Jason C. Furtado Advisor: E. Di Lorenzo School of Earth & Atmospheric Sciences

18
in Tropical SST Reconstructions Derived From Tropical Precipitation Records? Jason C. Furtado Advisor: E. Di Lorenzo School of Earth & Atmospheric Sciences Georgia Institute of Technology EAS Graduate Student Symposium 2 November 2007

description

What Uncertainties Exist in Tropical SST Reconstructions Derived From Tropical Precipitation Records?. Jason C. Furtado Advisor: E. Di Lorenzo School of Earth & Atmospheric Sciences Georgia Institute of Technology EAS Graduate Student Symposium 2 November 2007. Previous Work. - PowerPoint PPT Presentation

Transcript of Jason C. Furtado Advisor: E. Di Lorenzo School of Earth & Atmospheric Sciences

Page 1: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

What Uncertainties Exist in Tropical SST

Reconstructions Derived From Tropical

Precipitation Records?

Jason C. FurtadoAdvisor: E. Di Lorenzo

School of Earth & Atmospheric SciencesGeorgia Institute of TechnologyEAS Graduate Student Symposium

2 November 2007

Page 2: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Previous Work

Single Proxy Record

Multiple Proxy Records

Cobb et al. 2003

Evans et al. 2002

Palmyra Coral

ReconstructedLeading SST Mode

Page 3: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Aims of the Study

•Compare two popular climate field reconstruction methods.

•Examine the uncertainties associated with each method.

•Evaluate the performance of a paleo-precipitation proxy network.

Use tropical precipitation records to reconstruct tropical SSTs.

Page 4: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Data & Methods

•Precipitation

•CMAP (Xie and Arkin 1997)

•Output from International Center for Theoretical Physics (ICTP) AGCM (Molteni 2003)

•ERA-40

•SSTs - NOAA ER SSTs (Smith and Reynolds 2003)

•Annual-mean anomalies used

•Spatially smoothed and detrended

•Reconstructions are done from 1979 - 2000.

Page 5: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Reconstruction Methods

EOF METHOD

Premise: SSTs and precipitation are dynamically(and statistically) linked in the tropics.

Regression Coefficient

1) EOFs are time invariant2) 3)

Page 6: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Reconstruction Methods

MULTIPLE REGRESSION

Premise: SSTs and precipitation are dynamically(and statistically) linked in the tropics.

Least-squares fitting (obtain optimal linear estimator E).

Only retain first few covariability modes.

Cross-validation method to test for robustness.

EOF METHOD

Regression Coefficient

1) EOFs are time invariant2) 3)

Page 7: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

How Good Are The Reconstructions?

•RMS Error:

•Skill:

•Spatial Correlation:

Averaged over all22 reconstructions

Page 8: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Evaluation - EOF MethodRMS ERROR SKILL

CMAP

ICTP MODEL OUTPUT

ERA-40

Page 9: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Evaluation -Multiple Regression

RMS ERROR SKILL

CMAP

ICTP MODEL OUTPUT

ERA-40

Page 10: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Spatial CorrelationsCorrelation

EOF MethodMulti-Regression

Mean r = 0.73

ICTP

EOF MethodMulti-Regression

Mean r = 0.75

CMAP

ERA-40

EOF MethodMulti-Regression

Mean r = 0.76

Mean r = 0.45 Mean r = 0.52

Mean r = 0.45

Page 11: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Why is Multiple Regression Better?

1st Left Singular Vector

2nd Left Singular Vector

1st Right Singular Vector

2nd Right Singular Vector

Dynamical

Response

to ENSO

Dipole (Tripole) in Precipitation

SST

SST

Precip.

Precip.

Page 12: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Proxy Network

Tree RingsCoralsMarine

SedimentsLake SedimentsSpeleothemIce Cores

Use multiple regression method with

CMAP data from only these points for SST

reconstructions

Page 13: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Evaluation - Proxy Network

~20% decrease in skill in the tropical Pacific and ~50% in the Indian OceanDesigning an Ideal Paleo-Precipitation Network

Use the adjoint (ET) for sensitivity study.

Page 14: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

But What About Stationarity?1950 - 1978 1950-2000

SST

RSV-2(Precip)

LSV-2(SST)

Out-of-phase relationship b/t Indian and E Pacific (1950-1978)In-phase relationship b/t Indian and E Pacific (1950-2000)

Page 15: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Conclusions• Multiproxy tropical precipitation records effectively reconstruct tropical SSTs.

• The multiple regression method outperforms the EOF method, with a 20-30% improvement in skill in the tropical Pacific and much more in the Indian Ocean.

• The paleo-precipitation proxy network recovers almost 50% of the observed variance in tropical SSTs and 80% of the skill vs. the full tropical precipitation field.

• Is there a reconstruction technique that can account for the nonstationarity in the ENSO statistics / covariability modes?

Page 16: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Thank You!

Questions?

Page 17: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Error Propagation Analysis•Add an error term to the precipitation in

the linearized relationship:

•Define: ; sn = signal-to-noise ratio

Page 18: Jason C. Furtado Advisor:   E. Di Lorenzo  School of Earth & Atmospheric Sciences

Error Propagation Analysissn = 10 sn = 2

Nonzero np

everywhere

np = 0 in

Eastern Hemisphe

renp = 0 in

Western Hemisphe

re