Initialization Techniques in Seasonal Forecasting

Transcription

1 Initialization Techniques in Seasonal Forecasting Magdalena A. Balmaseda Contributions from Linus Magnuson, Kristian Mogensen, Sarah Keeley ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting

2 Outline The importance of the ocean initial conditions in SF Ocean Model initialization The value of observational information: fluxes, SST, ocean observations The difficulties The traditional Full Initialization approach: pros and cons. Other approaches. Assessment Full Initialization, Anomaly Initialization ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 2

3 The basis for extended range forecasts Forcing by boundary conditions changes the atmospheric circulation, modifying the large scale patterns of temperature and rainfall, so that the probability of occurrence of certain events deviates significantly from climatology. Important to bear in mind the probabilistic nature of SF The boundary conditions have longer memory, thus contributing to the predictability. Important boundary forcing: Tropical SST: ENSO, Indian Ocean Dipole, Atlantic SST Land: snow depth, soil moisture Sea-Ice Mid-Latitude SST Atmospheric composition: green house gases, aerosols, ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 3

4 Impact of Sea Ice: CI. Ice Cover anomaly Observed Z500 anomaly CI Dm ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting Courtesy of Sarah 5 Keely

5 Impact on Z500: 2007 ey98-ev9o: Z500 JA 2007: (May mon3-mon4 Obs-Clim Ice Z ) ey98-ev9o: (May mon3-mon4 ey98-ev9o: Z500 SLP JA 2008: (May 2007) mon3-mon4 Obs-Clim Ice Z ) 2008 ey98-ev9 60 W E 40 E F Mean of 3 Uninitialised Analyses Valid: VT:00UTC July 2008 to 00UTC 3 July hPa Geopotential 40 W 20 N 20 N N 70 N 60 N 50 N 40 N 30 N E 40 E 40 W 20 W N 20 N 20 E 00 E 80 E 60 E W ECMWF Mean of 3 Uninitialised Analyses Valid: VT:00UTC July 2008 to 00UTC 3 July hPa Geopotential 40 W 60 W E 40 E 20 N 00 W Atmos model 80 W -2 ObsIce-ClimIce 60 W 20 N N 70 N 60 N 50 N 40 N 30 N W 20 W 0 20 E 40 E N 20 N 20 E 00 E 80 E 60 E Low over Western Europe and Greenland high: similar response in both years. Consistent with observations The response was conditioned by the SST, in particular the North Atlantic (Gulf Stream region), pointing towards the need of high resolution ocean models (or flux corrections). The question of the predictability of the sea-ice anomaly remains. From Balmaseda et al 200 ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 6

6 Potential Energy for Tropical Cyclones ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 7

7 ENSEMBLE GENERATION PROBABILISTIC CALIBRATED FORECAST Anomaly (deg C) End-To-End Seasonal forecasting System COUPLED MODEL Tailored Forecast PRODUCTS ECMWF Seasonal Forecast Tropical Storm Frequency Forecast start reference is 0/06/2005 Ensemble size = 40,climate size = 70 JASON Significance level is 90% 20 E FORECAST 40 E 60 E 80 E 00 E 20 E 40 E CLIMATE 60 E W 40 W 20 W 00 W 80 W 60 W 40 W 20 W 80 N 80 N 70 N 70 N 60 N 60 N OCEAN 50 N 50 N 40 N 40 N 30 N 30 N 20 N 20 N N 0 N S 0 S 20 S 20 S 30 S 30 S 40 S 40 S NINO3.4 SST anomaly plume 50 S 50 S ECMWF forecast from Jan S 60 S Monthly mean anomalies relative to NCEP adjusted OIv climatology 70 S 70 S 80 S System 3 80 S 20 E 40 E 60 E 80 E 2 00 E 20 E 40 E 60 E W 40 W 20 W 00 W 80 W 60 W 40 W 20 W 2 No Significance 90% Significance 95% Significance 99% Significance JUL 2006 AUG SEP OCT NOV DEC JAN 2007 Produced from real-time forecast data FEB MAR APR MAY JUN JUL AUG SEP Initialization Forward Integration Forecast Calibration ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 8

