Introduction of the North American Ensemble Forecast System (NAEFS)

1
Introduction ofthe North American Ensemble
Forecast System(NAEFS)

• Yuejian Zhu
• Zoltan Toth (NCEP) and Louis Lefaivre (CMC)
• Environmental Modeling Center
• NOAA/NWS/NCEP
• Acknowledgements
• Richard Wobus and Bo Cui EMC/NCEP
• Ed OLenic, David Unger CPC/NCEP
• Brent Gorden, David Michaud NCO/NCEP
• Richard Verret, Lawrence Wilson CMC/MSC

2
PROJECT HISTORY MILESTONES

• February 2003, Long Beach, CA
• NOAA / MSC high level agreement about joint
  ensemble research/development work
• (J. Hayes, L. Uccellini, D. Rogers, M. Beland, P.
  Dubreuil, J. Abraham)
• May 2003, Montreal (MSC)
• 1st NAEFS Workshop, planning started
• November 2003, MSC NWS
• 1st draft of NAEFS Research, Development
  Implementation Plan complete
• May 2004, Camp Springs, MD (NCEP)
• Executive Review
• September 2004, MSC NWS
• Initial Operational Capability implemented at MSC
  NWS
• November 2004, Camp Springs
• Inauguration ceremony 2nd NAEFS Workshop
• Leaders of NMS of Canada, Mexico, USA signed
  memorandum
• 50 scientists from 5 countries 8 agencies
• March 2006, MSC NWS
• 1st Operational Implementation
• Bias correction
• Climate anomaly forecasts

3
INAUGURATIONCEREMONY
4
NAEFS ORGANIZATION
Meteorological Service of Canada National Weather
Service, USA MSC NWS
PROJECT OVERSIGHT
Michel Beland, Director, ACSD Pierre Dubreuil,
Director, AEPD Jim Abraham, MRB
Louis Uccellini (Director, NCEP/NWS) Greg Mandt
(Director, OST/NWS) Steve Lord, EMC
PROJECT CO-LEADERS
Louis Lefaivre (Implementation) Gilbert Brunet
(Science)
Zoltan Toth (Science) David Michaud / Brent
Gordon (Impl.)
JOINT TEAM MEMBERS
Meteorological Research Branch MRB Peter
Houtekamer, Herschel Mitchell, Lawrence
Wilson Canadian Meteorological Center CMC Yves
Pelletier, Gerard Pellerin, Richard Verret, Alain
Patoine, Manon Lajoie 
Environmental Modeling Center EMC Yuejian Zhu, Bo
Cui, Richard Wobus NCO Maxine Brown, Scott
Jacobs HPC Peter Manousos Storm Prediction
Center David Bright Climate Prediction Center
CPC Ed OLenic,Mike Halpert , David Unger NWS
Richard Grumm, Fred Branski
National Meteorological Service of Mexico (NMSM)
joined in Nov. 2004 Acknowledgements to J.
Whitaker, T. Hamill, Y. Gel, R. Krzysztofowicz
5
CONCEPT OF OPERATIONS

• Exchange 50 selected variables
• Use GRIB2 to reduce volume of data
• Generate basic products using same
  algorithms/codes
• Reduce systematic error
• Bias estimation
• Combine two ensembles
• Determine weights
• Express forecast in terms of climatological
  anomalies
• Compare forecast with reanalysis climate
  distribution
• Generate center-specific end products NCEP
• Graphical products (NA, NH, Caribbean, South
  America, and AMMA areas)
• NCEP official web site (gif)
• NCEP Service Centers (NAWIPS metafile)
• Gridded products
• NAWIPS grids
• NCEP Service Centers
• GRIB2 format
• Products of general interest (Possible ftp
  distribution)
• NDGD (10-50-90 percentile forecast value
  associated climate percentile)

