US CLIVAR Symposium – 14 July 2008 (Photo credit: Arthur Greene) Model Predictions/Projections for 2018: What is Being Planned and What Could They Tell.

US CLIVAR Symposium – 14 July 2008

(Photo credit: Arthur Greene)

Model Predictions/Projections for 2018Model Predictions/Projections for 2018: : What is Being Planned and What is Being Planned and What Could They Tell Us? What Could They Tell Us?

[Summary of AGCI workshop][Summary of AGCI workshop]


Lisa Goddard – PPAIInternational Research Institute for Climate & Society

The Earth Institute of Columbia University


OUTLINE

Summary of AGCI Workshop

Efforts at specific research institutions

Other coordinated efforts within the EU

Experimental [Dynamical] Decadal Predictions


AGCI Workshop – Climate Prediction to 2030: Is it possible, what are the scientific issues, and how would those predictions be used?

Meeting Goals:1) Experimental design originally discussed in 2006 for AR5

that explicitly included short-term climate predictions to be performed for assessment by the international climate modeling community. The 2008 AGCI session carried this concept to the next level by tackling the formidable science issues involved with designing and running short term climate projections (now more commonly referred to as "decadal prediction”).

2) Address the important issues of the utility and applications of this information for decision support and impacts research.

25 June 2008 US CLIVAR Symposium – 14 July 2008

Global Climate Change Projections

Source: IPCC 4th Assessment Report, Working Group 1: The Physical Science Basis for Climate Changehttp://ipcc-wg1.ucar.edu/wg1/wg1-report.html

2030


Coordinated Decadal Prediction for AR5

Basic model runs:

1.1) 10 year integrations with initial dates towards the end of 1960, 1965, 1970, 1975, 1980, 1985, 1990, 1995 and 2000 and 2005 (see below).- Ensemble size of 3, optionally increased to O(10)- Ocean initial conditions should be in some way representative of the observed anomalies or full fields for the start date.- Land, sea-ice and atmosphere initial conditions left to the discretion of each group.

1.2) Extend integrations with initial dates near the end of 1960, 1980 and 2005 to 30 yrs.- Each start date to use a 3 member ensemble, optionally increased to O(10)- Ocean initial conditions represent the observed anomalies or full fields.



Additional model runs:

1.3) 10 year integrations each year in Argo era from near end of 2001, 2002, 2003, 2004, 2006 (2007, ..)

1.4) For models w/ 20th century runs, run additional ensemble members that extend to 2035. These runs form a “control” against which the value of initializing short-term climate and decadal forecasts can be measured.

1.5) For models which do not have 20th century and other standard runs, suggest making a 100 year control integration, and a 70 year run with a 1% per year increase in CO2. These integrations will allow an evaluation of model drift, climate sensitivity and ocean heat uptake, and give some idea of the natural modes of variability of the model.

2) Further studies which would be of interest• Comparison of initialization strategies • Repeat of the 1.1 2005 forecast with a high and/or low anthropogenic aerosol

scenario• Repeat of the 1.1 2005 forecast with an imposed “Pinatubo” eruption in 2010• Impact of Interactive Ozone chemistry• Air quality



Participating (/represented) Modeling Groups:BoM - Australia

CCCMA - Canada

COLA - USA

GFDL - USA

UKMO/Hadley Centre – UK

IMF-GEOMAR – Germany

MPI - Germany

NCAR - USA

RSMAS - USA

JMA & U. Tokyo - Japan



Main Scientific Issues:• Initialization

- Assimilation issues/products; initialize ocean models with anomalies vs full values vs forced by atmosphere; etc.

• Ensemble Generation Strategy

- perturb ocean and/or atmosphere; perturb model physics

• Ensemble Size

• External Forcing, particularly volcanoes

• Verification

- Modes of variability (e.g. ocean - AMO, PDO), regional surface climate, probabilistic v deterministic, ‘trend’ v ‘natural variability’


OUTLINE





Predictability of MOC• In COLA CGCM, MOC decadal variability is forced by

weather noise.– If this conclusion is general, the only paths towards improving

prediction of MOC and related surface climate variability are:• More accurate ocean initial conditions

• Improved models on climate time scales

– Reduced biases in climate statistics

– More realistic coupled feedbacks

(Source: Ed Schneider, COLA)

US CLIVAR Symposium – 14 July

2008

MOC in 20th Century Ensemble Integrations

PI CONTROL

(Courtesy: Joe Tribbia, NCAR)

NCAR:Short-term Simulations/Forecasts

• Use higher 0.5o resolution atmosphere and land.

