1 / 71

How Climate Can Be Predicted

How Climate Can Be Predicted. Why Some Predictability Exists, and How Predictions Can Be Made Accessing Data and Tools from IRI. Much of the predictable part of seasonal climate comes as a result of anomalies of sea surface temperature (SST) in tropical ocean basins.

ravery
Download Presentation

How Climate Can Be Predicted

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. How Climate Can Be Predicted Why Some Predictability Exists, and How Predictions Can Be Made Accessing Data and Tools from IRI

  2. Much of the predictable part of seasonal climate comes as a result of anomalies of sea surface temperature (SST) in tropical ocean basins.

  3. ENSO: The strongest source of tropical SST variation Total SST Anomaly of SST

  4. El Nińo La Nińa

  5. Normal El NińoLa Nińa

  6. Nińo3.4 region: 5ºN-5ºS, 120º-170ºW Stronger El Niño El Nińo Dec La Nińa StrongerLa Niña

  7. El Nińo episodes often begin in April, May or June, and end in about 10-12 months, in February, March, April, or May.

  8. http://iridl.ldeo.columbia.edu/SOURCES/.IRI/.Analyses/.ENSO-RP/.ver1950-2002/http://iridl.ldeo.columbia.edu/SOURCES/.IRI/.Analyses/.ENSO-RP/.ver1950-2002/ warm

  9. http://iridl.ldeo.columbia.edu/SOURCES/.IRI/.Analyses/.ENSO-RP/.ver1950-2002/http://iridl.ldeo.columbia.edu/SOURCES/.IRI/.Analyses/.ENSO-RP/.ver1950-2002/ . cold

  10. warm

  11. . cold

  12. warm

  13. . cold

  14. warm

  15. . cold

  16. Is the ENSO phase predictable?

  17. A Brief History of LDEO Model • LDEO1:Original Cane and Zebiak model (Cane et al., Nature, 1986) • LDEO2:LDEO1 plus coupled initialization (Chen et al., Science, 1995) • LDEO3:LDEO2 plus sea level data assimilation (Chen et al., GRL,1998) • LDEO4:LDEO3 plus statistical bias correction (Chen et al., GRL, 2000) • LDEO5:LDEO4 plus additional correction on SST (Chen et al., Nature, 2004) LDEO5 Forecast Skill LDEO4 0.8 --------------------------------------------------------------------------------------------------- LDEO2 LDEO3 0.7 --------------------------------------------------------------------------------------------------- LDEO1 --------------------------------------------------------------------------------------------------- 0.6 6 month lead; 1970-1985 --------------------------------------------------------------------------------------------------- 0.5 0.4 1995 2005 1985 1990 2000

  18. ***** poor forecast: late onset of El Nińo . poor forecast: early dissipation of El Nińo due partly to MJO poor forecast: Onset of La Nińa at unusual time of year

  19. ***** acceptable forecast: late onset of El Nińo . acceptable forecast: early dissipation of El Nińo due partly to MJO acceptable forecast: Onset of La Nińa at unusual time of year

  20. Correlation Skill for NINO3 forecasts Skill of LDEO3 (Zebiak-Cane) simple dynamical model, 1970-2000 for NINO3 Region Useful long-lead skill Northern Spring barrier Correlation between forecast and obs

  21. ENSO Predictability: Improvement from Mid-1980s to Today Improvements were large in the late 1980s, small to moderate in the 1990s, and not much in the 2000s. We do not know the upper limit of ENSO predictability. We still have a big problem predicting ENSO from the early part of a calendar year to the middle of that calendar year. The potential for better pre- dictions may be quite large, but it is also possible that it is only slightly better than what we can do now.

  22. Climate prediction designs: Statistical – based on historical observed data for the predictand (e.g. rainfall, temperature) and for relevant predictors (e.g. SST, atmospheric pressure). Dynamical – using prognostic physical equations 2-tiered systems (first predict SST, then climate). 1-tiered systems (predict ocean and atmosphere together)

  23. Prediction Systems: statistical vs. dynamical system ADVANTAGES Based on actual, real-world observed data. Knowledge of physical processes not needed. Many climate relationships quasi-linear, quasi-Gaussian ------------------------------------ Uses proven laws of physics. Quality observational data not required (but needed for val- idation). Can handle cases that have never occurred. DISADVANTAGES Depends on quality and length of observed data Does not fully account for climate change, or new climate situations. ------------------------------ Some physical laws must be abbreviated or statis- tically estimated, leading to errors and biases. Computer intensive. Stati- stical ------- Dyna- mical

