PROGNOSTIC DISCUSSION FOR 6 TO 10 AND 8 TO 14 DAY OUTLOOKS NWS CLIMATE PREDICTION CENTER CAMP SPRINGS, MD 300 PM EDT FRI AUGUST 21 2009

THE OPERATIONAL 00Z AND 06Z GFS MODEL SOLUTIONS BEGIN TO BREAK DOWN THE RIDGE OVER THE PACIFIC NORTHWEST WHILE THE HIGH RESOLUTION 00Z ECMWF MAINTAINS A STRONG RIDGE THERE. TELECONNECTIONS FROM THE UPSTREAM TROUGH OVER THE GULF OF ALASKA AND WESTERN ALEUTIAN RIDGE BOTH SUPPORT RIDGING OVER WESTERN NORTH AMERICA WHICH AGREES MORE WITH THE 00Z ECMWF SOLUTION. THIS, IN COMBINATION WITH VERY HIGH 500-HPA ANOMALY CORRELATIONS EXHIBITED DURING THE PAST 60 DAYS, RESULTED IN THE ECMWF-BASED SOLUTIONS BEING FAVORED IN TODAYS OFFICIAL 500-HPA

HEIGHT BLEND CHART. TODAY'S OFFICIAL 500-HPA BLEND CONSISTS OF 10% OF TODAY'S OPERATIONAL 6Z GFS CENTERED ON DAY 8...20% OF TODAY'S GFS SUPERENSEMBLE MEAN CENTERED ON DAY 8...20% OF TODAY'S OPERATIONAL 0Z ECMWF CENTERED ON DAY 8...40% OF TODAY'S 0Z ECMWF ENSEMBLE MEAN CENTERED ON DAY 8...AND 10% OF

TODAY'S 0Z CMC ENSEMBLE MEAN CENTERED ON DAY 8.

Generic Levels of EM Uncertainty – Lessons from geophysical

modeling and current studies of climate impacts

Nicholas Bond1

Kerim Aydin2, Anne Hollowed2, James Overland3 and Muyin Wang1

1 University of Washington/JISAO

2 NOAA/AFSC3NOAA/PMEL

Techniques for Incorporating Model Ensembles

• Simple means• Means w/ individual bias corrections• Means w/ collective bias corrections• Regularization via EOFs• Bayesian techniques

Forecast error (24-h Temp.)

Rixen et al. (J. Mar. Sys., 2009)

Time series of uncertainty

IPCC I.D. Country Atmosphere Resolution

Ocean Resolution

# of Control runs

# of 20c3m runs

# of A1B runs

1 BCCR-BCM2.0 Norway T63L31 (0.5 -1.5°) x 1.5°L35 2 1 1

2 CCSM3 USA T85L26 (0.3 -1.0°) x 1.0°L40 1 1

3 CGCM3.1(T47) Canada T47 L3 1 1.9° x 1.9°L29 2 5 5

4 CGCM3.1 (T 63 ) Canada T63L31 1.4 ° x 0.9 °L29 1 1 1

5 CNRM-CM3 France T42L45 182x152L31 3 1* 1

6 CSIRO -Mk3.0 Australia T63L18 1.875° x 0.925° L31 3 3 1

7 ECHAM5/ MPI -OM Germany T63L31 1.5°x1.5°L40

8 FGOALS -g1 .0 (IAP) China T42L26 1°x1°xL30 9 3 3

9 GFDL-CM2.0 USA 2.5°x2. 0° L24 1°x1°L50 5 3 1

10 GFDL-CM2.1 USA 2.5°x2. 0° L24 1°x1°L50 5 5 1

11 GISS -AOM USA T42L20 1.4°x1.4°L43 2 2 2

12 GISS -EH USA 5°x4°L20 2°x2° *cos(lat) L16 4 5 3

13 GIS S-ER USA 5°x4°L13 5°x4°L33 1 9 5

14 INM-CM3.0 Russia 5°x5°L21 2°x2.5°L33 2 1 1

15 IPSL-CM4 France 3.75°x2.5° L19 2°x1 °L31 3 1 1

16 MIROC3.2(hires) Japan T106 L56 0.28°x0.188° L47 1 1 1

17 MIROC3.2(medres) Japan T42 L20 (0.5° -1.4°)x 1.4° L44 3 3 3

18 ECHO -G (MIUB) Germany/Korea T30L19 T42L20 1 3 3

19 MRI-CGCM2.3.2 Japan T42 L30 (0.5° -2. 5°) x 2° L23 3 5 5

20 PCM USA T42L18 (0.5 -0.7°) x 0.7° L32 1

21 UKMO -HadCM3 UK 3.7°5x2.5° L15 1.25°x1.25° L20 2+1* 1 1

22 UKMO -HadGem1 UK 1.25°x1.875°L38 (0.33 -1.0°) x 1.0° L40 1+2* 2 1*

Sum 55 40

Models Contributed to IPCC AR4Models Contributed to IPCC AR4

Walsh et al. (2007)

Bayesian Model Averaging (BMA)

• Considers an ensemble of plausible models• Key Idea - The models vary in their skill, and

calibration of this skill produces better forecasts • Forecast PDF estimated through weighting the

PDFs of the individual models, with weights determined by posterior model probabilities

• BMA possesses a range of properties optimal from a theoretical point of view; works well in short-term weather prediction

p(y) is the forecast PDF; fk is the kth forecast model; wk is the posterior probability of forecastk being the best; gk(y | fk) is the PDF conditional on fk being the best forecast.

