Improving ExtendedRange Forecasts through Analog Ensemble Calibration Edward OLenic and Scott Handel

1
Improving Extended-Range Forecasts through
Analog Ensemble CalibrationEdward OLenic and
Scott Handel NOAA-NWS-Climate Prediction
Center
2

• Purpose to improve the skill of probabilistic
  6-10-day and 8-14-day forecasts of 2m T.
• 10 best analogs in the historical record
  (Re-analysis) to the operational GFS 500-hPa
  height forecast for each of the 23 members of the
  forecast system in a 35-day window centered on
  the forecast date.
• The corresponding 230 2m T maps are used to
  formulate probabilistic 3-class forecasts.
• Compare the reliability of these forecasts with
  that for 3-class probability forecasts of two
  meter temperature composed from un-calibrated raw
  GFS ensemble 2m temperature forecasts.
• The analog ensemble forecasts are much better
  calibrated than those from the raw surface
  temperature forecasts.
• Graphs of spread-skill indicate that the analog
  ensembles
• Have rmse about twice that of the GFS.
• Show a wider range of spread than the GFS, but
  only in the warm season.
• Categorical skill of analog ensemble 6-10 day
  surface temperature is comparable to that from
  the CDC re-forecast.

3
Official 6-10-day 2m Temperature
4
Average of the ten best analogs to the pattern.
The top 5 analog dates are listed in the lower
right-hand corner. The correlation of this
10-analog mean map with the original forecast map
over the PNA region is listed in the upper left.
Forecast made Oct 12, 2004.
Correlate the GFS forecast 500-hPa height
pattern, over the Pacific North American Region
(20-90N, 175E-60W), with observed patterns in the
NCAR-NCEP Re-analysis from 1950-1999 in a 35-day
seasonal window centered on the forecast date.
5
Probability of each of three classes of 2m
temperature. Forecast made Oct 12, 2004. Top
un-calibrated GFS 23-member ensemble. Bottom
230-member analog ensemble.
6
10-analogs to GFS ENS mean Z500
Uncalibrated 23-member GFS
2-meter Temperature
Reforecast
230-analogs
7
500-hPa height Oct 2003-Oct 2004
8
500-hPa height Oct-Mar 2003-4
500-hPa height Apr-Sep 2004
9
6-10-day 2m Temperature Categorical Skillvs
Analogs and ReforecastS ((C-E)/(T-E))100.
Summer 24.98 Winter 17.41 Summer 26.43 Winter
23.99
Summer 19.95 Winter 25.49 Summer 26.43 Winter
23.99
10
Summary

• 230 member ensembles of the 10 best analogs to
  the 500-hPa height forecast over the PNA region
  to each of the 23 GFS ensemble members improves
  the calibration of probabilistic 6-10-day 2m T
  forecasts, compared with un-calibrated T
  forecasts from the 23 members, and with official
  T forecasts.
• Analogs also improve the sharpness of the 2m T
  forecasts over official forecasts.
• Graphs of 500-hPa spread-skill indicate that the
  analog ensembles
• Have rmse about twice that of the GFS.
• Show a wider range of spread than the GFS, but
  only in the warm season.
• Categorical skill of analog 2m T forecasts is
  higher than CDC reforecast in
• summer and lower in winter, over the 1-year
  sample.