  24. In Dynamical Prediction System: 2-tiered vs. 1-tiered forecast system ADVANTAGES Two-way air-sea interaction, as in real world (required where fluxes are as important as large scale ocean dynamics) -------------------------------------- More stable, reliable SST in the prediction; lack of drift that can appear in 1-tier system Reasonably effective for regions impacted most directly by ENSO DISADVANTAGES Model biases amplify (drift); flux corrections Computationally expensive ------------------------------ Flawed (1-way) physics, especially unacceptable in tropical Atlantic and Indian oceans (monsoon) 1-tier ------ 2-tier

  25. Climate forecasts need to be expressed probabilistically, because there is a wide distribution of possibilities even with our better-than-chance accuracy.

  26. What probabilistic forecasts represent Near-Normal BelowNormal AboveNormal Historical distribution (climatological distribution) (33.3%, 33.3%, 33.3%) FREQUENCY Forecast distribution (15%, 32%, 53%) NORMALIZED RAINFALL Historically, the probabilities of above and below are 0.33. 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. (Courtesy Mike Tippett)

  27. Abbreviating a predicted shift in the probability distribution: Terciles (Below normal,, near normal, above normal) Climatological probabilities= 1/3 33%33%33% Below|Near| Below|Near| Above Below|Near| | || |||||||.| || | | || | | |.| | | | | | | | | | Data: 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 Rainfall Amount (mm) (30 years of historical data for one station and season)

  28. Example of a climate forecast with a strong probability shift 10% 25% 65% Below| Near | Below| Near | Above Below| Near | 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 Rainfall Amount (mm)

  29. Example of a climate forecast with a weak probability shift 25% 35% 40% Below| Near | Below| Near | Above Below| Near | 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 Rainfall Amount (mm)

  30. Example of a climate forecast with no probability shift 33% 33% 33% Below| Near | Below| Near | Above Below| Near | 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 Rainfall Amount (mm)

  31. Tercile probabilities for various correlation skills and predictor signal strengths (in SDs). Assumes Gaussian probability distri- bution. Forecast (F) signal = (Predictor Signal) x (Correl Skill). *0.3 **0.04

  32. Predicting the atmospheric climate, based on the expected SST anomaly patterns.

  33. IRI’s Forecast System IRI is presently (in 2007) using a 2-tiered prediction system to probabilistically predict global temperature and precipitation with respect to terciles of the historical climatological distribution. We are interested in utilizing fully coupled (1-tier) systems also, and are looking into incorporating those. Within the 2-tiered system IRI uses 4 SST prediction scenarios, and combines the predictions of 7 AGCMs. The merging of 7 predictions into a single one uses two multi-model ensemble systems: Bayesian and canonical variate. These give somewhat differing solutions, and are presently given equal weight.

  34. IRI DYNAMICAL CLIMATE FORECAST SYSTEM 2-tiered 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 SCENARIOS TROP. PACIFIC: THREE (multi-models, dynamical and statistical) TROP. ATL and INDIAN (2 and 3 multi-models) EXTRATROPICAL (damped persistence) 12 Forecast SST Ensembles 3/6 Mo. lead 24 model weighting 24 30 12 30 30

  35. IRI DYNAMICAL CLIMATE FORECAST SYSTEM 2-tiered OCEAN ATMOSPHERE MULTIPLE GLOBAL ATMOSPHERIC MODELS ECPC(Scripps) ECHAM4.5(MPI) CCM3.6(NCAR) NCEP(MRF9) NSIPP(NASA) COLA2 GFDL PERSISTED GLOBAL SST ANOMALY FORECAST SST TROP. PACIFIC: 3 scenarios, based on mean of predictions of 1) CFS model, 2) LDEO model, 3) Constr. Analog model TROP. ATL, and INDIAN oceans Same as Trop. Pacific, except no LDEO model, and plus a CCA for Indian Ocean EXTRATROPICAL damped persistence

  36. Collaboration on Input to Forecast Production Sources of the Global Sea Surface Temperature Forecasts Tropical Pacific Tropical Atlantic Indian Ocean Extratropical Oceans NCEP Coupled NCEP Coupled NCEP Coupled Damped PersistenceLDEO Coupled CPTEC Statistical* IRI Statistical (CCA) Constr Analogue Constr Analogue Constr Analogue *when skillful Atmospheric General Circulation Models Used in the IRI's Seasonal Forecasts, for Superensembles Name Where Model Was Developed Where Model Is Run ECHAM 4.5 MPI, Hamburg, Germany IRI, Palisades, New YorkNSIPP NASA/GSFC, Greenbelt, MD NASA/GSFC, Greenbelt, MDCOLA COLA, Calverton, MD COLA, Calverton, MD ECPC SIO, La Jolla, CA SIO, La Jolla, CA CCM3.6 NCAR, Boulder, CO IRI, Palisades, New York GFDL GFDL, Princeton, NJ GFDL, Princeton, NJ

