Transcript KEY PONTS
ENSEMBLE
FORECASTING
COMAP Symposium 00-1
Presented by Steve Tracton
Wednesday, 15 December 1999
GIVE ME ODDS
KEY POINTS
•
•
•
•
THERE ARE INEVITABLE UNCERTAINTIES IN NWP DUE
TO UNCERTAINTIES IN INITIAL CONDITIONS AND
MODEL FORMULATION
WEATHER FORECASTING, THEREFORE, IS INHERENTLY
STOCHASTIC, NOT DETERMINISTIC IN NATURE
ENSEMBLE PREDICTION - REVOLUTIONARY CHANGE IN
THE THRUST OF OPERATIONAL NWP (“WAVE OF THE
FUTURE”) - CONSISTS OF MULTIPLE PREDICTIONS
FROM SLIGHTLY DIFFERENT INITIAL CONDITIONS
AND/OR WITH VARIOUS VERSIONS OF MODELS, THE
OBJECTIVES BEING TO:
–
IMPROVE SKILL THROUGH ENSEMBLE AVERAGING,
WHICH ELIMINATES NON-PREDICTABLE
COMPONENTS
–
PROVIDE RELIABLE INFORMATION ON FORECAST
UNCERTAINTIES (E.G., PROBABILITIES) FROM THE
SPREAD (DIVERSITY) AMONGST ENSEMBLE
MEMBERS
REALITY - POSITIVE RESULTS ON BOTH COUNTS
WITH OPERATIONAL GLOBAL MODEL ENSEMBLE
SYSTEM; EXPERIMENTAL REGIONAL MODEL ENSEMBLES
ENCOURAGING (OPERATIONAL EARLY 2000?)
–
•
NET RESULT - ENHANCE UTILITY OF NWP FOR
VIRTUALLY ALL APPLICATIONS
REALIZING THE PRACTICAL UTILITY OF ENSEMBLES
ACCOMPLISHED VIA A VARIETY OF NEW PRODUCTS
DESIGNED TO CONDENSE AND MAXIMIZE
INFORMATION CONTENT FOR USERS; USER FEEDBACK
ESSENTIAL AND ENCOURAGED!!!
KEY CONSIDERATIONS
•
STRATEGIES FOR CREATING ENSEMBLES
–
PROCDEDURES FOR GENERATING INITIAL STATE
PERTURBATIONS
• RANDOM
• TIME LAGGING
• ANALYSES FROM OTHER CENTERS
• “BREEDING”
• SINGULAR VECTORS
–
PERTURBING MODEL (E.G., CONVECTIVE
PARAMETERIZATION) AND/OR MULTI-MODEL
ENSEMBLES
–
MODEL CONFIGURATION?
• RESOLUTION
• PHYSICAL SOPHISTICATION
• DOMAIN
–
ENSEMBLE SIZE
NOTE: OPTIMUM STRATEGY UNKNOWN (NO CONCENSUS)!!
IDEAL: EFFECTIVE/EFFICIENT SAMPLING OF ALTERNATIVE
SCENARIOS, I.E., PROBABILITY DISTRIBUTIONS.
LIMITED COMPUTER RESOURCES GENERALLY REQUIRE
COMPROMISES RELATIVE TO PERCEIVED OPTIMUM,
E.G., MODEL RESOLUTION VERSUS ENSEMBLE SIZE)
•
KEY CONSIDERATIONS(CONT.)
PRODUCT DEVELOPMENT
OBJECTIVE:
CONDENSE LARGE AMOUNTS OF OUTPUT INTO A
“USER FRIENDLY” FORM THAT PROVIDES RELIABLE
ESTIMATES OF THE RANGE AND LIKLIHOOD OF
ALTERNATIVE SCENARIOS
–
PRODUCTS CAN RANGE FROM DISPLAY OF ALL
FORECASTS THROUGH MEANS/SPREAD AND
CLUSTERS TO FULL PROBABILITIY
DISTRIBUTIONS DISPLAYED IN VARIOUS
FORMATS
•
•
•
•
•
STATISTICAL POSTPROCESSING (E.G.,
BIAS CORRECTIONS, CALIBRATION OF
PROBABILITIES
ENSEMBLE OUTPUT STATISTICS
ADDITIONAL/ALTERNATIVE PRODUCTS
CONTINUAL INTERACTION AMONGST
DEVELOPERS AND USERS
VALIDATION
–
–
–
STANDARD SKILL SCORES
MEASURES OF SPREAD
MEASURES OF RELIABILITY
EDUCATION AND TRAINING
– COMET SYMPOSIUM
– TRAINING MODULES
– ON SITE VISITS
– WEB BASED
– ??