6
CONFIGURATION, DATA EXCHANGE

• Ensemble configuration
• Resolution - Increased for days 8-16 in Aug 05
• Membership - Increase from 10 per cycle to 20
  in 2 steps
• Add 4 members Feb. 2006
• Computing limitations to further expansion
• Add additional 6 members Feb. 2007 at the latest
• Data exchange
• Currently internet is used to exchange ensemble
  data between MSC NWS
• 1-hr transmission time
• Assess reliability, pursue alternative solution
  if needed
• GRIB2
• NCEP getting ready
• Consistent with TIGGE requirements

7
CONFIGURATION, OUTPUT CHARACTERISTICS
2005, 2006, 2007, 2008
8
BASIC PRODUCTS

• NAEFS basic product list
• Bias corrected members of joint MSC-NCEP ensemble
• 40 members, 35 of NAEFS variables, GRIB2
• Bias correction against each centers own
  operational analysis
• Weights for each member for creating joint
  ensemble
• 40 members, independent of variables, GRIB2
• Weights depend on geographical location (low
  precision packing)
• Climate anomaly percentiles for each member
• 40 members, 19 of NAEFS variables, GRIB2
• Non-dimensional unit, allows downscaling of
  scalar variables to any local climatology
• Issues Products to be added in future years
• Bias correction on precipitation some other
  variables not corrected yet)
• Use CMORPH satellite-based analysis of
  precipitation rates
• CPC collaborators (J. Janowiak)
• Climate anomalies for missing variables
• Need to process reanalysis data to describe
  climatology for missing variables

9
RAW BASIC PRODUCT AVAILABILITY
2005, 2006, 2007, 2008
10
END PRODUCTS

• End product generation
• Can be center specific
• Need to conform with procedures/requirements
  established at different centers
• End products generated at NCEP
• Based on prioritized list of requests from NCEP
  Service Centers
• Graphical products (including Caribbean, South
  American, and AMMA areas)
• NCEP official web site (gif NA, Caribbean, SA,
  AMMA)
• NCEP Service Centers (NAWIPS metafile)
• Gridded products
• NAWIPS grids
• NCEP Service Centers (list of 661 products)
• GRIB2 format
• Products of general interest (Possible ftp
  distribution, no decision yet on products)
• NDGD (10-50-90 percentile forecast value
  associated climate percentile)
• End products generated at MSC
• TBD
• End products generated jointly
• Experimental probabilistic Week-2 forecast
• Fully automated, based on basic products bias
  corrected, weighted climate anomalies

11
ENSEMBLE PRODUCTS - FUNCTIONALITIES
List of centrally/locally/interactively generated
products required by NCEP Service Centers for
each functionality are provided in attached
tables (eg., MSLP, Z,T,U,V,RH, etc, at
925,850,700,500, 400, 300, 250, 100, etc hPa)
FUNCTIONALITY CENTRALLY GENERATED LOCALLY GENERATED INTERACTIVE ACCESS
1 Mean of selected members Done
2 Spread of selected members Done
3 Median of selected values Sept. 2005
4 Lowest value in selected members Sept. 2005
5 Highest value in selected members Sept. 2005
6 Range between lowest and highest values Sept. 2005
7 Univariate exceedance probabilities for a selectable threshold value FY06?
8 Multivariate (up to 5) exceedance probabilities for a selectable threshold value FY06?
9 Forecast value associated with selected univariate percentile value Sept. 2005 - FY06?
10 Tracking center of maxima or minima in a gridded field (eg low pressure centers) Sept. 2005, Data flow FY06?
11 Objective grouping of members FY08?
12 Plot Frequency / Fitted probability density function at selected location/time (lower priority) FY07?
13 Plot Frequency / Fitted probability density as a function of forecast lead time, at selected location (lower priority) FY07?
Potentially useful functionalities that need
further development - Mean/Spread/Median/Ranges
for amplitude of specific features -
Mean/Spread/Median/Ranges for phase of specific
features
Additional basic GUI functionalities - Ability
to manually select/identify members - Ability to
weight selected members Sept. 2005
12
ENSEMBLE PRODUCT REQUEST LIST NCEP SERVICE
CENTERS, OTHER PROJECTS
13
Example of PQPF from NCEP and CMCs GEFS
14
Example of PQPF and CPQPF from NCEP GEFS
15
Ensemble size 10 members
16
(No Transcript)
17
Based on raw forecasts, no climate and current
analysis correction
18
NDGD FORECAST UNCERTAINTY - DOWNSCALING

