2 abingdon
The role of ECMWF in delivering advances in seasonal-to-decadal climate predictions With thanks to Magdalena Balmaseda, Judith Berner, Paco Doblas-Reyes, Mark Rodwell, Antje Weisheimer
Coapec/ncas/met office/ecmwf
Slide 1
ECMWF
The role of ECMWF in delivering advances in seasonal-to-decadal climate predictions ECMWF Seasonal Forecast System - Operational system, provision to Member States - Applications (eg Southern Africa Malaria Outlook Forum, Flood Forecasting for Bangladesh Met Department)
Multi-model ensembles: from DEMETER to EUROSIP ENSEMBLES (EU FP6) - Seasonal and decadal predictions - Comparing multi-model, stochastic physics, perturbed parameters methodologies of representing model uncertainty
Development of “seamless” approach to weather and climate prediction. Coapec/ncas/met office/ecmwf
Slide 2
ECMWF
Cellular Automaton Stochastic Backscatter Scheme (CASBS)
smooth scale
Cellular Automaton state
streamfunction forcing shape ! function
#% = $ ! " ( x, y ) ! D #t
D = sub-grid energy dissipation due to numerical diffusion, mountain drag and convection α = dimensional parameter
Coapec/ncas/met office/ecmwf
G.Shutts, 2005 Slide 3 ECMWF
The development of stochastic parametrisation is being aided by coarse-grain budget analyses from cloud resolving models (Shutts and Palmer, J. Clim, to appear)
Coapec/ncas/met office/ecmwf
Slide 4
ECMWF
1. Seasonal
Coapec/ncas/met office/ecmwf
Slide 5
ECMWF
∆ηµητηρ
Development of a European Multi-Model Ensemble System for Seasonal to Interannual Prediction
Reliability: 2m-Temp.>0 0.049 0.902 0.147
0.058 0.904 0.151
0.099 0.923 0.176
-0.007 0.886 0.107
0.075 0.921 0.153
-0.055 0.838 0.107
0.068 0.903 0.164
0.222 0.994 0.227
-- high malaria years -- low malaria years
-- high malaria years -- low malaria years
ROC Score
Precipitation
Event
CMAP
Incidence DEMETER
CMAP
DEMETER
Low
1.00 (1.00-1.00) 0.95 (0.82-1.00) 1.00 (1.00-1.00) 1.00 (1.00-1.00)
High
1.00 (1.00-1.00) 0.52 (0.25-0.78) 0.94 (0.80-1.00) 0.84 (0.65-0.98)
Reliability: Precip>0.43σ -0.141 0.855 0.004
-0.115 0.880 0.004
-0.108 0.885 0.007
-0.086 0.910 0.004
-0.107 0.887 0.006
-0.137 0.862 0.001
-0.115 0.881 0.004
-0.022 0.969 0.009
Blocking frequency in seasonal hindcasts Northern Hemisphere blocking frequency for DEMETER hindcasts November start, 1959-2001, 9-member ensembles Top row: November (first month) Bottom row: January (third month) ERA40 Single models CNRM
ECMWF
Met Office
era-40 stoch phys control
Coapec/ncas/met office/ecmwf
Slide 12
ECMWF
Possible impact of Stochastic Parametrisation - a nonlinear paradigm for diagnosing the causes of model bias Eg ball bearing in potential well.
• Without smallscale “noise”, this regime is too dominant
•
Coapec/ncas/met office/ecmwf
Without small-scale “noise”, this minimum might be inaccessible
Jung et al 2006 Slide 13
ECMWF
There will now be a focus on the winter 2005/2006 in RT1 of the ENSEMBLES project.
Coapec/ncas/met office/ecmwf
Slide 14
ECMWF
If seasonal forecast models are generally under-dispersive in terms of blocking activity, can we be sure that consistency of multi-model climate-change predictions of mid-lat precipitation is an indication of forecast reliability?
