Bias correction methods 1: adjusting moments or pdf

Document Sample
Bias correction methods 1: adjusting moments or pdf Powered By Docstoc
					      Bias Correction / Weighting
Review of Ongoing Development /Testing


  Bo Cui1, Zoltan Toth2, Yuejian Zhu2, Richard Verret3, David Unger4

                1SAIC at Environmental Modeling Center, NCEP/NWS
                    2Environmental Modeling Center, NCEP/NWS

          3Canadian Meteorological Centre, Meteorological Service of Canada

                      4Climate Prediction Center, NCEP/NWS




                          Acknowledgements
   Stéphane Beauregard, Richard Wobus, Dingchen Hou, Malaquias Pena

                                                                              1
                           Ongoing Development

 Bias correct NCEP ensemble mean
   • Current system: choose a fixed globally decaying weight 2 % and apply it on all 35 variables
     (1x1 degree, 4 cycle per day: 00z, 06z,12z and 18z)
   • Daily update: bias estimation
   • Result check: raw, bias-corrected ensemble mean and mean bias comparison; time series
     of raw and bias corrected ensemble mean at two points
     ( http://www.emc.ncep.noaa.gov/gmb/wx20cb/Bias_Correction_Algorithm/1st_2nd_Moment
     s/Training_1month/NAEFS/test)


 Bias correct CMC member individually
   • Need analysis data with 06 interval ( available? )
   • Use initial data of the control test as analysis to:
        • estimate the bias of individual member


 Estimate bias of the CDAS reanalysis for anomaly forecast
   • Daily update the difference between CDAS reanalysis and CMC oper. analysis
   • Daily update the difference between CDAS reanalysis and NCEP oper. analysis

                                                                                           2
                            Ongoing Testing

   Weighting
    1. Best member stat. method (Yuejian Zhu): count over a period of time how often
       each member is closest to the verifying analysis and accumulate this stat for
       each point separately. Use the frequency data for weighting the members.
       Similar to the decaying average algorithm, update the prior every day.
         • choose total energy norm (involves variables U,V and T for all levels)
         • choose the mean of NCEP and CMC analysis as the verifying analysis
         • choose the mean of all NCEP member frequency to weight all NCEP
           member
         • choose individual CMC member frequency to weight individually

    2. Default: use equal weight for all members in the combined ensemble for next
       spring implementation




                                                                                3
                           Computer Usage
 1 fcst. lead time, 2 member mean, 35 variables (update bias estimation)

   real 0m28.15s
   user 0m8.38s
   sys 0m1.01s

 29 fcst. lead time, 2 member mean, 35 variables (update bias estimation)

   real 4m55.80s
   user 0m27.49s
   sys 0m4.26s

 29 fcst. lead time, 35 variables (debias)

   real 1m26.56s
   user 0m19.65s
   sys 0m4.20s


                                                                        4

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:0
posted:10/1/2012
language:English
pages:4