190 likes | 204 Views
This study assesses the performance of the ECMWF High-Resolution Global Model during the 2006 Northern Hemisphere season and its influence on consensus forecasting quality. The analysis includes comparisons with peer models, error compensation mechanisms, and timing aspects in forecasting Tropical Cyclones. The study highlights the model's strengths and weaknesses across different basins, emphasizing the impact on predicting storm intensity. Speculation for future model improvements and basin-specific performance variations are also discussed based on the findings.
E N D
Performance of the ECMWF High-Resolution Global Model during the 2006 Northern Hemisphere SeasonandImpact on CONsensus Mike Fiorino1 michael.fiorino@noaa.gov National Hurricane Center Miami, FL 7 March 2007 1CDR USN(RC), CNE-C6F DET 802 Atlanta
Full Disclosure… • was “seconded” to ECMWF 1998-99 for their ERA-40 reanalysis project • Developed TC data assimilation techniques and TC model verification “schemes” • ECMWF is very much O2R vice R2O • Research is (wholly) driven by Operations except for the proverbial “42.9 stone” member of staff (e.g., Tim Palmer) • Operations = medium-range weather (10-15 d) = 5-d 500 AC + probability seasonal… nothing to do with TCs…
One of my takeaways from the 61st IHC… from my ONR S&T Program 38 days So What? Who Cares? from Bill Gray This is all well and good Mike, but why are you doing this? BLUF – Bottom Line Up Front BS Leverages/Losses yoUFunding a corollary
Answer to Bill Gray … NHEM because model (track) skill varies with (low-freq) synoptic situation …… potential for dynamical model intensity forecast skill in high resolution solutions …
LANT 2006 :: BLUFhuge trend in ECMWF skill with tau @ d+3/4/5 >> peers 72-h Vmax - Mean Abs Error - Bias d+3 d+5 ECMO06 > ECMO12 > CONU > OFCL
EPAC 2006:: BLUFECMWF ~ peers except > d+4/5 OCFL > all models and CONU strong error compensation GFS -CTE bias (equatorward) GFDL +CTE bias (poleward) ECMWF actually better than in previous years in EPAC persistent slow bias for WNW moving storms…
WPAC 2006:: BLUF ECMWF ~ peers except > d+3/4/5 ECMO12 as at the “level of incompetence” d+3 JTYM06 > ECMO12 >
Summary of 2006 NHEM Model Errors • LANT • ECMWF medium-range track (MRT) skill high with low “perishability” – even +12h tracker has value • bias-corrected intensity forecast shows a glimmering of skill at the medium-range – benefit of high spatial resolution? yes and no… • EPAC • all models had poor skill, but ECMWF shows some MRT skill • strong error compensation between GFS and GFDL due to poor vortex initialization CON >> than individual model • WPAC • better and less variation in MRT skill of the models • ECMWF shows good MRT skill, but dreadful intensity errors ??? (meteorology)
GTS ECMWF BUFR trackers – Data flowlatency ~ 0.5 h ECMWF JTWC UKMO FNMOC TOC NCEP/NCO NHC
Timing Issuese.g., 00Z model -> 06Z forecast with 09Z initial posit • ECMWF Integrated Forecast System (IFS) • +6.0 h – Tl799L91 (N400, dx~25km) 1 deterministic run • +8.0 h – Tl399L62 (N200, dx~50km) 51-member EPS run • +8.1 h – TC TRACKERS for DET and EPS run • +8.5 h – TC trackers reach TOC/NCEP/FNMOC/NHC/JTWC • ECMWF 1deg fields at NCEP NHC • +7.0 h – 120-h 1deg DET solutions available • NHC • forecast process starts +6.5 need CON by + 6.5 h • JTWC • forecast process starts +7.0 need CON by + 7.5 h • “on time” model trackers ≤ 6.5 h • “late” model trackers > 6.5 h
Summary of ECMWF timing and prospects for 2007 NHEM season… ECMWF trackers “late”NCEP/NHC tracker of ECMWF 1deg semi “on-time” ECMWF working to decouple TC trackers from the EPS run, may be “on time” for 2007
Impact on CON… The Goerss Laws of CON_orthe conditions when medium-range forecast error of CON_> models 1) models have similar skill – applies in both directions… 2) error tend to be decorrelated
CONM – LANT 2006ontime v late v all (CONM) 72 h ecmo06 > ecmo12 > conm >> conu contributes to CON, but not always … sampling problem late taus: model >> peers, CON < best model early taus: model << peers, CON degraded
CONM – EPAC 2006ontime v late v all (CONM) ecmo06 & ecmo12 ~ other models but contributes because of error decorrelation model ~ peers, CON > best modelerror compensation
CONM – WPAC 2006ontime v late v all (CONM) big impact at d+5 model ~ peers, big contribution to CON at all taus
Summary and some Speculation… • ECMWF makes positive contribution to CON > current baseline CON at the medium-range • exception was the LANT because it was much better than other members (Goerss CON rule 1) • Performance varies by basin synoptic pattern subtropical ridge/midlat baroclinic activity • ECMWF very good for storms influenced by the midlats • hi-res solution showing intensity prediction skill at the medium range, again for midlats • Speculation – ECMWF will NOT do as well in 2007 in the LANT as in 2006…but will improve CON in all basins at the medium range …
Feb 2005Commander Fiorino, what’s your forecast for my 2005 WESTPAC season?COMPACFLT Meteorologist It won’t be like 2004 Captain…
NHEM TC activity anomalies 2003-2006 • 2006 ∟ 2005 • WPAC year-to- year shifts > EPAC/LANT • 2006 WPAC: strong STR – 5 hits on the PI • 2006 LANT: weaker STR, more midlat • 2006 EPAC: N tracks, synoptic-scale land effects
sometimes the magic works, and sometimes it doesn’t…The Goerss Second Law of TC model skill … partly because model skill does vary with the low-freq synoptic pattern implied by the TC activity anomalies …