CDEP Consortium Ed Schneider (COLA) and Chaojiao Sun (GMAO)

CDEP Consortium
Ocean Data Assimilation Consortium for Seasonal-to-Interannual Prediction
(ODASI)
COLA, GFDL, IRI, LDEO, NCEP, GMAO(NSIPP)
Ed Schneider (COLA) and Chaojiao Sun (GMAO)
Michele Rienecker, Steve Zebiak, Tony Rosati
Jim Kinter, Alexey Kaplan, Dave Behringer
http://nsipp.gsfc.nasa.gov/ODASI
CDEP Consortium
Ocean Data Assimilation Consortium for Seasonal-to-Interannual Prediction
(ODASI)
COLA, GFDL, IRI, LDEO, NCEP, GMA
GMAO
Michele Rienecker
Chaojiao Sun
Jossy Jacob
Robin Kovach
Anna Borovikov
IRI
Steve Zebiak
Eli Galanti
Michael Tippett
GFDL
COLA
Tony Rosati
Matt Harrison
Andrew Wittenberg
LDEO
Alexey Kaplan
Dake Chen
Jim Kinter
Ed Schneider
Ben Kirtman
Bohua Huang
NCEP
Dave Behringer
http://nsipp.gsfc.nasa.gov/ODASI
ODASI Themes:
1. ODA product intercomparisons (models, assimilation
methodologies, assimilation parameters) using a common forcing
data set and common QC’d in situ data streams
Models: MOM4, MOM3, Poseidon, Cane-Patton, LDEO4
Methodologies: 3DVAR, OI, EnKF, Reduced state KF and optimal
smoother, bias correction strategies
Coupled Forecast Sytems: CGCMs, Hybrid models, Intermediate models
2.
3.
Development of observational data streams
Validation of assimilation products in forecast experiments
4.
Observing system impacts - focused on TAO:
TAO array was established for S-I forecasting.
•
•
•
Is it effective in its present configuration?
Could it be modified to provide better support for S-I
forecasts?
what is its role c.f. other elements of the ocean observing
system?
Coupled Data Assimilation Workshop, Portland, April 2003:
• Assimilation of subsurface temperature improves Niño-3
forecast skill (usually), but we aren’t sure why (initialization of
state, anomalies)
• Forecast errors are dominated by coupled model shocks and
drifts
• It is not yet clear as to the “best” method for forecast
initialization
• consistent with observed state
• consistent with CGCM climatological biases
• initialize the model’s coupled modes
Can we use seasonal forecast skill to comment on observing
system issues?
The Experiments:
** initial conditions for 1 January and 1 July, 1993 to 2002
** Forecast duration: 12 months
** 6-member ensembles for each system
** The observations: assembled and QC'd by Dave Behringer at NCEP
— historical XBTs from NODC, MEDS
— TAO from PMEL web site
— Argo profiles from GODAE/Monterey server
** Surface forcing: assembled by GFDL
— NCEP GDAS daily forcing: momentum, heat, freshwater
— surface wind climatology replaced by Atlas’s SSMI surface wind
analysis
— include a restoration to observed SST and SSS
The Experiments (ctd):
Initial conditions for forecast experiments prepared using
1. All in situ temperature profiles, including the full TAO array
2. Western Pacific (west of 170W) TAO moorings
3. Eastern Pacific TAO moorings
Hypothesis: the Eastern Pacific data important for shorter lead forecasts
and the Western Pacific data important for longer lead forecasts.
Address uncertainty in the results by use of
•
•
•
•
ensembles
different assimilation systems
different CGCMs
different classes of models (CGCMs, hybrid, intermediate)
Outline
• Niño 3 SST anomaly Forecast skill
— from different models, assimilation systems,
observational constraints
— January consensus forecast from CGCMs
— Reynolds SST is verification
• Ensemble spread
• Skill in the equatorial band (analysis is verification)
• Impacts on the Analysis
• Conclusions
Niño-3 SST anomalies
CGCM1
CGCM2a
CGCM2b
January Starts
All TAO moorings
West TAO moorings
East TAO moorings
Obs (Reynolds)
hybrid1
hybrid2a
hybrid2b
Intermed1
Niño-3 SST anomalies
CGCM2a
CGCM2b
July Starts
All TAO moorings
West TAO moorings
Intermed1
East TAO moorings
Obs (Reynolds)
hybrid1
hybrid2a
hybrid2b
Intermed
CGCM Forecast skill - January starts - multimodel ensemble
All TAO moorings
West TAO moorings
East TAO moorings
Obs (Reynolds)
January starts
Niño3
Niño4
July starts
CGCM2a - forecast anomaly correlations
SST - July start
3mo
6mo
HC - July start
HC - Jan start
Jan
Jul
Analysis: Average Temperature in upper 300m
XBT profiles available per month
Dec 1996: 1440
Jun 1997: 2021
Seasonal drift of NSIPP CGCMv1 as a function of forecast lead time
June for each initialization month.
January for each initialization month
Niño 3 anomaly correlation of 0.9.
Vintzileos et al. (GSFC)
Coupled Data Assimilation Workshop, Portland, April 2003:
• Assimilation of subsurface temperature improves Niño-3
forecast skill (usually), but we aren’t sure why (initialization of
state, anomalies)
• Forecast errors are dominated by coupled model shocks and
drifts
• It is not yet clear as to the “best” method for forecast
initialization
• consistent with observed state
• consistent with CGCM climatological biases
• initialize the model’s coupled modes
Can we use seasonal forecast skill to comment on observing
system issues?
Conclusions:
Early stage of the analysis - we have to study the results in
more detail
Statistical significance of results - need more ensemble
members and more cases of both warm and cold events for
robust conclusions
• Eastern array definitely improves forecast skill
• Western array improves skill in central Pacific
• Entire array
— best results
— probably associated with atmospheric response across
the entire Pacific
— some indication that get a tighter spread
• results are subtle - complicated by coupled model shocks and
drifts