220 likes | 373 Views
National Scale Probabilistic Storm Forecasting for Aviation Planning Talk Focus - Storm Coverage. Presenter: Dr James O. Pinto, NCAR/RAL Collaborators: Rasmussen, Steiner, Megenhardt, Rehak, Dixon, Phillips Acknowledgements : FAA AWRP. Playbook: Revised at 2000 UTC. Playbook at 1200 UTC.
E N D
National Scale Probabilistic Storm Forecasting for Aviation PlanningTalk Focus - Storm Coverage Presenter: Dr James O. Pinto, NCAR/RAL Collaborators: Rasmussen, Steiner, Megenhardt, Rehak, Dixon, Phillips Acknowledgements : FAA AWRP
Playbook: Revised at 2000 UTC Playbook at 1200 UTC Verification Need for Reliable Storm Nowcasts of 8+ hrs for Strategic Planning 6 hr Probability Forecast Fcst valid : 2040 – 0240 UTC
Outline • Motivation • Need for probability of storm orientation, organization and coverage (e.g., storm permeability) • Current Probabilistic Storm Forecast Systems • Current Research – Storm Coverage Forecasts • Assess explicit model prediction of storm coverage and echo top heights • Use model and obs climatologies to determine relationship between storm coverage and forecast valid time, environmental conditions, location.
Outline • Motivation • Need for probability of storm orientation, organization and coverage (IOW: storm permeability) • Current Probabilistic Storm Forecast Systems • Current Research – Storm Coverage Forecasts • Assess explicit model prediction of storm coverage and echo top heights • Use model and obs climatologies to determine relationship between storm coverage and forecast valid time, environmental conditions, location.
Deterministic Forecasts of Precipitation - Extrap Extrapolation Verification Successive fcsts valid at same time. 12 13 14 15 16 17 18 UTC 6 hr fcst 3 hr fcst 1 hr fcst
Probabilistic Forecast of Convective Hazard 2 h P NCWF2 Hazard Detection with Probabilistic Fcsts 1 h Probabilities determined via spatial filter and a VIL threshold. Spatial filter increases with fcst lead time (after Germann and Zawadzki 2004) Probabilities also influenced by observed trends, environmental conditions and climo. P Probabilities may be interpreted as: -likelihood of storms exceeding 35 dbz at a given time and location -the coverage of storms exceeding 35 dbz (if reliable).
Probabilities based on: Spatial coverage of convective precip predicted by the RUC-20 model 3-member time-lagged ensemble Square filter of 180 km Precipitation rate threshold for convection (1-2 mm/hr) Tuned using 40 km truth data set RUC Convective Prob. Forecast P . Probabilistic Forecasts Systems 7 hr fcst Valid: 21 UTC Successive fcsts valid at same time. 12 13 14 15 16 17 18 19 20 21 UTC } 9 hr fcst 8 hr fcst 3-member ensemble valid at 21 UTC 7 hr fcst
Verification Comparison with Operational 2 hour TSTM Fcst Products From Seseke et al. 2006 QA Report NOAA/Earth System Research Lab
Outline • Motivation • Need for probability of storm orientation, organization and coverage (IOW: storm permeability) • Current Probabilistic Storm Forecast Systems • Current Research – Storm Coverage Forecasts • Assess explicit model prediction of storm coverage and echo top heights • Use model and obs climatologies to determine relationship between storm coverage and forecast valid time, environmental conditions, location.
Methodology for Improving Storm Coverage Fcsts • Storm-resolving forecasts using WRF • Compare model and obs distribution of storm coverages (regionally dependent) • Focus on Southeastern US where scattered storms common • Assess predictability of storm coverage for each region as a function of environmental conditions, time of day, etc. • Verification • standard skill scores to assess reliability • Need for more descriptive skill scores such as those available in MODE (object-based verification). • E.g. storm spacing
Deterministic Forecasts of Convection WRF-ARW (4km) – 08hr fcst WRF-ARW (4km) – 20hr fcst 00 ……12 13 14 15 16 17 18 19 20 UTC 20 hr fcst 8 hr fcst Case Study- 2006 July 19 WRF Model Reflectivity Verification – WSI Mosaic REPLACE – July 19th • Storm-resolving realtime fcsts run in collab. with Wang and Weismann • WSM Microphysics, MYJ PBL, Noah LSM • BCs: NAM – 40 km grid 212 • No Data Assimilation • Run Jun/Jul 2005 (0Z) & 2006 (00,12 Z) Successive fcsts valid at same time.
Deterministic Forecasts of Storm Coverage WRF-ARW (4km) – 08hr fcst WRF-ARW (4km) – 20hr fcst 00 ……12 13 14 15 16 17 18 19 20 UTC 20 hr fcst 8 hr fcst Case Study- 2006 July 19 Verification – WSI Mosaic WRF vs WSI Coverages WSI Refl - 2000 UTC obs model Successive fcsts valid at same time. • Threshold = 35 dBZ • Impressive accuracy of timing and location of max coverage areas • Coverage forecast improves as fcst length decreases
Titan Storm Detections – WSI Reflectivity r =100 km • WSI reflectivity mosaic from WSR-88D radar • Degraded to 4 km using spatial filter • 35 dBZ and 75 km2 thresholds
Storm Spacing – WRF Reflectivity (20 hr fcst) Update with WRF image r =100 km 100 km • WRF reflectivity – max in column from 00 UTC run • 35 dBZ and 75 km2 thresholds
Conclusions • User needs (e.g., aviation planning) drive the system requirements • Likelihood of storm at a given location not enough info for users in decision making • Need for PDF expressing likelihood of coverages and joint PDFs of coverage / echo top likelihood. • Current technology in predicting storm coverages have limited reliability. • Convection resolving simulations may offer hope in predicting storm coverages and spacing.
NRC Report on Weather Forecasting Accuracy for FAA Traffic Flow Management “Because accurate deterministic 2- to 6-hour forecasts are not available, it is necessary to develop probabilistic forecasts that can readily be used by both humans and automated air traffic management decision support tools.” NRC Report, 2003
NCWF2 Convective Hazard Detection (NCWD) Unisys VIL Vaisala CG Ltg Unisys Echo Tops Data Feeds
Convective Hazard Detection (NCWD) VIL (Echo Tops > 15 kft contoured) VIL (Stratiform areas blue) VIL (Echo > 15 kft removed) VIL (Stratiform Removed) Hazard Detection (filtered VIL + Ltg) Unisys VIL Inputs: Unisys VIL & Echo Tops, NLDN C-to-G Ltg Steps to produce NCWD 1) Stratiform Filter (Steiner et al. 1995) 2) Echo tops Filter (remove echo < 15 Kft) 3) Combine with Lightning
Current Skill of Nowcasting Technologies 1.0 .8 Accuracy of Rainfall Nowcasts >2 mm/h Forecast Skill .6 Goal - idealized (CSI) .4 NWP Extrapolation .2 0 1 2 3 4 5 6 Forecast Length, hours
Algorithm for Blending Probabilistic Forecasts (e.g., 4 hr forecast) Preprocessing Statistical Performance RUC Probabilistic Convection Forecasts Weights Climato- logical Interest Blending Frontal Interest Probabilistic Extrapolation Fcsts Thermo- dynamic Mask Merged Forecast * * Amenable to Forecaster Modification