• Run from 1980 – 2000 using observed forcing, and then from 2000 – 2030 using the A1B scenario.

• Have just interpolated 1980 atmosphere and land ICs from 20th Century run using ~2o resolution.

• Do need to initialize the ocean for these runs?

• Idea is to improve near-term projections over USA.

(Source: Peter Gent, NCAR)

GFDL: Decadal Predictability-Related PlansGFDL: Decadal Predictability-Related Plans Participate in Hindcast/Decadal Forecast Coordinated

Experiment protocol developed at the July ‘08 Aspen workshop(Stockdale, et. al) with a focus on evaluating hindcast skill. Use GFDL CM2.1 model (AR4 vintage) for short-term climate predictions.

Continue exploring questions of decadal climate predictability (e.g., mechanisms underlying decadal variability, model dependence,

regionality, initialization/coupled assimilation methods, skill levels desired/needed by stakeholders.) Ongoing collaborations

with NCAR, MIT on mechanisms of Atlantic variability.

CONSIDERED AN OPEN RESEARCH QUESTION AT GFDL…CONSIDERED AN OPEN RESEARCH QUESTION AT GFDL…

LOW LOW potential predictabilitypotential predictability HIGH HIGH

HOW MUCH “DECADAL” SKILL CANHOW MUCH “DECADAL” SKILL CAN ? BE GAINED FROM INITIALIZATIONBE GAINED FROM INITIALIZATION ?

• A near-term prediction up to 2030 with a high-resolution coupled AOGCM

– 60km Atmos + 20x30km Ocean– w/ updated cloud PDF scheme, PBL, etc– advanced aerosol/chemistry

• Estimate of uncertainty due to initial conditions– 10(?)-member ensemble– For impact applications

• water risk assessment system• impacts on marine ecosystems• etc.

• Test run w/ 20km AOGCM (in 2011)

Ensemble hindcast/forecast

Assimilation/Initialization

110km mesh model

60km mesh model

5-min topography

Japanese CLIMATE 2030 Project

(Source: Masahide Kimoto, U. Tokyo)

Decadal Prediction Research in Reading1. Analysis of sources of uncertainty in decadal

climate predictions using CMIP data (Hawkins & Sutton, submitted to BAMS, 2008)

2. Estimation of Singular Vectors for Decadal Predictions

3. Analysis of UK Met Office Decadal Prediction system

4. Decadal predictability studies:

• Sensitivity of predictability, for ocean and climate variables, to the initial ocean state

• Potential predictability of rapid THC changes (Hawkins & Sutton, 2008, GRL)

• Impact of higher resolution on simulation of decadal variability and predictability using UK-HiGEM model

(~1 degree atmosphere, 1/3 degree ocean)

(Source: Ed Hawkins, U. Reading)

© Crown copyright Met Office

Impact of ocean observations on 30-year forecast

•Forecast from March 2007

•Sub-sampled = with 1980s or 1960s obs

•5-year running means

•Shading = confidence of ensemble mean

•10 members DePreSys and sub-sampled, 4 members NoAssim

•Max overturning at 30N

2007 obs 1980 obs 1960 obs

(Source: Doug Smith, Hadley Centre)

Future activities at IFM-GEOMARFuture activities at IFM-GEOMAR

1. Investigate methods to extend simple initialisation schemes:

– Perfect model experiments to develop better understanding of the utility of SST restoring

– Investigate methods to account/include salinity variations

– Investigate statistical methods for using SST data

2. Understand the mechanisms for Atlantic multi-decadal variability using model hierarchy

(Source: Noel Keenlyside, IFM-GEOMAR)

Prospect for future climate prediction studies in Hamburg

CLISAP: Integrated Climate System Analysis and Prediction Large scale 5 year research project at U. Hamburg with participation of

MPI-M Plans to develop a climate monitoring and prediction system

CSC: Climate Service CentreBMBF funded 5-year project, possible tasks: dissemination of data Regionalization large scale or quasi-operational simulations for the German research

community

BMBF program for climate prediction (scheduled for 2009 – ? )

“COMBINE” proposal to European commission: Develop/test different initialization and bias correction methods, climate predictions and sensitivities to model improvements (stratosphere, …)

“Storm” project of German research consortium: Explore benefits of high spatial resolution for climate simulation and climate prediction (Atmosphere > T250, Ocean ~0.1°)

(Source: Marco Giorgetta, MPI)


OUTLINE






● Three systems: multi-model (ECMWF, GloSea, DePreSys, Météo-France, IfM-Kiel, CERFACS, INGV), stochastic physics (ECMWF) and perturbed parameters (DePreSys).