Coapec/ncas/met office/ecmwf
Slide 15
ECMWF
Coapec/ncas/m et office/ecm wf
Slide 16
ECMWF
Probability of 1-in-20 year wet winter based on AR4 multi-model ensemble
Weisheimer and Palmer, 2006)
systematic bias 1991-2001 DJF (Nov start): precipitation control – GPCP
CASBS – control
CASBS – GPCP
2. Decadal
Slide 19
ECMWF
THC: Atlantic Meridional Transport (30N) 2.0·10
7
1.8·10
7
1.6·10
7
1.4·10
7
1.2·10
7
0001 ATL30NU 132
upper 1000m
1970
1980
1990
2000
1990
2000
Time -7.0·10
6
-8.0·10
6
-9.0·10
6
-1.0·10
7
-1.1·10
7
-1.2·10
7
0001 ATL30NI 132
1000m-3000m
1970
-1.0·10
7
-1.2·10
7
-1.4·10
7
-1.6·10
7
-1.8·10
7
1980
Time
0001 ATL30ND 132
3000m-5000m
Slide 20
Magdalena A. Balmaseda 1970
1980
1990 Time
Values from Bryden 2000et al 2005
THC: Atlantic Meridional Transport (30N) 0001 ATL30NU 132
2.0·107
1.8·107
1.6·107
1.4·107
1.2·10
upper 1000m
7
1970
1980
1990
2000
1990
2000
Time
0001 ATL30NI 132
-7.0·106 -8.0·106
1000m-3000m
-9.0·106 -1.0·107 -1.1·107 -1.2·107 1970
1980
Time
0001 ATL30ND 132 -1.0·107
-1.2·107
3000m-5000m
-1.4·107
-1.6·107
Values from Bryden et al 2005
-1.8·107 1970
1980
1990 Time
2000
decadal forecasts: forecast anomalies
global SST multi-model
start dates: 1965 and 1994
stochastic physics start dates: 1965 and 1994
perturbed physics start dates: 1991 - 1994
3. More benefits of a seamless NWP Climate approach
Slide 22
ECMWF
Stainforth et al (2005)
Previously unknown risk of catastrophic warming
IPCC (2001) range
Probability of Global Warming
Climate: Error vs Sensitivity
All Stainforth models accepted Highest sensitivity for low entrainment models
Circles: AGCM + Mixed-Layer model results from Stainforth et al. (2005) show combined RMSE of 8 year mean, annual mean T2m, SLP, precipitation and ocean-atmosphere sensible+latent heat fluxes (equally weighted and normalised by the control). Diamonds: AGCM results from Rodwell & Palmer (2006) show RMSE from 39 year mean, annual mean T850, SLP and precipitation (equally weighted and normalised by the control). Coapec/ncas/met office/ecmwf
Slide 24
ECMWF
January 2005 Initial T Tendencies 12 36
Rodwell and Palmer, 2006 12 36
96
96
CONTROL Pr e s s u r e (a p p r o x )
Pr e s s u r e (a p p r o x )
539
353
539
728
728
884
884
979 1012
979 1012 -9
-6
CLOUD
202
353
-3 Kday
0 (K for bias)
-1
3
6
9
Radiative Vertical Diffusion Cumulus Convection Large Scale Precipitation
202
Pressure (approx)
12 36
202
Dynamic
96
Total D+5 Bias Cloud Frac
353
539
CLOUD better for T, worse for q
728 -9
-6
-3 Kday
0 (K for bias)
-1
3
6
9
884 12 36
12 36
ENTRAIN/5
96
202
Pr e s s u r e (a p p r o x )
Pr e s s u r e (a p p r o x )
202
979 1012
ENTRAINx3
96
353
539
-6 353
539
728
728
884
884
979 1012
979 1012 -9
-6
-3
0
3
6
ENTRAIN/5 and ENTRAINx3 are completely out of balance: reject or down-weight? -3
-9
-6
-3
0
3
6
By D+5, interactions between processes (non-linearity) leads to completely different balance
Amazon = [300oE-320oE, 20oS-0oN]. 70% confidence intervals shown. Model = 29R1,T159,L60,1800S. Slide 25 Coapec/ncas/m et office/ecm wf
ECMWF
