 |
| Titel |
The Performance of the Supermodeling Approach In the Real Imperfect Model Scenario |
| VerfasserIn |
Carsten Grabow, Thomas Stemler, Juergen Kurths |
| Konferenz |
EGU General Assembly 2014
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 16 (2014) |
| Datensatznummer |
250100745
|
| Publikation (Nr.) |
 EGU/EGU2014-16738.pdf |
|
|
|
|
|
| Zusammenfassung |
| The novelty of the supermodeling (SUMO) approach is that it is an interconnected ensemble
of existing imperfect models of a real, observable system. The connections between the
models can be learned from observational data using methods from machine learning. The
supermodel outperforms the individual models in simulating the behaviour of the real system
since it has learned to combine the strengths of the individual models. The concept of
supermodeling is based on a new combination of insights from climate science, nonlinear
dynamical systems, and machine learning.
This SUMO framework provides a possible new approach to forecasting in the imperfect
model scenario (IPMS). Numerical investigations found that coupling several imperfect
models resulted in better forecasting skill than that of the uncoupled models or their averaged
outputs. In the initial SUMO studies - essentially showing a proof of concept - a very limited
type of model error had been investigated. Here we present our recent studies which include
an extension of the types of model error that were previously investigated within the SUMO
framework. In addition, we introduce a new measure based on the separation time of the
different models aimed at validating the super models. All of this will shed new light
and yield further understanding of the SUMO setup and assist in quantifying its
performance. |
|
|
|
|
|
|