● Hindcasts in two streams:

o Stream 1: hindcast period 1991-2001, seasonal (May and November start dates), annual (November start date) and 2 decadal (1965 and 1994), 9 member ensembles.

o Stream 2: As in Stream 1 but over 1960-2005, with 4 start dates for seasonal hindcasts, at least 1 for annual and at least one 3-member decadal hindcast every 5 years.

o Additional simulations: DePreSys_PPE carries out a 10-year hindcast every year and a 30-year hindcast every 5 years + lots of sensitivity experiments from the other contributors.

Seasonal-decadal prediction in the EU ENSEMBLES project

(Source: James Murphy, Hadley Centre)


Further EU Projects (1)

COMBINE (Comprehensive Modelling of the Earth System for Better Climate Prediction and Projection)

• Use new climate model components developed since AR4• Input to AR5• Decadal and centennial timescales• Initialisation Work Package: assess different initialisation strategies

- assimilate full values and remove bias calculated from hindcasts

- anomaly initialisation

- empirical model error correction diagnosed from assimilation runs



Further EU Projects (2)

THOR (thermohaline overturning at risk?)• analyse mechanisms driving the THC• assess skill on decadal timescales (using ENSEMBLES hindcasts)• assess relative impact of greenhouse gases and initial conditions

using following hindcasts using:– A 1965 initial conditions, observed GHGs (including aerosols) from 1965

– B 1994 initial conditions, observed GHGs from 1994

– C 1965 initial conditions, observed GHGs from 1994

– D 1994 initial conditions, observed GHGs from 1965

• Idealised experiments to assess impact of observations of predictability of THC



OUTLINE







Few Pioneers1. Hadley Centre (Smith et al, 2007 - Science)

2. IFM-GEOMAR (Keenlyside et al, 2008 - Nature)

3. MPI/Hadley Centre (Pohlman et al, 2008 submitted)

Uncertainty (how to present)

Validation/verification

Source(s) of predictability

© Crown copyright 2007 Page 25

1. Hindcast experiments to assess skill Decadal Prediction System (DePreSys)(Smith et al, 2007)

• 10 year hindcasts started from 1st March, June, September and December in each year from 1982 to 2001 (80 start dates)

• 4 ensemble members, starting from consecutive days

• Do we achieve additional skill by starting the model from observed initial conditions ?

• Test by making a set of hindcasts (NoAssim) parallel to DePreSys

• NoAssim includes the same external forcings as DePreSys but omits the assimilation of anomalous observed initial conditions.



1. Sub-surface ocean analysis(Smith et al, 2007)

Optimal interpolation using covariances computed directly from HadCM3 transient integration

Analyses computed off-line, stored in ancillary files

Relax model to analyses (6 hour timescale)



2. Initialisation and decadal hindcasts using ECHAM5/MPI_OM coupled model(Keenlyside et al, 2008)

• 3x 20th century transient simulations (anthro, solar, volcanic forcing)

• 3 x 20th century simulations assimilating SST anomalies (same forcing)

• Decadal hindcasts started every 5 years from 1955-2005. 3 ensemble members; (anthro forcing, repeated solar cycle, no volcanoes)

2. Decadal hindcast/forecast strategyDecadal hindcast/forecast strategy (Keenlyside et al, 2008)

• Model: ECHAM5/MPIOM Climate model (IPCC AR4 version)

• Initial conditions: Coupled model SST anomalies restored to observations

• Boundary conditions: 20th century/A1B radiative forcing Nudging constant varies with latitude

0.25 days-1

Linear transition

0 days-1 (fully coupled)

Linear transition

0 days-1 (fully coupled)

30˚N

30˚S

60˚N

90˚N

60˚S

90˚S

(Source: Noel Keenlyside, IFM-GEOMAR)

