Extreme waves in the ECMWF operational wave forecasting system Jean-Raymond Bidlot Peter Janssen Saleh Abdalla European Centre for Medium range Weather Forecasts Shinfield Park, RG 9AX, Reading, United Kingdom +--999-7 jean.bidlot@ecmwf.int 9th int. workshop on wave hindcasting and forecasting Slide
Outline ECMWF wave forecasting system. Impact of recent improvements. Extreme wave forecasting: - Ensemble prediction system (EPS) - Freak wave 9th int. workshop on wave hindcasting and forecasting Slide
Operational wave forecasting at ECMWF A version of WAM cycle (ECWAM) is used at ECMWF. It is an integral part of the Centre s forecasting systems. The wave model benefits from improvements in the atmospheric model via the forcing winds. Work on improving the wave model physics, dynamics and data assimilation is continuing. Several operational configurations exist. 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF wave model configurations ) Limited area model (LAW) From 9 N to N and 9 W to E km grid spacing ( km soon). frequencies and directions Forced by m wind fields from the global atmospheric model. Data assimilation of altimeter wave heights (ENVISAT and Jason) and ASAR spectra (ENVISAT). daily forecasts (from & Z) extending to day. 9 W W 7 W W W Tuesday March W UTC ECMWF Forecast t+ W VT: Wednesday March W W W UTC 7 W 7 E Surface: E E E significant E wave Eheight W W E E Forecast wave height on // UTC. N 7 N N N N N N N E E 9 E E 7 E E E E E.. 9.7 9. 7..7...7...7 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF wave model configurations ) Global models Global from S to N Coupled to the atmospheric model (IFS) with feedback of the sea surface roughness change due to waves. The interface between WAM and the IFS has been generalised to include air density and gustiness effects on wave growth and more recently neutral winds. Data assimilation of ENVISAT and Jason altimeter wave heights and ENVISAT ASAR spectra. 7 N N N N S S S S S S S S S 7 S Tuesday March UTC ECMWF Forecast t+ VT: Wednesday March UTC Surface: significant wave height E N N N N N N E E E E E E E E E E E E E E E W W W W W W W W W W W W Forecast wave height on // UTC. 7 N N N N S S S 7 S.9 9.7 9. 7..7...7...7 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF wave model configurations Deterministic model km grid spacing. frequencies and directions. Coupled to the TL799 model (~km resolution). Analysis every hrs and day forecasts from and UTC and day forecasts from and UTC. Probabilistic forecasts km grid spacing. frequencies and directions. Coupled to the TL99 model (~km resolution). (+) day forecasts from and Z. 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF wave model configurations Monthly forecasts. x. grid. frequencies and directions. Coupled to the TL9 model. Deep water physics only. Seasonal forecasts. x.. frequencies and directions. Coupled to the T9 model. Deep water physics only. 9th int. workshop on wave hindcasting and forecasting Slide 7
ECMWF wave model configurations ERA reanalysis (97-). x. grid. frequencies and directions. Coupled to the TL9 model. Deep water physics only. years of wave analysis The data were analysed at knmi. http://www.knmi.nl/waveatlas Interim reanalysis (99 to present). x. grid. frequencies and directions. Coupled to the latest TL model. Shallow physics only. Production has started. Early assessment indicates a much improved data set. 9th int. workshop on wave hindcasting and forecasting Slide
Monitoring: ECMWF analysis against GTS in-situ data.. BIAS (model obs).. S.I. BIAS(m) -. -. -. -. S.I. (%) -. -. 99 J9 j J9 j J9 j J9 j J97 j J9 j J99 j J j J j J j J j J j J j J j months ( month running average) J9 j J9 j J9 j J9 j J97 j J9 j J99 j J j J j J j J j J j J j J j months ( month running average) GTS in-situ wave data locations 99 S.I.: standard deviation of error normalised by the mean of the observations 9th int. workshop on wave hindcasting and forecasting Slide 9
ECMWF analysis versus buoy spectral data Jason alt. + ENVISAT ASAR Revised dissipation Equivalent wave height bias at all US and Canadian buoys. Operational analysis. Jun- Feb- Oct- Jun- Feb- Oct- Jun- Feb- Oct- Jun- Feb- Oct- Jun- Feb- Oct- Jun- Feb- Oct- Jun- Feb- Oct-99 Jun-99 Feb-99 Oct-9 Jun-9 months Bias (m) : model - buoy.-.7 m.-..-..7-..-.7.-..-..7-..-.7.-. -. -.- -.--. -.7--. -.--.7 -.--. -.--. -.7--. -.--.7 Feb-9 7.. 7. Wave period (s) 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF forecasts versus GTS in-situ data months to August, all buoys: wave height rmse expver= 99_9 99_9 99_9 Operational forecasts 99_97 979_9 99_99 999_ 9_ 9_ 9_ 9_ 9_ 9_. Hs RMSE (m).7.. 7 9 9 Forecast range (hours) forecast from UTC 9th int. workshop on wave hindcasting and forecasting Slide
