Hurricane Forecast Improvement Project (HFIP):Where do we stand after 3 years?
Bob Gall – HFIP Development Manager
Fred Toepfer—HFIP Project manager
Frank Marks – HFIP Research Lead
Ed Rappaport – HFIP Operations Lead
March 6, 2013
2
The HFIP Project – Vision/Goals
• Visiono Organize the hurricane community to dramatically
improve numerical forecast guidance to NHC in 5-10 years
• Goalso Reduce numerical forecast errors in track and
intensity by 20% in 5 years, 50% in 10 yearso Extend forecasts to 7 dayso Increase probability of detecting rapid intensification
at day 1 to 90% and 60% at day 5
HFIP Baselines and Goals:Track
3
HFIP Baselines and Goals:Intensity
4
HFIP Overall Strategy
• Use global models at as high a resolution as possible to forecast track out to 7 days
• Use regional models at 1-3 km resolution to predict inner core structure to meet intensity goals out to 5 days including rapid intensification
• Hybrid DA for both regional and global using as much satellite and aircraft data as possible
• Both regional and global models run as ensemblesBoth regional and global models run as ensembles
• Statistical post processing of model output to further increase forecast skill
Track Error of Models (2010-2011)(% Improvement over HFIP baseline)
Impact of Aircraft Data(% improvement over D-SHIFOR)
7
How are we doing?
• The HFIP goals are for model products delivered from NCEP to NHC.– The delivery date for these goals is hurricane
season 2014
• The following show the operational models (Global and Regional) performance for hurricane track and intensity in the Atlantic for latest hurricane season (2012)
8
Baseline skill
5-year skill goal
GFS
HWRFGFDL
Comparison of 2012 NCEP Operational Models to the 5 Year HFIP
Goal: Track
Baseline skill
GFS
HWRF
Comparison of 2012 NCEP Operational Models to the 5 Year HFIP
Goal: Intensity
GFDL
HFIP 5 year Goal
11
Stream 1.5 Results for 2012
AHW
HWRF
FSU
FIMGFS
NOGAPS
TVCA
GFDL
ECMWF
UKMET
Canadian Model
AHW
HWRF
FSU
Intensity Consensus
Wisconsin
GFDL
DSHPLGEM
SPC3
TC-COAMPS
• The upgrade to the 3km triple-nested HWRF is a result of multi-agency efforts under HFIP support
– EMC - Computational tuning to speed up the model, nest motion algorithm, physics improvements, 3km initialization and pre-implementation T&E
– HRD/AOML - multi-moving nest, nest motion algorithm, PBL upgrades, interpolation routines for initialization and others.
– DTC - code management and maintain subversion repository
– ESRL - Physics sensitivity tests and idealized capability– NHC - Diagnose the HWRF pre-implementation results– URI - 1D ocean coupling in Eastern Pacific basin
2012 HWRF Upgrades
14
2012 3km HWRF Operational Upgrade Summary
HOPS: oper. HWRF
H212: 2012 HWRF
ATL Tracks• Significant Improvements of H212
– Track/intensity forecast skills for 2011/2010 seasons on Atlantic basin 20-25% improvement against HOPS
– Track forecast skills of H212 of Eastern Pacific basin maximum 25% over the HOPS in 2011 season, but little degradation at day 4 and 5 in 2010 season mainly due to Hurricane Frank
– Intensity of 2011 EP basin with over 40% to HOPS. Significant improvements in intensity bias is noted for both Atlantic and Eastern Pacific, for both 2010-2011 seasons.
– The storm structure in terms of storm size and PBL height significantly improved
– Much improved wind-pressure relationship in high wind speed regime
15
ATL IntensityHOPS: oper. HWRF
H212: 2012 HWRF
20-25% improvement
16
Impact of Radar Data
Impact of TDR data assimilation to hurricane intensity forecast
2.2.2 (EMC)TDR assimilation
OPR HWRF
HWRF TDR
Cross section at initial time
17
With TDR
Impact of TDR Data In Operational HWRF
Without TDR
Without TDR
With TDR
Track Error Intensity Error
• Questions?
19
Extra Slides
Comparison of 2012 NCEP Operational Models to the 5 Year HFIP
Goal: Track
Comparison of 2012 NCEP Operational Models to the 5 Year HFIP
Goal: Intensity
Statistical Post Processing
• Statistical Post Processing can add skill to dynamical forecasts.
• There are a number of techniques based on ensembles or individual models.
• One method is shown in the following figure
– From the FSU Multi-Model Ensemble (MMEN) which forms a weighted mean of the many global and regional models run both operationally and by HFIP in real time.
2012 all storms
Genesis
25
Verification of model genesis for operational global models
• All models have a bias towards over-prediction, caused by both false alarms as well as genesis occurring in the forecast long (>>48h) before observed genesis.
• 4-ensemble consensus close to reliable up through 50-60%.26
NHC Hurricane Genesis Statistics
27