3. Pohlmann et al., submitted to J. Climate.

Improving Decadal Climate Predictability through the Initialization of a Coupled Model with the GECCO Oceanic Synthesis

Holger Pohlmann(1,2), Johann Jungclaus(1), A. Köhl(3), D. Stammer(3), J. Marotzke(1)

(1) Max Planck Institute for Meteorology, Hamburg, Germany(2) Now at Met Office Hadley Centre, Exeter, UK(3) Institute of Oceanography, ZMAW, Univ. Hamburg, Germany

ECHAM5/MPI-OM climate model (~ MPI-M IPCC AR4 model) GECCO ocean synthesis 1952-2001 10 year hindcasts and forecasts


Pohlmann et al.: Experimental design

Experiments Initialization Forcing Amount Period

Control In 1900, 1910 and 1920 from an IPCC AR4 20th century simulation

GHG + aerosol 3 1900 – 2011

Assimilation In 1952 from Control (initialized in 1900)

GHG + aerosol and T + S from ECCO

1 1952 – 2001

Hindcast At the end of every year from Assimilation

GHG + aerosol 49 / 40 10 years duration

Forecast At the end of Assimilation

GHG + aerosol 7 2002 – 2011

Control

Assimilation

Hindcast

Forecast

123

40

19521900 2002 2011



Predictions? / Projections?

DOI: 10.1126/science.1139540, 796 (2007); 317 Science


Smith et al (2007)

Figure 2 Figure 4

CONs• Global average• Little to no evidence of [predictable] LF climate variability at long lead PROs:• Improved projections relative to original system• View of change in uncertainty with time scale1) Uncertainty in decadal-average2) Uncertainty through a decade due tointerannual variability3) Realization of natural variability throughdecade


Smith et al (2007)

Regionality?• Ts projections improved

over many regions

Climate variability?• Ts projection worse over

N.Atlantic• Much improvement in

regional T is associated with improvement in regional H, which bears striking resemblance to regions where T is dominated by externally-forced signal.

Figure 5

Ratio of Externally-forced to Total Variance

(Courtesy: M. Ting et al, J.Climate, submitted)


Keenlyside et al (2008)

“… the initialized prediction indicates a slight cooling relative to 1994-2004 levels, while the anthropogenic-forcing-only simulation suggests a near 0.3 K rise.”

PRO:• Focus on mode(s) of

natural climate variability

CONs:• Statements/conclusions

seem at odds with evidence (ie. fcst evolution)

• Uncertainty given by spread of 3 ensemble members

• Demonstration of natural climate variability (AMOC) not obvious

Figure 4


Keenlyside et al (2008)

Regionality?• New method seems to have

greater errors in most places, especially the N. Atlantic

• What does improved performance in eastern Pacific suggest for ENSO variability?

Climate variability?• Lack of verifying observations,

so don’t really know truth• But – according to available

truth, hindcast has no skill

Supp. Figure 2c

Figure 3a

Maximum MOC Strength

Difference in RMSE (deg. K)

Pohlmann et al.: North Atlantic annual mean SST (°C)

HadISST

Assimilation

h/f castyear 1-5

Control

h/f castyear 5-10

h/f castyear 1

h/f castyear 1-10

Time series Anom. Cor. | RMSE

H-cast/HadISST

H-cast/AssimCtrl/Assim

95% CL

Persistence



Predictions? / Projections?

Projections?• Yes – seems possible to provide better

estimates of near-term anthropogenic climate change (at least T), due largely to correcting biases in ICs

Predictions?• Not yet – Some evidence of potential

predictability (perfect model/ICs) and slight evidence of real experimental predictability, but very little available at regional scales (and nothing yet demonstrated for precipitation).


SUMMARY

• Considerable national and international efforts

• Numerous scientific questions remain, particularly on initialization, mechanisms and model validation

• Decadal prediction/projection is promising but in VERY EARLY stages. The climate community must first assess what we have (and don’t have) before invoking its direct use by applications and decision makers.

US CLIVAR Symposium – 14 July 2008 (Photo credit: Arthur Greene) Model Predictions/Projections for 2018: What is Being Planned and What Could They Tell.

Documents

climate society

climate sensitivity

year integrations

short term climate

shortterm climate predictions

climate change http

coordinated decadal

short term climate projections