  37. IRI’s monthly issued probability forecasts of seasonal global precipitation and temperature We issue forecasts at four lead times. For example: NOV | Dec-Jan-Feb Jan-Feb-Mar Feb-Mar-Apr Mar-Apr-May Forecast models are run 7 months into future. Observed data are available through the end of the previous month (end of October in example above). Probabilities are given for the three tercile-based categories of the climatological distribution.

  38. Skill of forecasts at different time ranges: 1-2 day weather good 3-7 day weather fair Second week weather poor, but not zero Third week weather virtually zero Fourth week weather virtually zero 1-month climate (day 1-31) poor to fair 1-month climate (day 15-45) poor, but not zero 3-month climate (day 15-99) poor to fair At shorter ranges, forecasts are based on initial conditions and skill deteriorates quickly with time. Skill gets better at long range for ample time-averaging, due to consistent boundary condition forcing

  39. Lead time and forecast skill Weather forecasts (from initial conditions) Forecast Skill super 0.9 good 0.6 fair 0.3 poor 0.0 Potential sub-seasonal predictability Seasonal forecasts (from SST boundary conditions) (from MJO, land surface) 10 20 30 60 80 90 Forecast lead time (days)

  40. Combining the Predictions of Seasonal Climate by Several Atmospheric General Circulation Models into a Single Prediction

  41. RPSS Skills of Individual Models: JAS 1950-97 How do we combine their forecasts?

  42. Goals To combine the probability forecasts of several models, with relative weights based on the past performance of the individual models To assign appropriate forecast probability distribution: e.g. damp overconfident forecasts toward climatology

  43. Favorable results of application of Bayesian consolidation are evidenced in an analysis of reliability (the correspondence between forecast probability and relative observed frequency of occurrence). Simple pooling (assignment of equal weights to all AGCMs) gives more reliability than that of individual AGCMs, but the Bayesian method results in still much more reliability. Note that flattish lines show model overconfidence; 45º line shows perfect reliability. This is for JAS precipitation, for grid points between 30N and 30S. Above-Normal Below-Normal perfect reliability perfect reliability Bayesian Pooled Observed relative Freq. Observed relative Freq. Individual AGCM Forecast probability Forecast probability (3-model) JAS Precipitation, 30S-30N See also Barnston et al. (2003), Bull. Amer. Meteor. Soc., 1783-1796, Fig 7 (page 1793). from Goddard et al. 2003 EGS-AGU-EGU Joint Assembly, Nice, France, 7-11 April

  44. Summary of Merging of AGCM Forecasts IRI is presently using a 2-tiered prediction system, but is interested in using fully coupled systems also. Within its 2-tiered system it uses 4 SST prediction scenarios, and combines the predictions of 6 AGCMs. The merging of 7 predictions into a single one uses two multi-model ensemble systems: Bayesian and canonical variate. These give somewhat differing solutions, and are presently given equal weight.

  45. A “strong” shift of odds in rainfall forecast for Kenya during El Nino OND | | | | | | | | 13% 29% 59%

  46. A “strong” shift of odds in rainfall forecast for Kenya during El Nino OND Steps in finding probabilities of each of the tercile- based categories (below, near and above normal). 1. Use regression to make a deterministic (single point) forecast. 2. Determine standard error of estimate to represent the uncertainty of the deterministic forecast. 3. Use standard error of estimate to form a forecast distribution (i.e., make the red curve). 4. Find what value of z on the forecast distribution coincides with the tercile boundaries of the climatological distribution (33%ile and 67%ile on the black curve). Then use z-table to get the probabilities associated with these z values.. | | | | | | | | 13% 29% 59%

  47. Skill results for IRI real-time climate forecasts from 1997-2008

  48. Skill of IRI’s SST forecasts, late 1997 to present [ r = 0.5-mo lead(3.5-mo lead) ]; 5% significance in bold r = 0.88 (0.75) r = 0.36 (0.47) r = 0.69 (0.27) r = 0.44 (0.21) r = 0.18 (-0.10) Nino3.4 Indian Atlantic 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

More Related