ENSEMBLE FORECASTING

1
ENSEMBLE FORECASTING

• Zoltan Toth
• Ensemble Probabilistic Guidance Team
• Environmental Modeling Center

2
OUTLINE

• User requirements / Decision support
• Ensemble system upgrade plans
• GEFS
• SREF
• Ensemble processing plans
• Bias correction
• Downscaling
• Products

3
USER REQUIREMENTS - SIMPLIFIED

• Applications affected by weather
• Must consider information on weather to
• Minimize losses due to adverse weather
• Optimal user decision threshold equals
• Probability of adverse weather exceeding
• Cost / loss ratio of decision situation
  (simplified decision theory)
• Probability of weather events must be provided
• Only option in past, based on error statistics of
  single value forecasts
• Eg, MOS POP
• Now can be based on ensemble
• Users act when forecast probability exceeds their
  cost/loss ratio
• Advantages
• Limited set of products (eg, 10 / 50 / 90
  percentile forecast) doable today
• Problems

4
USER REQUIREMENTS - ADVANCED

• Some (many?) applications affected by
• Joint probabilities
• Eg, probability of significant convection in
  vicinity of 2 specific hubs between 3-6pm
• Many variables, many probability / critical value
  decision thresholds
• Access to multiple (n) plausible weather
  scenarios critical
• Statistically reliable ensemble weather data
• 6D-Cube - space (3D) time variables
  ensemble
• Expanded NDFD future official NWS weather /
  climate / water forecast database
• Develop user application model simulating optimal
  weather related operations including
• Assuming weather is known What are actions /
  costs / benefits?
• Run application model n x n times with
• Weather scenario from each ensemble member using
• User operations (oi) optimized for each ensemble
  member (ei)
• Take ensemble mean of economic outcome (costs
  losses) for each set of user procedures
• Choose operating procedures (oi) minimizing costs
  losses in expected (ensemble mean) sense

5
RESPONSIBILITIES OF ENSEMBLE TEAM

• Assess, model, communicate uncertainty in
  numerical forecasts
• Represent uncertainty in numerical forecasting
• Tasks
• Design, implement, maintain, and continuously
  improve ensemble systems
• Topics
• Initial value related uncertainty
• Model related forecast uncertainty
• Ensemble systems
• Global NAEFS / GEFS
• Regional SREF
• Climate Contributions to future CFS
  configuration
• Statistical correction of ensemble forecasts
• Tasks
• Correct for systematic errors on model grid
• Downscale information to fine resolution grid
  (NDFD)
• Combine all forecast info into single
  ensemble/probabilistic guidance

6
NEXT GEFS IMPLEMENTATION (2009)

• Increase horizontal resolution from T126 to T190
• 4 cycles per day, 201 members per cycle
• Up to 384 hours (16 days)
• Add stochastic perturbation scheme to account for
  random model errors
• Increased ensemble spread and forecast skill
  (reliability)
• Use 8th order horizontal diffusion for all
  resolutions
• Improved forecast skill and ensemble spread
• Introduce ESMF (Earth System Modeling Framework)
• Version 3.1.0r
• Allows concurrent generation of all ensemble
  members
• Needed for efficiency of stochastic perturbation
  scheme
• Add new variables (26 more) to pgrba files
• Based on NAEFS user request
• From current 52 to future 78 variables
• For NAEFS ensemble data exchange
• Yuejian Zhu

7
GEFS 2009 IMPLEMENTATION TESTS
CRPS NH 850 hPa Temp
RMSE / SPREAD 500 hPa Height
Extend current 5-day skill to 6.5-day
Yuejian Zhu, Dingchen Hou, Bo Cui, Dick Wobus
8
GEFS IMPLEMENTATION PLANS FOR FY2010

• Real time generation of hind-casts
• Make control forecast once every 5th day
• T190L28, out to 16 days
• Use new reanalysis (30 yrs)
• Why
• Increase sample of analysis forecast pairs for
  statistical corrections
• Benefits
• Improved bias correction beyond 5 days
• Potential for regime / situation dependent bias
  correction
• Adaptive modification of initial and stochastic
  model perturbation variances
• Based on recursive average monitoring of
• Forecast errors
• Ensemble spread
• Why
• Avoid having to tune perturbation size after each
  analysis/model/ensemble change
• Benefit
• Improved performance
• Easier maintenance

