MultiModel ENSO Prediction with CFS and CCSM3

1
Multi-Model ENSO Prediction with CFS and CCSM3

• Ben Kirtman
• University of Miami-RSMAS
• Center for Ocean-Land-Atmosphere Studies (COLA)

Acknowledgement Dughong Min
2
Why?

• Multi-Model Methodologies Are a Practical
  Approach to Quantifying Forecast Uncertainty Due
  to Uncertainty in Model Formulation
• And, Apparently Improve Forecast Quality
• US National Multi-Model Capability
• Results with CCSM3 Encouraging
• Use Existing Operational Analyses for
  Initialization (CFSRR for Retrospective Forecasts)

No Determination of Which Model is Better -
Depends on Lead- Time and Start Month and the
Debate is Counter Productive
3
Outline

• Mean Simulation Errors
• Annual Mean
• ENSO Simulation Errors
• Forecast Quality
• Forecast Initialization
• Nino3.4 Forecast Assessment
• Random Thoughts

4
Equatorial Pacific SST
OISST
CFS
CCSM
5
Equatorial Pacific SST Variance
CFS
OISST
CCSM
6
Forecast Initialization

• Atmosphere-Land-Sea Ice Initialization via AMIP
  e.g., No Observational Data
• GFDL MOM3 Ocean Data Assimilation Interpolated to
  CCSM3 (POP)
• Number of Technical Issues
• Future Use CFSRR and Operational Analysis

7
Nino3.4 Multi-Model Forecast Assessment

• Initial Conditions
• CFS 5-Member Ensembles, December 30-January 3
  (ODA-January 1) 1982-1998
• CCSM3 6-Member Ensembles January 1 1982-1998
• Similarly with April and July Starts
• Multi-Model Better than Random Numbers?
• Monte-Carlo Red-Noise Simulations
• Multi-Model Better then Just More Ensemble
  Members?
• All Possible 5-Member Ensembles
• Are We Resolving the PDF Better?

8
January Cases
9
January Starts Nino3.4 Systematic Evolution
OISST
CFS
CCSM3
10
(No Transcript)
11
Is Multi-Model Better than Noise?
Noise Red Noise with Same Variance and Lag-1
Autocorrelation
12
Lead Time 5-Months
Model B
AB
Model A
ANoise
13
Is Multi-Model Better than Just a Larger Ensemble?
Consider All Possible 6-Member Ensembles with at
Least 2-Ensemble Members from Model A and Model B
14
5-Member Model B
5-Member AB
5-Member Model A
Lead Time 5-Months
15
(No Transcript)
16
Nino3.4 Estimated PDFs Lead-Time 5 Months
Observations
CFS
CCSM3
CFSCCM3
17
Nino3.4 Conditional PDFs Lead-Time 5 Months
Obs 0.5
CCSM Obs 0.5
CFS Obs 0.5
CFSCCSM Obs 0.5
18
Nino3.4 Conditional PDFs Lead-Time 5 Months
CFS Obs Obs CCSM Obs CFSCCSM Obs 19
Random Thoughts

• Multi-Model is Promising
• Generally (but not always) Better Forecast
  Quality
• More Than Just a Larger Ensemble
• Dont Need to Ask Which Model is Better?
• Livezey Question Does it Matter?
• In Practice Multi-Model is Ad-Hoc
• Orthogonal Skill
• Only Two Models
• Initialization
• Land, Atmosphere
• CFSRR
• Multi-Model Should not be Used as an Excuse to
  Avoid Model Improvement

20
(No Transcript)
21
(No Transcript)
22
(No Transcript)