1. Page 1© Crown copyright 2004 From DEMETER to EURO-SIP: The European Seasonal to Inter-annual Prediction project Richard Graham acknowledgements to DEMETER and EURO-SIP partners (Francisco Doblas-Reyes, Laura Ferranti (ECMWF)) NOAA 30 th Climate Annual Climate Diagnostics and Prediction Workshop, October 24-28 th, State College, Pennsylvania

2. Page 2© Crown copyright 2004 Talk Outline  DEMETER project (EU FP5) – completed September 2003  European seasonal multi-model system (hindcasts)  Application/impact models (crops, health)  forecast system sensitivity studies  EURO-SIP  operational real-time implementation of DEMETER  ENSEMBLES (EU FP6)  prediction/impacts seasonal to decadal (and longer) timescales  new representations of model and initial condition uncertainty  5 year project, now one year in  Met Office European winter forecast 2005/06

3. Page 3© Crown copyright 2004 The DEMETER multi-model ensemble system  DEMETER system: 7 coupled global circulation models  6 installed at ECMWF  Hindcast periods:  1980-2001 (22yrs) – all models  1959-2001 (43yrs) – 3 models  9 member ensembles  ERA-40 initial conditions  SST and wind perturbations  4 start dates per year  6 months hindcasts (GloSea)

4. Page 4© Crown copyright 2004 Multi-model ensemble system: hindcasts Feb 87 May 87 Aug 87 Nov 87 Feb 88... 7 models x 9 ensemble members 63 member multi-model ensemble DEMETER system: 7 coupled global circulation models CNRM (FR) ECMWF (INT) INGV (IT) LODYC (FR) MPI (DE) UKMO (UK) CERFACS (FR)

5. Page 5© Crown copyright 2004 Multi-model ensemble system: hindcasts Feb 87 May 87 Aug 87 Nov 87 Feb 88... DEMETER system: 7 coupled global circulation models CNRM (FR) ECMWF (INT) INGV (IT) LODYC (FR) MPI (DE) UKMO (UK) CERFACS (FR)

6. Page 6© Crown copyright 2004 Multi-model ensemble system: hindcasts Feb 87 May 87 Aug 87 Nov 87 Feb 88... DEMETER system: 7 coupled global circulation models CNRM (FR) ECMWF (INT) INGV (IT) LODYC (FR) MPI (DE) UKMO (UK) CERFACS (FR)

7. Page 7© Crown copyright 2004 Multi-model ensemble system: hindcasts Feb 87 May 87 Aug 87 Nov 87 Feb 88... 63 member multi-model ensemble = 1 hindcast DEMETER system: 7 coupled global circulation models CNRM (FR) ECMWF (INT) INGV (IT) LODYC (FR) MPI (DE) UKMO (UK) CERFACS (FR)

8. Page 8© Crown copyright 2004  Evaluation of DEMETER multi- model hindcast skill  Applications/impact models  Crop yield  malaria  Downscaling studies  Statistical & dynamical  Sensitivity studies  e.g. CGCM Vs AGCM Results DEMETER Special Issue 2005 Tellus 57A, No. 3, 21 contributions

9. Page 9© Crown copyright 2004 Brier skill score Single vs. multi-model: Brier skill score ‘basket’ of regions, start dates and leadtimes Resolution skill score=Res/BS c Reliability skill score=1-Rel/BS c

10. Page 10© Crown copyright 2004 Impact of number of models on multi-model skill: Tropics, JJA precip, 1m lead: RPSS Multi-model realizations Single-model realizations

11. Page 11© Crown copyright 2004 France Germany Denmark Greece Application models: wheat yield predictions for Europe From P. Cantelaube and J.-M. Terres, JRC SIMULATIONWEIGHTED YIELD ERROR (%) ± STANDARD ERROR JRC February7.1 ± 0.9 JRC April7.7 ± 0.5 JRC June7.0 ± 0.6 JRC August5.4 ± 0.5 DEMETER (Feb. start)6.0 ± 0.4 DEMETER multi-model predictions (7 models, 63 members, Feb starts) of average wheat yield for four European countries (box-and-whiskers) compared to Eurostat official yields (black horizontal lines) and crop results from a simulation forced with downscaled ERA40 data (red dots).

