3rd WCRP Conference on ReanalysisJanuary 28 - February 1, 2008

The GMAO’s Ensemble Kalman Filter Ocean Data Assimilation System

Michele Rienecker, Christian Keppenne, Robin KovachJossy Jacob, Jelena Marshak

Global Modeling and Assimilation Office (GMAO)NASA/Goddard Space Flight Center

ODAS-1Algorithms:- Univariate optimal interpolation (UOI) - Multivariate EnKF

Model:Poseidon v4 OGCM (Schopf and Loughe, 1995) :

• Quasi-isopycnal vertical coordinate• Prognostic variables are H, T, S, u and v• Sea surface height (SSH) is diagnostic• 1/3° x 5/8° x L27

Observations:- T(z) from XBTs/Moorings + synthetic S(z) from T-S climatology- T(z), S(z) from Argo drifters - SSH from Topex/Poseidon and Jason (only for EnKF)

Forcing:- SSM/I and QuikSCAT surface wind stress products (Atlas & Ardizzone)- NCEP reanalysis surface heat fluxes- GPCP monthly precipitation- Reynolds & Smith SST relaxation- Levitus SSS relaxation

ODAS-2• Implemented with ESMF under GEOS-5 modeling system• MOM4

GMAO Ocean Data Assimilation Systems

Ocean EnKF

• Multivariate: updates T, S, u & v • Compactly supported background covariances• Temporal and spatial smoothing of covariances for small ensembles• System noise representation: Model-error and forcing-error model

(1) Perturbed physics(2) Perturbed forcing

(3) Perturbed state

where ωj are from pre-computed ensembles

• Online bias correction used in SSH anomaly assimilation

• Tests with 9-, 17-, 33- and 65-member ensembles

τ x,y(i) (tk ) = τ x,y

(i) (tk ) + μδτ x,y(i) (tk )

δτ x,y(i) (tk ) = τ x,y

(i) (tk−1) +γ(τ x,y(i) (tk1) − τ x,y

(i) (tk2))

δi (tk ) = γij (tk )ω jj

∑

EnKF-33: filterSchur(C,P) @(0N,156E,150m)

H-section z=150m V-section x=156E

03/31/01

06/31/01

09/31/01

12/31/01

Corr(T,T)

Mar. 01

Jun. 01

Sep. 01

Dec. 01

Temporal evolution of Kalman gain for T obs.Ocean stateOcean state--dependent covariances with the dependent covariances with the EnKFEnKF

ODAS-1Experiments: 1993-present- Univariate optimal interpolation (UOI) - temperature and

salinity analyses are independent

- Multivariate EnKF: temperature assimilation also corrects salinity and currents

- Argo impacts (2003 →)

- Seasonal Forecasts with GMAO CGCMv1

GMAO Ocean Data Assimilation Experiments

EnKF - OIOI Analysis is better

EnKF Analysis is better

EnKF - OI

OI Analysis is better

EnKF Analysis is better

Generally too salty

Weak SSS relaxation; Inadequate E-P fluxes

Surface Salinity, February 2006

Western Equatorial Pacific: 165-170E, 0-5S ( are Argo data)

EnKF with Argo EnKF w/o Argo Control

OI with Argo OI w/o Argo Levitus clim

Salinity Variability along the Equatorial Pacific24.5kg/m3 Density Surface

Salinity on 24.5kg/m3 Density Surface: March to July 2006

Salinity and currents on 24.5kg/m3 Density Surface: March to July 2006

GMAO CGCMv1 (Tier1) Forecast EnsemblesGMAO CGCMv1 (Tier1) Forecast Ensembles

12 month Coupled Integrations: 6-30 ensemble members

AGCM (AMIP forced with Reynolds SST; NCEP Analyses)

Ocean DAS (Surface wind analysis, GPCP precipitation; Reynolds SST, Temperature profiles; synthetic salinity profiles; Argo; altimetry)

Ocean state estimate perturbations:δ’s randomly from snapshots

Atmospheric state perturbations: δ’s randomly from previous integrations

AGCM: NSIPP1 AGCM, 2 x 2.5 x L34LSM: Mosaic (SVAT)OGCM: Poseidon v4, 1/3 x 5/8 x L27, with embedded mixed layer physicsCGCM: Full coupling, once per day

ODAS: Optimal Interpolation; Ensemble Kalman Filter“LDAS”: Offline forced land states (recalibrated)

1-monthforecast

3-monthforecast

6-monthforecast

EnKF with SSH EnKF w/o SSH OI

1-monthforecast

3-monthforecast

6-monthforecast

EnKF with SSH EnKF w/o SSH OI

1-monthforecast

3-monthforecast

6-monthforecast

EnKF with SSH EnKF w/o SSH OI

Summary

ODAS-1 Multivariate EnKF generally outperforms the OI implementation - analysis and forecasts

Argo - an invaluable additional data set to correct salinity

Next steps:

Use MERRA atmospheric state replay in GEOS-5 coupled model with ODAS-2

- generate better balanced IC for seasonal forecasts

AGCMFinite-volume dynamical core (S.J. Lin)Moist physics (J. Bacmeister, S. Moorthi and M. Suarez)Physics integrated under the Earth System Modeling Framework (ESMF)Generalized vertical coord to 0.01 hPaCatchment land surface model (R. Koster)Prescribed aerosols (P. Colarco)Interactive ozonePrescribed SST, sea-ice

AnalysisGrid Point Statistical Interpolation (GSI from NCEP)Direct assimilation of satellite radiance data using

JCSDA Community Radiative Transfer Model (CRTM)

Variational bias correction for radiances

AssimilationApply Incremental Analysis

Increments (IAU) to reduce shock of data insertion (Bloom et al.)

IAU gradually forces the model integration throughout the 6 hour analysis period

GEOS-5 Atmospheric DAS for MERRA

∂qn

∂t⎛

⎝ ⎜

⎞

⎠ ⎟

total

= dynamics (adiabatic ) + physics (diabatic ) + Δq

Model predicted change Correction from DASTotal “observed change”

Analysis

Background (model forecast)Raw analysis (from GSI)

Assimilated analysis(Application of IAU)

03Z 06Z 09Z 12Z 18Z15Z 21Z 00Z 03Z

Initial States for CorrectorAnalysis Tendencies for CorrectorCorrector Segment (1- and 3-hrly products)

ODAS-2Algorithms:- UOI, MvOI, EnKF

Models:- MOM4, but accommodates other models as well

Why a new system?:- Adherence to ESMF gridded component paradigm - Model independent & enhanced portability- Main new features:

• Faster analysis• Supports multi-model, multi-resolution ensembles• Adaptive localization for EnKF• Incorporation of breeding for MvOI• Implementation under GEOS-5 system allows coupling with our atmospheric analysis

• MERRA (2º) replay with ocean analysis

The Next GMAO Ocean Data Assimilation System