letkf on an eu/mediterranean scale - inverse problems4_torrisi_kenda_miniws.pdf · francesca...
TRANSCRIPT
Francesca Marcucci and Lucio Torrisi
CNMCA, National Meteorological Center,Italy
KENDA MINI-WORKSHOP Feb 28, 2014
LETKF on an EU/Mediterranean scale
Outline
Overview of last changes in operational suite:
dynamical land emissivity retrieval for AMSUA
pseudo-RH variable (J. Liu,2007)
reduced horizontal length for humidity obs
Experimental suites of the CNMCA-LETKF system: MHS observation assimilation
self evolving additive noise
7 km40 v.l.
2.8 km65 v.l.
- compressible equations- parameterized convection
- compressible equations- explicit convection
CNMCA NWP SYSTEM since 1 June 11CNMCA NWP SYSTEM since 1 June 11LETKF analysis ensemble (40+1 members) every 6h using RAOB (also 4D), PILOT, SYNOP, SHIP, BUOY, Wind Profilers, AMDAR-ACAR-AIREP, MSG3-MET7 AMV, MetopA-B/Oceansat2 scatt. winds, NOAA/MetopA-B AMSUA radiances+ Land SAF snow mask, IFS SST analysis once a day
Local Area Modelling: COSMO
Ensemble Data Assimilation:
COSMO-ME (7km) ITALIAN MET SERVICE
10 km45 v.l.
Control StateAnalysis
COSMO-ME EPS(pre-operational)
LETKFAnalysis
Changes in CNMCA LETKF systemChanges in CNMCA LETKF system
Recent changes in the operational suite are: COSMO model (tuning and adaptation)Space and time displacement in radiosoundings (only BUFR messages)Humidity bias correction for Vaisala RS AMSU-A radiances over sea and land Additive noise from IFS forecasts instead from model climatology
Dynamical land emissivity retrieval for AMSUAPseudo-RH variableReduced horizontal length for humidity obsHumidity bound check
Dynamical land emissivity retrievalDynamical land emissivity retrieval
The method proposed in Karbou et al. (2005) is used to improve the specification of land surface emissivity The emissivity is dynamically retrieved from suitable window channels, using background information to estimate the required terms in the radiative transfer equation. The retrieved emissivity is then applied for the forward calculations for the sounding channels The method is applied to AMSU-A and MHS data: AMSU-A channel 3 and MHS channel 1 are used to estimate the emissivity for the other sounding channels
BIAS STDV SPREAD
AMSU-A OBSERVATION INCREMENT STATISTICS
•A reduction of AMSU-A temperature bias is observed if the dynamical land A reduction of AMSU-A temperature bias is observed if the dynamical land emissivity retrieval is appliedemissivity retrieval is applied• No significant impact on standard deviation No significant impact on standard deviation
Period:Period:16-09-2012 16-09-2012
to to 05-10-2012 ) 05-10-2012 )
Dynamical land emissivity retrievalDynamical land emissivity retrieval
•The impact is evaluated through the relative difference (%) in RMSE computed against IFS analysis with respect to the configuration without dynamical emissivity retr. (MHS+AMSU-A assimilation) for 00 UTC COSMO runs from 16-09-2012 to 05-10-2012 • A clear positive impact is observed at day 2
Dynamical land emissivity retrievalDynamical land emissivity retrieval
+36h +48h +36h +48h +36h +48h
TEMPERATURE WIND VECTORTEMPERATURE REL. HUMIDITY
PSEUDO-RH VARIABLE
Two options were evaluated:1- LIU pseudo-RH normalization qv
i / qv_satbg
specific humidity normalized by the mean bg saturation specific humidity 2- pseudo-RH variable qv
i / qv_satbg
i
change of variable (from specific humidity to pseudo-RH) using the bg saturation specific humidity for each i-member
“Among the other choices of humidity variable types, the best result is from pseudo-RH assimilation. The error distribution of pseudo-RH is more Gaussian than specific humidity observations. It has similar error distribution as the relative humidity observations, but unlike relative humidity observations, it has no error correlation with the other observation variables” (J. Liu, 2007, PhD thesis )
Changes in CNMCA LETKF systemChanges in CNMCA LETKF system
Relative difference (%) in RMSEcomputed against IFS analysisfor 00 UTC COSMO runs from16-09-2012 to 05-10-2012negative value = positive impact
