Seasonal Forecasting Current Status and Future Prospects

1
Seasonal Forecasting Current Status and Future
Prospects

• Magdalena A. Balmaseda

2

• Typical Single Forecasting System
• Better than persistence
• Non reliable
• RMS gt Spread

3
End to End Forecasting System
4
Initialization

• Uncertainties
• Different Strategies

5
Most common practiceUncoupled initialization
of ocean and atmosphere

• Atmosphere Initialization (from NWP or AMIP)
• atmos model (atmos obsassimilation
  system)prescribed SST
• Ocean Initialization
• ocean model ocean obs assimilation system
  prescribed surface fluxes
• So far mainly subsurface Temperature, and
  altimeter.
• Salinity from ARGO is used in the new ECMWF
  system.
• Atmospheric Fluxes are a large source of
  systematic error in the ocean state.
• Data Assimilation struggles to correct the
  systematic error

6
Consistency between historical and real-time
initial conditions
Quality of reanalysis affects the climatological
PDF
To be taken into account when doing OSEs
7
Data coverage for Nov 2005
The Observation coverage is getting global
Data coverage for June 1982
and although a changing observing system is a
challenge for consistent reanalysis
?Moorings SubsurfaceTemperature ? ARGO floats
Subsurface Temp and Salinity XBT Subsurface
Temperature
Todays Observations will be used in the years to
come
8
Data Assimilation reduces uncertainty in forcing
fluxes
but, what about uncertainty in the
assimilation procedure?And uncertainty in
variables not so well observed?
9
ENACT
Multi-model, multi assimilation methods ocean
reanalysis
10
We can now quantify the current level of
uncertainty in theclimate reanalysis
Uncertainty/Variability in T300
Uncertainty/Variability in S300

• Smallest Uncertainty is for T300 in the Eastern
  Pacific
• But still is 25 of the signal!
• Assimilation is better in most cases

• The Uncertainty is S300 is gt 50
• Salinity from ARGO more than welcome

Ongoing effort?
11
Does Ocean data assimilation also improves the
forecast skill? Sometimes yes(Alves et al 2003)
beginners luck?
12
Initialization
Uncoupled Most common
Other Strategies

• Advantages
• It is possible
• The systematic error during the initialization is
  small(-er)
• It can be used in a seamless system
• Disadvantages
• Model is different during the initialization and
  forecast
• Possibility of initialization shock
• No synergy between ocean and atmospheric
  observations

• Full Coupled Initialization
• No clear path for implementation in operational
  systems
• Need of a good algorithm to treat systematic
  error
• Coupled Anomaly Initialization
• Weakly-coupled initialization?
• Atmosphere ocean mixed layer
• Ocean Atmosphere boundary layer
• Simplified coupled models?

13
Expamle of Anomaly Initialization DePreSys
DePreSys Decadal Prediction System Glosea
Seasonal Prediction (full
uncoupled initialization)
1y forecasts DePreSys is comparable Glosea
Courtesy of Doug Smith DePreSys Group
14
Coupled Model
15
So far, so good
Can we guarantee or guarantee this standard? Can
we extend the lead time?
16
NINO 3.4
The atmosphere model has the strongest impact on
the forecast performance (after month 3)
NINO3.4

• Typical ensemble from perturbations to initial
  conditions ( wind/SST)
• Atmospheric model version
• Ocean model (parameters,)
• Different assimilation parameters

17
The atmosphere model has the strongest impact on
the forecast performance (after month 3)

• Known Sensitivities
• Horizontal Resolution (T159 does better than T95)
• Vertical Resolution (62 levels better than 40
  levels)
• Aerosol climatology (Atlantic Sector)
• Convection/clouds properties/radiation/boundary
  layer physics
• And the list goes on
• Non Linearities and Time-Scale Interaction
• Bias affects the interannual variability
• A posteriori bias correction or flux correction?
• Intraseasonal influence the interannual
• Stochastic
  Optimal?

18
Possible non-linearity influence of mean state
in interannual variability
Warmer bias leads to weaker interannual
variability
19
10m zonal wind bias reduction T95 vs T159 MJJ
(1987-2003) month2-4
Cy 29r1 uncoupled
Cy 29r1 coupled
T 95
T 159
20
The warmer the bias, the strongest the spring
barrier
21
Possible non-linearity influence of the
intrasesonal variability in interannual
variability
These models have very poor MJO is that the
reason for weak interannual variability?
22
Ensemble generation Adding perturbations on line

• Atmosphere model is deficient in reproducing some
  parts of the observed spectrum (example MJO) that
  may influence the ocean interannual variability
• Add perturbations to the coupled model to account
  for known deficiencies (on a similar spirit as
  the Stochastic Physics)
• How should be these perturbations?

• If simplified MJO
• Is there an optimal period?
• Is there a right position and spatial structure?
• Should the perturbations propagate? At which
  speed?
• May be it is needed to use OPTIMAL Methods
• Stochastic optimals
• Forcing Singular Vectors

23
Adding perturbations on line
Results Ensemble spread -Slight growth only
visible at months 2-3 -The saturation value of
the spread remains unchanged. RMS error and
Correlation -Improved for april/july starts.
Significant?
24
Stochastic Physics and Stochastic Perturbations
Need for constraints
25
MJO Experiments
Impact of the vertical resolution of the
mixed-layer model
MJO CORRELATION Effect of Coupling
MJO CORRELATION Effect of ML resolution

Woolnough et al, Submitted
26
Ensemble Generation (single model)

• Most common strategy is the initial Condition
  perturbation
• Wind perturbations, SST perturbations, sequential
  days
• Not designed to be optimal
• Online perturbations?
• Stochastic Optimals
• Stochastic Physics (different flavours)
• Perturbed parameters
• Do we need more pptimal methods for ensemble
  generation
• Coupled Breeding Vectors? , SV /SO from
  simplified models?