N-AWIPS GRAPHICAL PRODUCTS
(GEMPAK META FILES)
•
•
•
SPAGHETTI CHARTS
– 500 Z
– 1000Z
– 1000/500 TCK
– MSLP
– 850 T
– 700 RH
SPREAD
– 1000 Z
– 500 Z
CLUSTERS
– 1000 Z
– 500 Z
•
PROBABILITIES
•
MSLP CENTERS
–
–
–
–
–
500
700
TCK
250
850
Z > THRESHOLDS
RH > 70%
<540
V > THRESHOLDS
T > 0C
PRODUCT DEVELOPMENT INCLUDES
•
PROBABILITIES
– VIRTUALLY ALL RELEVANT AND MODEL DERIVED
PARAMETERS, E.G.,
• SEVERE WEATHER INDICES
• AVIATION WINDS > THRESHOLD
• SENSIBLE WEATHER ELEMENTS (MODEL
DERIVED/INFERRED
• CIRCULATION INDICES (E.G., BLOCKING)
•
EXPANDED CLUSTERED PARAMETERS AND FOR
SPECIALIZED REGIONS
•
VERTICAL PROFILES
•
METEOGRAMS
•
ENSEMBLE DERIVED MOS
•
TROPICAL STORM TRACKS
–
–
DIRECT FROM ENSEMBLES
BACKGROUND FOR GFDL MODEL ENSEMBLES
SOME APPLICATIONS
•
•
FORECASTS OF ENSEMBLE MEAN, SPREAD,
PROBABILITY DISTRIBUTIONS, ETC. OF ANY
MODEL FIELD/PARAMETER OR QUANTITIES
DERIVED THEREFROM
ENHANCE THE
UTILITY OF FORECASTS
APPLICABLE TO MODELS FROM VERY SHORT
RANGE CLOUD SCALE THROUGH REGIONAL
MESOSCALE SHORT RANGE AND GLOBAL MEDIUM
RANGE TO COUPLED OCEAN/ATMOSPHERE
CLIMATE PREDICTION SYSTEMS
•
IMPROVE DATA ASSIMILATION SYSTEMS
•
ADAPTIVE/TARGETED OBSERVATIONS
•
DATA SETS FOR FUNDAMENTAL RESEARCH ON
PREDICTABILITY ISSUES
NOTE: LARGE CURRENT USER COMMUNITY
(OPERATIONAL GLOBAL SYSTEM) INCLUDES NCEP
SERVICE CENTERS, WFO’S, USAF, OH,
PRIVATES/BROADCASTERS
CLUSTER ANALYSIS
OBJECTIVELY GROUP TOGETHER FORECASTS WHICH
ARE SIMILAR ACCORDING TO SOME CRITERIA
GOAL: IDENTIFY EXTREMES, GROUPINGS (CLUSTERS)
WITHIN ENVELOPE OF POSSIBILITIES
(“ATTRACTORS”)
•
ISSUES:
– QUANTITY
• MSLP
• 500 Z
• ETC.
