Berlin 2014 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
UP: Fachverband Umweltphysik
UP 7: Methoden - Messverfahren und Datenauswertung
UP 7.2: Talk
Wednesday, March 19, 2014, 11:45–12:00, MAG 100
Deciphering transitions in climate time series using Bayesian inference — •Nadine Berner1, Martin H. Trauth2, and Matthias Holschneider1 — 1Focus Area for the Dynamics of Complex Systems (DYCOS), Universität Potsdam — 2Institute of Earth and Environmental Science, Universität Potsdam
The estimation of transition events in environmental time series challenges analysis methods and modeling concepts. Commonly such changes are considered as isolated singularities in a more regular background indicating the transition between two regimes governed by different internal dynamics or external forcings.
Our aim is to derive a probabilistic expression quantifying multiple transition events in a data set on different temporal scales. Therefore we model a change in terms of the underlying regular dynamics and the evolution of the scedasticity of the data. We combine these two aspects in a linear mixed model which speeds up computations considerably. Furthermore we accomplish the estimation of the model's parameters via Bayesian inference and obtain associated uncertainty levels in a natural way. By applying a kernel based approach, we formally localize the resulting joint probability density of a transition. Thus we are able to investigate highly complex signals explicitly for trend changes in the statistical properties, without enlarging the dimensionality of the assumed underlying model.
We discuss our method's performance on well studied hydrological time series and present preliminary results of long scale, highly complex palaeo-signals from Northern Africa.