Improvements%20in%20Skill%20of%20CPC%20Outlooks%20%20Ed%20O

1
Improvements in Skill of CPC OutlooksEd
OLenic and Ken Pelman, NOAA-NWS-Climate
Prediction Center 33rd Climate Diagnostics and
Prediction Workshop, October 21-24, 2008,
Lincoln, Nebraska
2
Introduction

• This paper discusses recent improvements in the
  skill and coverage of CPC T, P Outlooks.
• The heidke skill score and the percentage of
  non-EC probabilities are the performance
  measures.
• s ((c-e)/(t-e))100 , -50 lt s lt 100
• s is the percent improvement over random forecasts

3

• CPC 3-Month Outlook map lines show the
  probability that the indicated category, B, N, or
  A, will occur.
• In blank regions, the probabilities of B, N, A
  are equal at 1/3 each (EC), and give no forecast.
  
• Lines show Non-EC (potentially useful) forecast
  regions.
• On a line, probabilities of B and A vary
  simultaneously and inversely above and below
  33.33, while that of N usually stays at 33.33.
  
• The 3 sum to 100 at every point on the map.

4
How CPC Outlooks are Made

• CPC 3-month outlooks are currently made using a
  combination of at least 5 tools, in consultation
  with partners.
• From 1995- 2004 these tools were weighted
  subjectively. In 2006, an objective
  consolidation (CON) was introduced, which weights
  the tools by skill history and spread (Unger et
  al, 2008).
• Retrospective verification of CON forecasts shows
  them to be much more skillful than official (OFF)
  1995-2004 outlooks (OLenic et al, 2008), in both
  categorical U.S. average skill, and in coverage
  by non-equal-chances (non-EC) forecasts,
  properties users want.

5
NEW OTLK
6
NEW OTLK
7
HSS
OFF
½ MO LEAD PRECIPITATION RESULTS Heidke Skill
Score (HSS, lines) and Percent Non-EC (colors),
Map average Non-EC. A. OFFICIAL FORECAST
(OFF) B. CONSOLIDATION (CON) C. DIFFERENCE,
CON-minus-OFF, US average
Non-EC CON raises US annual average HSS from 9
to 12 compared with OFF
Area non-ec14
Area non-ec27
A
Map color legend,
Area non-ec20
Area non-ec33
HSS
CON
DIF
Area non-ec35
Area non-ec32
8
18
B
C
Area non-ec53
Area non-ec36
20
16
8
HSS
OFF
½ MO LEAD TEMPERATURE RESULTS Heidke Skill Score
(HSS, lines) and Percent Non-EC (colors), Map
average Non-EC. A. OFFICIAL FORECAST (OFF) B.
CONSOLIDATION (CON) C. DIFFERENCE CON OFF US
average Non-EC CON raises US annual average
HSS from 22 to 26 compared with OFF
Area non-ec47
Area non-ec46
A
Map color legend,
Area non-ec27
Area non-ec41
HSS
CON
DIF
Area non-ec78
Area non-ec57
11
31
B
C
40
Area non-ec96
55
Area non-ec67
9
GPRA Score Official Skill Metric48-Mo. Running
Mean U.S. Average T HSS
10
SUMMARY

• - Outlook prepared subjectively 1995-2004
• - Objective consolidation begun 2006
• - Retrospective verification shows significant
  increase in CON skill over OFF
• - Western and Eastern P forecasts better than
  many areas
• - Forecasts are better, more objective
• - Higher categorical skill
• - Far fewer EC forecasts
• - P HSS rose from 9 (OFF) to 12 (CON) (US ann.
  mean)
• - T HSS rose from 22 (OFF) to 26 (CON) (US ann.
  mean)
• - Non-EC rises in all seasons, gt30 for P, gt50
  for T
• - Official T skill rose starting in 2006 due to
  use of CON.

11
Forecast Evaluation Tool Example of a Means to
Address Gaps

• What FET and CLIDDSS provide
• User-centric forecast evaluation and data access
  and display capability.
• Leveraging of community software development
  capabilities.
• Opportunity to DISCOVER, collect, and invest in
  user requirements.

12
FUTURE Implement FET at CPC
13
FUTURE Implement FET at CPC
14
FETA Wide Variety of Skill Renderings
A
A
B
B
A
B
B
A
P
T
B
B
A
A
15
FUTURE of the FET

• Next 6 months
• Finalize and implement FET project plan at CPC.
• Ellen Lay (CLIMAS) to train CPC personnel on FET
  version control and bug tracking at CPC, November
  12-14, 2008.
• Necessary software (APACHE TOMCAT, JAVA, Desktop
  View) acquired and installed at CPC.
• Forecast, observations datasets in-place at CPC.
• FET code ported to CPC, installed, tested.
• FET installed to NWS Web Operations Center (WOC)
  servers