2. Climate Risk Management: Seasonal Climate Prediction
Sylwia Trzaska
IRI: Steve Zebiak, Lisa Goddard, Simon Mason, Tony Barnston, Madeleine Thomson, Neil Ward, Ousmane N’Diaye and many others
ECMWF: Magdalena Balmaseda
Meteo-France: J.-P. Céron

3. Climate Risk Management: Seasonal Climate Prediction
IRI: Linking Climate and Society
Climate Prediction
Seasonal Climate Forecast
Use of Ocean Data
Importance of ARGO data
Climate Information & Climate Prediction Tool

4. Linking Science to Society

5. The IRI’s mission Motivation Activities
To enhance society's capability to understand, anticipate and manage the impacts of seasonal climate fluctuations, in order to improve human welfare and the environment, especially in developing countries.
Motivation
Research and practical experience already gained with many collaborators has convinced us that achievement of global (sustainable) development goals is strongly dependent on recognition of the role of climate, and effective use of climate information in policy and in practice.
Activities
With many partners, developing the capacity to manage climate-related risks in key climate-sensitive sectors: agriculture, food security, water resources management, public health, disasters
Climate knowledge/information as a resource
! Uptake of climate information is NOT trivial

6. Relationship of overall GDP, agricultural GDP and rainfall in Ethiopia (Grey and Sadoff, 2005)

7. Semi-arid areas in Africa prone to negative, anti-development outcomes
Figure 1
Figure 2
Semi-arid areas in Africa prone to negative, anti-development outcomes
hunger (figure 1),
disasters (figure 2),
epidemic disease outbreaks (figures 3-4).
climate impacts across many sectors =>ripple through the economy
Figure 3
Figure 4

8. Climate Prediction

9. Example: Time Scales of Variability

10. Weather & Climate Prediction
Initial & Projected Atmospheric Composition
Initial & Projected State of Ocean
Climate Change
Initial & Projected State of Atmosphere
Current Observed State
Decadal
Uncertainty
Time Scale, Spatial Scale

11. Basis of Seasonal Climate Prediction:
Changes in boundary conditions, such as SST and land surface characteristics, can influence the characteristics of weather (e.g. strength or persistence/absence), and thus influence the seasonal climate.
Will emphasize SST, as this is overall the dominant factor in regional predictability.
So, want to predict changes in SST field, and use those to predict changes in the seasonal climate

12. Influence of SST on tropical atmosphere
- SST gradients/pressure gradients drive low-level winds
- Winds want to move towards warmest waters (lower pressures)  convergence  atmos. Heating & upper level divergence, influencing remote response

13. What we can foresee now
Effective management of climate related risks (opportunities) for improved:
Agricultural production
Stocking, cropping calendar, crop selection, irrigation, insurance, livestock/trade
Water resource management
Dynamic reservoir operation, power generation, pricing/insurance
Food security
Local, provincial, regional scales
Public health
Warning, vaccine supply/distribution, surveillance measures,…
Natural resource management
Forests/fire, fisheries, water/air quality
Infrastructure development

14. Epidemic Malaria = Interannual variability => Climate control
Example 1: Malaria Early Warning System
Epidemic Malaria = Interannual variability => Climate control
Temperature:
“highland malaria”
Precipitation:
“desert-fringe malaria”
Awareness, use of prevention measures (bednets)
(timely) Availability & access to health care/diagnostic/treatment
Lags in intervention implementation (esp. if remote resources)

15. Malaria and Rainfall
The disease is highly seasonal and follows the rainy season with a lag of about 2 months

16. Biological Mechanism for the Relationship of Malaria Incidence to Rainfall
Increases in rainfall => increase breeding site availability => increase in malaria vector populations
Increases in rainfall ~ increases in humidity => higher adult vector survivorship => greater probability of transmission.
Precise numerical models of host/vector/parasite cycle and/or population/epidemics exist but require very fine environmental data (breeding sites, rainfall, temperature, humidity…)
Scale/info mismatch between environmental conditions forecast/monitoring and such models
Frequent lack of evidence of links btwn large scale epidemics and climate for public health services
Many other factors: accuracy of the data, access to drugs/health services, intervention policies, population migration

17. Incidence-based decisions
Purchase of drugs
interventions
Report national level
Threshold in
malaria cases
Drugs/interventions available at district

18. Rainfall-based decisions
Report national level
Purchase of drugs
interventions
Threshold in
Rainfall amounts
Drugs/interventions available at districts

19. Forecast-based decisions
Drugs/interventions available at national level
Purchase of drugs
interventions
Report national level
malaria monitoring
Predicted rainfall
Rainfall monitoring
Drugs/interventions available at districts
Match between scale/accuracy/confidence/lead
of the information and decision/interventions
More effective use of limited resources
Interactions with end-users are crucial

20. Exemple 2: Senegal River Basin
Manantali Dam, Senegal River
Multi-user dam
Hydropower,
flow regulation: flood control, irrigation,
water for flood recession agriculture,
minimum ecological impact

