WED am Weygandt 5 31

Steve WeygandtStan BenjaminForecast Systems LaboratoryNOAA: 

RUC Convective Probability Forecasts using Ensembles and Hourly Assimilation Steve Weygandt Stan Benjamin Forecast Systems Laboratory NOAA

Slide2: 

Background on Rapid-Update Cycle U.S. operational model, short-range applications, situational awareness model Used by aviation, severe weather and general forecast communities 1-h update cycle, many obs types including: ACARS, profiler, METAR, radar Full cycling cloud/precip Grell/Devenyi ensemble cumulus parameterization Benjamin, Thurs. 9:30 talk

Research Background: 

Research Background Problem: Thunderstorm likelihood information needed by aviation traffic community for strategic planning (Collaborative Convective Forecast Product) Goals: Utilize outputs from RUC hourly output cycle to provide guidance for aviation forecasters. Blend model-based probabilities with observation- based probabilities (Pinto, next talk) Collaboration: NCAR Research Applications Lab (Mueller, Poster 5.21) National Weather Service Aviation Weather Center

Model-based Convective Probability Forecasts : 

Model-based Convective Probability Forecasts Principle: Convective forecasts at specific model grid points from a single deterministic model run less likely to be correct than ensembles of model outputs. Ensemble Approaches: Adjacent model gridpoints (2003) Time-lagged ensembles (2004) Cumulus parameterization closures Procedure: Use model 1-h parameterized precipitation Specify length-scale and precipitation threshold Bracketing 1-h model outputs from successive cycles

RUC convective precipitation forecast: 

RUC convective precipitation forecast 5-h fcst valid 19z 4 Aug 2003 3-h conv. precip. (mm)

RUC convective probability forecast: 

% 10 20 30 40 50 60 70 80 90 Prob. of convection within 120 km RUC convective probability forecast 5-h fcst valid 19z 4 Aug 2003 Threshold > 2 mm/3h Length Scale = 120 km Box size = 7 GPs 7 pt, 2 mm (gridpoint ensemble)

Slide7: 

Time-lagged ensemble Model Init Time Eg: 15z + 2, 4, 6 hour RCPF forecast Forecast Valid Time (UTC) 11z 12z 13z 14z 15z 16z 17z 18z 19z 20z 21z 22z 23z 13z+4,5 12z+5,6 11z+6,7 13z+6,7 12z+7,8 13z+8,9 12z+9 RCPF 2 4 6 18z 17z 16z 15z 14z 13z 12z 11z 6 7 5 6 7 8 9 10 4 5 6 7 8 9 Model runs used model has 2h latency

Slide8: 

Precipitation threshold adjusted diurnally and regionally to optimize the forecast bias Use smaller filter length-scale in Western U.S. Forecast Valid Time GMT EDT Higher threshold to reduce coverage Lower threshold to increase coverage Multiply threshold by 0.6 over Western U.S. Bias corrections

CSI by lead-time, time of day: 

.24, .25 .22, .23 .20, .21 .18, .19 .16, .17 .14, .15 .12, .13 .10, .11 CSI by lead-time, time of day Forecast Valid Time Diurnal cycle of convection 6-h 4-h 2-h 6-h 4-h 2-h 6-h 4-h 2-h RCPF v2004 RCPF v2003 CCFP (Verifiation 6-31 Aug. 2004) Fcst Lead Time GMT

CSI by lead-time, time of day: 

.24, .25 .22, .23 .20, .21 .18, .19 .16, .17 .14, .15 .12, .13 .10, .11 CSI by lead-time, time of day Forecast Valid Time Diurnal cycle of convection 6-h 4-h 2-h 6-h 4-h 2-h 6-h 4-h 2-h RCPF v2004 RCPF v2003 CCFP (Verifiation 6-31 Aug. 2004) Fcst Lead Time GMT Quick spin-up 18z init

CSI by lead-time, time of day: 

.24, .25 .22, .23 .20, .21 .18, .19 .16, .17 .14, .15 .12, .13 .10, .11 CSI by lead-time, time of day Forecast Valid Time Diurnal cycle of convection 6-h 4-h 2-h 6-h 4-h 2-h 6-h 4-h 2-h RCPF v2004 RCPF v2003 CCFP (Verifiation 6-31 Aug. 2004) Fcst Lead Time GMT Quick spin-up 18z init

Bias by lead-time, time of day: 

Bias by lead-time, time of day 6-h 4-h 2-h 6-h 4-h 2-h 6-h 4-h 2-h 2.75-3.0 2.5-2.75 2.25-2.5 2.0-2.25 1.75-2.0 1.5-1.75 1.25-1.5 1.0-1.25 0.75-1.0 0.5-0.75 v2004 v2003 CCFP (Verifiation 6-31 Aug. 2004) Forecast Valid Time Diurnal cycle of convection Fcst Lead Time GMT

Slide13: 

2005 Sample RCPF and CCFP 25 – 49% 50 – 74% 75 – 100% Verification 00z 8 Mar 2005 NCWD CCFP 18z + 6h Forecast RCPF Verification from FSL Real-Time Verification System (Kay, Thurs. 12:48 talk)

Slide14: 

Height (ft x 1000) RUC 4-h Forecast Potential Echo Top Observed Composite Radar Reflectivity/ EchoTops

Slide15: 

A-S M-Con CAPE Grell Use of Ensemble Cumulus Closure Information Normalized 1-h avg. rainrates From different closure groups VERIFICATION 2100 UTC 26 Aug 2005 RCPF 8-h fcst

Slide17: 

Relative Operating Characteristic (ROC) curves Show tradeoff: “detection” vs. “false-alarm” “Left and high” curve best Does gridpoint ensemble add skill? POD POFD ----- gridpoint ensemble ----- deterministic forecast Sample: 5-h fcst from 14z 04 Aug 2003 Low prob Low precip High precip High prob detection false detection 9 pt, 4 mm 25%

Slide18: 

CSI = 0.22 Bias = 0.99 RCPF – 20 AUG ’05 11z+8h Scores for 40% Prob. NCWD valid 19z 20 AUG 05 RCPF20 RCPF13 CSI = 0.15 Bias = 1.19 25 – 49% 50 – 74% 75 – 100%

Slide19: 

Sample 3DVAR analysis with radial velocity 500 mb Height/Vorticity * Amarillo, TX Dodge City, KS * * Analysis WITH radial velocity * * Cint = 2 m/s * * Cint = 1 m/s K = 15 wind Vectors and speed 0800 UTC 10 Nov 2004 Dodge City, KS Vr Amarillo, TX Vr * * Analysis difference (WITH radial velocity minus without)