11 th international meeting on statistical climatology, 12 july 2010, edinburgh

28
Federal Department of Home Affairs FDHA Federal Office of Meteorology and Climatology MeteoSwiss Climate Change Projections for Switzerland: A Bayesian Multi-Model Combination using ENSEMBLES Regional Climate Models 11 th International Meeting on Statistical Climatology, 12 July 2010, Edinburgh Andreas Fischer, Andreas Weigel, Mark Liniger, Christoph Buser, Christof Appenzeller

Upload: daryl

Post on 12-Jan-2016

31 views

Category:

Documents


0 download

DESCRIPTION

Climate Change Projections for Switzerland: A Bayesian Multi-Model Combination using ENSEMBLES Regional Climate Models. Andreas Fischer, Andreas Weigel, Mark Liniger, Christoph Buser, Christof Appenzeller. 11 th International Meeting on Statistical Climatology, 12 July 2010, Edinburgh. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss

Climate Change Projections for Switzerland: A Bayesian Multi-Model Combination using ENSEMBLES Regional Climate Models

11th International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

Andreas Fischer, Andreas Weigel, Mark Liniger, Christoph Buser, Christof Appenzeller

Page 2: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

2 Climate Services, IMSC Edinburgh | 12 July [email protected]

ENSEMBLES R2TB

AOGCMs RCMs@25km

HadCM3

HIRHAM (Met.No)

REMO (MPI)

SRES A1B

ECHAM5

Low sens.

High sens.

Standard sens.

ARPEGE

CGCM3

BCMRCA (SMHI)

HadRM3 (Met Office)RCA (SMHI)

HadRM3 (Met Office)RCA3 (C4I)

CLM (ETHZ)PROMES (UCLM)

HIRHAM (DMI)RACMO (KNMI)

RCA (SMHI)

ALADIN v1 (CNRM)

HIRHAM (DMI)

REGCM3(ICTP)

CRCM (OURANOS)

RRCM (VMGO)

IPSL CLM (GKSS)

HadRM3 (Met Office)

ALADIN v2 (CNRM)

HIRHAM (Met.No)

HIRHAM (DMI) Final Report (2009)

RCMs@25kmAOGCMs

1950 - 2050

8 AOGCMs / 21 Model Chains 6 AOGCMs / 15 Model Chains

2050 - 2100

HadCM3

REMO (MPI)

ECHAM5

Low sens.

High sens.

Standard sens.

ARPEGE

BCMRCA (SMHI)

HadRM3 (Met Office)

RCA (SMHI)

HadRM3 (Met Office)

RCA3 (C4I)

CLM (ETHZ)

HIRHAM (DMI)

RACMO (KNMI)

RCA (SMHI)

HIRHAM (DMI)

REGCM3(ICTP)

HadRM3 (Met Office)

ALADIN v2 (CNRM)

HIRHAM (DMI)

Page 3: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

3 Climate Services, IMSC Edinburgh | 12 July [email protected]

Derivation of Probablistic Scenarios

Modelled Climate Change Signals

PDF

?Bayes Algorithm(Buser et al., 2009)

Assumptions transparent

Page 4: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

4 Climate Services, IMSC Edinburgh | 12 July [email protected]

Bayesian Multi-Model Combination (Buser et al., 2009)

Obs NOW

Models NOW

Models FUTURE

„Obs“ FUTURE

Seasonally averaged 30yr time periods

Page 5: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

5 Climate Services, IMSC Edinburgh | 12 July [email protected]

Bayesian Multi-Model Combination (Buser et al., 2009)

Obs NOW

Models NOW

Models FUTURE

„Obs“ FUTURE

Mean Climate Shift Model Projection Errors

NOW FUTURE

Page 6: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

6 Climate Services, IMSC Edinburgh | 12 July [email protected]

Bayesian Multi-Model Combination (Buser et al., 2009)

Obs NOW

Models NOW

Models FUTURE

„Obs“ FUTURE

Mean Climate Shift Model Projection Errors

NOW FUTURE

• μ and βi non identifiable

• Assumption has to be taken about projection error Δβi ~ N(0; σ2

β)

Page 7: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

7 Climate Services, IMSC Edinburgh | 12 July [email protected]

Bayesian Multi-Model Combination (Buser et al., 2009)

Prior p(x)

Posterior p(x|data)

Obs NOW

Models NOW

Models FUTURE

„Obs“ FUTURE

Likelihood p(data|x)

P(x|data) p(x) * p(data|x)

Gibbs Sampler

Page 8: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

8 Climate Services, IMSC Edinburgh | 12 July [email protected]

Sensitivity Experiments: Effect of Likelihood

Climate ChangeSignal

Likelihood affects variance and location of posterior distribution

All prior distributions set to be uninformative

Page 9: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

9 Climate Services, IMSC Edinburgh | 12 July [email protected]

Bayesian Multi-Model Combination (Buser et al., 2009)

Prior p(x)

Posterior p(x|data)

Obs NOW

Models NOW

Models FUTURE

„Obs“ FUTURE

Likelihood p(data|x)

P(x|data) p(x) * p(data|x)

