Probabilistic Forecasts of Extreme Precipitation Events for the
Probabilistic Forecasts of Extreme Precipitation Events for the U. S. Hazards Assessment Kenneth Pelman 32 nd Climate Diagnostics Workshop Tallahassee, Florida
Outline • • • Current Hazards Assessment Motivation for Probabilistic Forecast Details of Objective Probabilistic Tool Verification Results Conclusions Future Work
Current CPC Hazards Assessment • Made each Monday-Friday and covers Days 3 -14 • Designed to take current state of climate and predict hazardous weather conditions in support of CPC’s mission • Hazards include heavy rain (a proxy for flooding), severe weather, extreme heat and cold, and severe drought • Hazard forecasts are subjective and deterministic
Motivation for a Probabilistic Hazards Assessment • Looking to improve on subjective scores • Probabilistic forecasts provide more information about uncertainty to users • Puts Hazards Assessment in same format as other popular CPC products, such as 610/8 -14 day, monthly, and seasonal forecasts
Objective Probabilistic Heavy Precipitation Tool • Uses 0 z, 6 z, and 12 z GFS ensemble members to forecast for 881 grid points across the CONUS • Rainfall totals not bias corrected or calibrated • 1, 2, and 3 -day Hazards are forecast out to 384 hours • A Hazard is defined as the greater of 1 inch/day or the 95 th percentile value Forecast probability contours in solid green, hazard thresholds in dashed black (in mm) • Climatology (1971 -2000) derived from U. S. precipitation dataset (Higgins et. al. 2000)
Reliability Diagrams (Using Total Precipitation in Period)
Reliability Diagrams (Using only 1 Day in Period)
a c b d 1 -Day Event Contingency Table Scores (10% threshold) POD a/(a+c) Threat Score a/(a+b+c) FAR b/(a+b) Bias a+b/a+c Day 3 0. 32 0. 14 0. 80 1. 62 Day 4 0. 26 0. 12 0. 81 1. 45 Day 5 0. 21 0. 10 0. 85 1. 38 Day 6 0. 14 0. 07 0. 87 1. 10 Day 7 0. 09 0. 05 0. 88 0. 83 Day 8 0. 05 0. 04 0. 90 0. 55 Assessment 0. 12 0. 08 0. 79 0. 57
a c b d 2 - Day Event Contingency Table Scores (10% threshold) Total Precipitation in Period POD a/(a+c) Threat Score a/(a+b+c) FAR b/(a+b) Bias (a+b)/(a+c) Days 2+3 0. 21 0. 10 0. 85 1. 38 Days 3+4 0. 17 0. 08 0. 86 1. 18 Days 4+5 0. 16 0. 08 0. 85 1. 19 Days 5+6 0. 15 0. 08 0. 86 1. 09 Assessment 0. 12 0. 08 0. 79 0. 57 1 Day in Period POD Threat Score FAR Bias Days 2+3 0. 13 0. 09 0. 76 0. 55 Days 3+4 0. 12 0. 08 0. 77 0. 52 Days 4+5 0. 10 0. 07 0. 78 0. 48 Days 5+6 0. 05 0. 81 0. 35 Assessment 0. 12 0. 08 0. 79 0. 57
a b c d 3 -Day Event Contingency Table Scores (10 % threshold) Total Precipitation in Period POD a/(a+c) Threat Score a/(a+b+c) FAR b/(a+b) Bias (a+b)/(a+c) Days 1+2+3 0. 08 0. 05 0. 91 0. 95 Days 2+3+4 0. 09 0. 05 0. 90 0. 91 Days 3+4+5 0. 08 0. 05 0. 88 0. 67 Assessment 0. 12 0. 08 0. 79 0. 57 1 Day in Period POD Threat Score FAR Bias Days 1+2+3 0. 05 0. 04 0. 80 0. 24 Days 2+3+4 0. 04 0. 81 0. 22 Days 3+4+5 0. 03 0. 82 0. 17 Assessment 0. 12 0. 08 0. 79 0. 57
ROC Diagram
Conclusions • Limited usefullness of this tool in a strict probabilistic sense • 1 -Day Hazard forecasts show the best improvement over human-made Assessments • Tool can immediately be used by forecasters as a first guess • There is validity in converting the Hazards Assessments into a probabilistic forecast
Future Work • Determine best definition of a hazard • Generate contingency tables using different thresholds • Use calibrated precipitation forecasts and calibrated probabilities. • Use more ensembles in forecast (e. g. CAN, ECMWF, CFS)
Reference • Higgins, R. W. , W. Shi, E. Yarosh and R. Joyce, 2000: Improved United States Precipitation Quality Control System and Analysis. NCEP/Climate Prediction Center Atlas No. 7.
- Slides: 14