9
REAL-TIME GENERATION OF HIND-CAST DATASET
Todays Julian Date TJD
TJD 30
TJD - 30
Actual ensemble generated today
2006
Time
2005
2004
2003
1968
1967
Hind-casts for TJD30 generated today
Hind-casts (or its statistics) for TJD/- 30
saved on disc
BACK
10
SREF DEVELOPMENT PLAN

• Stochastic physics
• Convective parameterization of Teixeira et al -
  NRL
• Stochastic feedback from Grell-Devenyi
• Ensemble Transform initial perturbations
• Consistent with boundary perturbations from GEFS
• River flow, wave, etc ensemble applications
• Explore use in targeting observations, too
• Resolution
• Next upgrade 20km
• Infrastructure
• Model diversity under NEMS
• Jun Du

11
HIGH IMPACT ENSEMBLE SYSTEM

• Motivation
• Support probabilistic high impact event
  forecasting
• Potential applications
• Severe / fire / aviation weather, AQ, tropical /
  winter storm, etc
• General framework
• Relocatable / movable mesh embedded into SREF
• Model diversity etc configured differently for
  each application
• Deployment
• On demand, based on user request / priority list
  of the day
• Like Winter Storm Reconnaissance requests for
  observations
• Operational execution
• High impact window hurricane window?
• Potential partners
• HWTB, RR, HFIP, ensemble-DA, etc
• Status
• Initial planning phase
• Mesh capability being worked on
• Possible testing from 2010?

12
grid221 (30km)
13
ENSEMBLE DATA PROCESSING STEPS

• Bias correction / combination of all info
• Remove lead time dependent model bias on model
  grid
• Calibrate higher moments of ensemble
• Combine information from all sources into single
  set of ensemble
• Forecaster modification
• Subjective changes to ensemble data (over US
  only?)
• Proxy for truth
• Create observationally based fine resolution
  analysis for use in downscaling
• Downscaling
• Interpret bias corrected ensemble on user
  relevant grid NDFD
• Additional variables
• Derive further variables from bias corrected /
  downscaled NWP output
• Visualization / Derived products
• Interrogate bias corrected / downscaled ensemble
  dataset

14
BIAS CORRECTION OF FORECASTS

• Method
• Bayesian processor
• Combines information from all sources
• Fuses forecast data with climatological
  distribution (prior)
• Adjusts spread according to skill observed in
  forecast sample
• Outputs statistically corrected distribution
  (posterior)
• Ensemble members adjusted to represent posterior
  distribution
• Data sources
• Reanalysis as prior (use new reanalysis when
  available)
• Sample of past forecasts - most recent 60-90 days
• Include control hind-casts when available
• Latest analyzed or observed data
• Current status
• 35 NAEFS SREF variables bias corrected
• 1st moment corrected only
• NAEFS SREF processed separately
• CMC NCEP ensembles GFS hires control combined

15
List of Variables for Bias Correction,
Weightsand Forecast Anomalies for CMC NCEP
Ensemble
Yuejian Zhu
16
ESTIMATED BIAS 2m Temperature, 5-d forecast
BEFORE AFTER BIAS CORRECTION
Bo Cui
17
IMPACT OF BIAS CORRECTION ON ESTIMATED
SYSTEAMTIC ERROR PROBABILISTIC SCORES
NH 500hPa height
850hPa temperature
Before bias correction (1x1)
NH 500hPa height
After bias correction (1x1)
NH 2m temperature
Tropics 500hPa height
Before Bias Correction
After Bias Correction
Bo Cui
18
PROXY FOR TRUTH

• Method
• Observationally based data assimilation
• If NWP first guess is used
• Consider bias correcting first guess
• Do not cycle first guess from observationally
  based analysis
• Allows to draw closer to data
• Data sources
• Near surface observations
• NWP first guess (if good quality available)
• Current status
• Real Time Meso-scale Analysis (RTMA)
• On NDFD grid
• RUC as first guess
• 2d version of GSI as DA procedure
• Fields
• 2m temp, 10m wind (u, v, speed, direction), sp
• Domains

