Upgraded LAMP Convection and Total Lightning Probability and Potential Guidance for CONUS Jerome Jess Charba presenting Co contributers Fred Samplatsky Andrew Kochenash Phil Shafer Judy Ghirardelli Bob ID: 773644
Download Presentation The PPT/PDF document "Upgraded LAMP Convection and Total" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Upgraded LAMP Convection and Total Lightning Probability and "Potential" Guidance for CONUSJerome (Jess) Charba (presenting) Co-contributers: Fred Samplatsky, Andrew Kochenash*, Phil Shafer, Judy Ghirardelli, Bob Glahn, Chengie Huang*, and Mike Allard National Weather ServiceMeteorological Development Laboratory (MDL)* NOAA affiliate, KBRwyle, Inc.Webinar for Total Lightning Implementation Working Group (TLIWG)12 May 2017
2Background: Localized Aviation MOS Program (LAMP)LAMP is statistically-based model to produce “weather” forecast guidance for aviation, public, and other usersApplies fine scale observations and NWP models each hour to add skill and detail to large scale MOS forecasts out to 25 hours LAMP convection and lightning probability and “potential” guidance operational in NWS since April 2014 Two-hour valid periods; on 20-km grid Some users have indicated need for increased spatial and temporal resolution MDL recently developed experimental ( upgraded) convection and lightning products with higher resolution One-hour valid periods; on 10-km grid Focus of talk
3Main TopicsUpgrades to LAMP convection and lightning predictands and predictors Convection and lightning probability skill Derived “potential” product Forecast performance: case studies
Convection event: radar reflectivity ≥ 40 dBZ and/or ≥ 1 lightning flashesLightning event: ≥ 1 lightning flashesPredictand upgrades: upgraded databases + doubled resolutionConvection and Lightning Predictands: Operational versus Upgraded Operational Upgraded Reflectivity data RCM * MRMS ** Lightning data cloud-to-ground (CG) total lightning (TL) ***Valid period 2 h 1 hGrid box size 20 km 10 km * RCM = Radar Coded Messages ( replicated from MRMS ) ** MRMS = Multi-Radar Multi-Sensor (MRMS) composite reflectivity (MDL supplemental quality control applied) *** TL = CG + in-cloud flashes from Earth Networks Total Lightning Network
Highlights of Upgraded Convection / Total Lightning (TL) Probability Regression Equations Equations developed from 4.5 year sample ending 05/31/2016 Unique predictors for –Each hourly cycleSpring, summer, and cool season Seven overlapping regionsRegion overlap avoids discontinuities in probabilities Four forecast projection ranges
Latest fine scale observational data (obs)MRMS reflectivity parameters TL flash count parameters Obs grids advected to time-match LAMP forecastsFine scale hourly HRRR (High Resolution Rapid Refresh) model outputUse HRRR predictors from two latest cycles Large scale 6-hourly NWP-based predictorsUse GFS-based and NAM-based MOS convection probabilities for two latest cycles Upgraded Predictor Databases
Predictor Type versus Forecast Projection Predictor type1 -12 hours11-14 hours14-17 hours16-25 hoursMRMS + total lightning observations * (fine scale) HRRR forecasts ( fine scale ) ** ** GFS/NAM MOS (large scale) ** ** * - initial + advected ** - two cycles
Sensitivity of Convection Probability Skill to Predictor Type CONUS mean for 18z and 06z cyclesIndependent sample = 216 quasi-random days from Sep 2012 – May 2016 Observations (obs) strongly augment HRRR + MOS to 3 hours
Sensitivity of Total Lightning Prob. Skill to Predictor Type CONUS mean for 18z and 06z cyclesIndependent sample = 216 quasi-random days from Sep 2012 – May 2016 Obs predictors dominate to 3 hoursHRRR/MOS predictors weak
Convection Probability Skill versus Projection/SeasonCONUS mean over 24 cycles (dependent sample) Obs HRRR MOSSpring best for short rangeCool best for medium and long rangeSummer worst for medium and long range Predominant predictors Skill fall due to loss of HRRR predictors
