Verifications of CPC Seasonal Forecasts from 1995-2007. Arun Kumar Peitao Peng Climate Prediction Center 32 nd CDPW, Tallahassee, FL. CPC’s Seasonal Forecasts. Have been issued in their current format since January 1995 - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Verifications of CPC Seasonal Forecasts Verifications of CPC Seasonal Forecasts from 1995-2007from 1995-2007
Arun Kumar Arun Kumar
Peitao PengPeitao Peng
Climate Prediction CenterClimate Prediction Center
• Have been issued in their current format since January Have been issued in their current format since January 19951995
• Are issued with a lead time of ~15 days, and are made for Are issued with a lead time of ~15 days, and are made for next 13 overlapping seasonsnext 13 overlapping seasons
• Are for surface temperature and precipitation (over the Are for surface temperature and precipitation (over the contiguous U.S. & Alaska)contiguous U.S. & Alaska)
• Are based on a combination of empirical tools (CCA, OCN, Are based on a combination of empirical tools (CCA, OCN, SMLR) and a dynamical coupled model (currently the CFS)SMLR) and a dynamical coupled model (currently the CFS)
• Major changes have beenMajor changes have been
– 2 upgrades in the dynamical prediction system since 2 upgrades in the dynamical prediction system since 19951995
– Introduction of an objective consolidation technique in Introduction of an objective consolidation technique in January 2006January 2006
OutlineOutline
• Format of Seasonal ForecastsFormat of Seasonal Forecasts
• Verifications of seasonal forecastsVerifications of seasonal forecasts
– Different skill measuresDifferent skill measures
– Temporal and spatial distribution of skill Temporal and spatial distribution of skill measuresmeasures
– Attributing skill to different predictorsAttributing skill to different predictors
– Cost-Loss analysis of forecastsCost-Loss analysis of forecasts
• Future plansFuture plans
Format of Seasonal ForecastsFormat of Seasonal Forecasts
Today’s Special at the Café Verification Today’s Special at the Café Verification are…are…
• Surface temperatureSurface temperature over the contiguous U.S. over the contiguous U.S.
• Various Skill MeasuresVarious Skill Measures
– HeidkeHeidke
– RPSSRPSS
– ReliabilityReliability
– ROCROC
• Sample size: All (145 seasons); Sample size: All (145 seasons); Warm and cold Warm and cold season (60 seasons);season (60 seasons); individual season (12 individual season (12 seasons)seasons)
RPSS: Spatial and Temporal RPSS: Spatial and Temporal VariationsVariations
BinBin
NumberNumber
Forecast Forecast Probability Probability (Frequency)(Frequency)
Observed Observed OccurrencesOccurrences
Observed Non-Observed Non-OccurrencesOccurrences
Observed Observed Frequency of Frequency of OccurrenceOccurrence
11 0-10(%)0-10(%) 66 2525 0.190.19
22 10-20(%)10-20(%)2929
3030 0.490.49
33 20-30(%)20-30(%) 684684 851851 0.450.45
44 30-40(%)30-40(%) 98809880 1076610766 0.480.48
55 40-50(%)40-50(%) 17291729 12411241 0.580.58
66 50-60(%)50-60(%) 474474 205205 0.700.70
77 60-70(%)60-70(%) 115115 4343 0.730.73
88 70-80(%)70-80(%) 88 66 0.570.57
99 80-90(%)80-90(%) 00 00 NANA
1010 90-100(%)90-100(%) 00 00 NANA
Frequency Distribution: Surface Temp; Upper Tercile (above normal)Predicted and Observed Frequency Distribution for the Predicted and Observed Frequency Distribution for the Above Normal CategoryAbove Normal Category
Heidke Skill Score: Official vs. ConsolidatedHeidke Skill Score: Official vs. Consolidated
60% Improvement
11 MRM
Heidke Skill Score: Official vs. ConsolidatedHeidke Skill Score: Official vs. Consolidated
Future PlansFuture Plans
• Complete the verification summary (precipitation; Complete the verification summary (precipitation; verification for individual tools;…)verification for individual tools;…)
• Update the verification summary after the calendar year Update the verification summary after the calendar year is overis over
• Error bars on skill scoresError bars on skill scores
• Cost-Loss Analysis for different (decision) probability Cost-Loss Analysis for different (decision) probability thresholdsthresholds
• Consider how to intelligently use this information from Consider how to intelligently use this information from the the user’suser’s and forecaster’s perspective… and forecaster’s perspective…
Farmer Jones’ DilemmaFarmer Jones’ Dilemma
Multiple Sources!
What is the relevance of past skill information in the context of me using this SPECIFIC forecast?
Solution to Farmer Jones’ Dilemma? Solution to Farmer Jones’ Dilemma? Conditional Skill Estimates and Reliable Conditional Skill Estimates and Reliable