19
Mike Charles
RFC 4X4 km grid
CPC 1/8 grid
Combined, 5x5 km grid
20
DOWNSCALING

• Method
• Perfect prog
• Establish relationship (donwscaling vector)
  between
• Proxy for truth (high resolution observationally
  based analysis)
• NWP analysis (used as reference in bias
  correction step)
• Level of sophistication
• Climatological (statitical)
• Regime dependent (statistical)
• Case dependent (dynamical, using LAM) most
  expensive
• Sub-NWP-grid resolution variance
• Need to be stochastically added in statistical
  methods
• Outputs ensemble members statistically consistent
  with
• Bias corrected forecasts on NWP grid
• Proxy for truth on fine resolution grid
• Data sources
• Sample of
• Fine resolution observationally based analysis
  fields
• Corresponding NWP analysis fields
• Current status

21
MDL GMOS NAEFS Downscaled Forecast Mean
Absolute Error w.r.t. RTMA Average For Sept.
2007
Valery Dagostaro, Kathy Gilbert, Bo Cui, Yuejian
Zhu
24-h GMOS Forecast
24-h NAEFS Forecast
For CONUS NAEFS(1.45) GMOS(1.72) 19 impr.
over GMOS
22
Surface Temperature MAE CONUS, Sept. 2007
12Z NDFD vs. 00Z MOS/GMOS/NAEFS
Surface Temperature MAE CONUS, Sept. 2007
00Z GMOS vs. 00Z NAEFS
RTMA Analysis
METAR obs. 1221 sites
GMOS forecast
0.5C
NAEFS products
0.5C
Valery Dagostaro, Kathy Gilbert
Bo Cui, Yuejian Zhu
23
Surface Temperature Area Mean Bias CONUS,
Sept. 2007 12Z NDFD vs. 00Z MOS/GMOS/NAEFS
Surface Temperature Pointwise Bias CONUS,
Sept. 2007 00Z GMOS vs. 00Z NAEFS
Pointwise Bias
Area Mean Bias
RTMA Analysis
METAR obs. 1221 sites
GMOS forecast
0.6C
0.6C
NAEFS products
Valery Dagostaro, Kathy Gilbert
Bo Cui, Yuejian Zhu
24
DERIVED VARIABLES

• Objective
• Generate variables not carried in NWP models
• Input data
• Bias corrected and downscaled ensemble data (NWP
  model output)
• Methods
• Model post-processing algorithms
• Apply after downscaling for variables affected by
  surface processes
• SMARTINIT
• NDFD weather element generator
• Other tools?
• Text generation, etc?

Geoff Manikin et al.
25
STATISTICAL PROCESSING OF PRECIPITATION

• Problem
• Precipitation is discontinuous
• Traditional approach to calibrated PQPF
• Devise method to calibrate POP
• Devise method to calibrate conditional QPF
• Solution
• Define new continuous variable pseudo
  precipitation
• Equals precipitation when pgt0
• Proportional to moisture deficit in vertical
  column
• Use generic Bayesian ensemble processor
• Status
• Generic Bayesian ensemble processor in collab.
  w R. Krzysztofowicz
• Algorithm complete - in collaboration with
• Significant part of codes transferred from Univ.
  VA
• Next - Assemble and test codes
• Pseudo precipitation data in collaboration with
  Paul Schultz
• Combined CPC RFC observed precip dataset
  complete in collab. w. OHD

26
Application of Pseudo-precipitation

• With a single continuous distribution, can bias
  correct in one step
• Shift entire ensemble distribution to remove bias
  in PP
• For example, if some members never precipitate in
  an area with frequent drizzleBias correction
  shifts the PP distribution to the right, forcing
  some of those members into the positive PP
  (forcing them to precipitate)

