BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Machine learning seminar series - Exploring Machine Learning for D
ata Assimilation
DTSTART;VALUE=DATE-TIME:20200507T100000Z
DTEND;VALUE=DATE-TIME:20200507T153000Z
DTSTAMP;VALUE=DATE-TIME:20200708T113538Z
UID:indico-event-197@events.ecmwf.int
DESCRIPTION:\n\n\nHost\n\nMassimo Bonavita\n\nSpeaker\n\nAlban Farchi\n\nA
lban Farchi is a recently hired permanent researcher at CEREA. He works in
the field of data assimilation for the geosciences with application to at
mospheric chemistry. Currently\, he is a visitor of the ECMWF where he wor
ks on machine learning applications to numerical weather forecasts. \n\nA
bstract\n\nRecent developments in machine learning (ML) have demonstrated
impressive skills in reproducing complex spatiotemporal processes by effic
iently using a huge amount of data. ML methods rely on flexible and parall
elisable tools to enable optimisation in high dimension. However\, contrar
y to data assimilation (DA)\, the underlying assumption behind ML methods
is that the system is fully observed and without noise\, which is rarely t
he case in numerical weather prediction. In order to circumvent this issue
\, it is possible to embed the ML problem into a DA formalism characterise
d by a cost function similar to that of the weak-constraint 4D-Var (Bocque
t et al.\, 2019\; Bocquet et al.\, 2020). In practice ML and DA are combin
ed to solve the problem: DA is used to estimate the state of the system wh
ile ML is used to estimate the full model. This approach has been implemen
ted and successfully tested with low-order one-dimensional models. Using a
sufficiently long trajectory of the model\, they are able to reconstruct
the model dynamics.\n\nIn realistic systems\, the model dynamics can be ve
ry complex and it may not be possible to reconstruct it from scratch. An a
lternative could be to learn the model error of an already existent model
using the same approach combining DA and ML. The feasibility of the method
is first tested using the QG model developed in OOPS. In this presentatio
n\, we briefly describe the QG model and the kind of model error that will
be learnt. We then show the results of preliminary ML experiments\, and w
e present what will be the next steps of the study. \n\nBocquet\, M.\, Br
ajard\, J.\, Carrassi\, A.\, and Bertino\, L.: Data assimilation as a lear
ning tool to infer ordinary differential equation representations of dynam
ical models\, Nonlin. Processes Geophys.\, 26\, 143–162\, 2019\n\nBocque
t\, M.\, Brajard\, J.\, Carrassi\, A.\, and Bertino\, L.: Bayesian inferen
ce of chaotic dynamics by merging data assimilation\, machine learning and
expectation-maximization\, Foundations of Data Science\, 2 (1)\, 55-80\,
2020\n\n\nhttps://events.ecmwf.int/event/197/
LOCATION:11:00 BST
URL:https://events.ecmwf.int/event/197/
END:VEVENT
END:VCALENDAR