NATO Undersea Research Centre Partnering for Maritime Innovation

  • Slides: 56
Download presentation
NATO Undersea Research Centre Partnering for Maritime Innovation Multi-model Super-Ensembles Applied to Dynamics of

NATO Undersea Research Centre Partnering for Maritime Innovation Multi-model Super-Ensembles Applied to Dynamics of the Adriatic NRL Stennis 15 -17 November 2006 Michel Rixen rixen@nurc. nato. int NATO UNCLASSIFIED

Ensembles… • • Ensemble (single model) – Initial conditions – Boundary conditions – Statistics/parameterization

Ensembles… • • Ensemble (single model) – Initial conditions – Boundary conditions – Statistics/parameterization Super-ensemble (multi-model of the same kind) – Least-squares: weather+climate (Krishnamurti 2000, Kumar 2003) – Max likelihood+ regularization by climatology : tropical cyclones (Rajagopalan 2002) – Kalman filters: precipitation (Shin 2003) – Probabilistic: precipitation (Shin 2003) ‘Hyper’-ensemble (multi-model of different kinds) – e. g. combination of ocean+atmospheric+wave models? General aim: forecast + [uncertainty/error/confidence estimation] 2 particular research lines relevant to MILOC/EOS/NURC/NATO • Acoustic properties • Surface drift 2 NATO UNCLASSIFIED

Super-Ensembles (SE)… Models • • Weights Data Simple ensemble-mean Individually bias-corrected ens. -mean Linear

Super-Ensembles (SE)… Models • • Weights Data Simple ensemble-mean Individually bias-corrected ens. -mean Linear regression (least-squares) Non-linear regression (least-squares) – Neural networks (+regularisation) – Genetic algorithms Compute optimal combination from past model-data regression, then use in forecast-mode 3 NATO UNCLASSIFIED

MREA 04: sound velocity (100 m) 4 NATO UNCLASSIFIED

MREA 04: sound velocity (100 m) 4 NATO UNCLASSIFIED

SE Weights 5 NATO UNCLASSIFIED

SE Weights 5 NATO UNCLASSIFIED

Forecast errors on sound velocity Analysis HOPS IHPO HOPS HRV 6 HOPS HRV FINE

Forecast errors on sound velocity Analysis HOPS IHPO HOPS HRV 6 HOPS HRV FINE NCOM COARSE 2 HOPS NCOM FINE 2 NCOM Single models 4 models SE NATO UNCLASSIFIED

SE Sound speed profile errors 7 NATO UNCLASSIFIED

SE Sound speed profile errors 7 NATO UNCLASSIFIED

HOPS-IHPO (1) HOPS-Harv. (2) Coarse NCOM (3) Fine NCOM (4) SE (2) SE (4)

HOPS-IHPO (1) HOPS-Harv. (2) Coarse NCOM (3) Fine NCOM (4) SE (2) SE (4) SE (1, 2) SE (3, 4) SE (1 to 4) 8 NATO UNCLASSIFIED

MREA 04: DRIFTERS 9 NATO UNCLASSIFIED

MREA 04: DRIFTERS 9 NATO UNCLASSIFIED

Hyper-ensembles Ocean 10 Meteo HOPS ALADIN FR NCOM COAMPS Hyper-ens. Linear HE Non-linear HE

Hyper-ensembles Ocean 10 Meteo HOPS ALADIN FR NCOM COAMPS Hyper-ens. Linear HE Non-linear HE NATO UNCLASSIFIED

Drifter tracks True drifter Ocean advection 48 h forecast Rule of thumb Hyper-ensembles 11

Drifter tracks True drifter Ocean advection 48 h forecast Rule of thumb Hyper-ensembles 11 NATO UNCLASSIFIED

Hyper-ensemble statistics 12 Julian day NATO UNCLASSIFIED

Hyper-ensemble statistics 12 Julian day NATO UNCLASSIFIED

Strong Wind Event (Bora) R. Signell 13 NATO UNCLASSIFIED

Strong Wind Event (Bora) R. Signell 13 NATO UNCLASSIFIED

Standard vs refined turbulence scheme 14 R. Signell NATO UNCLASSIFIED

Standard vs refined turbulence scheme 14 R. Signell NATO UNCLASSIFIED

ADRIA 02 -03 drifters (Jan-Feb) 15 NATO UNCLASSIFIED

ADRIA 02 -03 drifters (Jan-Feb) 15 NATO UNCLASSIFIED

Analysis: 14 Feb 2003 16 NATO UNCLASSIFIED

Analysis: 14 Feb 2003 16 NATO UNCLASSIFIED

ADV WIND Ro. T Indiv. Forecast err. : 14 Feb 2003 (12 Feb 2003+

ADV WIND Ro. T Indiv. Forecast err. : 14 Feb 2003 (12 Feb 2003+ 48 h) ADV+WIND Ro. T 17 ADV+WIND+STOKES NATO UNCLASSIFIED

