Documents
Resources
Learning Center
Upload
Plans & pricing Sign in
Sign Out
Get this document free

Gridded MOS

VIEWS: 8 PAGES: 40

									  AN OVERVIEW OF MOS
         AND
RECENT GOINGS ON AT MDL

…or, “what do MOS developers
        do all day?”

                  Michael N. Baker
             michael.n.baker@noaa.gov
        Meteorological Development Laboratory
             Statistical Modeling Branch
                      7 April 2010
                  Outline
          (wide range of topics!)
• Overview of MOS
• GFS MOS Refresh
    – Snow development & use of NCDC COOP data (for
      example)
•   NAM-NMM MOS Development
•   Gridded MOS
•   Day 8-10 GFS MOS
•   Future Work
•   Acknowledgements/References
    Model Output Statistics (MOS):
             Overview
•    Relates observed weather elements (PREDICTANDS)
     to appropriate variables (PREDICTORS) via a
     statistical approach.
•    Predictors are obtained from:
    – Numerical Weather Prediction (NWP) Model
        forecasts
    – Prior Surface Weather Observations
    – Geoclimatic Information (lat/lon, elevation, climo)
•    Current Statistical Method:
    – Multiple Linear Regression (Forward Selection)
                MOS Properties
• Mathematically simple, yet powerful
• Need historical record of observations at forecast points
   – Hopefully a long, stable one!
   – ~3-5 year sample at least.
• Equations are applied to future run of similar forecast
  model
   – Single station and regional eqn. developments
• Non-linearity can be modeled by using NWP variables
  and transformations
• Probability forecasts possible from a single run of NWP
  model
• Other statistical methods can be used
   – e.g. Polynomial or logistic regression
        MOS Properties (cont.)
• ADVANTAGES
  –   Recognition of model predictability
  –   Removal of some systematic model bias
  –   Optimal predictor selection
  –   Reliable probabilities
  –   Specific element and site forecasts
• DISADVANTAGES
  – Short samples
  – Changing NWP models
  – Availability and quality of observations
      Operational Station MOS:
           A Little Trivia!
• 9 million regression equations
• 75 million forecasts per day
• 1200 products sent daily

• 400,000 lines of code – mostly FORTRAN
• 180 min. supercomputer time daily

• All developed and maintained by ~ 12 MDL /
  SMB meteorologists!
    “TRADITIONAL” MOS TEXT
          PRODUCTS
• Variables include:
  – Temperature, dewpoint, max/min
    temperature, wind speed/direction, POP,
    QPF, snow, ceiling, cloud cover,
    visibility/obstructions, prob of
    thunderstorm/severe, PTYPE
• CONUS, Alaska, Hawaii, Puerto Rico
        TEXT PRODUCTS (cont.)
• GFS MAV: 6-72h for 00Z, 06Z, 12Z, 18Z cycles
• GFS MEX: 24-192h for 00Z and 12Z cycles
• NAM: 6-72h for 00Z and 12Z cycles
   – Replaced old Eta MOS product in December 2008
   – Based upon the new NAM-NMM (more details later)
• Additional text products for:
   – COOP-based Max/Min
       • NCDC cooperative observer sites
   – Marine MOS
       • Buoy locations
   – Western Pacific MOS
• BUFR files also contain all MOS probabilities that might
  not appear in text messages
           GFS MOS REFRESH
• Within past year, equations for almost all elements of GFS-based
  MOS were re-developed
   – Temperature, dewpoint, max/min, wind speed/direction, ceiling
     height, thunderstorm/severe, PTYPE, snow…new equations
   – POP/QPF, visibility/obst. to vision have same previous equations
   – Sky cover was re-developed prior to the larger refresh

