Expressing Forecast Uncertainty

Document Sample
Expressing Forecast Uncertainty Powered By Docstoc
					Expressing Forecast
   Uncertainty
          Jeff Craven
 Science and Operations Officer
  NWS Milwaukee/Sullivan WI
NWA Annual Meeting, October 2006
    Expressing Forecast
   Uncertainty at a NWS
      Forecast Office
 Routine Temperature and Precipitation
Forecasts at NWS Milwaukee/Sullivan WI
      (Began at NWS Jackson MS)
           (a work in progress)
Part of WAS *IS (Weather and
  Society Integrated Studies)

   A subgroup of WAS *IS social and
physical scientists working to find ways to
 express uncertainty in both routine and
hazardous weather forecasts and warnings
WAS *IS (Weather and Society
    Integrated Studies)
    http://www.sip.ucar.edu/wasis/
      Process and Methods
• 1) Discussed at NWS Jackson staff
  meeting. Divergence of opinions (and
  passion) was much greater than
  anticipated

• 2) Formed a local team composed of
  two non-meteorologists, three
  forecasters, an intern, and myself to
  mold a plan
       Process and Methods
• 3) Also had a mini-focus group with group of
  8 Emergency Managers and 1 TV
  Meteorologist to discuss plan

• 4) Had a break out discussion group at Annual
  Emergency Manager/Media workshop at
  Grenada MS on expressing uncertainty in both
  routine and hazardous weather forecasts

• 5) First cut in March 2006. Final draft after
  input from various WAS *IS and NWS
  scientists in August 2006
       Process and Methods

• Did research to see what the typical
  error is in our High and Low
  Temperature forecasts at KMKE
  (Milwaukee) and MSN (Madison MS).
         Example MAE for KMKE (October)
                         October 2004/2005 MAE for MKE Point Forecasts

         7                                                                                     6.56

         6                                                                               5.6
                                                                              5.3 5.21
         5                                                 4.79 4.91 4.97
                                                    4.44
                                            4.08
         4
MAE oF




                                3.14 3.29
         3          2.57 2.74
             2.37
         2


         1


         0
              1      2     3     4    5      6        7      8       9   10   11   12    13     14
                                                   Forecast Period
      Process and Methods
• Found that it varied from 2-5 degrees
  in July/August to 3-8 degrees in
  December. Error gets larger as time
  goes on in forecast period (chaos and
  predictability, as expected)

• Used that information to create “error
  bars” that increase with time
      Process and Methods
• Utilize GFS Ensemble Guidance (MEN)
                 Definitions
• Forecast Degree of Difficulty/Uncertainty
• Think Diving Competition

• Degree of spread from GFS ensemble
  standard deviation (MENMKE, MENMSN)

•   Very Low :     0 degree,    0-3%
•   Low:           1 degree,    4-6%
•   Medium :       2 degrees,    7-9%
•   High:          3 degrees,  10-12%
•   Extreme:       4+ degrees,    13+%
Deliverables (example)
Deliverables (example)
    Feedback from Focus Group (EM)

•    I, personally, very much appreciate the
     “honesty in forecasting” that the Jackson
     Office has always expressed. I think it
     enhances your credibility when you are
     up front about your uncertainty. This
     product would be another useful tool to
     me when advising local authorities,
     citizens, and others about weather
     events.
    Feedback from Focus Group (Media)

•    I think this is a nice step towards a more
     "realistic" forecast. Many people tend to
     place too much stock in the exact details
     of a seven day forecast. While your
     point about retaining credibility is
     certainly valid, I think a graphical
     expression of uncertainty would help
     educate some about the pitfalls of the 5-
     7 day period (and in some cases, the 1-3
     day period).
 Expressing Forecast
Uncertainty at a NWS
   Forecast Office
Hazardous Weather Probability
 Forecasts at NWS Jackson
     (work already done)
http://www.srh.noaa.gov/jan/HWO.php
Cheat Sheet
   Verification of Forecasts
• Example from April 6th Tornado
  Outbreak in Jackson MS CWA

• Rather subjective
         Comments, Questions?
• Thanks for your interest and kind attention
http://www.crh.noaa.gov/mkx/?n=experimental-fcst-uncertainty




                                  WI Fall Colors