CMIP5 decadal experiments at CERFACS: Initialisation and ... · CMIP5 decadal experiments at CERFACS: Initialisation and preliminary results Aspen CMIP5 workshop,Colorado, June 2011

CMIP5 decadal experiments at CERFACS: Initialisation and preliminary results

Aspen CMIP5 workshop,Colorado, June 2011

Christophe Cassou, Emilia Sanchez-Gomez Elodie Fernandez Laurent Terray

Prediction or forecast?

Predic'on = prédic'on Forecast = prévision 

Meteorologist, climatologist Miss Sun, astrologist

Outline

OUTLINE 

1.  Descrip'on of the CERFACS model and CERFACS  contribu'on to CMIP5 database  

2. Impact of the ini'alisa'on technique upon model ini'al shock and driL 

3. Very preliminary results about AMO predictability 

The Cerfacs-CNRM contribution to the CMIP5 database

All simulations are completed and model outputs have been posted last week end on ESG nodes (except sea-ice data)

Core experiments

« Tier 1 »: 1.  Increase ensemble sizes from 3 to 10 members

2. Forecast with 2010 Pinatubo eruptions (10 members)

3. Work on initialization techniques

« Bonus »: 10 more dates starting one year before the requested CMIP5 experiments

The decadal forecast initialization

The coupled model = CNRM-CM5

Land surface, ISBA

SURFEX Interface

Atmosphere ARPEGE-Climat v5.2

T127 (1.4°), 31 levels

Ocean NEMO v3.2 1°

42 levels

River Routing TRIP

Sea Ice GELATO v5

OASIS v3 (24h coupling)

24h

The COMBINE ECMWF Ocean Reanalysis

Full Initialization: Ocean Only In a coupled mode from 1958

To 2008

3D Nudging as a function of depth and space

β = f(depth, space) Reanalysis

Current

No 3D nudging within the

Equatorial band (1oN-1oS) and near the coast

(300km) (1/ β) =0

In the mixed layer (1/ β) =0

Deep Ocean β = 360 days

Below thermocline β = 10 days

Sea surface restoring

Heat flux:

Heat flux at the surface

feedback term. SSTobs= observations

Feedback coefficient = -40W/m2/K

Fresh water flux:

Fresh water budget at the surface

Feedback term. SSSobs= observations

Feedback parameter = -167 mm/day

At the surface

Outline

Sensi'vity of the ini'al shock and driL to the 3D‐nudging domain of applica'on 

Two types of CMIP5 experiments

GLOB  No 3D nudging within the 1°S–1°N band 

•  Nudging to the ECMWF ocean reanalysis NEMOVAR (ORCA1°) 1958 -2008 •  Several tests have been performed to set the optimal parameters for dQ/Dt and β

EXTROP No 3D nudging  Within the 15°S–15°N band 

Ocean Heat Content

Mean 1960-2005

ΔH EXTROP GLOB

NEMOVAR

Mean 1960-2005

Along the equatorial thermocline

(GLOB-EXTROP) (HIST-NEMOVAR) Mean [1960-2005] Temperature difference averaged over the 2oN-2oS band

Impact of initialization on mean ocean heat transport

GLOB: perturbation of the Northward heat transport around 10o Of latitude

GLOB: perturbation of the Northward heat transport around 10o Of latitude in the Atlantic

EXTROP: close to HIST up to 20oN close to NEMOVAR northward

Impact of initialization on prediction

Global Mean SST

GLOB / EXTROP

Strong warm initial shock mainly due to very rapid austral ocean warming

Annual mean SST biases (HIST-ERSST (1960-2005)

6 months

2-3 years

> 10 years

Impact of the initialization on predictions

Global Mean SST (from 35oS to 90oN)

1-yr warming much drift less pronounced

GLOB warm shock stronger than EXTROP

GLOB / EXTROP

Thermocline depth prediction

3-member ensemble Mean D20 isotherm depth over the 2oN-2oS band 1 spaghetti = 1 forecast date

GLOB/EXTROP

HIST

ENSO variability

NINO34 SST

Ensemble mean (dates+members) Nino34 SST index

Yr1 Yr2 Yr3 Yr4

Perturbation of the tropical climate up to 4yr (systematic NINO the 1st and 3rd year in GLOB)

Teleconnection

Difference between GLOB and EXTROP ensemble mean (dates+members) Z500 and precipitation for the 1yr-forecast winter

Strong difference in rainfall drift between GLOB and EXTROP, presence of ENSO Teleconnection in Z500 drift

Color=precip / contour=Z500 (significance hatching)

Does the drift affect skills?

2 questions:

1. Do the strength and physical structures of the initial shock and subsequent drift affect the skill of the forecast at decadal time-scale (yr 2-5?)

2. What is the relevance of linear debiasing methods in presence of drifts that DO project on natural modes of variability that are clearly nonlinear (convection and teleconnection associated with ENSO)?

Outline

Preliminary results on model  skill and AMO predictability 

26 June ‐1 July 2011  CMIP5 Decadal Predic8on Workshop, Aspen 

SST Correlation skill (with trend)

Decadal Forecast

Historical 20th simulations

26 June ‐1 July 2011  CMIP5 Decadal Predic8on Workshop, Aspen 

SST Correlation skill (without trend)

Decadal Forecast

Historical 20th simulations

26 June ‐1 July 2011  CMIP5 Decadal Predic8on Workshop, Aspen 

Global (60°S-60°N) SST Predictability

OBS=ERSST3 DEC HIST

26 June ‐1 July 2011  CMIP5 Decadal Predic8on Workshop, Aspen 

Detrended global (60°S-60°N) SST

OBS=ERSST3 DEC HIST

26 June ‐1 July 2011  CMIP5 Decadal Predic8on Workshop, Aspen 

AMO Predictability

OBS=ERSST3 DEC HIST

Conclusions

•  We need to be extremely cautious in the way ocean 3D fields are initialized

•  Model drift correction is a central question (interaction with the forcing and the natural variability modes)

We need to be very cautious with the estimation of the decadal forecast skill (choice of the statistics etc.) and with the way users are going to deal with the decadal forecasts

The Cerfacs-CNRM data are available! Feel free to use them!

IPO/PDO ERSST3 1890-2010

IPO/PDO Predictability