Ocean Data Assimilation Activities at NOAA/GFDL

Current status and future directionsMatthew Harrison, Ants Leetmaa,

Anthony Rosati, Andrew Wittenberg, Shaoqing Zhang

Ocean Modeling Needs Data Assimilation

• Uncertainty in air-sea fluxes • Uncertainty in model physics. • ODA produces consistent ocean states serving

as initial conditions for model forecasts• The reconstructed time series of ocean states

with a 3D structure aids further understanding of the dynamical and physical mechanisms of ocean evolution

• Ocean analyses for model simulation or hindcast verification

… however

• Ocean data assimilation products do not all agree reflecting uncertainty in models and data assimilation methods

• ODA does not necessarily lead to model improvement nor to increased understanding

ODA Components

• Quality Control

• Model

• Assimilation algorithm

Ocean Observations

• GODAE Server near real-time data stream (http://www.usgodae.org)– VOS XBT– CTD– Argo– TAO/PIRATA– Altimetry– maintain data center QC flags

• In-house quality control

Ocean Model

• MOM4 OM3– Re-engineered for multi-processor platforms– Utilize FMS infrastructure for communications, i/o and coupling– Global 1 degree x 50 level z-coordinate with 1/3 degree tropical

resolution– Tripolar grid (no polar filter required, full Artic Ocean)– KPP Vertical mixing– SWEBY advection scheme (courtesy of A. Adcroft )– Rotated isopycnal diffusion with G-M thickness flux

• Coupled with Ice Model

Ocean Data Assimilation

• 3DVar– Error covariance is stationary in time.

• Kalman Filter– Error covariance evolves based on model

linearization

• Ensemble Adjustment Kalman Filter – Use full model equations to propagate error

• 4DVar– “strong constraint” to model equations– No source/sink terms

Towards understanding the ocean’s role in climate

• GFDL will be expanding its ocean data assimilation activities starting this year through partnership with the ECCO group

• Estimating the Circulation and Climate of the Ocean (http://www.ecco-group.org)

• MIT (Carl Wunsch, Patrick Heimbach, Alistair Adcroft) /JPL (Ichiro Fukimori, Tony Lee) / Harvard (Eli Tziperman, Jake Gebbe)

OVERVIEW- ECCO Collaboration

Ocean Data Assimilation for Climate Testbed

Indicates common/additions from ECCO

OBSCommon metadata

3D-variational

Ensemble filter

Start 4d-var

Common infrastructureMOM4+

ENSO forecasts

GODAE-global change

NCEP Operations – when mature

Kalman filter

Routine Products

•Heat & salt storage•Sea level rise•Carbon storage•Initializations dec-cen forecasts

3DVar

http://nomads.gfdl.noaa.gov LAS/DODS

1980-present analyses

Routinely used for SI forecast initialization

Tropical Pacific is well constrained in upper 300m

Mid-to-high latitudes are more problematic– Sparse data outside of trade routes– Few salinity measurements – Use T/S covariance?– Multivariate using ARGO Salinity?

Ensemble Adjustment Kalman Filter (EAKF)

• Initial tests in the tropical Pacific basin for ENSO prediction

• Linearized atmospheric model with added noise• Each model realization is integrated between

analysis steps which determines the PDF of the model guess.

• Followed by a least square adjustment based on the model and observational PDF of the respective ensemble members.

Simulation

OI

EAKF

TAO

Seasonal Cycle of Temperature at 140W, Equator

Time Series of Temperature at 140W, Equator

Simulation

OI

EAKF

TAO

Simulation

OI

EAKF

TAO

1996-1999 Temperature at 140W, Equator

CDEP ConsortiumOcean Data Assimilation Consortium for Seasonal-to-Interannual Prediction

(ODASI)

COLA, GFDL, IRI, LDEO, NCEP, GMAO

http://nsipp.gsfc.nasa.gov/ODASIhttp://nsipp.gsfc.nasa.gov/ODASI

COLAJim KinterEd SchneiderBen KirtmanBohua Huang

GMAOMichele RieneckerChaojiao SunJossy JacobNicole KurkowskiRobin KovachAnna Borovikov

GFDLTony RosatiMatt HarrisonAndrew Wittenberg

IRISteve ZebiakEli GalantiMichael Tippett

LDEOAlexey KaplanDake Chen

NCEPDave Behringer

CGCM Forecast skill - January starts - multimodel ensemble

All TAO moorings

West TAO moorings

East TAO moorings

Obs (Reynolds)

Longer term change

• Globally averaged heat content

• Blue line is the GFDL R30 Climate Model

• Red Line is R30 Model with added aerosol forcings including volcanic

• Black line is the Levitus analysis

• Dashed line is GFDL 3DVar analysis

zonal average 1980-1999 temperature trend from 3DVar

Analysis

1992-1997 SSH Trend

• JPL Kalman Filter smoother (top)

• GFDL OI (middle)• ECCO 1deg Iter69

(bottom)• We need to

understand the differences between these analyses

Overview of Early Activities: GFDL, NCEP, Goddard, MIT(AER), JPL, Harvard

Year 1•Begin global state estimation at GFDL with ECCO•Produce and utilize routine data streams including ARGO T&S•Tangent linear/adjoint model. Ocean model and model parameterization development•Derivation of Kalman filter and smoother for MOM4 and Poseiden

Year 2•Global estimates with ECCO continue•Optimization of ECCO and GFDL “ECCO” like models for Dec/Cen applications•Experiments for S/I forecasts utilizing different assimilation schemes/models•Experiments with new data types (GRACE) & comparison to independent data types like length of day and earth polar motion•Adjoint sensitivity analyses to various controls and observing systems

Year 3

Year 4