Paul Schultz, Mike Charles
26
27
END-TO-END FORECAST PROCESS
WHO / WHAT
HOW
WHY
Numerical modeling
Dynamical Prediction
NCEP, other centers Ensemble generation
Statistical calibration
Statistical consistency
MDL/NCEP, partners Ensemble processing
USERS
OBSERVATIONS
Human intervention
NCEP/SS ? WFOs QC, enhancements
Added value Interpretation
USERS Applications
Decision models
28
BACKGROUND
29
USER REQUIREMENTS

• Decisions regarding operations affected by
  weather
• Must consider information on weather to
• Minimize losses due to adverse weather
• Maximize benefits from favorable weather
• Different applications affected by
• Various weather parameters
• Optimal user decision threshold equals
• Probability of adverse weather exceeding
• Cost / loss ratio of decision situation
  (simplified decision theory)
• Decisions have their unique cost / loss
  environment
• Joint probabilities needed when multiple weather
  parameters used
• Decision maker must know probability of adverse
  weather
• Probability of weather events can be derived from
  ensemble
• Probabilities for all variables affecting users,
  incl. joint probabilities
• Ensemble dataset as primary media/carrier of
  forecast info
• All relevant variables must be included

30
RTMA EVALUATION

• Experimental setup
• Cross-validation against data withheld from
  analysis
• Average for seven cases for May 3 2007, 12Z - 18Z
• RMS difference between independent observations
  and
• 1-hr RUC forecast (First Guess)
• RTMA analysis
• Benchamrk Cressman analysis with one pass
• Results
• 10m wind (m/s)
• Guess RTMA Baseline (Cressman)
• 2.96 1.43 2.41
• 2m temperature (K)
• Guess RTMA Baseline (Cressman)
• 2.20 1.26 1.73
• 2m specific humidity (g/kg)
• Guess RTMA Baseline (Cressman)

Manuel Pondeca
31
RTMA EVALUATION - 2
STATISTICS
BCKG ANL
BCKG ANL
(all obs) (all obs)
(cval) (cval)
Note Results represent 10-day averages from 26
Oct through 4 Nov 2008. Columns 2 and 3 show
results from the original gsi runs, ie., they
display statistics computed over the entire set
of obs used in the analysis. Columns 4 and 5
display results averaged over 5 independent
cross-validation datasets. Each cross-validation
set contains roughly 10 of all the data
Manuel Pondeca
32
CONNECTING DECISION MAKING WITH FORECAST INFO

• Option 1 - Weather forecaster prepares
  probabilistic products for users
• Eg, probability of significant convection in
  vicinity of 2 specific hubs between 3-6pm
• Only option in past, based on error statistics of
  single value forecasts
• Problems
• Ever growing list of special products needs to be
  generated
• Still not covering all situations Decision
  making compromised
• Estimate cost/loss ratio associated w. decision
  regarding adverse weather event
• User acts if probability of adverse event higher
  than cost/loss ratio
• Limitations
• Assumes prior analysis knowledge of
• Weather sensitive operations associated
  cost/loss ratio
• Sub-optimal decisions in cases deviating from
  analyzed situations
• Option 2 - Weather forecaster presents user with
  multiple scenarios
• Ensemble weather data in NEXGEN
• Space (3D) plus time plus variables plus ensemble
  6D-Cube
• Develop general model simulating user operations
  including costs
• Optimize weather related operations based on
  climatology
• Vary procedures in operations that are sensitive
  to weather

33
PARTNERS / OUTREACH

• Multi-center ensemble combination
• NAEFS - Env. Canada TIGGE GIFS (WMO/THORPEX)
  Other ensemble NWP centers/users
• Statistical processing Bayesian processor for
  ensembles
• Roman Krzysztofowicz, Univ. VA
• RTMA, Smartinit
• WFOs
• Hindcasting
• Tom Hamill, PSD/ESRL
• Pseudo precipitation, downscaling
• Paul Schultz, GSD/ESRL
• Precipitation analysis
• PingPing Xie, CPC/NCEP, DJ Seo, OHD
• Probabilistic verification

34
WORK ARRANGEMENTS

• Develop environment conducive of joint work
• Communication among partners critical
• NFUSE team on ensemble / probabilistic processing
• Encourage complimentary efforts
• Modular design
• Enable inter-comparison of competing methods
• Common/shared
• Verification tools
• Datasets
• What is role of OSIP?
• Suited to guide innovative development work?