12. Page 12© Crown copyright 2004 Summary  DEMETER: proof of multi-model benefit using 7 coupled ocean-atmosphere models  More than a multi-model exercise…encouraging results for integration with application models  Mutual understanding between prediction and application communities – ‘end-to-end’ approach  Infrastructure and standardisation developed  coordination of model output content (with application partners)  model output formats and units (archiving at ECMWF).  downscaling techniques/experience  centralised model diagnosis/verification system at ECMWF  data download and plotting facility http://data.ecmwf.int/data/

13. Page 13© Crown copyright 2004 EURO-SIP real-time multi-model ensemble  Three models running at ECMWF:  Met Office: GloSea (version of HadCM3) (41-ensemble)  ECMWF: system2 (40-ensemble)  Météo-France – Arpège (41-ensemble)  Unified system with common development strategy  Products in development:  2-model products available on Met Office website  ECMWF release of web products is expected soon

14. Page 14© Crown copyright 2004 Probability of precipitation – outer-quintile categories, GHACOF region, SON 2005 P(well-above) P(well-below) Yellow/red: enhanced chance of category (relative to climatology) Blue: diminished chance White:near climatological probabilities GloSea based on 1987-2001 climatology EURO-SIP (80 ensemble)

15. Page 15© Crown copyright 2004 Meteo-France ECMWFMet Office (GloSea) Probability of lower temperature tercile category: NDJ 2005 (from September) Forecast is made available on the 15th of each month. EURO-SIP multi-model

16. Page 16© Crown copyright 2004 Tropical Storms Forecasts starting on 1 st June 2005: JASON ECMWF Met Office Météo-France Obs 1 st May- mid-Sep Atl W-PacE-Pac Multimodel Forecast No. = 18 Climatology = 11.6

17. Page 17© Crown copyright 2004 After DEMETER…ENACT: enhanced ocean assimilation EN2 QC’d dataset and objective analysis, 1956-2004 http://www.hadobs.org

18. Page 18© Crown copyright 2004 ENSEMBLES: seasonal to decadal & improved representing model uncertainties 1.7 model multi-model installed at ECMWF - includes Met Office decadal forecast system (anomaly assimilation) - realistic GHGs, aerosols Met Office decadal forecast (DePreSys)

19. Page 19© Crown copyright 2004 ENSEMBLES: perturbed physics From Murphy et al, Nature 2004 2. Many models from perturbing physics parameters, eg. RH crit for cloud formation (ref Murphy et al, 2004; and climateprediction.net)

20. Page 20© Crown copyright 2004 ENSEMBLES: stochastic physics E(n) n = spherical harmonic order missing energy E(n) Stochastic physics experiment with ECMWF system: Cellular Automaton Stochastic Backscatter Scheme (CASBS) Without CASBS With CASBS From Tim Palmer, ECMWF 3. Stochastic ‘backscatter’ experiment (ECMWF)

21. Page 21© Crown copyright 2004 Summary and plans  EURO-SIP: Real-time follow up to DEMETER  products becoming available  Meteo-France model fully implemented early 2006  plans for model upgrades, convergence on the periods and ensemble size of hindcasts  ENACT:  enhanced ocean data assimilation – ocean ‘reanalyses’  QC’d ocean data and objective analyses  ENSEMBLES:  builds on DEMETER and ENACT  seasonal to decadal timescales (and longer)  further investigations of model uncertainties  calibrating and combining model ensembles

22. Page 22© Crown copyright 2004 Met Office forecast: European winter 2005/06 NAO index: difference between normalised pressure anomaly Azores - Iceland N. Atlantic May SST anomaly associated with +ve NAO N. Atlantic May SST anomaly 2005 Predicts correct sign in 2 out of 3 winters Empirical NAO prediction Rodwell & Folland, 2002

23. Page 23© Crown copyright 2004 September surface and sub-surface anomalies and GloSea prediction GloSea prediction from September 60-90m anomalies SST anomalies normalised actual