Total Ltng. (TL) Probability Skill versus Projection/SeasonCONUS mean over 24 cycles (dependent sample) Obs HRRR MOSTL skill superior for short rangeWeak skill for medium and long range Otherwise similar to convection Skill fall due to loss of HRRR predictors
Top Ranked Convec. / Total Ltng. Prob. Predictors for 1-12 H Range Ranking Convection Total Lightning_ 1 MRMS cref (init) HRRR ltg threat 2 HRRR cref GFS MOS prob 3 GFS MOS prob TL 60 min flash cnt ( adv ) 4 MRMS VIL (init) HRRR VIL 5 MRMS cref (adv) TL 30 min flash cnt (init) 6 HRRR 1h precip amt NAM MOS prob 7 HRRR VIL TL 60 min flash cnt (adv) 8 NAM MOS prob (GFS X NAM ) MOS prob Abreviations : init = initial time; adv = advected ; ltg threat = lightning threat 3 index; prob = probability; cnt = count; cref = composite reflectivity; VIL = vertically integrated liquid; precip amt = total precipitation amount Colors: fine scale observation; fine scale model; large scale model Ranking averaged over 24 LAMP cycles, 3 seasons, and 7 geographical regions
Convection and Lightning Probability Examples
SPC Storm Reports for 12/23/2015
Ex. Upgraded Probability: Convection versus LightningHigh detail and sharpness in probabilitiesLightning has less areal coverage Lightning valid 18-19 z 1 h from 18z 12/23/2015Convection valid 18-19 z
Ex. Convection Probability: Operational versus Upgraded 8 h from 18z 12/23/2015 Operational valid 00-02 z Upgraded valid 01-02 zMRMS composite reflectivity at 0130 z Upgraded has better focus in east and better pattern in west
Lightning complements convectionUpgraded has much better focusLightning flash count during 0100-0200 zEx. Lightning Probability: Operational versus Upgraded 8 h from 18z 12/23/2015 Operational valid 00-02 z Upgraded valid 01 - 02 z
Probability Skill: Convection and Lightning 1-h upgraded versus 2-h operationalIndependent sample: 246 random days from 05/06/2014 – 05/31/2016LAMP cycles: 18 and 06 z (combined)ConvectionLightningUpgraded better than operationalConvection better than lightning beyond 2 hours
Specifying “Potential” from ProbabilityFour risk (“potential”) levels specified from three pre-determined probability thresholdsThresholds derived by maximizing threat score with bias constrained to narrow range Threshold Probability Potential Average Bias low low 2.70 – 2.83 * (high) medium medium 1.03 – 1.13 ** (~unbiased) high high 0.38 – 0.43 (low)Potential complements probabilityProbability varies, especially with forecast projection (problem)Potential has known average bias (solution)* Applies to low and above potential (combined)** Applies to medium and above potential (combined)
Example of Convection Probability PotentialProbabilityPotentialPotential has known bias 8 h forecast f rom 18z 12/23/2015
More Example Forecasts with Verification
SPC Storm Reports for 04/26/2017
Probability (no verification)Potential with green and black verification colors superimposed Hit = Medium or High potential; convection occurrence (green)Miss = No or Low potential; convection occurrence (black)False Alarm = Medium or High potential; convection non-occurrence Convection
Convection2017/04/26 LAMP 20z cycle valid 20-21 z 00-01 h forecastTotal lightningStrong performance in 1 - 13 h range
Convection2017/04/26 LAMP 20z cycle valid 22-23 z 02-03 h forecastTotal lightning
Convection2017/04/26 LAMP 20z cycle valid 00-01 z 04-05 h forecastTotal lightning
Convection2017/04/26 LAMP 20z cycle valid 04-05 z 08-09 h forecastTotal lightning
Convection2017/04/26 LAMP 20z cycle valid 08-09 z 12-13 h forecastTotal lightningPerformance fall-off beyond 13h
Convection2017/04/26 LAMP 20z cycle valid 14-15 z 18-19 h forecastTotal lightning
Convection2017/04/26 LAMP 20z cycle valid 20-21 z 24-25 h forecastTotal lightning
06- 07 March 2017 Case1800 and 0000 z cycles
