Recent developments in long-range forecasting at ECMWF 1 Recent developments in long-range...

Recent developments in long-range forecasting at ECMWF 1

Recent developments in long-range

forecasting at ECMWF: steps towards

seamless predictions

Franco Molteni, Frederic Vitart, Tim Stockdale,

Laura Ferranti, Magdalena Balmaseda European Centre for Medium-Range Weather Forecasts, Reading, U.K.


ECMWF Seasonal forecast system (Sys-3)

OASIS-2HOPE

1.4 deg. lon1.4/0.3 d. lat.

TESSEL

IFS 31R11.1 deg. 62 levels

4-D variational d.a.

Multivar. O.I.

Gen. ofPerturb.

System-3CGCM

Initial Con. Ens. Forecasts


•COUPLED MODEL (IFS + OASIS2 + HOPE)•Recent cycle of atmospheric model (Cy31R1)•Atmospheric resolution TL159 and 62 levels•Time varying greenhouse gasses.•Includes ocean currents in wave model

•INITIALIZATION•Includes bias correction in ocean assimilation.•Includes assimilation of salinity and altimeter data. •ERA-40 data used to initialize ocean and atmosphere in hindcasts•Ocean reanalysis back to 1959, using ENACT/ENSEMBLES ocean data

•ENSEMBLE GENERATION•Extended range of back integrations: 11 members, 1981-2005.•Revised wind and SST perturbations. •Use EPS Singular Vector perturbations in atmospheric initial conditions.

•Forecasts extended to 7 months (to 13 months 4x per year).

The seasonal forecast System-3 (implem. March 07)


Present TL159 monthly system:

Future 32-day VAREPS/monthly system:

Initial condition Day 32

Coupled forecast at TL159

Planned unified VAREPS/monthly system

Initial condition

EPS Integration at T399

Ocean only integration

Heat flux, Wind stress, P-E

Coupled forecast at TL255

Day 32Day 10


New products from Sys-3: ocean reanalysis


Products from Sys-3: ’tercile summary’


Climagrams : area-averages of 2mT and rainfall


Climagrams :monsoon indices / teleconnections


All India Rainfall: “climagram” from 1 May 2007


Products from Sys-3: “plumes” for El Nino indices


Verification for T_2m in DJF from 25-year hindcast set

Black dots for values significantly different from zero with 95% confidence ( 1000 samples)Threshold computed ranking the sampleHindcast period 1981-2005 with start in November average over months 2 to 4Near-surface temperature anomalies above the upper tercileROC Skill Score for ECMWF with 11 ensemble members and 12 bins

-0.8-1 -0.6-0.8 -0.4-0.6 -0.3-0.4 -0.2-0.3 -0.1-0.2 0.1-0.1 0.20.1 0.30.2 0.40.3 0.60.4 0.80.6 10.8


Verification for SST in JJA from 25-year hindcast set


0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18Calendar month

24

25

26

27

28

29

30

Ab

solu

te S

ST

NINO3.4 mean absolute SST

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18Calendar month

-3

-2

-1

0

1

2

3

Dri

ft (

de

g C

)

NINO3.4 mean SST drift

Fcast S3

MAGICS 6.11 cressida - net Mon Feb 5 17:34:33 2007

JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC

Verification month

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

An

om

aly

co

rre

latio

n

wrt NCEP adjusted OIv2 1971-2000 climatology

NINO3.4 SST anomaly correlation at 5 months


Verification month

0

0.2

0.4

0.6

0.8

1

1.2

Rm

s e

rro

r (d

eg

C)

Ensemble size is 11300 start dates from 19810101 to 20051201

NINO3.4 SST rms errors at 5 months

Fcast S3 Persistence Ensemble sd

MAGICS 6.11 cressida - net Tue Feb 6 09:26:44 2007

Nino 3.4


0 1 2 3 4 5 6 7Forecast time (months)

0.4

0.5

0.6

0.7

0.8

0.9

1

An

om

aly

co

rre

latio

n


ATL3 SST anomaly correlation

0 1 2 3 4 5 6 7Forecast time (months)

0

0.2

0.4

0.6

0.8

Rm

s e

rro

r (d

eg

C)


ATL3 SST rms errors




Verification month

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

An

om

aly

co

rre

latio

n


ATL3 SST anomaly correlation at 5 months


Verification month

0

0.2

0.4

0.6

0.8

1

Rm

s e

rro

r (d

eg

C)