35
EMC-MDL COLLABORATION

• Compare quality of current operational /
  experimental products
• Gridded MOS vs. Downscaled NAEFS
• Ongoing
• Kathy Gilbert, Val Dragostano Zoltan Toth, Bo
  Cui, Yuejian Zhu
• Proxy for truth issue unresolved
• Need observations independent of MOS
• MDL experimental ensemble guidance vs. Downscaled
  NAEFS
• 10/50/90 percentiles to be evaluated
• Matt Peroutka Zoltan Toth
• Proxy for truth issue
• Proxy for truth?
• Agree on best proxy for truth
• Collaborate on
• Improving RTMA, including bias correction for FG
• Creating best CONUS precipitation analysis
  archive
• Joint research into best downscaling methods?
• Climate, regime, case dependent methods

36
KEY POINTS

• 6D-Cube - Dataset of bias corrected/calibrated
  ensemble members
• Variable, space (3D), time, ensemble member
• All forecast information can be derived
• Mass production of bias corrected ensemble
  dataset
• Currently 35 variables, 200 planned
• Very strong external links to accelerate progress
• NAEFS, NUOPC, THORPEX TIGGE / GIFS, etc
• Methods for processing must be
• General versatile for mass production
• Modular for collaboration
• Transition to operations unique job
• Select best methods for each module in processing
• Test / implement in stream of jobs
• Seamless links needed for use in end-to-end
  process

37
BACKGROUND
38
CONNECTING DECISION MAKING WITH FORECAST INFO

• TRADITIONAL APPROACH
• Provide probabilistic products for users
• Only option in past, based on error statistics of
  single value forecasts (eg, MOS POP)
• Eg, probability of significant convection in
  vicinity of 2 specific hubs between 3-6pm
• Now can be based on ensemble
• Users act when forecast probability exceeds their
  cost/loss ratio
• Advantage
• Can derive probabilities from single forecast
• Problems
• Proliferation of number of products
• Different variables, probability/weather element
  thresholds, joint probabilities
• Limited usage
• Downstream applications severely limited (eg,
  wave, streamflow, etc, ensembles not possible)
• PROPOSED APPROACH
• Provide multiple (N) plausible weather scenarios
• Ensemble weather data
• 6D-Cube - space (3D) time variables
  ensemble
• Expanded NDFD future official NWS weather /
  climate / water forecast database

39
CONNECTING DECISION MAKING WITH FORECAST INFO

• TRADITIONAL APPROACH
• Provide probabilistic products for users
• Only option in past, based on error statistics of
  single value forecasts (eg, MOS POP)
• Eg, probability of significant convection in
  vicinity of 2 specific hubs between 3-6pm
• Now can be based on ensemble
• Users act when forecast probability exceeds their
  cost/loss ratio
• Problems
• Proliferation of number of products
• Different variables, probability/weather element
  thresholds, joint probabilities
• Limited usage
• Downstream applications severely limited (eg,
  wave, streamflow, etc, ensembles not possible)
• PROPOSED APPROACH
• Provide multiple (N) plausible weather scenarios
• Ensemble weather data
• 6D-Cube - space (3D) time variables
  ensemble
• Expanded NDFD future official NWS weather /
  climate / water forecast database
• Develop user application model simulating optimal
  weather related operations including
• Actions / costs / benefits

40
CONNECTING DECISION MAKING WITH FORECAST INFO

• TRADITIONAL APPROACH
• Provide probabilistic products for users
• Only option in past, based on error statistics of
  single value forecasts (eg, MOS POP)
• Eg, probability of significant convection in
  vicinity of 2 specific hubs between 3-6pm
• Now can be based on ensemble
• Users act when forecast probability exceeds their
  cost/loss ratio
• Problems
• Proliferation of number of products
• Different variables, probability/weather element
  thresholds, joint probabilities
• Limited usage
• Downstream applications severely limited (eg,
  wave, streamflow, etc, ensembles not possible)
• PROPOSED APPROACH
• Provide multiple (N) plausible weather scenarios
• Ensemble weather data
• 6D-Cube - space (3D) time variables
  ensemble
• Expanded NDFD future official NWS weather /
  climate / water forecast database
• Develop user application model simulating optimal
  weather related operations including
• Actions / costs / benefits