ECMWF forecasts versus own analysis Forecast scores are also obtained by verifying against own analyses as it is done with atmospheric fields.. WAVE_G FORECAST VERIFICATION UTC HEIGHT OF WAVES SURFACE LEVEL STANDARD DEVIATION OF ERROR FORECAST N.HEM LAT. TO 9. LON -. TO. M T+ T+ MA T+ T+ MA T+ T+ MA MA = Month Moving Average. Standard deviation of error for N.Hem..... Day Day. 99 99 997 99 999 Day 9th int. workshop on wave hindcasting and forecasting Slide
Ensemble forecasting for waves Beside its deterministic model, ECMWF also runs an ensemble of forecasts twice daily. These forecasts are initialised from a set of perturbed analyses, designed to represent the inherent uncertainty in the operational analysis. Similarly, the uncertainty in the model physical parametrisation is accounted for by stochastic perturbations. Because of limited computer resources and lack of predictability of small scales, these forecasts are run with a coarser resolution. A control forecast is also run at that reduced resolution. These (+) forecasts are used to derive forecast error estimate and forecast probabilities. Waves are in the EPS since June 99. 9th int. workshop on wave hindcasting and forecasting Slide
Old example: - November storm in the Norwegian Sea From the EPS wave forecasts it is possible to derive probabilities for certain wave conditions. Tuesday November UTC ECMWF EPS Probability Forecast t+ VT: Sunday November UTC Surface: significant wave height probability > W W W W E E E E E Significant wave height (m).7 E N MIKE HEIDRUN DRAUGEN W E 7 9 Forecast day N N E Forecasts from Nov, UTC at platform Heidrun W W W E E November UTC ECMWF EPS probability forecast t+ SIGNIFICANT WAVE HEIGHT GREATER than m Perturbed forecasts Control forecast Deterministic forecast 9th int. workshop on wave hindcasting and forecasting Slide
Katrina: SWH probability in the then operational system T (~km) from opereational analysis in +h fcs The top-left panel shows the significant wave analysis SWH > m height (SWH) in the T799 analysis (cont interval is m). The other panels show the probabilities that: SWH>m (t-right) SWH>m (b-left) SWH>m (bright) Prob cont iso are /////%. SWH > m SWH > m 9th int. workshop on wave hindcasting and forecasting Slide
Katrina: SWH probabilities in the new operational system T99 ( km) from operational analysis in +h fcs The top-left panel shows the significant wave height (SWH) in the T799 analysis (cont interval is m). The other panels show the probabilities that: SWH>m (t-right) SWH>m (b-left) SWH>m (b-right) Prob cont iso are /////%. analysis SWH > m SWH > m SWH > m 9th int. workshop on wave hindcasting and forecasting Slide
Katrina: SWH t+h fcs at buoy, T99 and T: resolution matters Buoy obs for UTC of 9 Aug and t+h forecasts from Aug UTC. Bottom-left panel: buoy measured SWH of m. ECMWF analysis at T and T799 produced SWH of m. EPS forecasts were up to ~m (top-right), while VAREPS forecasts reached 9m (bottom-right). 9th int. workshop on wave hindcasting and forecasting Slide 7
99 9 99 9 99 Recent example: extra-tropical Gordon landfall in Ireland ECMWF Analysis VT:Wednesday September UTC Surface: mean sea level pressure/surf: mtr v W W W E.m/s ECMWF Analysis VT:Wednesday September UTC Surface: mean sea level pressure/surf: mtr v W W W E.m/s Analysed mean Sea Level pressure and m winds W W 99 97 9 9 9 99 97 99 9 99 N N 99 E W W Sep, UTC 97 99 9 99 99 N N E Sep, UTC N N W ECMWF Analysis VT:Thursday September UTC Surface: mean sea level pressure/surf: mtr v W W W E.m/s ECMWF Analysis VT:Thursday September UTC Surface: mean sea level pressure/surf: mtr v W W W W E.m/s W 99 W 99 99 N 99 N 9 99 9 E E W W N N 99 Sep, UTC N N Sep, UTC W W 9th int. workshop on wave hindcasting and forecasting Slide
Extra-tropical Gordon landfall: -day forecasts Tuesday 9 September UTC ECMWF EPS Probability Forecast t+7 VT: Friday September UTC Surface: significant wave height probability > ECMWF EPS FOR: M 9 DATE: 99 Z LAT:.7 LONG: -.7 m 7 m/s Significant wave height (m) 7 9 m wind speed (m/s) T_FC Veps_CF Veps_EMem W W W W W W E Probability that SWH will exceed m N N N E. ECMWF EPS FOR: Sevenstones 7 DATE: 99 Z LAT:. LONG: -. 7 7 m 9 7 7 9 m wind speed (m/s) m/s Significant wave height (m) T_FC Veps_CF Veps_EMem s 7 9 7 7 9 Peak wave period (s) 7 9 Perturbed forecasts Control forecast Deterministic forecast s 7 9 7 7 9 Peak wave period (s) 7 9 9th int. workshop on wave hindcasting and forecasting Slide 9