27
The multi-model ensemble

• DEMETER successful research on Multimodel for
• seasonal forecasts.
• EURO-SIP The operational European Multimodel
• ECMWF-UKMO-MeteoFrance
• ENSEMBLES continuation of Demeter.
• Data Distribution
• Multi-annual predictions
• The Future distributed global multimodel?

28
Application to the Real Time Multimodel
EUROSIP ECMWF-UKMO-MeteoFrance
29
ENSEMBLES stream 1 multi-model ensemble system

• ENSEMBLES system 6 coupled GCMs running at ECMWF

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.

• Hindcast production period for 1991-2001

30
ENSEMBLES stream 2 multi-model ensemble system

• ENSEMBLES system 7 coupled GCMs running at ECMWF

9-member ensembles ERA-40 atmosphere and soil
initial conditions 4 seasonal (7 months),1 annual
(14 months) runs per year At least, one
multi-annual run every five years, except 4 per
year for DePreSys Realistic boundary forcings
GHGs, aerosols, solar forcing, damped volcanic
aerosols, etc.

• Hindcast production period for 1960-2001

31
Data archiving and public dissemination
32
Calibration and Combination of (multi-) model
outputForecast Assimilation (or Bayesian
techniques)
33
Conceptual forecast framework
Forecast Assimilation
Data Assimilation
Stephenson et al 2005
34
Example S. American rainfall anomaly composites
Forecast Assimilation
Obs
Multi-model
DEMETER 3 coupled models (ECMWF, CNRM,
UKMO) 1-month lead Start Nov DJF ENSO
composites 1959-2001 16 El Nino years 13 La
Nina years
ACC0.51
ACC0.97
ACC1.00
ACCAnomaly Correlation Coefficient Spatial
correlation of map with obs map
ACC0.28
ACC0.82
ACC1.00
Coelho et al 2005
(mm/day)
35
Why South American rainfall?

• Agriculture
• Electricity More than 90 produced by
  hydropower stations
• e.g. Itaipu (Brazil/Paraguay)
• Worlds largest hydropower plant
• Installed power 12600 MW
• 18 generation units (700 MW each)
• 25 electricity consumed in Brazil
• 95 electricity consumed in Paraguay

36
Itaipu
37
Example S. American rainfall anomaly composites
Forecast Assimilation
Obs
Multi-model
DEMETER 3 coupled models (ECMWF, CNRM,
UKMO) 1-month lead Start Nov DJF ENSO
composites 1959-2001 16 El Nino years 13 La
Nina years
ACC0.51
ACC0.97
ACC1.00
ACCAnomaly Correlation Coefficient Spatial
correlation of map with obs map
ACC0.28
ACC0.82
ACC1.00
Coelho et al 2005
(mm/day)
38
Application to the Real Time Multimodel
EUROSIP ECMWF-UKMO-MeteoFrance
39
DONE

• We are now able to perform historical ocean
  analysis and to quantify uncertainty
• Observation coverage is global
• Salinity is measured routinely
• Operational reliable seasonal forecasts at 6
  months
• Operational Multimodel
• Operational Calibration
• Forecast information applied to practical cases
• Extending the forecast range
• Longer range 1 year
• Shorter range The infrastructure is used for
  monthly forecast
• More comprehensive coupled models (waves,
  aerosols, )

40
TO DO LIST

• Continue exercise in climate reanalysis
• Quantifying uncertainty in forcing versus
  assim/model
• Higher resolution ocean models?
• Asses impact of salinity data. Methods to use
  geoid information
• Develop methods to deal with systematic error
• Initialization of soil moisture, snow, ice
• reanalysis and ensemble perturbations
• More comprehensive coupled models (ice, coupling,
  resolution)
• Data servers and data conventions
• ncdf to allow ensemble members and multi-model
  attributes
• Study relation between 1day error and climate
  errors

41
Challenges seamless prediction systems

• More interaction between ocean and atmosphere
  during the initialization
• Improve Single Model Forecasting system
  (resolution as well as reliability)
• Better initialization (including mixed layer
  initialization). Do we have data?
• Better models !!!
• Understanding physical processes gt
  parameterizations
• Increase resolution (when needed and possible)
• Use generalized stochastic physics to represent
  truncation errors
• More optimality in the ensemble generation
• Better calibration of probabilistic forecasts
  extreme events
• Extending the forecast range
• what is predictable y what is useful?

42
Questions

• Can the NWP benefit from developments in seasonal
  forecasting?
• Impact of coupling in MJO. Maybe others
• Can the Seasonal Forecasting Systems benefit from
  better NWP models?
• Look at relevant diagnostics in the short
  forecast range?
• Which current strategies/practices should be
  revised? (a-posteriori bias correction versus
  flux correction? separate initialization,?)
• Is (a truncated version of) a good NWP model a
  good seasonal/climate forecasting model?
• Is a good climate model a good seasonal
  forecasting model?
• Is there any reason why the NWP and the climate
  community should not find a meeting point

43
Spring barrier predictability
potential predictability ) seasonal recharge
oscillator ?
?
skill ECMWF operational forecast 1987-2001
?




Estimate of predictability with parameters and
noise properties from seasonal fit. From Gerrit
Burgers KNMI
44
Initialization Shock versus Bias
Warm Initial Trend
Cold bias
Courtesy of Antje Weisheimer