SPC Storm Reports for 03/06/2017
Legend: Forecast Performance with Best Obs and HRRR Predictor MapsProbability(no verification) Potential + Verification Best Obs (initial or advected) PredictorBest HRRR Predictor
LAMP Convection 2017/03/06 1800z cycle00-01 h probability (%) valid 1800-1900 z 00-01 h potential + verif valid 1800-1900 z MRMS cref (dBZ) obs valid 1814 z 2-h HRRR cref ( dBZ) valid 1900 z Strong obs predictor impact ≤ 3h
LAMP Convection 2017/03/06 1800z cycle02-03 h probability (%) valid 2000-2100 z 02-03 h potential + verif valid 2000-2100 z 3-h adv MRMS cref (dBZ) obs valid 2100 z 4 -h HRRR cref (dBZ ) valid 2100 z LAMP missed explosive convection devel . due to “misleading” obs predictors Growing HRRR convec . line; no impact ≤ 3h Strong obs predictor impact ≤ 3h
LAMP Convection 2017/03/06 1800z cycle05-06 h probability (%) valid 2300-0000 z 05-06 h potential + verif valid 2300-0000 z 6-h adv MRMS cref (dBZ) obs valid 0000 z 7-h HRRR cref (dBZ ) valid 0000 z Strong HRRR predictor impact ≥ 6h LAMP performance improving due to transition to HRRR predictor dominance Little or no obs predictor impact ≥ 6h
LAMP Convection 2017/03/06 1800z cycle08-09 h probability (%) valid 0200-0300 z 08-09 h potential + verif valid 0200-0300 z 9-h adv MRMS cref (dBZ) obs valid 0300 z 10-h HRRR cref (dBZ) valid 0300 z Dramatic impact of HRRR predictors on LAMP performance HRRR convec . line now mature
LAMP Convection 2017/03/06 1800z cycle11-12 h probability (%) valid 0500-0600 z 11-12 h potential + verif valid 0500-0600 z 12-h adv MRMS cref (dBZ) obs valid 0600 z 13-h HRRR cref (dBZ) valid 0600 z
Now 0000 z Cycle
LAMP Convection 2017/03/07 0000z cycle00-01 h probability (%) valid 0000-0100 z 00-01 h potential + verif valid 0000-0100 z MRMS cref (dBZ) obs valid 0014 z 2-h HRRR cref ( dBZ) valid 0100 z Excellent LAMP performance due to synergy of obs and HRRR predictors Strong obs predictor impact ≤ 3h
LAMP Convection 2017/03/07 0000z cycle05-06 h probability (%) valid 0500-0600 z 05-06 h potential + verif valid 0500-0600 z 6-h adv MRMS cref (dBZ) obs valid 0600 z 7-h HRRR cref (dBZ ) valid 0600 z HRRR predictor dominance; little or no impact of adv obs 6h and beyond Little or no obs predictor impact ≥ 6h Strong HRRR predictor impact ≥ 6h
Now for Total Lightning
LAMP Total Lightning 2017/03/06 1800z cycle00-01 h probability (%) valid 1800-1900 z 00-01 h potential + verif valid 1800-1900 z TL obs 30-min count ending 1815 z 2-h HRRR ltng threat index valid 1900 z Strong obs predictor impact ≤ 3h
LAMP Total Lightning 2017/03/06 1800z cycle02-03 h probability (%) valid 2000-2100 z 02-03 h potential + verif valid 2000-2100 z 3-h adv TL obs 30-min count valid 2100 z 4-h HRRR ltng threat index valid 2100 z LAMP missed explosive total lightning devel . due to “misleading” obs /HRRR predictors Strong obs predictor impact ≤ 3h No HRRR lightning threat
LAMP Total Lightning 2017/03/06 1800z cycle05-06 h probability (%) valid 2300-0000 z 05-06 h potential + verif valid 2300-0000 z 6-h adv TL obs 30-min count valid 0000 z 7-h HRRR ltng threat index valid 0000 z LAMP performance improving due to transition to HRRR predictor dominance HRRR predictor impact ≥ 6h Little or no obs predictor impact ≥ 6h
LAMP Total Lightning 2017/03/06 1800z cycle08-09 h probability (%) valid 0200-0300 z 08-09 h potential + verif valid 0200-0300 z 9-h adv TL obs 30-min count valid 0300 z 10-h HRRR ltng threat index valid 0300 z Dramatic impact of HRRR predictors on LAMP performannce Strong HRRR predictor impact
LAMP Total Lightning 2017/03/06 1800z cycle11-12 h probability (%) valid 0500-0600 z 11-12 h potential + verif valid 0500-0600 z 12-h adv TL obs 30-min count valid 0600 z 13-h HRRR ltng threat index valid 0600 z
