THOR CT 4Predictabilityof the THC

GOALS of CT4

•Predict the Atlantic Meridional Overturning Circulation (and associated climate state) at decadal time scales

•Separate forced (anthropogenic) component from natural variations

•Make suggestions for the ocean observing system

ObservationsTime seriesOverflows

Storage

(re)analysesGridded data

GECCONEMOVAR

PredictionsModels

ProbabilisticIPCC

VerificationsMetrics

Windows of Opportunity

Decadal Predictions of the AMOCInitialize atmosphere-ocean-sea ice models from observed/analyzed ocean state

Perturb initialized models to generate ensembles

Perform decadal hindcasts and forecasts

Verify the results against own analyses and independent observations

In general: CT4 starts with CMIP5 (IPCC 5AR). Delivery first model results summer 2010. Sensitivity runs in 2011.

•Full initialization (KNMI, ECMWF) - Drift - No spinup needed

• Anomaly initialization (UKMO, MPI-M, IFM-GEOMAR) - Need spinup - Choice for nudging (how strong, long, which variables)

Sea ice is a challenge: short records, hardly thickness information

Analyses used: GECCO (MPI-M), NEMOVAR (ECMWF, KNMI), DePreSys (UKMO)

Initialization in THOR

Anomaly initialisation (ECHAM5/OM1)

Full field initialisation (ARPEGE4/OPA)

Mean bias removed

Mean bias removed

Full fields

Full fields

Courtesy: Francisco Doblas-Reyes (ECMWF, now IC3)

• Atmosphere •But, seems not to produce enough spread

• Ocean perturbations • Use forecast error• Use analyses error• Lagged analysis• Perturbed parameters• Stochastic physics• Optimal perturbations (e.g. Singular Vectors)

UREAD activity in THOR

Perturbing in THOR

Models considered so far

Models planned

Only requirement is a long (>500 year) control integration

See talk Ed Hawkins

Optimal perturbations

VerificationUse ‘own’ analyses AND independent observations: yr 1, yr 2, yr 2-5, yr 6-10.

Always verify against simple statistical model (e.g. damped persistence)

Verify trend and deviations from trend [deal with ‘forced’ trend, ideally with a control run without initialization (no-assim) or subtract global mean signal (presuming that is unrelated to AMOC)]

Metrics based on list of Atlantic-panel of WCRP-CLIVAR CT 2/3 may have suggestions (integrated, long time series, monthly means)

Oldenborgh, Doblas Reyes, Wouters, Hazeleger, in prep

See talk Bert Wouters

THOR CT4.2: impact ocean observations on THC predictions

Dunstone and Smith, 2009, submitted

Initialisation with sub-surface temperature and salinity (idealized experiments)

Initialisation with sea surface temperatures (idealized experiments

THOR CT 4.2 Forecast skill of top 360m ocean temperature (5-yr mean; idealized experiments)

Dunstone and Smith, 2009, submitted

Milestones & deliverables THOR

Paper: Multi-model decadal predictions of the AMOC. Start writing late 2010 when data of CMIP5 becomes available (if possible earlier), to be in time for deadline IPCC. KNMI lead.

Paper: Relative impact of initial conditions and GHG in different coupled models . Start writing late 2011, lead MPI-M

Paper: Assessment of ocean observations on predictability UKMO lead

Summary/planning•Experimental multi-model set up for CT 4.1 clear (CMIP5/IPCC)

•Perform runs and deliver results in July 2010 (MPI-M, UKMO, ECMWF, IFM-GEOMAR, KNMI) ; first multi-model THOR-AMOC review paper fall 2010

•Decided on verifications directions. May need workshop in summer 2010 (CT 2?)

•Data management at World Climate Data Centre, compliant with CMIP5/IPCC.

To work out in more detail:

•Protocol for observing system experiments (first tests done). Experiments in 2011.

•Protocol for experiments separating initial state and radiative forcing. Experiments in 2011.

Reporting month 18Optimal ocean perturbations from long ocean runs

Decadal prediction runs prepared and ongoing

Metrics verifying decadal predictions (ensemble means and spread)

Comparison of ocean (re)analysis in assimilation and hindcast modes

Idealized observing system simulation experiments

CT interactionCT1: provide long coupled runs for generating optimal perturbations

CT2/3: Metrics (robust, integrated, long term) for verifying decadal predictions

CT2/3: Suggestions for large events (GSA, 90s warming) to use for verification (windows of opportunity)

CT2: Ocean analyses without specific ocean observing system (e.g. without ARGO), to use for observing system simulation experiments

CT 5.2 ?????

Offer: predictions in WCDC

Metrics for THOR (see contribution Geert Jan)Directions:

For verification of climatology and forecast skill

• Trend

• Fluctuations around trend: yr 1, yr 2 – 5 averaged, yr 6-10 averaged

Skill against simple model (trend only or with damped persistence)

Against independent observations and own analyses

Data managementWhy?

• Share data for multi-model studies with common data-standards

• Not all experiments will be in CMIP5-database (or not on time)

Where? DKRZ could host an Opendap-like system

• Follow CMIP5 data standards

• FP7 Combine would like to ‘join’

Potential Predictability of DEN overflowNCEP-forced (Poster: Daniela Matei)

Denmark Strait (DEN) overflow transport potentially predictable up to 6 years in advance

Initialization strategies at MPI-M

Approach 1: Nudging ocean reanalysis (eg GECCO, ECMWF ORA-S3, SODA, ...)

Approach 2: Nudging atmosphere reanalysis (eg ERA40)

Approach 3: Nudging both ocean and atmosphere reanalysis (eg ERA40 and GECCO/ORA-S3)

Approach 4: Nudging to a NCEP-driven ocean simulation

based on anomaly nudging of different types of data

Sensitivity of Assimilation to applied Ocean State Estimate

HadCRUT3ECMWF ORA-S3IFM GECCONCEP-MPIOM

5yr Mean North Atlantic SST [x=60W:0E,y=20N:80N]