8 % Initialization into Context A decade of progress on ENSO prediction S S2 S3 Steady progress: ~ month/decade skill gain How much is due to the initialization, how much to model development? Relative Reduction in SST Forecast Error ECMWF Seasonal Forecasting Systems TOTAL GAIN OC INI MODEL TOTAL GAIN OC INI MODEL Half of the gain on forecast skill is due to improved ocean initialization ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting Balmaseda et al 200, OceanObs 9

9 Importance of Initialization Atmospheric point of view: Boundary condition problem Forcing by lower boundary conditions changes the PDF of the atmospheric attractor Loaded dice Oceanic point of view: Initial value problem Prediction of tropical SST: need to initialize the ocean subsurface. o Emphasis on the thermal structure of the upper ocean o Predictability is due to higher heat capacity and predictable dynamics ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 0

10 Need to Initialize the subsurface of the ocean 2OC Isotherm Depth Eq Anomaly SST Eq Anomaly ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting

11 Initialization Problem: Production of Optimal I.C. Optimal Initial Conditions: those that produce the best forecast. Need of a metric: lead time, variable, region (i.e. subjective choice) Usually forecast of SST indices, lead time -6 months Theoretically, initial conditions should represent accurately the state of the real world and project into the model attractor, so the model is able to evolve them. Difficult in the presence of model error Practical requirements: Consistency between re-forecasts and real time fc Current Priorities: Need for historical ocean reanalysis o Initialization of SST and ocean subsurface. o Land/ice/snow ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 2

12 Dealing with model error: Hindcasts time Ocean reanalysis Real time Probabilistic Coupled Forecast Coupled Hindcasts, needed to estimate climatological PDF, require a historical ocean reanalysis Consistency between historical and real-time initial conditions is required. Hindcasts are also needed for skill estimation ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 3

13 Information to initialize the ocean Ocean model Plus: SST Atmospheric fluxes from atmospheric reanalysis Subsurface ocean information Time evolution of the Ocean Observing System XBT s 60 s Satellite SST Moorings/Altimeter ARGO ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 4

14 How do we initialize the ocean? To a large extent, the large scale ocean variability is forced by the atmospheric surface fluxes. Different ocean models forced by the same surface fluxes will produce similar tropical variability. Daily fluxes of heat (short and long wave, latent, sensible heat), momentum and fresh water fluxes. Wind stress is essential for the interannual variability.. Constrained by SST: Fluxes from atmospheric models have large systematic errors and a large unconstrained chaotic component 2. Constrained by SST+ Atmos Observations: Surface fluxes from atmospheric reanalysis Reduced chaotic component. But still large errors/uncertainty 3. Constrained by SST+Atmos Observations+Ocean Observations: Ocean re- analysis Changing observing system and model error ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 5

15 Uncertainty in Surface Fluxes: Need for Data Assimilation Large uncertainty in wind products lead to large uncertainty in the ocean subsurface Equatorial Atlantic: Taux anomalies ERA5/OPS ERA40 The possibility is to use additional Equatorial Atlantic upper heat content anomalies. No assimilation information from ocean data (temperature, others ) Questions:.Does assimilation of ocean data constrain the ocean state? YES 2.Does the assimilation of ocean data improve the ocean estimate? YES 3.Does the assimilation of ocean data improve the seasonal forecasts. YES Equatorial Atlantic upper heat content anomalies. Assimilation ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 6

16 Depth (metres) The Assimilation corrects the ocean mean state 0 HOPE gcm:: 000 ICODE=78 contoured every XXX Zonal section at 0.00 deg N Plot resolution is.4063 in x and 0 in y ( 3 day mean) difference from 0 ( 3 day mean) Mean Assimation Temperature Increment Interpolated in y Equatorial Pacific (x) O E 00 O E 50 O E 60 O W 0 O W 60 O W 0 O W Longitude MAGICS 6.9. hyrokkin - neh Tue Jul 25 9:9: Data assimilation corrects the slope and mean depth of the equatorial thermocline z Free model Data Assimilation ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 7

