Systematic Errors in the ECMWF Forecasting System ECMWF Thomas Jung.
Recent developments in long-range forecasting at ECMWF 1 Recent developments in long-range...
-
Upload
ashlie-perry -
Category
Documents
-
view
217 -
download
2
Transcript of 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.
Recent developments in long-range forecasting at ECMWF 2
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
Recent developments in long-range forecasting at ECMWF 3
•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)
Recent developments in long-range forecasting at ECMWF 4
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
Recent developments in long-range forecasting at ECMWF 5
New products from Sys-3: ocean reanalysis
Recent developments in long-range forecasting at ECMWF 6
Products from Sys-3: ’tercile summary’
Recent developments in long-range forecasting at ECMWF 7
Climagrams : area-averages of 2mT and rainfall
Recent developments in long-range forecasting at ECMWF 8
Climagrams :monsoon indices / teleconnections
Recent developments in long-range forecasting at ECMWF 9
All India Rainfall: “climagram” from 1 May 2007
Recent developments in long-range forecasting at ECMWF 10
Products from Sys-3: “plumes” for El Nino indices
Recent developments in long-range forecasting at ECMWF 11
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
Recent developments in long-range forecasting at ECMWF 12
Verification for SST in JJA from 25-year hindcast set
Recent developments in long-range forecasting at ECMWF 14
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
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
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
Recent developments in long-range forecasting at ECMWF 16
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
wrt NCEP adjusted OIv2 1971-2000 climatology
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)
Ensemble size is 11300 start dates from 19810101 to 20051201
ATL3 SST rms errors
Fcast S3 Persistence Ensemble sd
MAGICS 6.11 cressida - net Tue Feb 6 09:26:44 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
ATL3 SST anomaly correlation at 5 months
JAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC
Verification month
0
0.2
0.4
0.6
0.8
1
Rm
s e
rro
r (d
eg
C)
Ensemble size is 11300 start dates from 19810101 to 20051201
ATL3 SST rms errors at 5 months
Fcast S3 Persistence Ensemble sd
MAGICS 6.11 cressida - net Tue Feb 6 09:26:44 2007
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
Recent developments in long-range forecasting at ECMWF 17
Anomaly correlation of seasonal-mean rainfall
Recent developments in long-range forecasting at ECMWF 19
Predictability of teleconnection/EOF indices in S-3
Rainfall: East. Tropical Indian Ocean pattern (JJA)
Recent developments in long-range forecasting at ECMWF 21
Predictability of teleconnection/EOF indices in S-3
Rainfall: Sahel / Guinea coast dipole (JJA)
Recent developments in long-range forecasting at ECMWF 22
Predictability of AIR in S-3
JJASCC = .25
JASCC = .46
Recent developments in long-range forecasting at ECMWF 23
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. !
Recent developments in long-range forecasting at ECMWF 25
Predictability of AIR in S-3: EOF filtered JAS (2)
CC = .50
UnfilteredCC = 0.46
EOF proj.
CC = 0.59
Recent developments in long-range forecasting at ECMWF 26
Predictability of East Africa short rains: EOF filtered OND
UnfilteredCC = 0.04
EOF proj.
CC = 0.39
Recent developments in long-range forecasting at ECMWF 27
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
Recent developments in long-range forecasting at ECMWF 28
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
Recent developments in long-range forecasting at ECMWF 29
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
Recent developments in long-range forecasting at ECMWF 30
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
Ensemble size = 10 (real time =150)Calibration period = 1991-2006Forecast start reference is 15/05/yyyyJune Precip over India (70-85E,5-30N)ECMWF Monthly Forecast
RMS Error= 2.32( 3.41)Correlation= 0.76( 1.00)
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
Ensemble size = 10 (real time =150)Calibration period = 1991-2006Forecast start reference is 15/05/yyyyJune Precip over India (70-85E,5-30N)ECMWF Monthly Forecast
RMS Error= 1.25( 1.46)Correlation= 0.79( 1.00)
ERA40 FORECAST Standard Deviation
Recent developments in long-range forecasting at ECMWF 31
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.
Recent developments in long-range forecasting at ECMWF 32
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.
Recent developments in long-range forecasting at ECMWF 38
GPCP rainfall climatology: JJA 1981-2005