El Niño Forecasting Stephen E. Zebiak International Research Institute for climate prediction The basis for predictability Early predictions New questions.

El Niño Forecasting

Stephen E. Zebiak

International Research Institute for climate prediction

• The basis for predictability

• Early predictions

• New questions raised in the 1990s

• Beyond El Niño proper – seasonal climate prediction

• 2002 El Niño

• Summary and questions for future research

Southern Oscillation

ENSO wind and SST patterns

Strong trade winds

Westward currents, upwelling

Cold east, warm west

Convection, rising motion in west

Weak trade winds

Eastward currents, suppressed upwelling

Warm west and east

Enhanced convection, eastward displacement

Surface layer

Deep ocean; u=v=w=0

Active layer

50 m

150 m

Simplified Ocean and Atmosphere Models

Simplified form of equations for conservation of mass, momentum, energy

SSTA

Tropopause

Model ENSO

wind stress

- h

+ h

- h

Ocean wave dynamics and ENSO

Early Forecasting Methods

Build-up, then relaxation of trade winds

(Wyrtki; diagnostic)

Ocean dynamic response to observed wind patterns(Inoue & O’Brien; prognostic, but not coupled)

Identification of precursor patterns in sea level pressure, SST, winds from historical observations

(Graham, Barnett, …; statistical)

Simplified dynamical coupled models

(Cane, Zebiak, … ; prognostic)

Winds,Heat

fluxes

Ocean simulation

Ocean obs.

Ocean analysis

t t + t

SST forcing

Atmos. simulation

Atmos. obs.

Atmos. analysis

t t + t

Data assimilation

Initial Conditions, t=t0

Atmosphere model Ocean model

FORECAST

Forecast Initialization Procedures

Winds,Heat

fluxes

Ocean simulation

Ocean analysis

t t + t

SST forcing

Atmos. simulation

Atmos. analysis

t t + t

Data assimilation

Initial Conditions, t=t0

Atmosphere model Ocean model

FORECAST

Forecast Initialization Procedures

Correlation Skill for NINO3 forecasts

A real-time forecast

Retrospective Assessment of ENSO Prediction Skill over the period 1970-1992

Statistical prediction models Mixed statistical-dynamical models

Dynamical coupled models

Winds,Heat

fluxes

Ocean simulation

Ocean analysis

t t + t

SST forcing

Atmos. simulation

Atmos. analysis

t t + t

Data assimilation

Initial Conditions, t=t0

Atmosphere model Ocean model

FORECAST

Forecast Initialization Procedures

NINO3 forecasts initialized each month

Revised NINO3 forecasts initialized each month

Lamont Model; 1972-1992 validation period

Winds,Heat

fluxes

Ocean simulation

Ocean obs.

Ocean analysis

t t + t

SST forcing

Atmos. simulation

Atmos. analysis

t t + t

Data assimilation

Initial Conditions, t=t0

Atmosphere model Ocean model

FORECAST

Forecast Initialization Procedures

NCEP PREDICTION

Impact of ocean initialization on NCEP coupled model forecast system skill

Zonal Wind and zonal wind anomalies during 1996-97

U.S. Precipitation in four El Niño winters

PERSISTED

GLOBAL

SST

FORECAST SST

TROP. PACIFIC (NCEP dynamical)

TROP. ATL, INDIAN(statistical)

EXTRATROPICAL (damped persistence)

GLOBAL ATMOSPHERIC

MODELS2°- 3° lat-lon

18 -19 vertical layers

ECHAM3.6(MPI)

ECHAM4.5(MPI)

NCEP (MRF9)

CCM3.2(NCAR)

NSIPP(NASA)

COLA2.x

AGCM INITIAL CONDITIONS

UPDATED ENSEMBLES (10+)WITH OBSERVED SST

ForecastSST

Ensembles3/6 Mo. lead

PersistedSST

Ensembles3 Mo. lead

REGIONALMODELS

HISTORICAL DATA

Extended simulations

Observations

IRI DYNAMICAL CLIMATE FORECAST SYSTEM

POSTPROCESSING

-Statistics

-MultimodelEnsembling

-graphics

Probabilistic Skill Score – IRI Seasonal Temperature Forecasts

Probabilistic Skill Score – IRI Seasonal Precipitation Forecasts

IRI Global Precipitation Forecasts – Ranked Prob. Skill Score

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2 JMA

NSIPP

ECMWF

LDEO

BMRC

COLA-fc

NCEP

SIO

CSIRO

SNU

UBC-hc

COLA-ac

CCA-ncep

MK-ncep

CCA-ubc

CLIPER

CCA-sawb

N-Net-ubc

CA-ncep

CDC-lim

Obs. El Nino yrs

Forecasts for Jun-Jul-Aug 2002 NINO3.4 SST anomalies

Dynamical ocean/atm or hybrid Statistical models Obs. El Nino yrs

IRI ENSO Quick Look

IRI ENSO Quick Look

Summary

• There is a physical basis for (limited) predictability of El Niño– relies on “slow” ocean dynamics and strong coupling between ocean and atmosphere in tropical Pacific

• There are also clear limits to predictability– model errors– effects of lack of observations (e.g., salinity)– unpredictable “noise”

• Dynamical and statistical models comparable in performance– dynamical methods have more potential for improvement

• Ensemble-based predictions offer best hope for characterizing real uncertainties

• Current global seasonal climate forecast performance depends strongly on the state of ENSO

– necessarily probabilistic– also depend on other phenomena requiring further study

Summary

• early 2002 situation: critical preconditions for El Niño were in place, but development was considered uncertain

– El Nino “watch” issued

– consensus of forecasts was “correct”

• For future progress, we must:– understand the role of “noise” and how to address it in forecasting– reduce systematic errors that limit forecast skill– improve initialization methods for predictions– further develop ensemble methods for probabilistic forecasts