21. Manantali Dam, Senegal River
August 20 – reservoir management decision for water release for traditional agriculture Sept-Oct, given electricity and irrigation demands Sept-July
Management strategy using Aug-Oct seasonal forecast made at Meteo-France end of July
=> Forecast water stock in the reservoir at the end of the monsoon season

22. Seasonal Forecasts

23. Methods of Seasonal Forecats
Statistical Methods: identify statistical relationships in the past
Ex. 3 SST indices used in stat forecast of seasonal rainfall in JAS in the Sahel
Ex. Rainfall in East Africa vs Nino3.4 SST
Pbs.
Spurious relationship (SST correlated by chance)
Instability of relationships (e.g. Sahel-ENSO)

24. Methods of Seasonal Forecats
Dynamical Methods: General Circulation Models
Constrains on computing time= constrains on resolution
Typical grid size ~ 250x250km
Time step 15min
Sources of error :
Scale of numerous processes << resolved scale
Models of different sub-systems developped separately – pb when coupling

25. Weather & Climate Prediction
Initial & Projected Atmospheric Composition
Initial & Projected State of Ocean
Climate Change
Initial & Projected State of Atmosphere
Current Observed State
Decadal
Uncertainty
Time Scale, Spatial Scale

26. What probabilistic forecasts represent
Climatological Average
“SIGNAL”
Forecast Mean
The SIGNAL represents the ‘most likely’ outcome, but quantifying the UNCERTAINTY is an important part of the forecast. The UNCERTAINTY represents the internal atmospheric chaos, uncertainties in the boundary conditions, and random errors in the models.
“UNCERTAINTY”
Same figure, slightly different description.
Means, shift, uncertainty.

27. Probabilistic forecasts
Near-Normal
Below Normal
Above Normal
Historical distribution
FREQUENCY
Forecast distribution
Breakpoints of categories are determined by historical observations. The probabilities of this distribution are the climatological probabilities.
Forecast distribution (say of the ensemble members at a point, or over a region) represent a shift in the range of possibilities. Now categorical probabilities are not equal – they differ from climatology.
NORMALIZED RAINFALL
Historically, the probabilities of above and below are Shifting the mean by half a standard-deviation and reducing the variance by 20% changes the probability of below to 0.15 and of above to 0.53.

28. Example of seasonal rainfall forecast
Regional
3-month average
Probabilistic
3-category forecasts : climatological probabilities, by construction, are equal chances for any category
This map shows the likelihood/probability for seasonal climate to fall within each of the categories (must sum to 100%).
Far from the detailed “answer” most users are looking for
Regional scale, in part, determined by resolution of the global models – shown by the size of the squares on the map.

29. Regional Outlook Forum
PRES-AO (9)
GHACOF (18)
PRES-AC (3)
SARCOF (10)
PRESANOR
Operational Seasonal Climate Forecasts for main rainy seasons:
Country level
Consensus regional forecasts released
Blend of statistical and dynamical methods
E.g. PRESAO

30. Optimizing probabilistic information
Reliably estimate the good uncertainty
-- Minimize the random errors
e.g. multi-model approach (for both response & forcing)
Eliminate the bad uncertainty
-- Reduce systematic errors
e.g. MOS correction, calibration

31. Use of Ocean Data

32. IRI DYNAMICAL CLIMATE FORECAST SYSTEM
2-tier
OCEAN ATMOSPHERE
GLOBAL
ATMOSPHERIC
MODELS
ECPC(Scripps)
ECHAM4.5(MPI)
CCM3.6(NCAR)
NCEP(MRF9)
NSIPP(NASA)
COLA2
GFDL
PERSISTED
GLOBAL
SST
ANOMALY
Persisted
SST
Ensembles
3 Mo. lead 10
POST
PROCESSING
MULTIMODEL
ENSEMBLING 24 24 10
FORECAST SST
TROP. PACIFIC
(multi-models, dynamical
and statistical)
TROP. ATL, INDIAN
(statistical)
EXTRATROPICAL
(damped persistence) 12
Forecast
SST
Ensembles
3/6 Mo. lead
Only ECHAM4.5 and CCM3.6 are run at IRI. The other models are run at by their developing institution – with the exception of the NCEP-MRF9 model, which is run by QCCA (Queensland Centre for Climate Applications) in Australia. 24 24 30 12 30 30

33. ECMWF: Weather and Climate Dynamical Forecasts
Delayed Ocean Analysis ~11 days
Real Time Ocean Analysis ~8 hours New
ECMWF:
Weather and Climate Dynamical Forecasts
10-Day
Medium-Range
Forecasts
Seasonal
Monthly
Atmospheric model
Wave model
Ocean model
M.A. Balmaseda ( ECMWF)

34. Data Assimilation struggles to correct the systematic error
Most common practice for initialization of coupled forecasts: Uncoupled initialization of ocean and atmosphere
Atmosphere Initialization (from NWP or AMIP):
atmos model +(atmos obs+assimilation 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
M.A. Balmaseda ( ECMWF)