• All elements now use the same station list, and newly-developed
  elements are now using the same developmental sample for all
  model cycles
   – Consistent set of data among elements
   – Starting point: GFS model from 2002
   – ~2300 METAR & Marine sites
           NEW GFS MEX MESSAGE
           (BEFORE 5-6 FEB STORM)
KDCA GFSX MOS GUIDANCE      2/04/2010 1200 UTC
FHR 24 36| 48 60| 72      84| 96 108|120 132|144 156|168 180|192
    FRI 05| SAT 06| SUN   07| MON 08| TUE 09| WED 10| THU 11|FRI CLIMO
N/X 28 35| 28 33| 17      27| 16 26| 24 36| 28 42| 23 34| 23 26 46
TMP 29 32| 31 27| 17      24| 17 25| 25 33| 30 36| 24 30| 25
DPT 23 25| 21 14| 7       11| 8 12| 15 21| 21 21| 11 11| 12
WND 4 13| 26 24| 13       10| 9    7| 6 7| 7 15| 14 15| 14
P12 13 100|100 85| 3       4| 4 16| 27 41| 41 20| 17 19| 12 22 22
P24    100|    100|        4|     16|     47|     41|     23|       32
Q12 0    4| 5    2| 0      0| 0    0| 0 1| 2       0|       |
Q24      3|      5|        0|      0|      1|      1|       |
T12 0    3| 1    1| 0      1| 0    0| 1 1| 1       1| 2    1| 1
T24       | 3     | 1       | 1     | 1     | 2     | 3     | 2
PZP 14 16| 31 13| 16      14| 20 16| 21 28| 38 20| 23 19| 24
PSN 86 84| 57 86| 83      84| 77 80| 73 59| 44 59| 59 58| 52
PRS 0    0| 12   0| 0      0| 0    0| 0 4| 11      5| 1    3| 3
TYP S    S| Z    S| S      S| S    S| Z Z| Z       Z| Z    Z| Z
SNW      4|      8|        0|      0|      1|      1|       |
    Updated Snow Development
         (as an example)
• 24-h snowfall amount forecasts
  – September-May “snow season”
  – Recent GFS refresh: we used 2002-03
    through 2007-08
• Snow amount categories:
    •   0.1 to < 2” (cat 1)
    •   2.0” to < 4” (cat 2)
    •   4.0” to < 6” (cat 4)
    •   6.0” to < 8” (cat 6)
    •   ≥8.0” (cat 8)
   Snow Development (cont.)
• Actually 3 developments involved…all for
  the price of just one MOS developer!☺
  – Prob of precip (POP)
  – Conditional prob of snow (CPOS)
  – Conditional prob of snow amount (CSNOW)
• Unconditional prob of snow amount and
  “best” categories derived from these
   Snow Development (cont.)
• Added gridded monthly snow climatology
  (relative frequencies) as a predictor
  – based upon a 30-year NCDC climo at many
    COOP locations
  – Only over CONUS, still working on AK (very
    few sites available)
  – Uses same techniques as gridded MOS
  – Relative frequency thresholds for ≥0.1” and
    ≥2.0” (measurable and “more significant”)
January CONUS snowfall relative frequency, 0.1” threshold
   Snow Development (cont.)
• Regional development
  – Slightly modified snow regions from previous
  – Dev sites “pooled” into climatologically similar
    regions
  – Equation applied to any site within that region
  – Useful for “rare” events
        NCDC COOP Data
• MOS snow based upon NCDC COOP
  dataset (predictand)
 – Archived by MDL, data back to 1998
 – Used for COOP max/min as well as snow
 – Only dense data network we have to derive
   MOS snow
 – Out of ~14000 possible sites, only about half
   are used as “quality sites” that report
   consistently (the more sites, the better!)
           COOP Data (cont.)
• COOP sites do not all report at the same time
  – Must “group” obs and set as valid at particular time
  – Nominal 24-h periods defined as 00Z-00Z and 12Z-
    12Z (note that they overlap)
  – Developed independently, with different stations
  – GFS MAV (& NAM): use 12Z-12Z period
  – GFS MEX: uses 00Z-00Z period
  – Gridded MOS: uses 00Z-00Z and 12Z-12Z
 NAM-NMM MOS Development
• With the Eta model being replaced by the
  (now current) NAM-NMM, a corresponding
  MOS product had to be formulated
  – Applying old Eta equations to new NAM
    model data produced notable and
    unacceptable errors for some elements
  – In particular, over Alaska and for single-
    station equations (such as temperature, wind)
  – Regionalized equations not as badly affected
   Eta applied to NMM: Problems

                                          Mean Absolute Error - 00Z Temperature
                                             335 Stations - March-May 2006

                          5.00
                          4.50
            MAE (deg F)




                          4.00
                          3.50
                                                                                                ETAMOS
                          3.00
                                                                                                ETA-NMM
                          2.50
                          2.00
                                 6   12   18 24   30   36   42 48   54   60   66 72   78   84
                                                       PROJECTION (h)




