r. fablet - imt · pdf fileinstitut mines-télécom the upper ocean dynamics at...

Institut Mines-Télécom

The upper ocean dynamics at

high-resolution: applications

and challenges

R. Fablet Telecom Bretagne, département SC

UMR LabSTICC/TOMS

http://perso.telecom-bretagne.eu/ronanfablet


A variety of sensors and modalities

IMT,

November

2014 2

wide-swath sensors,

eg temperature

SAR sensors,

eg wind field

Satellite ocean sensing: context

narrow-swath sensors, eg altimeters


From low-resolution ……………. to high-resolution

Québec,

October 2013 3

Satellite ocean sensing: context

Low-resolution: ~25km (0.25°) eg, Altimetry (SSH), Radiometer (SST)

High-resolution: 1-4km (0.01°) or < 1km eg, SAR (currents), Infrared (SST, Ocean Colour)

Key issue How to deliver daily HR geophysical field anywhere and anywhen from

the irregular space-time sampling of satellite sensors ?

SAR

SST

SST

SSH

Chl-A

VIGISAT

Institut Mines-Télécom IMT,

November

2014 4

High-resolution ocean sensing: what for?


Eg, high-resolution monitoring of deepwater horizon spill

IMT,

November

2014 5


Collard et al., 2010


Eg, understanding physics-to-ecology interactions/transfer

IMT,

November

2014 6


Short trips

Long trips

Europa island

Top marine predators track

Lagrangian coherent structures,

Tew Kai et al., PNAS 2010.

GPS seabird tracks

Ocean dynamics

Need for

« coherent » spatial

observation scales


Eg, understanding physics-to-ecology interactions/transfer

IMT,

November

2014 7


Reconstructing and understanding seabass migration patterns from DSTs, Pontual et

al., ICES Symp. 2013.

Fish data DST data (T & D)

data Temp. and bathymetry

Satellite data

MARS3D data

Signal processing

(tracking model, HMM)


Eg, impacts of fine-scale physical processes onto ecosystem

dynamics (Bertrand et al., Nature Com. 2014)

IMT,

November

2014 8


Ocean turbulence creates « small » pelagic oasis for marine life

(plankton, pelagic fish, seabirds, ….).

Institut Mines-Télécom 9

The upper ocean dynamics at high-

resolution: from an irregular space-time

sampling to anywhere and anywhen?

A signal processing perspective:

challenges and expected contributions

IMT, November 2014

Institut Mines-Télécom 10

The upper ocean dynamics at high-resolution

anywhere and anywhen?

Space-time satellite

sampling

Missing data issues with multi-

sensor/multiscale sources

The challenge: an irregular multimodal space-time sampling

IMT, November 2014



resolution anywhere and anywhen?

The state-of-the-art: model-driven data assimilation

11

Limitations:

• Complexity of the numerical resolution

• Model complexity vs. Model « genericity »

• Objective: combine a prior ocean model and available observations

• Operational models: eg, NEMOVAR (ECMWF), Wavewatch III

(NOAA)

IMT, November 2014




Towards data-driven assimilation models: why?

Québec, October 2013 12 Key objective: learning new multi-scale/multi-modal representations of ocean dynamics from multi-sensor remote sensing archives

HR observations are irregularly

sampled in space and time.

But ….

1) low resolution observations are

generally available

2) we may learn a lot from past

observations.

Archived database of observation and simulation data

eg, low-resolution observations

eg, high-resolution observations




Towards data-driven assimilation models: our strategy

Québec, October 2013 13

Database of past multimodal observations and/or simulations

New observations Reconstruction of HR dynamics

Key objective: learning new multi-scale/multi-modal representations of ocean dynamics from multi-sensor remote sensing archives




Preliminary results (1):

Québec,

October 2013

Model-free/data-driven dynamics for the assimilation

of geophysical systems Tandeo et al, 2014 14

Model-free stochastic filters (e.g. EnKF, particle filter) Proof-of-concept validated on chaotic dynamics (Lorentz model)




Preliminary results (2): deterministic transfer function

Québec, October 2013 15

Learning ECMWF-to-SAR transfer functions

for HR wind field emulation He-Guelton et al, 2014

Institut Mines-Télécom Québec, October 2013

Learning stochastic LR-to-HR transfer functions

for HR SST emulation



Preliminary results (3): stochastic transfer function

16

Spectral density HR detail marginals Multifractal Spectrum

IHR− I LR

LR (AMSR) SST Simulated SST HR (METOP) SST

Fablet et al, 2013




Preliminary results (4): multimodal transfer function

Québec,

October 2013

Towards high-resolution sea surface currents

from a joint SST-SSH analysis

SQG-like assumption: local relationships between local SST patches and sea surface currents Learning from joint SST-SSH observations

Application to a one-year AMSR/SST-AVISO/SSH series

Tandeo et al, 2013 17




Preliminary results (4): big data challenge

Québec,

October 2013

Nephelae demonstrator: dedicated architecture to process terabytes-to-petabytes of data http://cersat.ifremer.fr/About-us/Infrastructure/Nephelae

Ex.: High-resolution Sea Surface Temperature 200Go daily to be processed to fill cloud gaps Processing time : • Months with traditional system • 8 hours with Nephelae architecture

Towards high-performance architecture for ocean-big-data

management and processing Piolle et al, 2013

18




EMOCEAN project (2013-2016 ANR grant (French NSF))

19

Partners: Telecom Bretagne (R. Fablet), Ifremer (B. Chapron),

Ocean Data Lab (spin-off, F. Collard), OUC (G. Chen)

Main tasks

• Learning multi-modal data-driven

representations of ocean surface

dynamics

• Stochastic reconstruction of HR ocean

surface geophysical fields from partial

observation series

• High-performance architecture for "real-

time" implementation

Case-studies

HR composite fields

Short-term dispersal scenarii

IMT, November 2014

Institut Mines-Télécom Merci pour votre attention

Acknowledgements

Joint work with B. Boussidi, P. Tandeo, R. Garello

(Telecom Bretagne), E. Autret, B. Chapron, L. He-Guelton,

H. de Pontual (Ifremer), F. Collard (Ocean Data Lab), A.

Bertrand (IRD)

r. fablet - imt · pdf fileinstitut mines-télécom the upper ocean dynamics at...

Documents