35. Real Time Ocean Observations
Moorings
ARGO floats
XBT (eXpandable BathiThermograph)
Satellite
SST
Sea Level
M.A. Balmaseda ( ECMWF)

36. Ocean Observing System
Data coverage for June 1982
Changing observing system is a challenge for consistent reanalysis
Data coverage for Nov 2005
Today’s Observations will be used in years to come
▲Moorings: SubsurfaceTemperature
◊ ARGO floats: Subsurface Temperature and Salinity
+ XBT : Subsurface Temperature
M.A. Balmaseda ( ECMWF)

37. Main Objective: to provide ocean Initial conditions for coupled forecasts
Ocean reanalysis
Real time Probabilistic Coupled Forecast
time
Coupled Hindcasts, needed to estimate climatological PDF, require a historical ocean reanalysis
Consistency between historical and real-time initial initial conditions is required
Quality of reanalysis affects the climatological PDF
M.A. Balmaseda ( ECMWF)

38. Importance of ARGO Data

39. Atlantic Anomalies: 2005 versus 2006
No Data assimilation:
Aug 2005
With subsurface data (mainly ARGO) the anomaly is stronger.
Aug 2005
Aug 2006
The temperature anomaly in the North Southtropical Atlantic is much weaker in 2006.
M.A. Balmaseda ( ECMWF)

40. Ocean Observing System Experiments (OSES): Effect of Argo
All – NoArgo:
mean
Surface Salinity (CI=0.1psu)
M.A. Balmaseda ( ECMWF)

41. Impact on Forecast Skill
No Data/ Data assim
Ocean Data Assimilation improves forecast skill in the Equatorial Pacific, especially in the Western Part
M.A. Balmaseda ( ECMWF)

42. Misc. TOGA-TAO failure in E Pacif June-Oct 2006
Long x depth cross sections in the Pacific 2S-2N
Nov 2006
June 2006
July 2006 ….

43. Research!

44. Loss of skill in AGCM due to imperfect predictions of SST
Simulation
Hindcast (persisted SSTa)
Loss of skill in AGCM
due to imperfect
predictions of SST
(Goddard & Mason ,Climate Dynamics, 2002)
Dominant pattern of
precipitation error
associated with
dominant pattern of
SST prediction error
Again, we have the potential to predict West Africa variability, if we can get the Atlantic SSTs right. If not, we lose that potential.
Figure (upper) shows correlation skill between model simulated precipitation and observations. On left the SSTs are as observed – good pcp skill. On right, the SSTs are persisted (simplest prediction, but better than most evolving SST forecasts) – most skill is lost.
Figure (lower) shows canonical correlation (dominant patterns of co-variability) between SST differences, from forcings used in experiments shown above, and differences in resulting precipitation fields. Message is that errors in the equatorial Atlantic are most to blame for error in predicted rainfall. So, for example, if we don’t predict the equatorial Atlantic to warm during the forecast season, we will fail to predict the wet conditions that are likely to develop.

45. Climate Variability in the Atlantic Sector
CLIVAR TAV
… so how accurate are the observations?

46. Surface Temperature composites of 4 phases of QB component (model)
Interannual Climate Variability in the South Atlantic: Linking Tropics and Subtropics
Coupled air-sea variability in S. Atlantic
Model: UCLA AGCM coupled to uniform depth mixed-layer ocean in the Atlantic, 34 yr run
LF 5yr QB
obs
SST
MTM spectrum
model
Similar spatial patterns and temporal scales despite absence of ocean dynamics in the model
5yr and QB component on red noise
Surface Temperature composites
of 4 phases of QB component (model)
Leading mode of SST- SLP covariability
Quasi-biennial component
mature
transition
Anomaly propagation from extratropics to tropics (also seen in obs), strongly tied to the seasonal cycle of convection
SST forcing on atmosphere in the tropics, atmospheric forcing of the SST in the subtropics via atmospheric bridge
Reversed surface flux feedback in the east vs west and ITCZ
East - dominated by shallow clouds - SST anomalies generated and maintained by SST- cloud/radiation feedback, damped by SST- wind/evaporation
West and ITCZ - deep convection - SST anomalies generated and maintained by SST- wind/evaporation, damped by SST- cloud/radiation feedback
Trzaska S., A.W. Robertson, J.D. Farrara and C.R. Mechoso, J. Climate, 2006: sub judice

47. CONCLUSION
Skillful climate prediction requires skillful SST prediction in the tropics.
Skillful SST prediction requires accurate GCMs
GCMs can be used for prediction and process studies if they do the right thing.
 We can really only assess what they do right and wrong if the observations used for verification are accurate with a good spatial and temporal coverage

48. Climate Information http://iri.columbia.edu
Data Library: numerous data incl. seasonal forecast, mapping &analysis tools
Tutorials and Manuals
Climate Prediction Tool