Note greater error applying Eta eqns. to NMM compared to its Eta MOS counterpart
Eta applied to NMM: Problems
• Several possible reasons for the issues:
  – Eta equations were rather old (developed ~5 years or
    more previously), used primarily older Eta model data
  – Eta itself underwent changes in time
  – Even though new NAM-NMM was “similar” to last
    incarnation of the Eta, changes at that point may have
    been enough to have problems appear
  – Primarily an issue with single station elements, not
    quite so much with regional ones
     Short Sample NAM-NMM
• MDL performed a study on the ability to create a
  new MOS system based on a short available
  sample of data
  – Cool season, October-March
  – Prototype system: combined one year of Eta (its last
    year) with one year of NMM model data
  – Results were positive: performed at least as well as
    the old Eta MOS, and in some instances better
  – Prototype system was significant improvement over
    applying old Eta equations to the NMM model
NAM-NMM Prototype
                             NAM MOS TEMPERATURE MAE
                                  PROTOTYPE TEST

                6.00

                                                                             Similar to old
  MAE (deg F)
                5.00
                                                                  PROTO
                                                                             Eta, much
                4.00                                              ETA-NMM
                                                                             better than Eta
                                                                  ETAMOS
                3.00                                                         on NMM
                2.00
                       6 12 18 24 30 36 42 48 54 60 66 72 78 84
                                   PROJECTION (h)            335 Sites

                               NAM MOS DEWPOINT BIAS
                                  PROTOTYPE TEST

                3.00
                2.50
 BIAS (deg F)




                2.00                                              PROTO
                1.50                                              ETA-NMM
                                                                            Note greatly
                1.00                                              ETAMOS
                                                                            improved dewpoint
                0.50                                                        bias!
                0.00
                       6 12 18 24 30 36 42 48 54 60 66 72 78 84
                                   PROJECTION (h)
                                                              335 Sites
       NAM-NMM Implemented
• We concluded that if the changed model is “substantively
  similar” to the previous version, combining these can
  provide a stable enough sample even if it’s relatively
  short
• Based on prototype results, we moved forward with a
  new operational NAM-based MOS product
   – New NAM MOS development as implemented was done using
     one year of Eta and two years of NMM, for several elements
     (e.g., temp, wind, POP)
   – Similarly positive verification results
   – Other elements not immediately re-developed can be revised in
     time
                Gridded MOS
• Goal of Gridded MOS:
  – Guidance on high-resolution grids
  – Correspond to NDFD elements, at same temporal
    and spatial resolution
  – Maintain comparable skill and accuracy to station-
    based MOS

• MOS station-based forecasts analyzed onto grid
  – BCDG (Bergthorssen, Cressman, Doos, Glahn)
    technique of successive corrections, distinguishes
    between land and water, and makes elevation
    adjustments on-the-fly based on the data.
  – There are many options available to fine-tune the
    analysis.
Station Distribution
      Gridded MOS Weather Elements
Guidance on a grid commensurate with the resolution being used
by local forecasters in producing their local grids (5 km or finer)


                                              •Surface Temperature
                                              •Dew Point
                                              •Max/Min Temperature
                                              •Relative Humidity
                                              •Wind Speed/Direction
                                              •Wind Gusts
                                              •Probability of Precipitation
                                              •Precipitation Amount
                                              •Opaque Sky Cover
                                              •Snowfall Amount
                                              •Thunderstorms
            Gridded MOS Milestones
•   August 2006 – Initial implementation over the CONUS (5 km)
     – max/min, temperature, dewpoint, relative humidity, wind
       speed/direction, (Prob) thunderstorms, Probability of Precipitation.
•   June 2007 – Added more CONUS elements
     – opaque sky cover, 24-hr snowfall amount, 6- and 12-h precipitation
       amount, and wind gusts.
•   June 2008 - Initial implementation over Alaska (3 km)
     – max/min, temperature, dewpoint, (Prob) thunderstorms, relative
       humidity
•   December 2008 – Added more Alaska grids:
     – sky cover, PoP, and wind grids
•   January 2010 – Added 6- and 12-h QPF and 24-h snowfall guidance to AK
•   March 2010 – Added extended range 24-h snowfall guidance
     – Both 00Z and 12Z cycle now go out to 162-h, CONUS and AK
• (Day 8-10 guidance, coming soon!)
      GRIDDED MOS QPF
• Use NPVU QPE data as predictand rather
  than METARs as currently done
• Will replace gridded QPF and POP over
  CONUS within next year
• Will NOT affect text products
• Jess Charba: jerome.charba@noaa.gov
• www.weather.gov/mdl/hrqpf
HPC Day 8-10 MOS Requirements

