Operational and Statistical Prediction of Rapid Intensity

Change

Mark DeMaria and Eric Blake, NCEP/NHCJohn Kaplan, AOML/HRD

Outline

• Evaluation of NHC forecasts and deterministic models for rapid intensification cases

• Operational methods for anticipating rapid intensification

• The SHIPS Rapid Intensification Index

Hurricane Marie 2014 Aug 23 2200 UTC to Aug 24 1300 UTC

Distribution of 24 hr Intensity ChangesAtlantic 1989-2006

From Kaplan et al (2010) WaF

RI Verification Procedure• Sample

– 2008-2014 Operational Forecasts– Tropical/Subtropical cases– Land cases removed

• Four intensity change thresholds – 20, 30, 40 and 50 kt

• Five forecast periods– 0-24, 24-48, 48-72, 72-96, 96-120 hr

• Forecast Techniques– OFCL, DSHP, LGEM, GHMI, HWFI

• Question 1: Do techniques forecast RI?• Question 2: How accurate are RI forecasts?

– Probability of Detection = % of observed RI cases correctly forecasted– False Alarm Ratio = % of RI forecasts that were not RI cases

Percentage of AL Cases with RI

Percentage of EP Cases with RI

POD and FAR for 0-24 hr 30 kt Threshold

Atlantic

East Pacific

Main Tools for RI Forecasting at NHC• SHIPS-RI Index

– 8 predictors used to estimate RI probabilities

• Global model predictions/fields– Favorable characteristic upper-level patterns

and/or the models showing sharp intensification.

• Microwave images – Research has shown that a closed ring on the 37

gHz channel with other favorable environment factors greatly increases the chance of RI.

RI GuidanceHurricane Rick (2009 - East Pacific)

(output below is available in real-time)* EAST PACIFIC SHIPS INTENSITY FORECAST ** GOES DATA AVAILABLE ** OHC DATA AVAILABLE ** RICK EP202009 10/16/09 18 UTC *

TIME (HR) 0 6 12 18 24 36 48 60 72 84 96 108 120V (KT) NO LAND 70 79 86 92 97 104 108 111 111 107 107 101 93V (KT) LAND 70 79 86 92 97 104 108 111 111 107 107 101 93V (KT) LGE mod 70 79 86 92 96 99 95 91 87 85 83 80 76

** 2009 E. Pacific RI INDEX EP202009 RICK 10/16/09 18 UTC **( 30 KT OR MORE MAX WIND INCREASE IN NEXT 24 HR)

12 HR PERSISTENCE (KT): 20.0 Range:-20.0 to 35.0 Scaled/Wgted Val: 0.7/ 1.6850-200 MB SHEAR (KT) : 6.0 Range: 15.2 to 1.6 Scaled/Wgted Val: 0.7/ 0.8D200 (10**7s-1) : 70.0 Range:-10.0 to 129.0 Scaled/Wgted Val: 0.6/ 0.4POT = MPI-VMAX (KT) : 96.7 Range: 46.6 to 134.3 Scaled/Wgted Val: 0.6/ 0.6850-700 MB REL HUM (%): 79.4 Range: 64.0 to 88.0 Scaled/Wgted Val: 0.6/ 0.2% area w/pixels <-30 C: 98.0 Range: 26.0 to 100.0 Scaled/Wgted Val: 1.0/ 0.5STD DEV OF IR BR TEMP : 8.3 Range: 35.4 to 2.7 Scaled/Wgted Val: 0.8/ 1.3Heat content (KJ/cm2) : 46.8 Range: 4.0 to 67.0 Scaled/Wgted Val: 0.7/ 0.4

Prob of RI for 25 kt RI threshold= 78% is 6.8 times the sample mean(11.5%)Prob of RI for 30 kt RI threshold= 71% is 9.3 times the sample mean( 7.7%)Prob of RI for 35 kt RI threshold= 66% is 12.6 times the sample mean( 5.2%)

Rapid IntensificationHurricane Rick (2009 - East Pacific)

NHC FORECAST POSITIONS AND MAX WINDS

INITIAL 16/2100Z 13.0N 100.0W 75 KT12HR VT 17/0600Z 13.2N 101.3W 90 KT24HR VT 17/1800Z 13.7N 103.3W 105 KT36HR VT 18/0600Z 14.3N 105.8W 115 KT48HR VT 18/1800Z 15.0N 108.1W 125 KT72HR VT 19/1800Z 16.5N 111.5W 120 KT96HR VT 20/1800Z 18.5N 113.0W 105 KT120HR VT 21/1800Z 20.5N 113.0W 85 KT

70 kt 135 kt

24 hrs

Difficulty with Rapid Intensity Change

• First round of RI (intensification to Cat 1 hurricane) relatively well anticipated in official NHC forecasts

• Second more extreme instance of RI was not well forecast• Rapid weakening after peak intensity not forecast well either

GFS SLP/200 mb winds

Marie

John Kaplan (NOAA/HRD), Mark DeMaria (NOAA/NHC), James Kossin (NOAA/NCDC), Christopher Rozoff (CIMSS), Charles Sampson(NRL), Christopher Velden (CIMSS), Joseph Cione (NOAA/ESRL/HRD), Jason Dunion (U. Miami/HRD), John Knaff (NOAA/NESDIS), Jeremy Solbrig (NRL), Jun Zhang (U. Miami/HRD), Jack Dostalek (CIRA), Jeffrey Hawkins (NRL), Evan Kalina (U. Colorado), Paul Leighton (NOAA/HRD)

Acknowledgements: Financial support provided by the NOAA/OAR USWRP Joint Hurricane Testbed and NOAA/NESDIS GOES-R Proving Ground programs.