17 Ocean Observing System Data coverage OBSERVATION for MONITORING June 982 X B T p r o b e s : p r o f i l e s 60 E 20 E W 60 W 0 60 N 60 N Changing observing system is a challenge for consistent reanalysis 30 N 30 N S 30 S Data coverage for Nov S 60 S 60 E 20 E W 60 W 0 60 E 20 E W 60 W 0 60 N 60 N Today s Observations will be used in years to come 30 N N 30 S 30 S 60 S 60 S Moorings: SubsurfaceTemperature ARGO floats: Subsurface Temperature and Salinity + XBT : Subsurface Temperature 60 E 20 E W 60 W ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 8 0

18 Impact of data assimilation on the mean EQATL Depth of the 20 degrees isotherm -70 Assim of mooring data ega8 omona.assim_an0 edp omona.assim_an0 CTL=No data Time PIRATA Large impact of data in the mean state leading to spurious variability This is largely solved by the introduction of bias correction ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 9

19 Latitude Need to correct model bias during assimilation Number of Temperature Observations Depth= meters x b a a x b f f b f K[ y H( x L[ y H( x f b f f b )] f )] 8*0 4 6*0 4 ALL XBT CTD MOOR ARGO ALL assimilated There is a model for the time evolution of the bias f f b k b k bk 4*0 4 2*0 4 This is an important difference with respect to the atmos data assimilation, where FG is assumed unbiased Time Temperature Bias Estimation from Argo: 300m-700m Without explicit bias correction changes in the observing system can induce Spurious signals in the ocean reanalysis Non-stationarity of the forecast bias, leading to forecast errors. Ideally, this information should be propagated during the forecast (for this the FG model and FC model should be the same, e.i. coupled model) 50N 0 50S 00E 60W 60W Longitude (C/h): Min= -.2e-03, Max= 7.5e-04, Int= 4.0e e e e e-05.2e e-04 Temperature Bias Estimation from WOA05: 300m-700m ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 20

20 Ocean Initialization at the ECMWF Ocean Reanalysis System 4 (ORAS4): 958 to present. 5 ens members Main Objective: Initialization of seasonal forecasts Historical reanalysis brought up-to-date => Useful to study and monitor climate variability Operational ORA-S4 NEMO-NEMOVAR ERA-40 daily fluxes ( ) and ERA-Interim thereafter Retrospective Ocean Reanalysis back to 958, 5 ensemble members Multivariate offline+on-line Bias Correction (pressure gradient, Temp,Sal, offline from recent period ) Assimilation of SST, temperature and salinity profiles, altimeter sea level anomalies an global sea level trends Balance constrains (T/S and geostrophy) Sequential, 0 days analysis cycle, 3D-Var FGAT. Incremental Analysis Update ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 2 Mogensen et al 202, Balmaseda et al 202

21 Depth (m) Latitude Depth (m) Depth (m) Depth (m) Time correlation with altimeter SL product rms EQ2 Potential Temperature 0 CNTL: NoObs rms EQIND Potential Temperature 0 NEMOVAR T+S rms EQ2 Potential Temperature rms TRPAC Potential Temperature N 40N 20N rms EQIND Potential Temperature ORAS4 T+S+Alti rms GLOBAL Potential Temperature correl (): fe5x sossheig ( ) CNTL NEMOVAR TS ORAS4 (TS+Alti) rms TRPAC Potential Temperature 20S 40S 60S E 60W 60W Longitude (ndim): Min= -0.37, Max=.00, Int= rms GLOBAL Potential Temperature ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 22

22 Impact on Seasonal Forecast Skill Consistent Improvement everywhere. Even in the Atlantic, traditionally challenging area ORAS4 CNTL ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 23