41
PROPOSED APPROACH

• SHORT TERM (2-3 years) Traditional approach as
  stopgap solution
• Users
• Identify key areas of weather sensitive
  operations
• Define limited set of associated probabilistic
  products
• Estimate related cost/loss ratios
• Forecasters
• Produce limited set of probabilistic products
  based on ensemble
• Users vary their action depending on
• Forecast probability AND
• Cost/loss ratio
• LONG TERM (5-10 years) New approach
• Forecasters
• Provide easy access to high quality ensemble data
  NEXGEN 6D-Cube
• Statistically calibrate raw ensemble data
• Expand current NAEFS / MOS capabilities
• Make accessible all statistically calibrated
  ensemble data
• Expand current TOC dataset
• Provide smart sub-setting interrogation tools

NFUSE - Doug Hildebrand
42
SREF DEVELOPMENT PLAN

• Expand bias correction from only CONUS domain
  (grid212) to Alaska and Hawaii two grids (grids
  243 and 216) or to bigger North American domain
  (grid221)
• Dynamically downscaling the 32km SREF using 12km
  NAM forecasts (dual-resolution ensemble approach)
• Statistically downscaling the 32km SREF using 5km
  RTMA
• Bias correcting SREF precipitation fields
• NEMS-only based SREF 20km, stochastic physics as
  well as regional ET (in a later time)
• Explore High-Impact Weather Ensemble Forecast
  system (HWEF)

43
NAEFS STATISTICAL CORRECTION / PRODUCT PLANS FOR
FY2009

• Downscaling
• CONUS
• Additional variables (Tmax and Tmin, wind speed
  and direction)
• Pending on RTMA availability
• Alaska
• 8 variables (T2m, Tmax, Tmin, Psfc, U10m, V10m,
  10 meter Ws and Wd)
• In testing
• In collaboration with HPC Alaska Desk
• Other regions
• Hawaii, Puerto Rico and Guam
• Pending on RTMA availability
• New NAEFS variables for data exchange
• 25 additional variables
• Use GRIB2 format faster transfer (30 min
  saving)
• Coordinated with CMC/MSC
• Dedicated line for NCEP and CMC NAEFS data
  exchange
• DS-3?
• Earlier access to NAEFS ensemble products

44
NCEP/GEFS raw forecast
8 days gain
NAEFS final products
From Bias correction (NCEP, CMC) Dual-resolution
(NCEP only) Down-scaling (NCEP,
CMC) Combination of NCEP and CMC
45
NAEFS STATISTICAL CORRECTION / PRODUCT PLANS FOR
FY2010

• Bias correction
• Develop, test, implement Bayesian bias correction
  technique
• Apply on all prognostic model variables
• Including pseudo-precipitation
• Downscaling
• Improve methods
• Apply to new variables

46
NAEFS EXPANSION PLANS FY2009/10

• Incorporate other ensemble data into NAEFS
• One year evaluation period
• Must coordinate with CMC
• Use mini-Bayesian for testing
• FNMOC global ensemble
• Need to resolve the problem for data transform
  (still missing forecast data for FNMOC)
• ECMWF global ensemble
• Need to coordinate with NCO for 00UTC data and
  6hr forecast intervals
• Probabilistic verification
• Unified probabilistic verification (3Q, 2009)
• Shared codes with SREF
• Partnership with OHD, etc

47
NAEFS future configuration Updated October 2008
48
CURRENT CONFIGURATION

• NORTH AMERICAN ENSEMBLE FORECAST SYSTEM (NAEFS)
• Definition
• Multi-center ensemble system combining
• 20 perturbed plus a control member out to 16 days
  twice a day on 1x1 lat/lon grid from
• NCEP - Global Ensemble Forecast System (four
  times per day)
• T126L28, ET initial perturbation technique, no
  model perturbations
• Canadian Meteorological Center (CMC)
• Operational steps
• Generation NCEP global ensemble
• Real time data exchange with CMC (50 variables)
• Bias correction of both ensembles (35 variables)
• Incorporation of information from high resolution
  GFS
• Downscaling of 1x1 bias corrected fields onto 5x5
  km NDFD grid (4 variables)
• Product generation see next page