• Ensemble uncertainty information
• Sent on NDGD grid for convenience (if no big
  overhead)
• Valid on model grids (32km for regional, 110 km
  for global ensemble)
• How to bridge gap between model and NDGD grids?
• Anomaly uncertainty information proposed
  methodology
• Establish reanalysis climatology
• In progress for global (NAEFS), methods can be
  transferred to regional reanalysis
• Bias correct ensemble forecasts (wrt operational
  analysis)
• Take 10-50-90 percentile values from bias
  corrected ensemble
• (For establishing anomaly forecasts, adjust
  10-50-90 percentile values to look like
  re-analysis)
• Check climatological percentile corresponding to
  10-50-90 forecast percentiles
• Provide climatological percentiles corresponding
  to 10-50-90 percentile forecast values as second
  set of guidance products

19
ENSEMBLE 10-, 50- (MEDIAN) 90-PERCENTILE
FORECAST VALUES (BLACK CONTOURS) AND
CORRESPONDING CLIMATE PERCENTILES (SHADES OF
COLOR)
Example of probabilistic forecast in terms of
climatology
20
EXPANSION OF NAEFS

• Discussions with other centers on expansion of
  NAEFS
• Experimental data exchange - March 2006
• UKMet, FNMOC, AFWA
• Operational status 2007-2008?
• Need to formalize
• Issues
• Name change from NAEFS to Global Ensemble
  Prediction System (GEPS)?
• Disc space requirements will grow
• Other centers that expressed interest in learning
  more about NAEFS
• ECMWF, NCMRWF, JMA, KMA, CMA
• Link with THORPEX Interactive Grand Global
  Ensemble (TIGGE)
• THORPEX research organized in 4 science Working
  Groups
• TIGGE data base supports ensemble-related
  research
• NAEFS GEPS provides
• Testing in and transition to operational use
• Real time forecast data for demonstration projects

21
NAEFS THORPEX

• Expands international collaboration
• Mexico joined in November 2004
• UK Met Office to join in 2006
• Provides framework for transitioning research
  into operations
• Prototype for ensemble component of THORPEX
  legacy forecast system
• Global Interactive Forecast System
  (GIFS)

RESEARCH
THORPEX Interactive Grand Global Ensemble (TIGGE)
THORPEX
RESEARCH
Articulates operational needs
Transfers New methods
North American Ensemble Forecast System (NAEFS)
OPERATIONAL
LEGACY (GIFS)
OPERATIONS
22
Performance Measures

• THORPEX focuses on high impact weather
• PQPF
• Extreme temperature
• Severe weather
• Tropical storm
• Common verification methods are still useful, but
• Specific and additional measurements are required
  for
• Probabilistic forecasts
• Extreme weather events
• One of THORPEXs goals to interactive with NAEFS
• Increase the rate of improvement of weather and
  water forecasts with the additional one day with
  useful skill of decade to two days (48 hours
  total) per decade by 2015
• Refer to day 7 weather forecast skill scores in
  FY2004
• Examples of NCEP GEFS performance based on 500
  hPa height
• PAC, RMS
• ROC, EV, RPSS and BSS

23
NCEP ensemble mean performance for past 6-year
24
Ranked probabilistic skill scores
NCEP ensemble probabilistic performance for past
6-year
Economic values for 110 cost/loss ratio
25
What is THORPEXs goal for 10 years ?
26
Additional information
27
Basic Products Summary

• Bias corrected forecasts
• Consider 35 variables in the first phase
• Statistical weights
• Consider 35 variables in the first phase
• Anomaly forecasts
• Consider 17 variables in the first phase
• NAWIPS grids and graphics
• NDGD grids