23 Quantifying the value of observational information The outcome may depend on the coupled system In a good system information may be redundant, but not detrimental. If adding more information degrades the results, there is something wrong with the methodology (coupled/assim system) Experiments conducted with the ECMWF S3 Balmaseda and Anderson 2009, GRL SST (SYNTEX System Luo et al 2005, Decadal Forecasting Keenlyside et al, 2008) SST+ Atmos observations (fluxes from atmos reanalysis) SST+ Atmos observations+ Ocean Observations (ocean reanalysis) ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 24

24 Initialization and forecast drift Different initializations produce different drift in the same coupled model. Warm drift in ALL caused by Kelvin Wave, triggered by the slackening of coupled model equatorial winds SST only has very little equatorial heat content, and the SST cool s down very quickly. ALL ATMOS+SST SST only SST+ATMOS seems balanced in this region. Not in others Sign of non linearity: The drift in the mean affects the variability ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 25

25 NINO3 NINO3.4 NINO4 TRPAC EQIND IND IND2 NSTRATL EQATL % Impact of real world information on skill: Increase (%) in MAE of SST forecasts from removing external information adding observational (-7 months) information 30 Reduction in Error (MAE) in SST SF by OC DATA 0 WINDS 5 DATA+WINDS The additional information about the real world improved the forecast skill, except in the Equatorial Atlantic Optimal use of the observations needs more sophisticated assimilation techniques and better models, to reduced initialization shock ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 26

26 Assessing the Ocean Observing System (S3). No observation system is redundant Not even in the Pacific, where Argo, moorings and altimeter still complement. Lessons for other basins? From Balmaseda and Anderson 2009 See also Fujii et al There were obvious problems in the Eq Atlantic: model error, assimilation, and possibly insufficient observing system Important to bear in mind. The assessment depends on the quality of the coupled model 2. Need records long enough for results to be significant => any observing system needs to stay in place for a long time before any assessment is possible. ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 27

27 Seasonal Forecasts Approach Some caveats SST FC bias CFS.v2 Kumar et al 20 MWR Non-stationary model error. It depends on starting date. For example, seasonal cycle dependence, which is known. There are other unknown dependences Drift depends of lead time. A large number of hindcasts is needed. This is even more costly in decadal forecasts. Initialization shock can be larger than model bias Non-linearities and non-stationarity can sometimes render the aposteriori calibration invalid ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 28

28 Perceived Paradigm for initialization of coupled forecasts Real world Model world Medium range Being close to the real world is perceived as advantageous. Model retains information for these time scales. Model attractor and real world are close? Seasonal? Decadal or longer Need to initialize the model attractor on the relevant time and spatial scales. Model attractor different from real world. At first sight, this paradigm would not allow a seamless prediction system. Seasonal traditional approach as in Medium range BUT (see next slide) Not clear how to achieve initialization in model attractor Anomaly Initialization (decadal forecasts, Smith et al 2007) Full initialization with coupled models of the slow component only Other more sophisticated (EnKF, coupled DA, weakly coupled DA) ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 29

29 Seasonal Approach Anomaly Initialization Long coupled integration Model climatology +observed anomaly As Medium range but: Model bias taken into account during DA. A posteriori calibration of forecast is needed. Calibration depends on lead time. The model for first guess during the initialization is different from the forecast model. Bias correction estimated during initializaiton can not be applied during the forecasts x b a a x b f f b f L[ y K[ y H( x H( x f b f f b )] f )] The model climatology does not depend of forecast lead time. Cheaper in principle. Hindcasts are still needed for skill estimation Acknowledgment of existence of model error during initialization. Model error is not corrected ( bias blind algorithm ): f K[( y y) H( x x)] ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 30 x a x f

30 Anomaly Initialization (Cont) Two flavours. One-Tier anomaly initialization (Smith et al 2007). Ocean observations are assimilated directly. Background error covariance formulation derived from coupled model (EOFs, EnOI, EnKF). Emphasis on large spatial scales 2. Two-Tier anomaly initialization (Pohlmann et al 2009). Nudging of anomalies from existing ocean re-analysis. The spatial structures are those provided by the source reanalysis. It assumes quasi-linear regime. Limitations Sampling: how to obtain an observed climatology equivalent to the model climate? ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 3

