1117c32160c13ae8730c2746b97b30b350ab BUD_20110426_0000.bufr
RADAR assim in AROME/HU (3)
EXP0426 → 1123_BU1117C32160c13ae8730c2746b97b30b350ab BUD_20110426_0000.bufr (12843)0a50b249103e46d11ae77022081e7bd2 NAP_20110426_0000.bufr (12892)7044d9c633ced3c72943c8ccb43a6d98 POG_20110426_0000.bufr (12921)
ALADIN/SHMU DA: plans
• Solve 2m T forecast problems• Increase the horizontal/vertical resolution • Test coupling with ECMWF (MF new schedule)• 3DVAR/ALADIN installation – check basic
impact• (optionally) 3VAR/AROME installation – to
continue with radar DA