![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
A data-driven prediction method for fast-slow systems |
VerfasserIn |
Andreas Groth, Mickael Chekroun, Dmitri Kondrashov, Michael Ghil |
Konferenz |
EGU General Assembly 2016
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 18 (2016) |
Datensatznummer |
250132451
|
Publikation (Nr.) |
EGU/EGU2016-12960.pdf |
|
|
|
Zusammenfassung |
In this work, we present a prediction method for processes that exhibit a mixture of variability
on low and fast scales. The method relies on combining empirical model reduction (EMR)
with singular spectrum analysis (SSA). EMR is a data-driven methodology for constructing
stochastic low-dimensional models that account for nonlinearity and serial correlation in the
estimated noise, while SSA provides a decomposition of the complex dynamics
into low-order components that capture spatio-temporal behavior on different time
scales.
Our study focuses on the data-driven modeling of partial observations from dynamical
systems that exhibit power spectra with broad peaks. The main result in this talk is that the
combination of SSA pre-filtering with EMR modeling improves, under certain
circumstances, the modeling and prediction skill of such a system, as compared
to a standard EMR prediction based on raw data. Specifically, it is the separation
into “fast” and “slow” temporal scales by the SSA pre-filtering that achieves the
improvement.
We show, in particular that the resulting EMR-SSA emulators help predict intermittent
behavior such as rapid transitions between specific regions of the system’s phase space. This
capability of the EMR-SSA prediction will be demonstrated on two low-dimensional models:
the Rössler system and a Lotka-Volterra model for interspecies competition. In either case,
the chaotic dynamics is produced through a Shilnikov-type mechanism and we argue that the
latter seems to be an important ingredient for the good prediction skills of EMR-SSA
emulators. Shilnikov-type behavior has been shown to arise in various complex geophysical
fluid models, such as baroclinic quasi-geostrophic flows in the mid-latitude atmosphere
and wind-driven double-gyre ocean circulation models. This pervasiveness of the
Shilnikow mechanism of fast-slow transition opens interesting perspectives for
the extension of the proposed EMR-SSA approach to more realistic situations. |
|
|
|
|
|