LAMP Total Lightning 2017/03/07 0000z cycle00-01 h probability (%) valid 0000-0100 z 00-01 h potential + verif valid 0000-0100 z TL obs 30-min count ending 0014 z 2-h HRRR ltng threat index valid 0100 z Excellent LAMP performance due to synergy of obs and HRRR predictors Strong obs predictor impact ≤ 3h
LAMP Total Lightning 2017/03/07 0000z cycle05-06 h probability (%) valid 0500-0600 z 05-06 h potential + verif valid 0500-0600 z 6-h adv TL obs 30-min count valid 0600 z 7-h HRRR ltng threat index valid 0600 z HRRR predictor dominance; little or no impact of adv obs 6h and beyond Strong HRRR predictor impact ≥ 6h Little or no obs predictor impact ≥ 6h
Principal FindingsMRMS and total lightning observations yield high detail and skill in first several hours of convection and lightning guidance productsHRRR extends detail and yields good skill to maximum HRRR projection (17 hours)Upgraded convection and lightning products have better resolution and skill than operational products Convection and lightning products –Are similar, but distinctComplement one another“Potential” product aids use of probabilities
Current Status and Path ForwardConvection and lightning products running experimentally for all hourly cycles since September 2016 (http://www.weather.gov/mdl/lamp_experimental)FAA and AWC feedback for convection quite positive …minimal field evaluation of lightning to dateAnticipated implementation September 2017
End
Reserved
Other Convective Guidance Products
Consolidated Storm Prediction for Aviation (CoSPA) ProductProduces convective cloud top height guidance forecast for CONUSOn 3 km gridUpdates every 5 minValid times15 min time increments to 2 hours1 hour time increments in 2 – 8 hour rangeUses fine scale observations and NWP (HRRR) outputDeveloped and maintained by MIT Lincoln LabsWebsite = https://cospa.wx.ll.mit.edu/login.jsf?p=LiveDisplaySupported by FAAHosted on FAA Weather Systems
CoSPA Convection Tops 8-h forecast Issued 2017/03/06 18z Valid 22z
SREF Calibrated CG Lightning Probability GuidanceExperimental calibrated CG lightning probability for CONUS40 km gridboxes3-h valid periods 0 – 36 hour rangeUpdates twice per day (00z and 12z cycles)Uses predictors from multi-Convection Allowing Models (CAMs) and multi-ensemble members from some CAMs (?)29 members (?)Probability maps –show large scale lightning threatPeak probabilities are generally < ~ 60 %Developed by NCEP/SPChttp://www.spc.ncep.noaa.gov/publications/bright/calthun.pdf
SPC CG Lightning Probability 24-h forecast Issued 2017/04/27 00z Valid 21-00 z
Trimmed
Convection Probability Skill Test: Regional vs “National” EquationsRegional = regionalized regression equationsNational = non-regionalized regression equations18z and 06z cycles combinedIndependent sample = 216 quasi-random days from Sep 2012 – May 2016Eastern USWestern USEast = small skill improvementWest = strong skill improvement
“Potential” Threat Score: Convection and Lightning“Yes” forecast = Medium + High potential; else “No”Independent sample: 246 random days from 05/06/2014 – 05/31/2016LAMP cycles: 18 and 06 UTC (combined)ConvectionLightningUpgraded better than operationalConvection better than lightning beyond 1 hourAvg bias 1 h = 1.1 2 h = 1.4Avg bias 1 h = 1.0 2 h = 1.2
Convection Probability Skill versus Projection/SeasonCONUS mean over 24 cycles (dependent sample) Obs HRRR MOSCool skill is bestSummer skill is worst Predominant predictors Skill fall due to loss of HRRR predictors
SPC Storm Reports for 04/26/2016
2-h Operational 1-h Upgraded 14-h Forecast Upgraded convection probability resolves double line in TX better than operational probability