Conclusions • Multi-model ensembles are not necessarily reliable on seasonal timescales. DEMETER/Eurosip models systematically undersimulate blocking – implications for climate change prediction? • Stochastic parametrisations increase the probability of occurrence of sub-dominant regimes in the ECMWF model • ECMWF analyses appear to have realistic THC variability. Initial decadal runs of the ECMWF model have been performed within the ENSEMBLES project • Very short-range tendencies of the ECMWF model can be used to constrain climatically-important fast-physics parameter perturbations • Overall ECMWF contributions support WCRP “seamless” strategy – unified approach to days, seasons, decades, centuries. Coapec/ncas/m et office/ecm wf
Slide 26
ECMWF
Coapec/ncas/met office/ecmwf
Slide 1
ECMWF
The role of ECMWF in delivering advances in seasonal-to-decadal climate predictions ECMWF Seasonal Forecast System - Operational system, provision to Member States - Applications (eg Southern Africa Malaria Outlook Forum, Flood Forecasting for Bangladesh Met Department)
Multi-model ensembles: from DEMETER to EUROSIP ENSEMBLES (EU FP6) - Seasonal and decadal predictions - Comparing multi-model, stochastic physics, perturbed parameters methodologies of representing model uncertainty
Development of “seamless” approach to weather and climate prediction. Coapec/ncas/met office/ecmwf
Slide 2
ECMWF
Cellular Automaton Stochastic Backscatter Scheme (CASBS)
smooth scale
Cellular Automaton state
streamfunction forcing shape ! function
#% = $ ! " ( x, y ) ! D #t
D = sub-grid energy dissipation due to numerical diffusion, mountain drag and convection α = dimensional parameter
Coapec/ncas/met office/ecmwf
G.Shutts, 2005 Slide 3 ECMWF
The development of stochastic parametrisation is being aided by coarse-grain budget analyses from cloud resolving models (Shutts and Palmer, J. Clim, to appear)
Coapec/ncas/met office/ecmwf
Slide 4
ECMWF
1. Seasonal
Coapec/ncas/met office/ecmwf
Slide 5
ECMWF
∆ηµητηρ
Development of a European Multi-Model Ensemble System for Seasonal to Interannual Prediction
Reliability: 2m-Temp.>0 0.049 0.902 0.147
0.058 0.904 0.151
0.099 0.923 0.176
-0.007 0.886 0.107
0.075 0.921 0.153
-0.055 0.838 0.107
0.068 0.903 0.164
0.222 0.994 0.227
-- high malaria years -- low malaria years
-- high malaria years -- low malaria years
ROC Score
Precipitation
Event
CMAP
Incidence DEMETER
CMAP
DEMETER
Low
1.00 (1.00-1.00) 0.95 (0.82-1.00) 1.00 (1.00-1.00) 1.00 (1.00-1.00)
High
1.00 (1.00-1.00) 0.52 (0.25-0.78) 0.94 (0.80-1.00) 0.84 (0.65-0.98)
Reliability: Precip>0.43σ -0.141 0.855 0.004
-0.115 0.880 0.004
-0.108 0.885 0.007
-0.086 0.910 0.004
-0.107 0.887 0.006
-0.137 0.862 0.001
-0.115 0.881 0.004
-0.022 0.969 0.009
Blocking frequency in seasonal hindcasts Northern Hemisphere blocking frequency for DEMETER hindcasts November start, 1959-2001, 9-member ensembles Top row: November (first month) Bottom row: January (third month) ERA40 Single models CNRM
ECMWF
Met Office
era-40 stoch phys control
Coapec/ncas/met office/ecmwf
Slide 12
ECMWF
Possible impact of Stochastic Parametrisation - a nonlinear paradigm for diagnosing the causes of model bias Eg ball bearing in potential well.
• Without smallscale “noise”, this regime is too dominant
•
Coapec/ncas/met office/ecmwf
Without small-scale “noise”, this minimum might be inaccessible
Jung et al 2006 Slide 13
ECMWF
There will now be a focus on the winter 2005/2006 in RT1 of the ENSEMBLES project.
Coapec/ncas/met office/ecmwf
Slide 14
ECMWF
If seasonal forecast models are generally under-dispersive in terms of blocking activity, can we be sure that consistency of multi-model climate-change predictions of mid-lat precipitation is an indication of forecast reliability?