Extra-tropical Gordon landfall: -day forecasts Wednesday September UTC ECMWF EPS Probability Forecast t+ VT: Friday September UTC Surface: significant wave height probability > ECMWF EPS FOR: M 9 DATE: 9 Z LAT:.7 LONG: -.7 m9 Significant wave height (m) T_FC Veps_CF Veps_EMem W W W W E. ECMWF EPS FOR: Sevenstones 7 DATE: 9 Z LAT:. LONG: -. Significant wave height (m) m9 T_FC Veps_CF Veps_EMem 7 N 7 W E 7 9 m wind speed (m/s) N 7 9 m wind speed (m/s) m/s 7 9 Peak wave period (s) Probability that SWH will exceed m W N m/s 7 9 Peak wave period (s) s s 7 9 7 Perturbed forecasts Control forecast Deterministic forecast 7 9 7 7 9 7 9 9th int. workshop on wave hindcasting and forecasting Slide
Extra-tropical Gordon landfall: -day forecasts Thursday September UTC ECMWF EPS Probability Forecast t+ VT: Friday September UTC Surface: significant wave height probability > ECMWF EPS FOR: M 9 DATE: 9 Z LAT:.7 LONG: -.7 m 7 m/s s 7 9 7 Significant wave height (m) 7 9 m wind speed (m/s) 7 9 Peak wave period (s) FC CF PF 7 9 W W W W W W E Probability that SWH will exceed m Perturbed forecasts Control forecast Deterministic forecast N N N E ECMWF EPS FOR: Sevenstones 7 9 DATE: 9 Z LAT:. LONG: -. 9 7 7 m 7 m/s s 7 9 7 Significant wave height (m) 7 9 m wind speed (m/s) 7 9 Peak wave period (s) FC CF PF 7 9 9th int. workshop on wave hindcasting and forecasting Slide
Extra-tropical Gordon landfall: analysis ECMWF Analysis VT:Friday September UTC Surface: significant wave height W.....9...7.... N... W 9th int. workshop on wave hindcasting and forecasting Slide
Freak waves Extreme conditions can arise on much smaller time scale as modelled by current spectral wave model: i.e. as freak waves. From Janssen (), it appears that nonlinear wavewave interaction is one of the possible mechanisms that will lead wave energy focussing. Waves needs to be sufficiently coherent and steep as quantified by the Benjamin-Feir Index (BFI), namely the ratio of the steepness of the waves and the width of the spectrum. A large BFI corresponds to favourable condition for freak waves. For freak waves prediction, probabilistic approach is feasible, following Janssen (), there is a link between BFI and the kurtosis of the sea surface elevation. 9th int. workshop on wave hindcasting and forecasting Slide
Wave model products: Freak waves Since October, new wave parameters have been produced that characterize extreme sea state: For example, The kurtosis of the sea surface elevation can be used to derive the enhanced probability that waves are larger than twice the significant wave height: Saturday 7 February UTC ECMWF Forecast t+ VT: Sunday February UTC Surface: **Wave spectral kurtosis Forecast enhancement probability factor of waves larger than Hs W W W N N N N N N N N W W W 7...... One day forecast of enhanced probability of extreme events (twice Hs) for th February. 9th int. workshop on wave hindcasting and forecasting Slide
Wave model products: Freak waves Enhanced probability of extreme events for. Wednesday September UTC ECMWF Forecast t+ VT: Thursday September UTC Surface: **Wave spectral kurtosis Thursday September UTC ECMWF Forecast t+ VT: Thursday September UTC Surface: **Wave spectral kurtosis W Forecast enhancement W probability factor of waves larger than W Hs E W Forecast enhancement W probability factor of waves larger than Hs W E..9.. W W N N Sep, UTC W E. W E. N N.. t+ hrs forecast. N. t+ hrs forecast. N. W W Wednesday September UTC ECMWF Forecast t+ VT: Friday September UTC Surface: **Wave spectral kurtosis W Forecast enhancement W probability factor of w aves larger than W Hs E. Thursday September UTC ECMWF Forecast t+ VT: Friday September UTC Surface: **Wave spectral kurtosis W Forecast enhancement W probability factor of w aves larger than W Hs E. W W N. N. Sep, UTC W E W E N. N. N. t+ hrs forecast. t+ hrs forecast. N. W 9th int. workshop on wave hindcasting and forecasting Slide W
Final words : Improved wave model and atmospheric model yields better forecasts. Probabilistic forecast can be useful in assessing likelihood of severe waves. Freak waves: ongoing validation of new parameters. 9th int. workshop on wave hindcasting and forecasting Slide