24. Page 24© Crown copyright 2004 Thank you! additional slides follow

25. Page 25© Crown copyright 2004 EURO-SIP real-time multi-model ensemble  Met Office (GloSea2)  Modified HadCM3  Atmos. 2.5° x 3.75° x 19L  Ocean (1.25° to 0.3°) x 1.25° x 40L  41 ensemble members (from mix of windstress and SST perturbations)  Hindcast: 1987-2001, 15 members  ECMWF (system2)  Atmos. TL95L40  Ocean (1.0° to 0.3°) x 1.0° x 29L  40 ensemble members  Hindcast: 1987-2001, 5 members  Meteo-France  Atmos. Arpège TL63L31  Ocean: ORCA (2.0° to 0.5°) 31L  41 ensemble members (mix of 5 ocean analyses and 8 lagged atmos)  Hindcast: 1993-2003, 5 members

26. Page 26© Crown copyright 2004 Parameter Perturbations Convection Entrainment rate Intensity of mass flux Shape of cloud (anvils) Cloud water seen by radiation Radiation Ice particle size/shape Sulphur cycle Water vapour continuum absorption Boundary layer Turbulent mixing coefficients: stability- dependence, neutral mixing length Roughness length over sea: Charnock constant, free convective value Dynamics Diffusion: order and e-folding time Gravity wave drag: surface and trapped lee wave constants Gravity wave drag start level Land surface processes Root depths Forest roughness lengths Surface-canopy coupling CO2 dependence of stomatal conductance Sea ice Albedo dependence on temperature Ocean-ice heat transfer Large Scale Cloud Ice fall speed Critical relative humidity for formation Cloud droplet to rain: conversion rate and threshold Cloud fraction calculation

27. Page 27© Crown copyright 2004 Met Office seasonal/multi-annual/decadal runs Model DePreSys (HadCM3) Current oper. range decadal assimilation method GloSea (HadCM3*) Seasonal (6months) Conventional (OI type) calibrated anomalies 9-ensemble experiments 1991-2001 pert. ODA pert. phys. lagged avge pert. phys. lagged avge Hindcasts pre-production period: 1991 - 2001 GloSea: ->7m:1 st /15 th May/Nov 1 st June/Dec ->14m:1 st May/June/Nov/Dec -> 10y:1 st May 1964, 1994 DePreSys: -> 10y:1 st May/Nov (all years) *Later HadGEM

28. Page 28© Crown copyright 2004 Multi-model evaluation: Niño3.4 SST anomaly: ASO at 3m lead: http://www.ecmwf.int/research/demeter/d/charts/verification/ Correlation: Multi-model = 0.83 Best single model = 0.81

29. Page 29© Crown copyright 2004 GloSea development and response to North Atlantic SST ‘tri-pole’ pattern: spring 1989 Jan 89 Feb 89 May 89 GloSeaHadISST MAM 1989 HadAM3 (+ pers Jan SSTA) GloSea ERA40

30. Page 30© Crown copyright 2004 ENSEMBLES: seasonal to interannual timescales (and longer); representing model uncertainties  Hindcast pre-production period: 1991-2001; later 1960-2001  9-member ensembles  ERA-40 atmosphere and soil initial conditions  ENACT-based ocean initial conditions with SST and wind perturbations  2 seasonal (7 months),1 annual (12-14 months) runs per year  Two multi-annual runs (1965, 1994) except 2 per year for DePreSys)  Realistic boundary forcings: GHGs, aerosols, solar forcing, etc.  ENSEMBLES system: 7 coupled GCMs running at ECMWF  upgraded models from DEMETER (eg. using ENACT developments etc)

31. Page 31© Crown copyright 2004 Sensitivity studies: AGCM Vs CGCM Nov. start for FMA (3-month lead) HadAM3 (2-tier) GloSea (1-tier) Much improved GloSea skill in tropical (south) Atlantic and Indian ocean, from representation of lagged responses to ENSO. ROC Skill for 2m temperature, upper tercile

32. Page 32© Crown copyright 2004 Met Office decadal forecast system (DePreSys): Forecast of global annual mean T s  Predict 0.36+/-0.23 o C (5-95%) warming for the coming decade  Unless a major volcanic eruption occurs!  Uses ‘anomaly assimilation’ (seasonal-range skill very competitive!)