Using Information Flow Networks for Characterizing Process and Model Dynamics (Invited Presentation)

Studies of natural systems are increasingly relying on direct observations of co-varying variables, often characterized as natural laboratory studies. This is in contrast to the approaches based on experimental laboratory studies, which rely on carefully designed experiments where the scientist constrains certain degrees of freedom while observing the others that remain unconstrained. Although, this approach has led to fundamental advances in our understanding of myriad processes, phenomena in nature are not subject to such constrained environments, where different variables interact with each other in a manner that may be best described as self-organized. This results in emergent characteristics, a term that indicates that the overall behavior could not have been predicted from the knowledge of the characteristics of the individual variables alone. This suggests that the inter-dependencies between variables create a whole that is greater than the union of the parts, which is interesting and relevant in its own right. Therefore, through the simultaneous measurement of several variables, it should be possible to unravel dependencies between these variables that are otherwise not possible. We show that by capturing the dynamics of propagation of fluctuations through the inter-dependent variables using information flow, it is possible to develop deeper understanding of such inter-dependencies. The goal of this talk is to expose methods to characterize information flow between variables interacting in a complex earth system framework, and develop an understanding of emergent dynamics. Since the goal of earth system modeling, a frontier challenge, is to capture these interdependencies to improve prediction of both phenomena and trajectories, we discuss how information flow based understanding can be used to assess and improve the isomorphism between model and natural systems.