Soutenance de Dario Lucente
| When |
Nov 09, 2021
from 02:00 to 04:00 |
|---|---|
| Where | Salle des thèses |
| Contact Name | Dario Lucente |
| Attendees |
Dario Lucente |
| Add event to calendar |
vCal
iCal
|
There is a large interest in predicting the occurrence of high impact climate events such as ENSO (El Niño Southern Oscillation) or rare events, for instance heat waves. Those are prediction problems at the predictability margin because the interesting time scale lies at the edge of the mixing time of the system. This thesis aims at introducing the relevant quantity for these prediction problems, the so-called committor function which is the probability for an event to occur in the future, as a function of the current state of the system. Computing the committor in a stochastic model for ENSO illustrates that the transition to strong El Niño regimes can have either intrinsic probabilistic predictability or unpredictability. The second goal is to illustrate how to compute and validate the committor function from observations, by discussing the analogue Markov chain which provides a way for learning effective dynamics from data. Starting from it, a new algorithm is developed, with the scope of computing the committor function more precisely than the other approaches, especially in case of lack of data. Moreover, it is shown, in the context of two stochastic systems, that coupling the learning of the com- mittor with a rare event algorithm improves the performance of the latter. Finally, this methodology is applied to a climate data-set, generated from a climate model, in order to study and predict the occurrence of extreme heat waves. After checking the consistency of the statistical quantities computed by the effective dynamics, a classifier based on the Markov chain is developed, with the capability of classifying heat waves better than other methods.