Reliability of Inference of Directed Climate Networks Using Conditional Mutual Information

Across geosciences, many investigated phenomena relate to specific complex 1 systems consisting of intricately intertwined interacting subsystems. Such dynamical com2 plex systems can be represented by a directed graph, where each link denotes an existence 3 of a causal relation, or information exchange between the nodes. For geophysical systems 4 such as global climate, these relations are commonly not known theoretically but estimated 5 from recorded data using causality analysis methods. These include bivariate nonlinear 6 methods based on information theory and their linear counterpart. A trade-off between 7 the valuable sensitivity of nonlinear methods to more general interactions and potentially 8 higher numerical reliability of linear method may affect inference regarding structure 9 and variability of climate networks. We investigate the reliability of directed climate 10 networks detected by selected methods and parameter settings, using stationarized model 11 of dimensionality-reduced surface air temperature data from reanalysis of 60-year global 12 climate records. Overall, all studied bivariate causality methods provided reproducible 13 estimates of climate causality networks; with linear approximation showing higher reliability 14 than the investigated nonlinear methods. On the example dataset, optimizing the investigated 15 nonlinear methods with respect to reliability increased similarity of the detected networks 16 to their linear counterparts, supporting the particular hypothesis of surface air temperature 17 climate reanalysis data near-linearity. 18 Version January 29, 2013 submitted to Entropy 2 of 20