Top Ranked Convection Probability Predictors for 1-12 H Range * Ranking Spring Summer Cool_ 1 MRMS cref (init) MRMS cref (init) MRMS cref (init) 2 GFS MOS prob MRMS vil ( init ) MRMS cref ( adv ) 3 HRRR cref GFS MOS prob HRRR cref 4 MRMS vil (init) HRRR 1h precip amt GFS MOS prob 5 NAM MOS prob HRRR cref HRRR vil 6 MRMS cref (adv) NAM MOS prob HRRR 1h precip amt 7 HRRR vil MRMS vil ( adv ) (GFS X NAM ) MOS prob 8 HRRR 1h precip amt TL 30 min flash cnt ( adv ) MRMS vil ( init ) Abreviations : init = initial time; adv = advected ; prob = convection probability; cref = composite reflectivity vil = vertically integrated liquid; precip amt = total precipitation amount ; TL = total lightning; cnt = count Colors: fine scale observation fine scale model large scale model * Ranking averaged over 24 LAMP cycles and 7 geographical regions
Top Ranked Total Lightning Probability Predictors for 1-12 H Range * Ranking Spring Summer Cool_ 1 HRRR ltg threat HRRR ltg threat GFS MOS prob 2 TL 30 min flash cnt (adv) TL 30 min flash cnt ( adv ) HRRR ltg threat 3 HRRR vil TL 60 min flash cnt (init) HRRR vil 4 TL 30 min flash cnt (init) TL 30 min flash cnt ( init ) TL 60 min flash cnt ( adv ) 5 TL 60 min flash cnt (adv) NAM MOS prob NAM MOS prob 6 GFS MOS prob HRRR vil TL 30 min flash cnt ( adv ) 7 NAM MOS prob (GFS X NAM ) MOS prob TL 30 min flash cnt ( init ) 8 ( GFS X NAM ) MOS prob GFS MOS prob (GFS X NAM ) MOS prob Abreviations : init = initial time; adv = advected ; ltg threat = lightning threat 3 index; TL = total lightning; prob = TL probability; cref = composite reflectivity; vil = vertically integrated liquid; precip amt = total precipitation amount ; cnt = count Colors: fine scale observation; fine scale model; large scale model * Ranking averaged over 24 LAMP cycles and 7 geographical regions
Case Performance of Convection and TL Probability and Potential Maps Verification Map Model for: 2017/04/26 20z LAMP cycleProbability (no verification)Potential + verification “No” forecast = low and below potential“Yes” forecast = medium and high potentialBlack = “no” forecast ; convection occurred (miss)Green = “yes” forecast ; convection occurred (hit) FARs = medium or high potential forecast and convection non-occurrence
Convection2017/04/26 LAMP 20z cycle valid 02-03 z 06-07 h forecastTotal lightning7
Convection2017/04/26 LAMP 20z cycle valid 06-07 z 10-11 h forecastTotal lightning11
Convection2017/04/26 LAMP 20z cycle valid 10-11 z 14-15 h forecastTotal lightning15
Convection2017/04/26 LAMP 20z cycle valid 12-13 z 16-17 h forecastTotal lightning17
Convection2017/04/26 LAMP 20z cycle valid 16-17 z 20-21 h forecastTotal lightning21
Convection2017/04/26 LAMP 20z cycle valid 18-19 z 22-23 h forecastTotal lightning23
LAMP Convection 2017/03/07 0000z cycle02-03 h probability (%) valid 0200-0300 z 02-03 h potential + verif valid 0200-0300 z 4-h HRRR cref (dBZ) valid 0300 z 3 -h adv MRMS cref ( dBZ ) obs valid 0300 z 3
LAMP Convection 2017/03/07 0000z cycle08-09 h probability (%) valid 0800-0900 z 08-09 h potential + verif valid 0800-0900 z 9-h adv MRMS cref (dBZ) obs valid 0900 z 10-h HRRR cref (dBZ ) valid 0900 z 9
LAMP Convection 2017/03/07 0000z cycle11-12 h probability (%) valid 1100-1200 z 11-12 h potential + verif valid 1100-1200 z 12-h adv MRMS cref (dBZ) obs valid 1200 z 13-h HRRR cref (dBZ) valid 1200 z 12
LAMP Total Lightning 2017/03/07 0000z cycle02-03 h probability (%) valid 0200-0300 z 02-03 h potential + verif valid 0200-0300 z 4-h HRRR ltng threat index valid 0300 z 3 -h adv TL obs 30-min count valid 0300 z 3
LAMP Total Lightning 2017/03/07 0000z cycle08-09 h probability (%) valid 0800-0900 z 08-09 h potential + verif valid 0800-0900 z 9-h adv TL obs 30-min count valid 0900 z 10-h HRRR ltng threat index valid 0900 z 9
LAMP Total Lightning 2017/03/07 0000z cycle11-12 h probability (%) valid 1100-1200 z 11-12 h potential + verif valid 1100-1200 z 12-h adv TL obs 30-min count valid 1200 z 13-h HRRR ltng threat index valid 1200 z 12
Probability (no verification)Potential with green and black verification colors superimposed Hit = Medium or High potential; convection occurrence Miss = No or Low potential; convection occurrenceFalse Alarm = Medium or High potential; convection non-occurrenceConvection