49
PRODUCTS

• NORTH AMERICAN ENSEMBLE FORECAST SYSTEM (NAEFS)
• Basic products on 1x1 lat/lon grid
• Bias corrected ensemble forecasts (42 members, 35
  variables)
• Climate anomaly forecasts for downscaling
  applications (42 members, 19 variables)
• Derived products
• Mean, mode, 10, 50 (median), 90 percentile
  forecasts, spread on 1x1 lat/lon grid
• Downscaled forecasts on 5x5 km grid (4 variables)
• Week-2 temperature forecasts (jointly with CPC)
• GEMPAK grids / images
• Ensemble / probabilistic displays
• Distribution of products
• Mainframe / NAWIPS for NCEP Service Centers
• AWIPS for WFOs
• WOC NCEP ftp servers for grids for external
  users
• Web sites for images
• NCEP official web site

50
CUSTOMERS

• NORTH AMERICAN ENSEMBLE FORECAST SYSTEM (NAEFS)
• NCEP Service Centers
• Gridded fields, NAWIPS manipulation tools, images
• NCEP International Desks
• South American Desk
• African Desk
• WFOs
• AWIPS grids, ftpd grids, images
• RFCs
• Grids (experimental)
• External users
• Ensemble grids, images
• NAEFS partners
• CMC
• National Meteorological Service of Mexico (NMSM)
  images
• International community
• Severe Weather Forecast Demonstration Projects
• Private companies

51
NWS Seamless Suite of ForecastProducts Spanning
Climate and Weather
NCEP Model Perspective
Forecast Uncertainty
Years
Seasons
Months
Climate Forecast System
2 Week
North American Ensemble Forecast System
Climate/Weather Linkage
1 Week
Global Forecast System
Short-Range Ensemble Forecast
Days
Ocean Model Hurricane Models
North American Forecast
Hours
Rapid Update Cycle for Aviation

• GFDL
• WRF

Dispersion Models for DHS
Minutes
Health
Maritime
Aviation
Agriculture
Recreation
Ecosystem
Commerce
Hydropower
Environment
Fire Weather
Life Property
Emergency Mgmt
Energy Planning
Space Operations
Reservoir Control
52
(No Transcript)
53
D50.357 2001
D60.334 2007
D50.356
D60.361
54
NCEP/GEFS raw forecast
4 days gain from NAEFS
NAEFS final products
From Bias correction (NCEP, CMC) Dual-resolution
(NCEP only) Down-scaling (NCEP,
CMC) Combination of NCEP and CMC
55
Latest retrospective run (full package)
NH 2-m temperature RMSE Spread
NH 500hPa height RMSE Spread
E20s T126L28 E20g T190L28 (0-180 only)
SH 500hPa height CRPSS
NH 500hPa height CRPSS
56
(No Transcript)
57
NH Anomaly Correlation for 500hPa HeightPeriod
August 1st September 30th 2007
GEFSg is better than GFS at 48 hours
GEFSg could extend skillful forecast (60) for 9
days 24 hours better than current GEFS 48 hours
better than current GFS
58
Plans for FY2010

• GEFS
• Configuration
• Variable resolutions
• T270L42 (0-180hr) (considering half-degree
  products)
• T190L28 (180-384hr)
• T126L28 (384-840hr)
• Full coupling with ocean model (assume in)
• Science
• Improving TS relocation
• Adopt all new developed TS relocation schemes
• Improving stochastic scheme
• Adaptive 2/3-demisional parameters adjustment