ADV WIND Ro. T SEs forecast err: 14 Feb 2003 (12 Feb 2003+ 48

ADV WIND Ro. T SEs forecast err: 14 Feb 2003 (12 Feb 2003+ 48 h) ADV+WIND Ro. T 18 ADV+WIND+STOKES NATO UNCLASSIFIED

ADV WIND ADV+WIND+STK Indiv. Mod. 19 SE 5, 10, 25 and 50 days NATO

ADV WIND ADV+WIND+STK Indiv. Mod. 19 SE 5, 10, 25 and 50 days NATO UNCLASSIFIED

20 NATO UNCLASSIFIED

20 NATO UNCLASSIFIED

Drifter tracks Ocean advection Ocean+Stokes Unbiased single models 21 SE 24 h forecast True

Drifter tracks Ocean advection Ocean+Stokes Unbiased single models 21 SE 24 h forecast True NATO UNCLASSIFIED

ADV WIND Ro. T Indiv. mod. uncertainty: 14 Feb 2003 (cross-validation) ADV+WIND Ro. T

ADV WIND Ro. T Indiv. mod. uncertainty: 14 Feb 2003 (cross-validation) ADV+WIND Ro. T 22 ADV+WIND+STOKES NATO UNCLASSIFIED

ADV WIND Ro. T SEs uncertainty on 14 Feb 2003 (cross-validation) ADV+WIND Ro. T

ADV WIND Ro. T SEs uncertainty on 14 Feb 2003 (cross-validation) ADV+WIND Ro. T 23 ADV+WIND+STOKES NATO UNCLASSIFIED

24 NATO UNCLASSIFIED

24 NATO UNCLASSIFIED

25 NATO UNCLASSIFIED

25 NATO UNCLASSIFIED

INDIV 26 SEs NATO UNCLASSIFIED

INDIV 26 SEs NATO UNCLASSIFIED

MS-EVA (JRP Harvard) New methodology utilizing multiple scale window decomposition in space and time

MS-EVA (JRP Harvard) New methodology utilizing multiple scale window decomposition in space and time of a model • multi-scale interactive • nonlinear • intermittent in space • episodic in time E. g. wavelet Selecting the right processes at the right time… 27 NATO UNCLASSIFIED

SE and MS-EVA=MSSE Model 1 Model 2 Model N 28 MSSE combines optimally the

SE and MS-EVA=MSSE Model 1 Model 2 Model N 28 MSSE combines optimally the strengths of all models at any time at different scales Note: Energy/vorticity/mass conservation issues Selecting the right processes from the right models at the right time… NATO UNCLASSIFIED

Lorenz equations 29 NATO UNCLASSIFIED

Lorenz equations 29 NATO UNCLASSIFIED

30 NATO UNCLASSIFIED

30 NATO UNCLASSIFIED

31 NATO UNCLASSIFIED

31 NATO UNCLASSIFIED

SEs 32 MSSEs SEs MSSEs NATO UNCLASSIFIED

SEs 32 MSSEs SEs MSSEs NATO UNCLASSIFIED

Dynamics of the Adriatic in Real-Time • Gulf of Manfredonia & Gargano Peninsula •

Dynamics of the Adriatic in Real-Time • Gulf of Manfredonia & Gargano Peninsula • Mid-Adriatic • Whole Adriatic • Critical mass of research and ressources 33 NATO UNCLASSIFIED

NURC-NRLSSC JRP GOALS • Assess real-time capabilities of monitoring (data) and prediction (models) of

NURC-NRLSSC JRP GOALS • Assess real-time capabilities of monitoring (data) and prediction (models) of small-scale instabilities in a controlled environment (operational framework) Produce a comprehensive data-model set of ocean and atmosphere properties (validation of fusion methods) • • • 1 A 5: ensemble modeling+uncertainty 1 A 2: air-sea interaction, coupling/turbulence 1 D 1: data fusion & remote sensing 1 D 3: geospatial data services • ONR projects: – – • 34 NRL-HRV on internal tides NICOP program on turbulence EOREA ESA (Sat. Ob. Sys/Flyby/ITN/NURC) NATO UNCLASSIFIED

PARTNERS • 33 institutions (on board+home institutions): • 10 USA, 15 ITA, 1 GRC,

PARTNERS • 33 institutions (on board+home institutions): • 10 USA, 15 ITA, 1 GRC, 1 DEU, 1 BEL, 2 FRA 35 Pf. P : 4 HRV, (1 ALB) NATO UNCLASSIFIED

Highlights IN-SITU • SEPTR (1 NURC, 3 NRL) • BARNY (2 NURC, 13 NRL,

Highlights IN-SITU • SEPTR (1 NURC, 3 NRL) • BARNY (2 NURC, 13 NRL, 2 HRV) • Wave rider, meteo stations • CTD chain • +Aquashuttle (NRL, Universitatis) MODELS • Ocean (6+3 to come) • Atmospheric (7) • Wave (4) REMOTE SENSING • NURC: HRPT, Ground station • NRL: MODIS • Sat. Ob. Sys: SLA 36 NATO UNCLASSIFIED

