Tropical cyclone track prediction through multimodel ensemble forecast
Tropical cyclone track prediction through multi-model ensemble forecast system Jiayi Peng*, Yuejian Zhu and Richard Wobus* *IMSG at Environmental Modeling Center /NCEP/NOAA, Camp Springs, MD 20746 Acknowledgements: EMC Ensemble Team Geophysical Fluid Dynamics Laboratory (Tim Marchok) National Hurricane Center (Christopher Sisko) Earth System Research Laboratory (Michael Fiorino) Hurricane Forecast Improvement Project (HFIP) [1]
Outline • • (1) Introduction (2) Models and data (3) Super-ensemble products (4) Conclusions [2]
21 TCs in Atlantic in 2010 12 hurricanes: Alex, Danielle, Earl, Igor, Julia(Category 4), Karl, Lisa, Otto, Paula, Richard, Shary, Tomas. http: //en. wikipedia. org/wiki/2010_Atlantic_hurricane_season [3]
12 TCs in East Pacific in 2010 3 hurricanes: Celia (Category 5), Darby, Frank. http: //en. wikipedia. org/wiki/2010_Pacific_hurricane_season [4]
19 TCs in West Pacific in 2010 7 typhoons: Conson, Chanthu, Kompasu, Fanapi, Malakas, Megi(885 mbar), Chaba. http: //en. wikipedia. org/wiki/2010_Pacific_typhoon_season [5]
Part(1): Introduction (1)National Hurricane Center TC track prediction: ----weighting for consensus ----weighting for non-consensus aid OFCL------Official NHC forecast OFCI------Previous cycle OFCL, adjusted (Interpolated) TVCN------Average of at least 2 of GFSI, EGRI, NGPI, EMXI, HWFI, GFNI, GHMI (Consensus) Global dynamical models GFSI----Previous cycle GFS, adjusted EGRI----Previous cycle EGRR (UKMET), adjusted NGPI----Previous cycle NGPS(Navy NOGAPS), djusted EMXI----Previous cycle EMX (ECMWF Model), adjusted Regional dynamical models HWFI----Previous cycle HWRF, adjusted GFNI----Previous cycle GFDN (Navy version of GFDL), adjusted GHMI----Previous cycle GFDL, adjusted No global ensemble product is used at NHC for TC track forecast! [6]
(2)Joint Typhoon Warning Center TC track prediction: ----weighting for consensus ----weighting for non-consensus aid JTWC----JTWC official forecast JTWI----Previous cycle JTWC, interpolated CONW----Consensus with AVNI, EGRI, ECMI, NGPI, JGSI, GFNI, JTYI, TCLI, WBAI Global dynamical models AVNI----NCEP AVN TC vortex tracker (AVNO, NHC interpolator) EGRI----Bracknell model ( UKMET) (NHC interpolator) ECMI----ECMWF model (Interpolated) NGPI----NOGAPS TC vortex tracker (NHC interpolator) JGSI----JGSM (NHC interpolator from ATCF JGSM, Japanese Global Spectral Model) Regional dynamical models GFNI----Navy version of GFDL model (NHC interpolator) JTYI----JTYM (NHC interpolator), Japanese Typhoon Model TCLI----TC LAPS, Australian TC-Limited Area Prediction System (NHC Interpolator) WBAI----Harry Weber model interpolated (barotropic track model) No global ensemble product is used at JTWC for TC track forecast! Our goal: Improve tropical-cyclone track prediction by using NCEP, CMC and ECMWF global ensemble forecast system products ! [7]
Part(2): Models and Data Resolution Members Daily Frequency Forecast Length 00, 06, 12, 18 UTC 16 days (384 hrs) NCEP ensemble GFS T 190 L 28 70 km - 20+1 CMC ensemble GEM(0. 9)L 58 -100 km 20+1 00, 12 UTC 16 days (384 hrs) ECMWF ensemble IFS T 639/319 L 62 -30/60 km 50+1 00, 12 UTC 15 days (360 hrs) NCEP deterministic GFS T 574 L 64 27 km 00, 16, 12, 18 UTC 192 hrs CMC deterministic GEM(0. 45 x 0. 3) L 58 -33 km 1 00, 12 UTC 180/240 hrs (12/00 z) ECMWF deterministic IFS T 1279 L 91 -16 km 1 00, 12 UTC 240 hrs 00, 12 UTC 16 days (384 hrs) - 1 FNMOC ensemble NOGAPS T 119 L 30 20 Data: 01/01~12/31/2010 NCEP, CMC and ECMWF tropical cyclone tracks; 08/25~12/31/2010 for FNMOC. [8]