59
Plans for FY2010 (Cont.)

• NAEFS
• New NAEFS component FNMOC global ensemble
• 4Q 2009 2Q 2010
• Pending on one year evaluation (May 2008 April
  2009)
• Using mini-Bayesian method for first moment
  correction
• Need to coordinate with CMC/MSC
• Improving NAEFS products
• Introduce full Bayesian model to calibrate high
  moments
• For precipitation forecast
• All variables
• Statistical down-scaling
• Precipitation
• Improving current method
• Adding new variables (pending on RTMA
  availability)
• TC related products
• Including bias correction
• Seamless weather-climate interface
• Merge GEFS and CFS

60
Background!!!
61
Acknowledgements
EMC Richard Wobus, Dingchen Hou, Bo
Cui Malaquias Pena, Weiyu Yang, Julia Zhu,
Mozheng Wei, Mike Charles, Yucheng Song, Jun
Du, Mark Iredell, John Ward and Steve Lord CPC
Qing Zhang, Jon Gottschalck, Jae Schemm NCO
Christine Caruso Magee, Joey Carr, Brent Gorden,
Daniel Starosta MDL Kathryn Gilbert MSC/Canada
Lewis Poulin and Andre Methot
62
CCS resources (estimated)

• Computation (Current)
• T126L28 out to 384 hours
• Assigned window (75min)
• Actually using 45 minutes
• Average 38 nodes
• Computation (future)
• T190L28 out to 384 hours
• Use 50 min
• Average 60 nodes
• 75 additional computer resources

• Space (current)
• T126L28 out to 384 hours
• Pgrba files
• 17 days on CCS for bias correction
• 55G (x4 per a day)
• Space (future)
• T190L28 out to 384 hours
• Pgrba files
• 17 days on CCS for bias correction
• 83G needed (x4 for a day)

63
NEXT NAEFS exchange pgrba files
64
NEXT NAEFS pgrba_bc files (bias correction)
65
Horizontal resolution change Ensemble control
only (deterministic) From T126 to T190 NH 500hPa
geopotential height
26 4
Gains from short waves
66
Horizontal diffusions Ensemble controls only
OPR(T126)-4th order NHD(T126)-8th order
May 2007
November 2007
67
Resolution and Diffusion for Global Ensemble
Without Stochastic
E20s T126 4th for all 16d (oper.) E20x T190
8th out to 16d E20e T190 8th (0-180h), then
T126 4th
When reducing resolution from T190 (8th order) to
T126 (4th order), the ensemble forecast
probabilistic skill score tends to t126
immediately, the example shows here for tropical
850hPa temperature. 8th order diffusion for t126
somewhat improves performance (not show here).
Therefore, both the resolution and diffusion play
an important role here.
68
Conclusion

• Based on two sets of retrospective runs (summer
  and winter 2007)
• New package improved the forecast skill (score)
  significantly
• For deterministic (ensemble mean)
• For probabilistic (ensemble distribution)
• The better results is benefited from
• Increase horizontal resolution (include
  diffusion)
• Stochastic perturbation scheme
• Better initial condition (analysis)
• Better forecast model (GFS)

69
Remain Issues

• Tune initial perturbation (may leave this to next
  implementation)
• Need to adjust the size of initial perturbation,
  due to
• Model resolution changed
• Model diffusion scheme changed
• Improved analysis
• Experiments are running, but very slow
• Due to limit computation resource
• Examine bias corrected forecast and down scaling
  forecast
• No enough CCS disk storage for 17d pgrb files on
  line
• Verify tropical storm tracks
• Working on 2007 summer season
• Planned for 2008 summer season
• Resource problem
• Personnel
• Computation and storage

70
Downstream Dependencies

• Sigma files
• SREF
• Yes
• It uses sigma forecast
• Wave ensemble
• No
• It uses bias corrected 10m winds
• Tracking
• No
• It uses pgrba file
• MDL GMOS
• No
• It uses pgrba and pgrbb files
• Public access
• No
• We dont post sigma files to public

• pgrb files (pgrba pgrbb)
• SREF
• No
• It produces pgrb file by itself
• Wave ensemble
• Yes
• But file has the same format for 10m wind
• Tracking
• Yes
• But it uses pgrba file only, the file has the
  same format
• MDL GMOS
• Yes
• It uses both pgrba and pgrbb
• Public access
• Yes
• pgrba and pgrbb