Coapec/ncas/met office/ecmwf
Slide 15
ECMWF
Coapec/ncas/m et office/ecm wf
Slide 16
ECMWF
Probability of 1-in-20 year wet winter based on AR4 multi-model ensemble
Weisheimer and Palmer, 2006)
systematic bias 1991-2001 DJF (Nov start): precipitation control – GPCP
CASBS – control
CASBS – GPCP
2. Decadal
Slide 19
ECMWF
THC: Atlantic Meridional Transport (30N) 2.0·10
7
1.8·10
7
1.6·10
7
1.4·10
7
1.2·10
7
0001 ATL30NU 132
upper 1000m
1970
1980
1990
2000
1990
2000
Time -7.0·10
6
-8.0·10
6
-9.0·10
6
-1.0·10
7
-1.1·10
7
-1.2·10
7
0001 ATL30NI 132
1000m-3000m
1970
-1.0·10
7
-1.2·10
7
-1.4·10
7
-1.6·10
7
-1.8·10
7
1980
Time
0001 ATL30ND 132
3000m-5000m
Slide 20
Magdalena A. Balmaseda 1970
1980
1990 Time
Values from Bryden 2000et al 2005
THC: Atlantic Meridional Transport (30N) 0001 ATL30NU 132
2.0·107
1.8·107
1.6·107
1.4·107
1.2·10
upper 1000m
7
1970
1980
1990
2000
1990
2000
Time
0001 ATL30NI 132
-7.0·106 -8.0·106
1000m-3000m
-9.0·106 -1.0·107 -1.1·107 -1.2·107 1970
1980
Time
0001 ATL30ND 132 -1.0·107
-1.2·107
3000m-5000m
-1.4·107
-1.6·107
Values from Bryden et al 2005
-1.8·107 1970
1980
1990 Time
2000
decadal forecasts: forecast anomalies
global SST multi-model
start dates: 1965 and 1994
stochastic physics start dates: 1965 and 1994
perturbed physics start dates: 1991 - 1994
3. More benefits of a seamless NWP Climate approach
Slide 22
ECMWF
Stainforth et al (2005)
Previously unknown risk of catastrophic warming
IPCC (2001) range
Probability of Global Warming
Climate: Error vs Sensitivity
All Stainforth models accepted Highest sensitivity for low entrainment models
Circles: AGCM + Mixed-Layer model results from Stainforth et al. (2005) show combined RMSE of 8 year mean, annual mean T2m, SLP, precipitation and ocean-atmosphere sensible+latent heat fluxes (equally weighted and normalised by the control). Diamonds: AGCM results from Rodwell & Palmer (2006) show RMSE from 39 year mean, annual mean T850, SLP and precipitation (equally weighted and normalised by the control). Coapec/ncas/met office/ecmwf
Slide 24
ECMWF
January 2005 Initial T Tendencies 12 36
Rodwell and Palmer, 2006 12 36
96
96
CONTROL Pr e s s u r e (a p p r o x )
Pr e s s u r e (a p p r o x )
539
353
539
728
728
884
884
979 1012
979 1012 -9
-6
CLOUD
202
353
-3 Kday
0 (K for bias)
-1
3
6
9
Radiative Vertical Diffusion Cumulus Convection Large Scale Precipitation
202
Pressure (approx)
12 36
202
Dynamic
96
Total D+5 Bias Cloud Frac
353
539
CLOUD better for T, worse for q
728 -9
-6
-3 Kday
0 (K for bias)
-1
3
6
9
884 12 36
12 36
ENTRAIN/5
96
202
Pr e s s u r e (a p p r o x )
Pr e s s u r e (a p p r o x )
202
979 1012
ENTRAINx3
96
353
539
-6 353
539
728
728
884
884
979 1012
979 1012 -9
-6
-3
0
3
6
ENTRAIN/5 and ENTRAINx3 are completely out of balance: reject or down-weight? -3
-9
-6
-3
0
3
6
By D+5, interactions between processes (non-linearity) leads to completely different balance
Amazon = [300oE-320oE, 20oS-0oN]. 70% confidence intervals shown. Model = 29R1,T159,L60,1800S. Slide 25 Coapec/ncas/m et office/ecm wf
ECMWF
Conclusions • Multi-model ensembles are not necessarily reliable on seasonal timescales. DEMETER/Eurosip models systematically undersimulate blocking – implications for climate change prediction? • Stochastic parametrisations increase the probability of occurrence of sub-dominant regimes in the ECMWF model • ECMWF analyses appear to have realistic THC variability. Initial decadal runs of the ECMWF model have been performed within the ENSEMBLES project • Very short-range tendencies of the ECMWF model can be used to constrain climatically-important fast-physics parameter perturbations • Overall ECMWF contributions support WCRP “seamless” strategy – unified approach to days, seasons, decades, centuries. Coapec/ncas/m et office/ecm wf
Slide 26
ECMWF
