Upload
tale
View
33
Download
0
Tags:
Embed Size (px)
DESCRIPTION
- PowerPoint PPT Presentation
Citation preview
Ensemble Data Assimilation and its ImpactEnsemble Data Assimilation and its Impacton Seasonal Hindcast Skillon Seasonal Hindcast Skill
Christian KeppenneChristian Keppenne11, Michele Rienecker, Michele Rienecker22, Nicole Kurkowski, Nicole Kurkowski33, Jossy Jacob, Jossy Jacob44 & Robin & Robin
KovachKovach55
1,3,4,51,3,4,5Science Applications International Corporation Science Applications International Corporation San Diego, CA 92121San Diego, CA 92121
22Global Modeling and Assimilation OfficeGlobal Modeling and Assimilation OfficeNASA Goddard Space Flight CenterNASA Goddard Space Flight Center
Greenbelt, MD 20771Greenbelt, MD 20771
AGU 2006 Spring Meeting, Baltimore, MDAGU 2006 Spring Meeting, Baltimore, MD
[email protected], [email protected], [email protected] , [email protected] , [email protected]
OutlineOutline• Poseidon OGCM and GMAO coupled forecasting systemPoseidon OGCM and GMAO coupled forecasting system• Ensemble Kalman filter (EnKF) implementation for PoseidonEnsemble Kalman filter (EnKF) implementation for Poseidon• Impact of T+SSH assimilation on CGCM seasonal hindcast skill Impact of T+SSH assimilation on CGCM seasonal hindcast skill
OGCM• Poseidon (Schopf and Loughe, 1995)• Quasi-isopycnal vertical coordinate• Prognostic variables are h, t, s, u and v• Sea surface height (SSH) is diagnostic: = ibuoyancy(ti, si) hi
• About 30 million prognostic variables at 1/3 x 5/8 x L27 resolution
GMAO CGCM coupled forecasting system- GMAO AGCM, 2 x 2.5 x L34- LSM: Mosaic (SVAT)- OGCM: Poseidon v4, 1/3 x 5/8 x L27
- Production ODAS: OI of in situ temperature profiles with salinity adjustment- Experimental ODAS: EnKFNote: ODAS is run on OGCM prior to coupling to AGCM & LSM
CGCM and OGCMCGCM and OGCM
System-noise modelling• Model errors: use model EOFs to generate state perturbations• Forcing errors: add random perturbations to wind stress forcing
Multivariate update• Multivariate compactly supported covariances• Update T, S, u & v• Layer thicknesses (h) adjust between analyses • Incremental update• Process SSH and T observations separately
Online bias estimation• Used in SSH assimilationHybrid (3DVAR + ensemble) covariances• Used in T assimilationSpatio-temporal filtering of covariances• To mitigate effect of small ensemble sizesComputational cost• About 4 hours/month on 240 SGI Altix CPUs for 16 members
EnKF implementation
true climate
model climate
obse
rvab
le
a) Standard assimilation
Assimilation alters the model
climatology
true climate
model climate
obse
rvab
le
b) Assimilation with Online bias estimation (OBE)
est.
biasEstimates of variability (biased
model state) and climatological error (bias) are evolved side by
side
Online bias estimationOnline bias estimation
EnKF-32: No filter, H-section
EnKF-32: No filter, V-section
EnKF-8: filter, H-section
EnKF-8: filter, V-section
EnKF-8: No filter, H-section
EnKF-8: No filter, V-section
Covariance Covariance filteringfiltering
•Temporal filter (exponential moving average) to reduce weather noiseTemporal filter (exponential moving average) to reduce weather noise•Spatial filter to lessen effect of running small ensemblesSpatial filter to lessen effect of running small ensembles
Assimilation experimentAssimilation experiment• Assimilate T/P anomalies + TAO & XBT temperature profiles 1/1/2001-12/31/2001Assimilate T/P anomalies + TAO & XBT temperature profiles 1/1/2001-12/31/2001• Online bias estimation in SSH assimilationOnline bias estimation in SSH assimilation• Random forcing from 32 OGCM leading EOFs + wind-stress perturbationsRandom forcing from 32 OGCM leading EOFs + wind-stress perturbations• Compare 8, 16 & 32 member EnKFCompare 8, 16 & 32 member EnKF• Compare with Compare with
• no-assimilation controlno-assimilation control• Production ODAS (temperature OI + salinity correction)Production ODAS (temperature OI + salinity correction)
ExperimentsExperiments
EnKF8
EnKF16
EnKF32OI+S(T)
Control
• Control is most biased • OI partly corrects SSH bias• EnKF runs have no noticeable SSH bias
SSH OMF and OMA statisticsmean OMFRMS OMF
mean OMA
RMS OMA
Note the differences in scales!
T OMF and OMA statistics
EnKF8 EnKF16 EnKF32OI+S(T)Control
(Hybrid covariances used in T assimilation with EnKF result in stronger data weights than OI+S(T))
Impact of assimilation on CGCM hindcast skillImpact of assimilation on CGCM hindcast skill• 16-member EnKF16-member EnKF• Assimilate T + SSH for February-April of each year of 1993-2002Assimilate T + SSH for February-April of each year of 1993-2002• Couple OGCM to AGCM & LSM after running ODASCouple OGCM to AGCM & LSM after running ODAS• 12-month CGCM hindcasts initialized using ocean states from EnKF runs12-month CGCM hindcasts initialized using ocean states from EnKF runs
(to save CPU time, CGCM hindcasts have only 5 ensemble member) (to save CPU time, CGCM hindcasts have only 5 ensemble member)
• Assess impact of assimilation on SST & SSH hindcast skillAssess impact of assimilation on SST & SSH hindcast skill• Compare to history of CGCMv1 May-start hindcasts (problematic month)Compare to history of CGCMv1 May-start hindcasts (problematic month)
Niño-3.4 SSTNiño-3.4 SST
EnKF T+SSH, OBEEnKF T+SSH, OBE
missed EN missed EN
NSIPP CGCMv1 tier 1NSIPP CGCMv1 tier 1
false LN alert false LN alert
missed EN
False EN alert
Niño-3 SSTNiño-3 SST
EnKF T+SSH, OBEEnKF T+SSH, OBEEN underestimated
missed EN missed EN
NSIPP CGCMv1 tier 1NSIPP CGCMv1 tier 1
false LN alertfalse LN alertmissed EN
False EN alert
Niño-3 SSHNiño-3 SSH
EnKF T+SSH, OBEEnKF T+SSH, OBE
GMAO CGCMv1GMAO CGCMv1
SSH hindcastsSSH hindcasts
Although minor 1995 warm event is Although minor 1995 warm event is missed and 1998 EN is missed and 1998 EN is underestimated, the correlation with underestimated, the correlation with TOPEX is generally higher in the TOPEX is generally higher in the hindcasts initialized with EnKF T+SSHhindcasts initialized with EnKF T+SSH
Niño-3.4 SSHNiño-3.4 SSH
ConclusionsConclusions
• Online bias correction important when T data & SSH anomalies are assimilated.
• Filtering of covariances effective in lessening the impact of running a low-rank EnKF
• EnKF-initialized hindcasts can miss or underestimate El Niño events but result in less false El Niño or La Niña alerts than production forecasts