ATL3 SST rms errors at 5 months



0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18Calendar month

24

25

26

27

28

29

30

Ab

solu

te S

ST

ATL3 mean absolute SST

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18Calendar month

-3

-2

-1

0

1

2

3

Dri

ft (

de

g C

)

ATL3 mean SST drift

Fcast S3

MAGICS 6.11 cressida - net Mon Feb 5 17:34:33 2007

Equatorial Atlantic


Anomaly correlation of seasonal-mean rainfall


Predictability of teleconnection/EOF indices in S-3

Rainfall: East. Tropical Indian Ocean pattern (JJA)


Predictability of teleconnection/EOF indices in S-3

Rainfall: Sahel / Guinea coast dipole (JJA)


Predictability of AIR in S-3

JJASCC = .25

JASCC = .46


Monsoon onset in India: seasonal fc. for June

1-month-lead fc. 0-lead fc. cc = 0.35 cc = 0.45

Forecast anomaly amplitude is ~ 2 x obs. !


Predictability of AIR in S-3: EOF filtered JAS (2)

CC = .50

UnfilteredCC = 0.46

EOF proj.

CC = 0.59


Predictability of East Africa short rains: EOF filtered OND

UnfilteredCC = 0.04

EOF proj.

CC = 0.39


Asian monsoon onset exp. with VarEPS-monthly

• VarEPS configuration with cycle 31r2• 45-day integration from 15 May to end of June

1991-2007

Verification of all-India rainfall vs. GPCP data • June mean rainfall• Pentad-mean rainfall


Monsoon onset in India: VarEPS-monthly fc.

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Year

-2

-1

0

1

2

Pre

cip.

per

day

(mm

)

-2

-1

0

1

2

Ensemble size = 10 (real time =140)Calibration period = 1991-2005Forecast start reference is 15/05/yyyyJune Precip over India (70-85E,5-30N)ECMWF Monthly Forecast

RMS Error= 0.88( 0.83)Correlation= 0.62( 0.99)

Observations (GPCP) FORECAST Standard Deviation

CC = 0.62


Anomaly correl. of pentad rainfall over India

1 2 3 4 5 6 7 8Pentads

-0.5

0

0.5

1

Co

rre

latio

n

Correlation EM/GPCPIndian Precip

Monthly Forecast

Persistence of pentad -1

Persistence of pentad 1


Monsoon onset predictions: early June pentads

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Year

-3

-2

-1

0

1

2

3

4

5

6

7

8

9

10

11

Pre

cip.

per

day

(m

m)

-3

-2

-1

0

1

2

3

4

5

6

7

8

9

10

11



ERA40 FORECAST Standard Deviation

Day 21-25: 6-10 June

CC = 0.76

Day 16-20: 1-5 June

CC = 0.79

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007

Year

-3

-2

-1

0

1

2

3

4

5

6

7

8

Pre

cip.

per

day

(m

m)

-3

-2

-1

0

1

2

3

4

5

6

7

8



ERA40 FORECAST Standard Deviation


Conclusions (1)

• SST predictions from the ECMWF seasonal forecast system-3 show

higher skill than those from previous system, particularly in the

tropical Pacific and eastern Indian Oc., but western Indian Oc. and

tropical Atlantic are still not better than persistence in NH summer.

• Substantial model errors affect rainfall variability over tropical land.

• Predictive skill for seasonal rainfall is generally good over the Pacific

and tropical S. America, poor along the coast of the Indian Ocean in

early summer. Skill for South Asian rainfall increases in the latter part

of the monsoon season.

• Seasonal forecasts over land can be improved by exploiting

teleconnections with adjacent ocean regions.


Conclusions (2)

• The VarEPS-monthly system represents an important step towards

merging the advantages of high-resolution simulations at medium

range with those of coupled simulations of low-frequency

intraseasonal variability.

• Predictions from the VarEPS-monthly system show useful skill at the

beginning of the South-Asian monsoon season, even on a pentad

time-scale.

• If results from VarEPS-monthly and seasonal forecast systems are

used on the appropriate scales, and minimal amount of statistical

post-processing is performed to correct model biases, dynamical

predictions from the ECMWF systems can be used into provide

skillfull, seamless information on atmospheric variability over a

number of tropical and extra-tropical regions.


GPCP rainfall climatology: JJA 1981-2005

Recent developments in long-range forecasting at ECMWF 1 Recent developments in long-range...

Documents

Transcript of Recent developments in long-range forecasting at ECMWF 1 Recent developments in long-range...