Knowledge discovery with networks for climate science

Abstract

The use of complex networks has been motivated in climate to understand attributes of large-scale dynamics, for example, correlations between variables and relations among climate oscillators. This talk addresses two specific questions that arose from discussions at the first Climate Knowledge Discovery Workshop held in Hamburg earlier this year. The first part revolves around the ability to represent the salient features of the climate system via networks. We show that the structure of networks constructed from outputs of physically-based climate models generally exhibit less intra-model variability (among members of an initial condition ensemble of the same model) than inter-model variability (between different models run with the same forcing). Moreover, we illustrate the stability of these patterns over a long-term integration of a climate model run. The second part deals with the effect of model resolution on large-scale phenomena. Specifically, we examine some high-level features (dipole patterns) extracted from two runs of the same climate model run at different resolutions using graph-based methods. The findings from both of these investigations lend additional credence to this growing field of network-based representation and analysis of complex systems.