SEPTR 37 NATO UNCLASSIFIED

SEPTR 37 NATO UNCLASSIFIED

SEPTR data in NRT on the web High bandwidth Ship-NURC satellite link NUR C

SEPTR data in NRT on the web High bandwidth Ship-NURC satellite link NUR C GEOS II Mirror GEOS II Time based scheduled synchronizations 38 NATO UNCLASSIFIED

Common box 39 NATO UNCLASSIFIED

Common box 39 NATO UNCLASSIFIED

Data and models: sound velocity 40 NATO UNCLASSIFIED

Data and models: sound velocity 40 NATO UNCLASSIFIED

Multi-scale super-ensemble (MSSE) Optimal combination of processes instead of models SEPTR TEMP Courtesy Paul

Multi-scale super-ensemble (MSSE) Optimal combination of processes instead of models SEPTR TEMP Courtesy Paul Martin (NRLSSC) NCOM TEMP ROMS TEMP 41 Courtesy Jacopo Chiggiato (ARPA) S-transform, multiple filter, wavelet Errors on sound velocity profile ‘Standard’ Super-ensemble (SE) Multi-scale Super-ensemble (MSSE) 4 -5 m/s 1 -2 m/s NATO UNCLASSIFIED

S-TRANSFORM (SVP, 20 m depth) ADRICOSM HOPS SEPTR NCOM 42 ROMS NATO UNCLASSIFIED

S-TRANSFORM (SVP, 20 m depth) ADRICOSM HOPS SEPTR NCOM 42 ROMS NATO UNCLASSIFIED

Sound velocity at 20 m SE MSSE 43 NATO UNCLASSIFIED

Sound velocity at 20 m SE MSSE 43 NATO UNCLASSIFIED

Hindcast skills: SE vs MSSE Co rre lat STD ion SE Skill 0. 1

Hindcast skills: SE vs MSSE Co rre lat STD ion SE Skill 0. 1 Skill 0. 9 MSSE SEPTR OBS. 44 NATO UNCLASSIFIED

Forecast skills: SE vs MSSE SE MSSE Skill 0. 1 Skill 0. 9 SEPTR

Forecast skills: SE vs MSSE SE MSSE Skill 0. 1 Skill 0. 9 SEPTR OBS. 45 NATO UNCLASSIFIED

Forecast: error on sound velocity SE MSSE 46 NATO UNCLASSIFIED

Forecast: error on sound velocity SE MSSE 46 NATO UNCLASSIFIED

Forecast: dynamic SE = KF+DLM Indiv models KF+uncertainty Forecast Sound velocity anomaly (m/s) 47

Forecast: dynamic SE = KF+DLM Indiv models KF+uncertainty Forecast Sound velocity anomaly (m/s) 47 NATO UNCLASSIFIED

Forecast: error on sound velocity ENSMEAN SE 48 UNBIASED ENSMEAN Kalman filter DLM+error evolution

Forecast: error on sound velocity ENSMEAN SE 48 UNBIASED ENSMEAN Kalman filter DLM+error evolution NATO UNCLASSIFIED

A priori forecast uncertainties ENSMEAN UNBIASED ENSMEAN Kalman filter DLM+error evolution 49 NATO UNCLASSIFIED

A priori forecast uncertainties ENSMEAN UNBIASED ENSMEAN Kalman filter DLM+error evolution 49 NATO UNCLASSIFIED

Forecast skill on sound velocity Whole period and water column UEM Best indiv. model

Forecast skill on sound velocity Whole period and water column UEM Best indiv. model EM SE KF 50 NATO UNCLASSIFIED

Conclusions • SE = paradigm for improved reliability and accuracy • NATO framework: cheap

Conclusions • SE = paradigm for improved reliability and accuracy • NATO framework: cheap (i. e. marginal cost) because model forecasts are available • “Relocatable science”: [ocean, atmosphere, wave, surf], [shallow, deep], [in-situ, remote], [linear, non-linear] 51 • Information fusion per-se, Recognized environmental picture • Uncertainty as a direct by-product (e. g. std of models) • Interoperability, network enabled capability • Information and decision superiority NATO UNCLASSIFIED

At the risk of repeating myself, WRT DART Thanks to NRL ! Thanks Jeff

At the risk of repeating myself, WRT DART Thanks to NRL ! Thanks Jeff ! Questions ? 52 NATO UNCLASSIFIED

53 NATO UNCLASSIFIED

53 NATO UNCLASSIFIED

54 NATO UNCLASSIFIED

54 NATO UNCLASSIFIED

Forecast errors Analysis COARSE NCOM SE FINE NCOM 55 Operational Models - no CTD

Forecast errors Analysis COARSE NCOM SE FINE NCOM 55 Operational Models - no CTD data ass. - two grids (coarse, fine) FINE NCOM SE COARSE+FINE NCOM NATO UNCLASSIFIED

Forecast errors Operational Models Single HOPS Model Runs - with CTD data ass. -

Forecast errors Operational Models Single HOPS Model Runs - with CTD data ass. - two training options Data Ass. SE I (using 2 models) Overall SE II (using 4 models) +2 NCOM models 56 NATO UNCLASSIFIED