Gibbs Sampler

Page 10: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

10 Climate Services, IMSC Edinburgh | 12 July [email protected]

Sensitivity Experiments: Effect of Prior

Projection Uncertainty

Mea

n C

lim

ate

Sh

ift

The uncertainty in Δμ is increased with a wider prior-setting for Δβi

Page 11: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

11 Climate Services, IMSC Edinburgh | 12 July [email protected]

CC Signal

Sensitivity Experiments: Effect of Prior

CC Signal

Outlier

Informative Prior Δβi

Non-Informative Prior Δβi

Central tendency of posterior distributions also affected by prior

CC Signal

Page 12: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

12 Climate Services, IMSC Edinburgh | 12 July [email protected]

Application of Algorithm using ENSEMBLES data

1. Estimation of Projection Uncertainty (σ2β)

2. Role of Internal Variability

3. Independent Model Data

Different considerations:

Page 13: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

13 Climate Services, IMSC Edinburgh | 12 July [email protected]

1. Estimating Projection Uncertainty

Assumption: Projection Uncertainty is fully sampled by range of available model simulations

ECHAM

HadCM3Q0

(2) RCM Uncertainty

8 different GCMs

(1) GCM Uncertainty

Smoothing of timeseries by polynomial fit (Hawkins & Sutton, 2009)

Page 14: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

14 Climate Services, IMSC Edinburgh | 12 July [email protected]

2. Internal Variability

(1) As a pre-processing step we remove internal variability from time-series

(2) Calculate posterior distributions with Bayes Algorithm

(3) Add internal variability to posterior distribution of μ

Page 15: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

15 Climate Services, IMSC Edinburgh | 12 July [email protected]

30-yr Running Mean

4th order polynomial fit

(Hawkins and Sutton, 2009)

Summer Temperature over CHNE (Model: ETHZ – HadCM3Q0)

2. Internal Variability

Page 16: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

16 Climate Services, IMSC Edinburgh | 12 July [email protected]

30-yr Running Mean

4th order polynomial fit

(Hawkins and Sutton, 2009)

Summer Temperature over CHNE (Model: ETHZ – HadCM3Q0)

2. Internal Variability

Page 17: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

17 Climate Services, IMSC Edinburgh | 12 July [email protected]

ECHAM

HadCM3Q0

3. Independent Model Data

ECHAM HadQ0 HadQ3 HQ16 ARP. BCM

ECHAM

HadQ0

HadQ3

HQ16

ARP.

BCM

DJF Temperature 1980-2009 (AL)

Average all RCMs driven by the same GCM

Page 18: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

18 Climate Services, IMSC Edinburgh | 12 July [email protected]

Probabilistic Climate Change Scenarios

Orography of Switzerland

Reference Period 1980 - 2009

Northeastern Switzerland

Page 19: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

19 Climate Services, IMSC Edinburgh | 12 July [email protected]

Swiss Climate Scenario (A1B)

GCM groups

203520602084

GCM-RCMchains

Temperature (K)

Internal Variability

Page 20: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

20 Climate Services, IMSC Edinburgh | 12 July [email protected]

Swiss Climate Scenario (A1B)

Relative Precipitation

GCM groups

GCM-RCMchains

203520602084

Internal Variability

Page 21: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

21 Climate Services, IMSC Edinburgh | 12 July [email protected]

Conclusions

The Bayes Algorithm by Buser et al. (2009) is a transparent tool for generating probabilistic climate change scenarios.

The uncertainty range in the climate change signal is highly dependent on the prior-settings of the projection uncertainty.

The Buser Algorithm does not account for internal variability. To circumvent this problem a pragmatic solution has been proposed.

The probabilistic climate change scenarios for Northeastern Switzerland show a continous increase in temperature over the 21st century. For precipitation only in summer a signal in the second half of the century is detectable.

Page 22: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

22 Climate Services, IMSC Edinburgh | 12 July [email protected]

Swiss Climate Scenarios: Precipitation

DJF Precipitation Change [%]

2035 2060 2084

JJA Precipitation Change [%]

2035 2060 2084

Page 23: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

23 Climate Services, IMSC Edinburgh | 12 July [email protected]

Effect correlated models

Delta Mu JJA T2 CHN

KNMI-ECHAM / ETHZ-HadQ0 / SMHI-HadQ3 / C4I-HadQ16 / CNRM-ARPEGE / SMHI-BCM / OURANOS

ECHAM av. / HadQ0 av. / HadQ3 av. / HadQ16 av. / CNRM-ARPEGE / BCM av. / OURANOS

Average of 1 GCM group / Rest as standard

Page 24: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

24 Climate Services, IMSC Edinburgh | 12 July [email protected]

Climate Scenarios

Global Mean Temperature wrt 1980-2009

B1

A1BA2

2035 2060 2084

comm

?

[K]

Page 25: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

25 Climate Services, IMSC Edinburgh | 12 July [email protected]

Pattern Scaling with CMIP A1B

Bayes Estimate 2035

Scaled from 2060

Scaled from 2084Temperature Relative Precipitation

Page 26: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

26 Climate Services, IMSC Edinburgh | 12 July [email protected]

Swiss Climate Scenarios

A2B1

A1B

Page 27: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

27 Climate Services, IMSC Edinburgh | 12 July [email protected]

Aim: Update of Probabilistic Scenarios

OcCC (2007)203020502070

Relative PrecipitationTemperature

Probabilistic Scenarios for Northern and Southern Switzerland based on PRUDENCE RCM simulations

Page 28: 11 th  International Meeting on Statistical Climatology, 12 July 2010, Edinburgh

28 Climate Services, IMSC Edinburgh | 12 July [email protected]

Model validation

CHS

CHWCHNE

Orography

Temperature (°C)

Temperature (°C)

Precipiation (mm/mt)

EOBS v3

EOBS v3

EOBS v3

(1980 – 2009)