31 phase space Initialization Shock and Skill Model Climate d e Initialization shock non-linear interactions important c b Real World a L Forecast lead time ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 32

32 phase space Initialization Shock and non linearities Model Climate b non-linear interactions important Empirical Flux Corrections Real World a Forecast lead time ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 33

33 Comparison of Strategies for dealing with systematic errors in a coupled ocean-atmosphere forecasting system as part of the EU FP7 COMBINE project Nature climate Flux correction Normal initialisation Anomaly initialisation Model climate Magnusson et al. 202, Clim Dyn Submitted. Also ECMWF Techmemo 658 Magnusson et al. 202, Clim Dyn Sumbitted. Also ECMWF Techmemo 676 ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 34

34 Coupled model error SST bias: model - analysis 0m winds: model - analysis Part of the error comes from the atmospheric component (too strong easterlies at the equator) The error amplifies in the couped model (positive Bjerkness feedback) Possibility of flux correction From Magnusson et al 202 Clim Dyn. Submitted ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 35

35 Different mean states Analysis Coupled Free Coupled Ucor Coupled UHcor Ucor: surface wind is corrected when passed to the ocean UHCor: surface wind and heat flux are corrected when passed to the ocean ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 36

36 Comparison of Forecast Strategies: Drift Nino 3 SST Drift -4 month forecast Analysis Full Ini Anomaly Ini U Correction U+H correction ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 37

37 Comparison of Forecast Strategies: Variability FC sdv / AN sdv Analysis Full Ini Anomaly Ini U Correction U+H correction ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 38

38 Nino3.4 SST forecasts November 995 November 998 Full Initialization Anomaly Initialisation U-flux correction U- and H-flux correction Linus Magnusson et al. ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 39

39 Impact on Forecast Skill (SST and Precip) Persistence Full Ini Anomaly Ini U Correction U+H correction The impact of initialization/forecast strategy depends on the region When the mean state matters (convective precip), the anomaly Initialization underperforms ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 40

40 What about Coupled Initialization? Advantages: Hopefully more balanced ocean-atmosphere i.c and perturbations. Important for tropical convection Framework to treat model error during initialization and fc If the FG and FC models are the same, the (3D) bias correction estimated during the initialization can (should) be applied during the forecast. Consistency across time scales (seamlessness): currently, weather forecasts up to 0 days use extreme flux correction, since SST is prescribed. For longer lead times a free coupled model is used. More gradual transition? Current Approaches Weakly Coupled Data assimilation: FG with coupled model, separate DA of ocean and atmos. Example is NCEP with CFSR: coupled reanalysis to initialized and calibrate seasonal forecasts Strongly Coupled Data assimilation: Coupled FG, Coupled Covariances. Usually EnKF Challenges: Different time scales of ocean atmosphere. Long window weak constrain? Cross-covariances. Ensemble methodology more natural? ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 4

41 Summary Seasonal Forecasting (SF) of SST is an initial condition problem Assimilation of ocean observations reduces the large uncertainty (error) due to the forcing fluxes. Initialization of Seasonal Forecasts needs SST, subsurface temperature, salinity and altimeter derived sea level anomalies. Data assimilation improves forecast skill. Data assimilation changes the ocean mean state. Therefore, consistent ocean reanalysis requires an explicit treatment of the bias The separate initialization of the ocean and atmosphere systems can lead to initialization shock during the forecasts. A more balance coupled initialization is desirable, but it remains challenging. Initialization and forecast strategy go together. The best strategy may depend on the model. The anomaly initialization used in decadal forecasts can have problems in seasonal ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 42

42 Evolution over the last 365 days ECMWF Seminar 202 Initialization Strategies in Seasonal Forecasting 43

43 Latest Conditions