PSEUDO-RH VARIABLE
+12h +24h +36h +48h+12h +24h +36h +48h
HORIZONTAL localization lengths of humidity observations
The use of different localization lengths for humidity obervations have been evaluated:
1- HUM loc. Length = 0.5 TUV loc. length
2-adapted from humidity correlation length
Changes in CNMCA LETKF systemChanges in CNMCA LETKF system
Relative difference (%) in RMSE computed against IFS analysis for 00 UTC COSMO runs from 16-09-2012 to 05-10-2012 negative value = positive impact
A positive impact is observed at day 2 changing the humidity-obs localization
+36h +48h+36h +48h
HUM obs horizontal localization lenghts
1- HUM loc. length = 0.5 TUV loc. length 2- NEW
MAXIMUM-BASED MAXIMUM-BASED METHODMETHOD
MHS are treated as MHS are treated as “single-level” obs“single-level” obs
Assign radiance to the Assign radiance to the pressure level obtained pressure level obtained by a weighted average by a weighted average using the normalized using the normalized weighting function weighting function (WF) larger than 0.8 (WF) larger than 0.8
Radiances Radiances from from RTTOVRTTOV
ny
Mean FG Mean FG time = ntime = n
Bias CorrectionBias CorrectionQuality ControlQuality Control
Observation Observation IncrementsIncrements
a a an n nx x X= +
Mean FG Mean FG Radiances from Radiances from RTTOV RTTOV
Observation Observation Increments withIncrements withMean FG Mean FG
LETKF LETKF Analysis Analysis time = ntime = n
Radiances Radiances
FG FG Ensemble Ensemble
bnx
ANALYSIS ANALYSIS Ensemble Ensemble
Obs from NOAA16-18-19 and Obs from NOAA16-18-19 and MetOpAMetOpA RTTOV v 10.2RTTOV v 10.2 Off Line Dynamic Bias Off Line Dynamic Bias CorrectionCorrection Obs Error 2 °K Obs Error 2 °K (no FOV 1-8 and 83-90)(no FOV 1-8 and 83-90) Horizontal thinning 120 kmHorizontal thinning 120 km Channels 3-4Channels 3-4 Rain check on CH 2:Rain check on CH 2: 5 °K 5 °K Discarded over high Discarded over high orographyorography
Use Ensemble Mean as Use Ensemble Mean as reference for BC and QCreference for BC and QC
)x(H bn)xH( b
n
)H(xbn
MHS rad. assimilation
Weighting function (transmittance vert. derivative)
Experimental suites
Relative difference (%) in RMSEcomputed against IFS analysisfor 00 UTC COSMO runs from16-09-2012 to 05-10-2012negative value = positive impact
MHS rad. assimilation
MHS+AMSU
AMSU
Self-Evolving Additive NoiseSelf-Evolving Additive Noise• A new additive inflaction formulation is needed, because IFS
additive noise is not consistent with COSMO model errors statistics. • The self-evolving additive inflaction (idea of Mats Hamrud –
ECMWF) was chosen. The idea is different from the evolved additive noise of Hamill and Whitaker (2010)
• Difference between ensemble forecasts valid at the analysis time is calculated. The mean difference is subtracted to yield a set of perturbations that are scaled and used as additive noise. The ensemble forecasts are obtained by the same ensemble DA system extending the end of the model integration.
• The self-evolving additive perturbations are both consistent with
model errors statistics and a flow-dependent noise • The error introduced during the first hours may have a component
that will project onto the growing forecast structures having probably a benificial impact on spread growth and ensemble-mean error
Experimental suites
IFS ADD SELF EV. ADD
Self-Evolving Additive NoiseSelf-Evolving Additive Noise
Relative difference (%) in RMSE,computed against IFS analysis, with respect to NO-ADDITIVE runfor 00 UTC COSMO runs from16-09-2012 to 05-10-2012negative value = positive impact
+12h +24h +36h +48h+12h +24h +36h +48h
• Self-evolving additive inflaction / Stochastics physics• COSMO-ME Short-Range EPS based on LETKF is experimentally running
• ATMS radiances and GPS zenith total delays assimilation tests• Improved drifting radiosounding thinning • Use of KENDA• Shorter assimilation window
Current and future developments
Thanks for your attention!