28
Bias Correction

• New products from bias correction method
• These new products is bias corrected products by
  using globally decaying weights 2 (approximately
  45-50 days) different of analysis and forecast
• New products will be
• For all 35 variables
• 1.01.0 degree resolution
• 4 cycles (00, 06, 12 and 18UTC)
• Up to 386 hours (16 days) forecasts

29
List of Variables for Bias Correction/weightsfor
CMC NCEP Ensemble
30
Climate Anomaly- Input climate/forecast data

• NCEP/NCAR reanalysis data
• 4 cycles (00UTC, 06UTC, 12UTC and 18UTC) per day
• 40 years (Jan. 1st 1959 Dec. 31th 1998)
• Need to consider the systematic difference
  between NCEP/NCAR reanalysis and current analysis
  (GDAS)
• Resolution and format
• 2.52.5 (lat/lon) grid, GRIB-1 format
• 1.01.0 (lat/lon) grid, GRIB-1 format (forecast
  only)
• Variables at levels (possible to add more)
• Height 1000hPa, 700hPa, 500hPa, 250hPa
• Temperature 2m, 850hPa, 500hPa, 250hPa
• Wind 10m, 850hPa, 500hPa, 250hPa
• PRMSL, max/min temperature

31
Climatology (Estimation)

• Mean (first moment)
• To use Fourier expansion from 40 years data and
  compare following four considerations
• Considering first four Fourier modes
• Fits to daily data to obtain annual, semi-annual,
  4-month and seasonal cycle
• Standard deviation (high moments)
• To get 40 years average daily standard deviation
  first
• To calculate monthly mean of standard deviation
  from daily
• To generate a slope from month to month
• To project to daily standard deviation from
  monthly mean

32
Products (plan)

• Based on 4 Fourier modes selection
• Assuming the normal distributions of the 40 years
  climate data
• PDF will be presented by first two moments (mean
  and standard deviation)
• Considering the systematic differences between
  NCEP/NCAR reanalysis and current GDAS
• Using bias corrected forecasts
• To calculate climate anomaly
• For 1x1 degree grid point globally.
• For all 19 variables (height, temperature, wind
  and etc.)
• For each ensemble member.
• Output in percentile (0-100, 50normal).

33
Weights (estimation-discussion)

• Best member statistic method
• Count over a period of time how often each member
  is closest to the verifying analysis and
  accumulate this stats for each point separately
• Use total energy norm which includes U,V and T
  for all vertical levels
• Verifying analysis
• Use the mean of NCEP and CMC analysis
• Use the frequency data for weighting the members
• Similar to the decaying average algorithm
• Update every day for each cycle
• Application
• Use the mean of NCEP member frequency to weight
  individual members
• Use individual CMC member frequency to weight
  individually

34
NAWIPS grids data, graphic and GIF images

• Mean of selected members
• Z500, z700, z850
• Spread of selected members
• U10m, V10m
• Exceeding probabilities for selected threshold
  values
• 10m wind speed thresh 20, 34, 50, and 64 kts
• Significant wave height at various values
• Spaghetti plots
• Height 200hPa, 300hPa and 500hPa
• Psml
• T2m 0c isotherm
• QPF 0.01, 0.25, 0.5, 1.0, 2.0, 3.0 and
  4.0 (for 6-h and 24-h)
• Snow 1, 2, 4, 6, 8, 12, 18, 24 (for 6-h
  and 24-h)
• Freezing rain 0.01, 0.1, 0.25, 0.5, 1.0
  (for 6-h and 24-h)

35
NDGD grids (see functionalities 9)

• Mean of selected members
• Z500, z700, z850
• Spread of selected members
• U10m, V10m
• Exceeding probabilities for selected threshold
  values
• 10m wind speed thresh 20, 34, 50, and 64 kts
• Significant wave height at various values
• Spaghetti plots
• Height 200hPa, 300hPa and 500hPa
• Psml
• T2m 0c isotherm
• QPF 0.01, 0.25, 0.5, 1.0, 2.0, 3.0 and
  4.0 (for 6-h and 24-h)
• Snow 1, 2, 4, 6, 8, 12, 18, 24 (for 6-h
  and 24-h)
• Freezing rain 0.01, 0.1, 0.25, 0.5, 1.0
  (for 6-h and 24-h)

