Identification and Statistics of Extreme Events (Present, Future and Trends) Using Markov Chains and Climate Networks
Standard statistical methods in climatology have certain weaknesses in detecting and analyzing present and future climate extremes. The goal of MarNet is therefore to reduce these deficiencies and complement conventional approaches by applying new and innovative methods, which are not commonly used in the context of extreme events, namely Climate Networks and Markov Chains. A Climate Network consists of nodes which are connected by edges. Nodes can either be observation sites or grid points of a climate model, and connections between such nodes are established if time series of meteorological data are there correlated. In this way, statistical similarities between nodes can be revealed, by which physical interrelations in the climate system can be analyzed. The method is therefore particularly suited to identify spatial (hot spots; changing locations) and temporal patterns in present and future extreme events. A Markov Chain is a time series of different states of a system (e.g. extreme or not-extreme) in which the probability of a system to change its state is incorporated in a transition matrix. By using different descriptors of a Markov Chain, certain characteristics of a system can be determined, such as the tendency of a system to stay in a given state, or to return to this state, and the predictability of a state transition. In this way, Markov Chains can be used to analyze the frequency, duration and regularity of present extreme events and describe their future change.
Institution: Karlsruhe Institute of Technology
Contact: Gerd Schäder, Markus Breil