Background

• Deterministic forecasts of RI have generally been unreliable.

• Statistically based model for estimating the probability of RI for a 24-h lead time (i.e., t=0 to t=24-h) using SHIPS environmental predictors (SHIPS-RII) was developed and implemented operationally at the NHC in 2004 (Kaplan and DeMaria 2003).

• Discriminant analysis version of the SHIPS-RII that employed additional environmental and new inner-core GOES-IR predictors was subsequently developed and became operational at the NHC in 2008 (Kaplan et al. 2010).

• Recent research has been conducted to improve statistical RI prediction by developing enhanced versions of the SHIPS-RII, Logistic regression, Bayesian, and Consensus RI models (Rozoff and Kossin 2011) for lead times of 12-h, 24-h, 36-h and 48-h.

Predictors used in new enhanced SHIPS-RII

Predictor Definition

DV12 Previous 12-h intensity change (+)

SHEAR 850-200-hPA vertical shear from 0-500 km radius with vortex removed (time-avg.) (-)

D200 200-hPA divergence from 0-1000 km radius (time-avg.) (+)

TPW Percent area with TPW < 45 mm within 500 km radius 90° up-shear (t=0 h) (-)

PC2 Second principle component of GOES-IR imagery within 440 km radius (t= 0 h) (-)

SDBT Std. dev. of 50-200 km GOES-IR brightness temperatures (t= 0 h) (-)

MPI Potential intensity (Current intensity – maximum potential intensity) (time-avg.) (+)

OHC Oceanic heat content (time-avg.) (+)

ICDA Inner-core dry air predictor (time-avg.) (-)

VMX0 Maximum sustained wind (t=0 h) (+)

SHIPS-RI Predictor Relative Weights

Atlantic E. Pacific

Verification Methodology

• Cross-validated leave one-year out versions of each of the Atlantic and E. Pacific RI models (SHIPS, Bayesian, Logistic regression, Consensus) were derived for all RI thresholds for the 10-year 2004-2013 sample.

• The new independent cross-validated RI models were rerun for all cases from 2004-2013 utilizing the operational GFS forecast fields and NHC storm input data archived for that 10-year period.

• The 2004-2013 cross-validated independent forecasts were verified for all tropical and subtropical over-water cases as determined based upon NHC best track data.

Skill of the probabilistic independent 2004-2013 RI model rerun forecasts

Atlantic E. Pacific

Reliability of the probabilistic consensus RI model 2004-2013 rerun forecasts

Atlantic E. Pacific

Sample consensus RI model Atlantic basin rerun forecasts

Wilma (2005) Michael (2012)

Deterministic RI verification

• All operational model, NHC, and consensus RI model rerun forecasts were verified for a homogenous sample of over-water tropical and subtropical forecast cases for the period 2008-2013.

• RI consensus model forecast probability cutoffs of ~3-3.5 times the climatological probability of RI used to predict RI determined based upon 2004-2007 rerun forecasts following the methodology described in Kaplan et al (2010).

• Operational intensity model and NHC OFCL forecasts determined to predict RI if forecasted intensity > RI threshold for a given case.

Atlantic (2008-2013) deterministic RI model forecast skill

E. Pacific (2008-2013) deterministic RI forecast skill

Improvements in NHC OFCL RI forecast skill

Future statistical RI model improvements

• NHC Structural information (Carrasco et al. 2014)

• Lightning data (DeMaria et al. 2012)

• Microwave satellite data (Rozoff et al. 2014)

• New HFIP diagnostics• Large-scale • Aircraft data (Rogers et al. 2014)

• Consensus methods• ECMWF version of RII• Other statistical techniques• Logistic regression, Bayesian

Extra slides

Percentile thresholds associated with the corresponding RI intensity change for the Atlantic/Eastern North Pacific basins RI thresholds (1995-2012).

RI threshold N NRI Probability of RI (kt) (%)

20 4387/3218 255/206 94.2/93.625 3768/2729 448/357 88.1/86.9 30 3768/2729 286/237 92.4/91.335 3768/2729 168/165 95.6/94.040 3768/2729 113/116 97.0/95.845 3224/2293 166/147 94.9/93.655 2767/1903 148/114 94.7/94.0

Sample Consensus RI model E. Pacific basin rerun forecasts

Rick (2009) Flossie (2007)