– MEASURE
•
•
•
•
ANOMALY CORRELATION
CIRCULATION PARAMETERS
PATTERN RECOGNITION
PHASE-SPACE MEASUREMENTS
– REGION
EVALUATION/VERIFICATION
•
•
SITUATIONAL AND PHENOMENOLOGICAL CASE
STUDIES (E.G., CYCLOGENESIS, FLOOD POTENTIAL)
STATISTICAL
–
–
–
–
–
–
STANDARD AC, RMS, SCORES (E.G., APPLIED TO
ENSEMBLE MEAN VS. CONTROL, RELATIVE
CLOSENESS OF MEMBERS TO ANALYSIS)
“TALAGRAND” (VERIFICATION RANK) DIAGRAMS MEASURES OF BIASES IN DISTRIBUTION OF
ENSEMBLE MEMBERS INCLUDING FREQUENCY OF
OUTLIERS)
BRIER, RANKED PROBABILITY SCORES
(PROBABILITY SKILL SCORES)
RELIABILITY DIAGRAMS (OBSERVED VERSUS
FORECAST FREQUENCIES; ENABLES CALIBRATION
OF PROBABILITIES)
MOS VERSUS ENSEMBLE POPS
RELATIVE OPERATING CHARACTERISTICS (ROC);
(EXPLICIT COMPARISON OF THE RELATIVE UTILITY
OF DETERMINISTIC AND ENSEMBLE PREDICTIONS)
SHORT RANGE ENSEMBLE FORECASTING (SREF)
•
OBJECTIVE: DEVELOP A REGIONAL MODEL, SHORTRANGE (0-3 DAYS) ENSEMBLE PREDICTION SYSTEM TO
PROVIDE OPERATIONALLY RELEVANT AND USEFUL
GUIDANCE ON THE PROBABILITY DISTRIBUTION OF
WEATHER ELEMENTS OR EVENTS, ESPECIALLY FOR QPF
•
GOAL: IMPLEMENT INITIAL OPERATIONAL
PRODUCTION OF A REGIONAL MODEL BASED ENSEMBLE
SYSTEM AND PRODUCT SUITE (SREF-I) BY ~ JANUARY,
2000
–
TARGET SYSTEM:
•
•
•
•
•
•
–
PRODUCT SUITE:
•
•
•
•
•
ENSEMBLE MEAN/SPREAD CHARTS
SPAGHETTI CHARTS
PROBABILITY CHARTS
METEOGRAMS
STATUS MILESTONES
•
•
•
ETA PLUS RSM MULTI-MODEL
10 MEMBER
40 KM RESOLUTION
~ETA DOMAIN
RUN TWICE PER DAY
PERTURBATIONS; REGIONAL “BREEDING”
CONDUCT PILOT STUDIES (10/96-3/98)
PARTICIPATE IN STORM AND MESOSCALE ENSEMBLE
EXPERIMENT (SAMEX) (3/98-11/98)
SOME ISSUES
–
–
–
–
–
–
ALTERNATIVE PERTURBATION STRATEGIES
TRADEOFFS; RESOLUTION, ENS SIZE/DOMAIN
PRODUCT DEVELOPMENT
VALIDATION PROCEDURES
DATA/PRODUCT DISSEMINATION
EDUCATION AND TRAINING
STATUS/MILESTONES
COMPLETED (PILOT STUDIES)
• MAJOR TASKS/ACCOMP (CONT.)
– ILLUSTRATE SIGNIFICANCE
OF UNCERTAINTIES IN SREF
– DEMONSTRATE THE
POTENTIAL OF SREF TO
PROVIDE OPERATIONALLY
USEFUL INFORMATION
– PROVIDE BASIS FOR A
PROTOTYPE OPERATIONAL
SYSTEM
STATUS/MILESTONES
COMPLETED (PILOT STUDIES)
•
SOME KEY FINDINGS
– ENHANCED DIVERSITY OF
SOLUTIONS (SPREAD) WITH:
• MULTI-MODEL ENSEMBLE
• HIGHER RESOLUTION
• GLOBAL BRED (VS “RANDOM”)
• REGIONAL ENHANCEMENT
STATUS/MILESTONES
COMPLETED (PARTICIPATE IN SAMEX)
• BOTTOM LINE:
– MUCH GAINED,
ACCOMPLISHED, LEARNED
– RESULTS GENERALLY
FAVORABLE
– SOME DISSAPPOINMENTS
RELATIVE TO
EXPECTATIONS, BUT WE
UNDERSTAND WHY
– REMAIN COMMITED TO
BASIC STRATEGY
STATUS/MILESTONES
COMPLETED (PARTICIPATE IN SAMEX)
•
SOME KEY FINDINGS
– IMPROVED ENS MEAN SKILL,
RELIABILITY, RPSS WITH
MULTI-MODEL APPROACH
– INSUFFICENT SPREAD
• DOMAIN TOO SMALL -
NEGATIVE IMPACT OF BC’S
– PRECIPITATION FORECASTS
“WOEFUL”
• WEAK FORCING, SUMMER
LIKE PATTERN
SAMEX DOMAIN
LARGE
SMALL
SAMEX SYSTEM:
MULTI-MODEL (ETA/RSM)
10 MEMBERS
32 KM RESOLUTION
SAMEX DOMAIN
REGIONAL ENHANCEMENT