             •maximum temperature every 24 hours
             •minimum temperature every 24 hours
             •dewpoint every 12 hours
             •12-h probability of precipitation every
             12 hours
             •velocity components every 6 hours
        Day 8-10 GFS MOS
• Plans afoot to implement extended range
  GFS MOS for days 8-10
  – No traditional text product
  – Station-based product supplied in BUFR
    • Should be operational in July 2010
  – Gridded MOS product in GRIB2
    • Development in progress
    • Currently making forecasts “in parallel”
    • Not available for viewing yet
        Day 8-10 GFS MOS
• Elements included:
  – Temperature, every 6-h
  – Dewpoint, every 6-h
  – Daytime maximum/nighttime minimum
  – RH, every 6-h
  – Wind speed/direction (develop u, v, speed),
    every 6-h
  – 12-h POP, every 12-h
         Current Development and
               Future Work
• Day 8-10 GFS MOS (coming soon)
• Hawaii/Puerto Rico gridded MOS
• 2.5-km CONUS gridded MOS
• Investigating use of OpenMP in MOS codes to
  accelerate computation time for finer grids
• Possible requirements for snow
    – 6-hourly snow amounts (how best to accomplish this?)
    – New categories for ≥10” and ≥12”?
• Weather grids to support NDFD
• Transition NAM MOS to upcoming NEMS NMM-B,
  replacing the WRF-NMM
           Acknowledgements
• Many within Statistical Modeling Branch of MDL
  contributed (to this talk and in general):
  –   Kathryn Gilbert: Branch chief
  –   Mark Antolik: QPF, Science lead, MOS software
  –   Kari Sheets: POPC, Geospatial task lead
  –   Joseph Maloney: POP, Operations task lead
  –   Phil Shafer: Thunderstorms/severe, Operations
  –   Eric Engle: Winds, Mesonet data archive
  –   Geoffrey Wagner: Gridded MOS, GIS
  –   Wei Yan: Clouds, Ceiling height, Visibility
  –   James Su: Pacific MOS
  –   Yun Fan: OpenMP, Gridded MOS
                                References
•   Antolik, M.S., and M.N. Baker, 2009: On the ability to develop MOS guidance with short
    dependent samples from an evolving numerical model. Preprints, 23rd Conf. on Wea. Analysis and
    Forecasting/19th Conf. on Numerical Wea. Prediction, Omaha, NE, Amer. Meteor. Soc.

•   Baker, M.N., K.L. Sheets, and G.A. Wagner, 2009: Using gridded MOS techniques to derive
    snowfall climatologies. Preprints, 23rd Conf. on Wea. Analysis and Forecasting/19th Conf. on
    Numerical Wea. Prediction, Omaha, NE, Amer. Meteor. Soc.

•   Gilbert, K.K, H.R. Glahn, R. Cosgrove, K. Sheets, and G. Wagner, 2009: Gridded model output
    statistics: Improving and expanding. Preprints, 23rd Conf. on Wea. Analysis and Forecasting/19th
    Conf. on Numerical Wea.

•   Glahn, H.R., and D.A. Lowry, 1972: The use of model output statistics (MOS) in objective weather
    forecasting. J. Appl. Meteor., 11, 1203-1211.

•   Glahn, H.R., K. Gilbert, R. Cosgrove, D.P. Ruth, and K. Sheets, 2009: The gridding of MOS. Wea.
    Forecasting, 24, 520-529.

•   Maloney, J., E. Engle, P. Shafer, and G. Wagner, 2009: The NMM MOS replacement for the Eta
    MOS. Preprints, 23rd Conf. on Wea. Analysis and Forecasting/19th Conf. on Numerical Wea.
    MOS & Gridded MOS Support
              (Links)
• MOS products page:
   – http://www.nws.noaa.gov/mdl/synop/products.php
• Gridded MOS description:
   – http://www.nws.noaa.gov/mdl/synop/gmos.php
• Gridded MOS QPF:
   – www.weather.gov/mdl/hrqpf
• MOS Info discussion group/forum (internal to NOAA
  only):
   – http://infolist.nws.noaa.gov/read/all_forums/subscribe?name=mo
     s_info
• Mike Baker e-mail:
   – michael.n.baker@noaa.gov
10 February 2010

								
To top