Weather Hazard Modeling Research and development: Recent Advances and Plans Stephen S. Weygandt Curtis Alexander, Stan Benjamin, Geoff Manikin*, Tanya Smirnova, Ming Hu NOAA Earth System Research Laboratory *NOAA National Centers for Environmental Prediction Friends and Partners of Aviation Weather NBAA Convention, Nov 2-3, 2016
High-resolution models of hazardous weather now making very realistic s Loop of radar reflectivity Observations 1 hour HRRR Storm details can be wrong Valid Given s are not always exactly right Missing radar observations NGM model guidance around 1990 How to improve decision making with inexact s? How can we continue to make s better? Very realistic s
Continued improvement for Rapidly updated, High-resolution models Improvement each year Critical Success Index HRRR Reflectivity Verification HRRR V1 22z + hr s HRRR V2 V2 03z V1 NCEP version Radar Observed
Continued improvement for Rapidly updated, High-resolution models RAP - upper-air verification Improvement to moisture / microphysics / precipitation, boundary layer / surface energy balance, ensemble assimilation Ongoing study to evaluate how improvements map to aviation decision-making improvements (GSD, AvMet)
Accurate prediction of hazardous weather Higher resolution models with better physics What are the key requirements? Reality 3-km 13-km Best use of observations (Ensemble data assimilation) Rawins Aircraft Surface Satellite Radar Profiler Rapid updating -- Fresher s -- Time-lagged ensembles Wind Errors 6-hr LAX ORD 3-hr LAX ORD
The benefits of rapid updating: June 16, 2014 NE twin tornadoes ~21z 16 June 14z + 7hr Successive HRRR runs converging to solution 1z + 6hr 16z + hr 17z + 4hr 21z 16 June radar obs
HRRR example of consistent solutions: May 11, 2014 NE squall-line 1z+7h 16z+6h 17z+h 22z 11 May Successive HRRR runs have consistent solution Storm Reports (obs) HRRR Updraft Helicity Time-Lagged Ensemble
Arkansas HRRR (and HRRR Tornadoes RAP) Milestones Future April Milestones 27 2014 Arkansas tornadoes Sunday 27 April 2014 8 fatalities 3 fatalities Tornado Track
Arkansas tornadoes: very challenging case 1z+10h HRRR Updraft Helicity 16z+9h Time-lagged ensemble of specific hazard indicator 17z+8h 18z+7h 19z+6h
Model ensembles: Key to assessing likelihood Deterministic 19 UTC Member 1 Member 2 Member 3 00 UTC Observations Supercell probability past hour 00 UTC
How do ensembles s help? Old way one deterministic 100% 80% Single snow Actual snowfall 60% 40% 20% 0% 0 1 2 3 4 6 7 8 Inches of snow
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% 0% 0 0 Forecast snow accumulation chances 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow Example: Run model 100 times with slightly different settings
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% 0% 0 0 Forecast snow accumulation chances Almost NO chance Of less than 2 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow Example: Run model 100 times with slightly different settings
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% 0% 0 0 Forecast snow accumulation chances 0 / 0 chance of or more 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow Example: Run model 100 times with slightly different settings
How do ensembles s help? Old way one deterministic 100% 80% Single snow Forecast snow accumulation chances New way run an ensemble of s get information about range of possible outcomes 60% 40% 20% 0% 0 0 Even without an ensemble 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow you still get the deterministic 21z 16 June radar obs
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% Single snow 0% 0 0 Forecast snow accumulation chances Users assume a range of possible outcomes 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow Forecast users will make their own assumptions about the range of possible outcomes
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% One more key benefit from ensembles: Old single snow 0% 0 0 Forecast snow accumulation chances Actual snowfall 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow
How do ensembles s help? Old way one deterministic New way run an ensemble of s get information about range of possible outcomes 100% 80% 60% 40% 20% 0% 0 0 One more key benefit from ensembles: ensemble data assimilation improves the deterministic Better single snow Forecast snow accumulation chances Actual snowfall 2 20 1 20 10 0 1 2 3 4 6 7 8 Inches of snow Need ensembles the most for the most challenging situations
More information on ensembles Types of ensembles: 1) Ensembles of opportunity -- Time-lagged ensemble (HRRR-TLE) -- Multi-model ensembles (SSEO) ADS: computationally inexpensive DISADS: model clustering, members not independent Example of model clustering Actual Model A Model B 2) Designed ensembles -- Perturbed members of single ensemble system (Global Ensemble Forecast System, HRRR-Ensemble) ADS: More useful spread, ensemble assimilation DISADS: computationally expensive
HRRR-TLE example: 18 April 2016 23z+13h 00z+12h 01z+11h accum. precip. HRRR-TLE probability of 6hr QPF exceeding 100 year average return interval (ARI) Obs. 12-hr precip.
HRRR Time-Lagged Ensemble (HRRR-TLE): GSD real-time product generation Real-Time Web Graphics (and grids via LDM/FTP) http://rapidrefresh.noaa.gov/hrrrtle Echo-Top > 2 kft Probability of condition) Echo-Top > 30 kft Echo-Top > 3 kft MVFR IFR
HRRR Ensemble (HRRRE) Web Graphics (real-time HRRRE to resume Nov. 2016) http://rapidrefresh.noaa.gov/hrrre Single core (ARW) Ensemble DA (GSI-EnKF) RAP mean + perturbations Planned Fall 2016 Refinements Add radar reflectivity data assimilation Stochastic physics Apply HRRR-TLE statistical processing Include lagged members? Assimilation Forecast 20 members 00z - Three mem to 30 hr 1 hr cycling 03z - Three mem to 27 hr 21 fcsts / day 12z - Six mem to 18 hr Start 21z day zero 1z - Eighteen mem to 1 hr End 18z day one 18z - Eighteen mem to 12 hr Proof-of-concept Real-time demonstration
Comparison: HRRR-TLE and HRRRE Updraft Helicity > 2 m 2 /s 2 from Different members HRRR-Ensemble 1z + 7 hr Time-Lagged Ensemble 1z-17z initializations
A challenge of ensembles: Displaying output / conveying information Who is the user? Postage stamp plots Snow accumulation with time Need more creative display methods Extract hazard specific information Simplify display, only most critical information Mean and spread Spaghetti plots Plume diagrams Paintball plots Probability regions Feed ensemble information into decision support systems?
Rapid Refresh and HRRR NOAA hourly updated models 13km Rapid Refresh (RAP) Version 3 -- NCEP implement 23 Aug 2016 Version 4 GSD Planned NCEP Early 2018 3km High Resolution Rapid Refresh (HRRR) Version 2 NCEP implement 23 Aug 2016 RAP (13-km) HRRR (3-km) Version 3 GSD Planned NCEP Early 2018 HRRR v3 plan HRRR-E (Ensemble)