Weighted ensemble mean of parameter y

Estimating Weights by Maximum Likelihood

• Yields parameter values (weights) that make observed data most likely

• Likelihood function maximized over time and space through determination of model weights for a particular parameter

• Method uses expectation-maximization (EM) algorithm, which resembles a “hotter-colder” game

• Final weights related to how often a particular model constitutes the best model

• Training data set consists of last 40 days of short-term weather forecasts

Ensemble Model Projections for North Pacific Marine Ecosystems

• Initial Selection - Pick models that replicate the observed character of the PDO in their 20th century hindcasts (12 of 22 pass test)

• Regional Perspective - Examine specific parameter(s) in region of interest; consider means, variances, seasonality, etc.

• Model projections - Use quasi-Bayesian method based on “distance” between hindcasts and observations; form weighted ensemble means

• Uncertainty/Confidence - Estimate based on a combination of inter-model and intra-model variances in projections

Present Application• Limited statistics for evaluation (there has been a single

outcome for the past climate) • Compare 20th century hindcast simulations by the climate

models to observations on a regional basis• Observations based on NCEP Reanalysis; good match of

spatial scales• Consider mean, variance, and other measures (trend,

seasonality, etc.) if appropriate• Estimate weights for projections following a scheme

developed for objective analysis of weather observations, Wk = exp(-Dk/Dm)

• Apply intra-model variance from models with 5+ runs to models with fewer runs

Parameters Evaluated• Bering Sea - Flow through Unimak Pass (Nutrient

supply); Spring Winds (Larval flatfish transports); Summer SST & Wind Mixing (Sustained productivity)

• Gulf of Alaska - Along-coast winds (Larval fish distribution and abundance); Precipitation (Upper ocean baroclinity and eddy generation)

• NE Pacific - Coastal upwelling (Productivity); Zonal winds (LTL communities); Pycnocline depth; Upper-ocean transports; Temperature cross-sections

Model Weights - GOA Along-Coast Winds

0

0.05

0.1

0.15

0.2

0.25

0 2 4 6 8 10 12

Model Number

Weight

MIROC_hi

MRI GFDL21

GFDL20 MPI

CCCMA_t63

MIROC_med

UKHadCM3

ECHO-G CCCMA_t47

UKHadGEM1

GOA Along-Coast Wind (Aug-Sep)

-2

-1

0

1

2

2000 2010 2020 2030 2040 2050 2060

Year

Wind (m/s)

Downwelling

Upwelling Weighted Ensemble Mean

Zooplankton on the Bering Sea Shelf - Coyle et al. (2008)

Compared water properties and zooplankton abundance and community structure between a cold and warm year

1999 2004------------------------------------------------------------------Upper Temp 7.0 12.6 (deg. C)Lower Temp 2.0 3.2

Oithona 348 1633 (#/m3)Pseudocalanus 404 1211Calanus 44 ~0Thysanoessa 0.33 0.05

Model Weights - SE Bering SST

0

0.04

0.08

0.12

0.16

0.2

0 2 4 6 8 10 12

Model Number

Relative Weight

CCCMA_t63

MRI

UKHadGEM1

MIROC_med

UKHadCM3

MIROC_hi

CCSM3

ECHO_G

CCCMA_t47

GFDL21 GFDL20

SE Bering Sea Summer SST (JAS)

0

0.5

1

1.5

2

2.5

2000 2010 2020 2030 2040 2050

Year

Deg. C Relative to 1950-2000 Mean

SLP AnomaliesStrong WindMixing Years

SLP AnomaliesWeak WindMixing Years

Sea Ice Area Anomaly for Bering Sea

Overland and Wang (2007)

Flatfish in the SE Bering

Larval Transport & Recruitment

n o r t h e r n r o c k s o l e r e c r u i t m e n t

0

1

2

3

4

5

6

1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002

y e a r c l a s s

recruitment (billions)

o n s h e l f w i n d d r i f t

o f f s h e l f w i n d d r i f t

m i d s h e l f p a t t e r n

Model Weights - SE Bering Wind

0

0.05

0.1

0.15

0.2

0.25

0 2 4 6 8 10

Model Number

Relative Weight

MRI

MIROC_med

GFDL21

UKHadCM3 MIROC_hi

GFDL20

CCCMA_t47

CCCMA_t63

ECHO-G

SE Bering Winds

-4

-2

0

2

4

2000 2005 2010 2015 2020 2025 2030 2035 2040 2045 2050

Year

Apr-Jun Mean Wind (m/s)