36
NDGD FORECAST UNCERTAINTY RECOMMENDATION

• Provide 3 ensemble-based guidance products for
  inclusion in NDGD
• 10, 50, and 90 percentile values
• SREF guidance out to day 3
• NAEFS guidance out to 16 days
• Use NDGD grid (5x5 km), with GRIB2 packing,
  minimal space overhead
• Approach
• Solicit comments on specific proposal from NCEP
  Service Centers and regions/field
• Use NAWIPS software (available soon?) to generate
  products
• Work with NAWIPS group to provide algorithm
• Simple counting of members with linear
  interpolation now
• Gaussian Kernel method in later implementation
• Factor of 3 increase in disc space
• D. Ruth positively inclined (WG member at NDFD
  Workshop)

37
ENSEMBLE-BASED PRODUCTS FOR NDGD

• National Digital Forecast Database (NDFD)
• Official NWS forecast, prepared by WFO offices
  (central guidance, coordination)
• 5x5 (2.5x2.5) km grid, out to 7 days
• Selected parameters (15)
• Available in digital format, query tools, etc
• No (minimal) provision for information on
  forecast uncertainty
• Recommendations from an NDFD workshop, Salt Lake
  City, 2003
• Interactive Forecast Preparation System (IFPS)
  offers tools to work with NDFD grids (forecasters
  can manipulate gridded data, etc)
• National Digital Guidance Database (NDGD)
• For posting numerical guidance products same way
  as NDFD
• New system, possibility to complement NDFD with
  forecast uncertainty info
• Based on global (NAEFS) and regional ensemble
  forecasts
• What forecast uncertainty info to post in NDGD?

38
NDGD FORECAST UNCERTAINTY REQUIREMENTS

• Compact (conveys uncertainty without posting all
  members)
• Add minimal new info
• Current disc, telecommunication, etc limitations
• Simple to understand and use by both trained and
  novice users
• Expand existing lines of work
• Informative without additional knowledge, tools,
  that are not yet available
• Solid scientifically based
• Can fit parametric pdf
• Allows to derive any univariate info
• Additional tools needed to use this feature
• Room for expansion
• Can easily be enhanced without major shift in
  direction
• More sophisticated methods can be added
• Possibly use Gaussian Kernel method of D. Unger

39
NDGD FORECAST UNCERTAINTY ALTERNATIVES

• Current status (in NDFD)
• Expected value (mean, median, or mode??) of
  distribution only
• Scenario 1 Add 1 variable
• Add spread to expected value (1 additional grid)
• Workshop WG felt that was not enough info
• Recommended adding 2 pieces of info
• Scenario 2 Add 2 variables
• Add info on spread on 2 sides of mean/median/mode
• 10/90 or 20-80 percentile values
• Preferred as opposed to variance (spread) info
  that is more abstract
• NDFD Workshop recommendation

40
NDGD FORECAST UNCERTAINTY QUESTIONS

• Use mean, mode, or median in NDGD?
• Mean Expected value
• Can fall around minimum in pdf
• Requires additional info (what percentile it
  corresponds with)
• Mode Most likely event
• Appealing heuristically (well defined meaning)
• Requires additional info (what percentile it
  corresponds with)
• Use in future when multiple modes can be
  considered?
• Median 50 percentile
• Heuristic meaning (half below, half above)
• Consistent with 10/90 (or 20/80) percentile
  approach
• Verifies similarly to ensemble mean
• No need for additional info
• Used by HPC in PQPF context
• Use 10/90 OR 20/80 percentile?
• 10/90 is more inclusive (covering explicitly 80
  of forecast distribution)