Part(3): Super-ensemble products. 33 TCs in Atlantic and East Pacific in 2010 1 Nautical Mile=1. 852 km Verification: 3 EMN and T 4 MN are much better than OFCL. NEMN----NCEP (GEFS) 20 -member mean 3 EMN----NCEP(20)+CMC(20)+ECMWF(50) 90 -member mean AVNO----NCEP (GFS) deterministic run (T 574 L 64) T 4 MN----3 EMN+(NCEP+CMC+ECMWF, three deterministic runs) OFCL----NHC official forecast [9]
21 Atlantic hurricanes (TCs) in 2010 Verification: 3 EMN and T 4 MN are much better than OFCL. NEMN----NCEP (GEFS) 20 -member mean 3 EMN----NCEP(20)+CMC(20)+ECMWF(50) 90 -member mean AVNO----NCEP (GFS) deterministic run (T 574 L 64) T 4 MN----3 EMN+(NCEP+CMC+ECMWF, three deterministic runs) OFCL----NHC official forecast [10]
12 TCs in East Pacific NEMN----NCEP (GEFS) 20 -member mean 3 EMN----NCEP(20)+CMC(20)+ECMWF(50) 90 -member mean AVNO----NCEP (GFS) deterministic run (T 574 L 64) T 4 MN----3 EMN+(NCEP+CMC+ECMWF, three deterministic runs) OFCL----NHC official forecast [11]
19 TCs in West Pacific Verification: 3 EMN and T 4 MN are much better than JTWC. NEMN----NCEP (GEFS) 20 -member mean 3 EMN----NCEP(20)+CMC(20)+ECMWF(50) 90 -member mean AVNO----NCEP (GFS) deterministic run (T 574 L 64) T 4 MN----3 EMN+(NCEP+CMC+ECMWF, three deterministic runs) JTWC----JTWC official forecast [12]
TCs: AL 06 -21, EP 09 -12, WP 07 -19 Cases 288 272 239 209 180 133 98 70 2 EMN---- NCEP(20)+CMC(20) 40 -member mean NAEF---- NCEP(20)+CMC(20)+FNMOC(20) 60 -member mean [13] 3 EMN----NCEP(20)+CMC(20)+ECMWF(50) 90 -member mean 4 EMN---- NCEP(20)+CMC(20)+ECMWF(50) +FNMOC(20) 110 -member mean
Strike probability AL 06: Danielle (2010) Ensemble mean track NEMN----NCEP 20 -member mean TEMN----CMC 20 -member mean UEMN---- ECMWF 50 -member mean 3 EMN----NCEP+CMC+ECMWF 90 -mean [14]
Strike probability WP 15: Megi (2010) Ensemble mean track NEMN----NCEP 20 -member mean TEMN----CMC 20 -member mean UEMN---- ECMWF 50 -member mean 3 EMN----NCEP+CMC+ECMWF 90 -mean [15]
Ensemble Spread VS. Forecast Error [16]
Part(4): Conclusions • The multiple model ensemble forecast system ( NCEP, CMC and ECMWF ensemble forecast) could significantly reduce the 2010 TC track forecast error in Atlantic and West Pacific, in comparison to the NHC and JTWC official forecast. • If the Navy FNMOC ensemble forecast is added into the multiple model ensemble forecast system, this will not improve the 2010 TC track forecast. [17]
Future plans: • Add mesoscale models (HWRF, GFDL, SREF, etc. ) into this multiple model ensemble forecast system. • Combine bias-correction TC-track products into this system, as we have the NCEP, CMC and FNMOC biascorrection in operational. • Develop new consensus /ensemble method. • Produce TC intensity multi-model ensemble forecast. [18]
Questions? [19]
Track forecast error for 2009 season (AL+EP+WP) Cases 240 223 196 169 144 110 75 42 [20]
2009 GEFS T 190 Parallel Run (AL/EP/WP, 00 Z) Case 255 235 208 184 159 120 91 62 [21]
- Slides: 21