P r o j e c t e d M e a n w i n d

- 1 7 2

- 1 7 0

- 1 6 8

- 1 6 6

- 1 6 4

- 1 6 2

- 1 6 0

2 0 0 1 2 0 0 6 2 0 1 1 2 0 1 6 2 0 2 1 2 0 2 6 2 0 3 1 2 0 3 6 2 0 4 1 2 0 4 6

ending 3 month drift longitude

2 0 0 0 2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0

0

1

2

3

4

5

y e a r

recruitment (billions)

2 0 0 0 2 0 1 0 2 0 2 0 2 0 3 0 2 0 4 0 2 0 5 0

0

1

2

3

4

5

y e a r

recruitment (billions)

Hollowed et al. (2009)

Projected Temperature Trends (2001-2050)

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

0 5 10 15 20 25

Model Run Number

Mean Trends (K/yr)

Series1

Weighted Mean

2

FEAST Higher trophic level model

NPZ-B-DLower trophic

level

ROMSPhysical

Oceanography

Economic/ecological model

Climate scenarios

BSIERP Integrated modeling

Observational Data

Nes

ted

mod

els

BEST

JELLYFISH

EUPHAUSIIDS

CALANUS NEOCALANUS

SMALL MICROZOOPLANKTON

LARGE MICROZOOPLANKTON

SMALL PHYTOPLANKTON

LARGE PHYTOPLANKTON

NITRATE AMMONIUM

DETRITUS

BENTHIC INFAUNA

BENTHIC DETRITUS

IRON

ICE ALGAE

NITRATE AMMONIUM

PSEUDOCALANUS

Excretion + Respiratio

n

Excretion + Respiratio

n

ICE

OCEAN

BENTHOS

BSIERP NPZmodel

MortalityPredationEgestionMolting

MortalityPredationEgestionMolting

*Both

*Z

**Z

*Z

*Transport

Transport*

Temperature X-Section

May Pycnocline Depth Errors

0

0.5

1

1.5

2

-4 -3 -2 -1 0 1 2 3 4

Mean Normalized Error

Variability

Newport

Alaska PeninsulaVancouver Is.

PAPA

Seward Line

*

Normalized Errors vs. Relative Variability (Summer)

0

0.2

0.4

0.6

0.8

1

1.2

-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5

Normalized Mean Error in Transport

Variability (Relative to Obs)

*

SODA ECHAM5 PCM1

MirocM MirocH MRI

Feb

Aug

Average Temperatures 1990-2000

SODA CCCMAT47_1 CCCMAT47_2

CCCMAT47_3 CCCMAT47_4 CCCMAT47_5

Feb

Aug

Average Temperatures 1990-2000

Absolute Model Errors

Parameter Mean Error Min Med Max

Observed Std. Dev.

Range Model SD

Other Measures

SE Bering SST (JAS)

0.8/ 1.9/ 4.2 (deg. C)

0.6 0.5 - 1.4 Trend Obs: 0.06/decade Models: -0.2 to 0.3

SE Bering Wind (AMJ)

0.02/ 0.4/ 1.0 (m/s)

1.1 0.7 – 1.5

GOA AlongShelf Wind (JAS)

0.04/ 0.3/ 0.9 (m/s)

0.7 0.5 – 1.0 Seasonality Obs: 3.2 Models: 0.7 to 3.3

NE Pacific Upwell SLP(JJA)

0.2/ 2.2/ 4.9 (mb)

1.1 0.9 – 1.7 Seasonality Obs: 7.8 Models: 6.9 to 14.5

Stn. PAPA Pycnocline Z

8.2/14/33 (m)

14 5.7 - 16 Seasonality Obs: -60 Models: -14 to -78

Alaskan Stream Currents

0.3/2.3/7.3 (cm/s)

3.1 0.2 – 3.0

Final Remarks• Global climate model simulations are being used for a

host of regional applications• There does not seem to be any single “best” method,

but the protocol should include evaluation of model hindcast simulations of key parameters in the region of interest.

• Multi-model ensembles represent a key tool for seasonal climate forecasts, and are being used increasingly for short-term weather prediction.

• On long time horizons, model structural uncertainty dominates initial condition sensitivity.

• From present to mid-21st century, climate change liable to be dominated by thermodynamic effects as opposed to dynamic effects (e.g., winds). The latter will be prone to interannual to decadal natural variability.

• The output from global climate models (perhaps subject to statistical downscaling) can complement that from vertically-integrated numerical models with full dynamics.