A Complex Network Approach to Understand Climate Variability

Abstract

Climate variability has considerable impact on our society throughout recorded history. If we would like to make informed decisions about our own future, it is essential that we could identify, quantify, understand such climate variability, and then eventually predict it so as to minimize its negative consequences and maximize its positive ones. Complex network theory has been successfully applied to many complex systems. The idea of complex network also has been recently extended to climate studies which is referred to as Climate Network (CN). The novelty of complex network theory is that it maps out the topological features that are related to the physics of the dynamical systems. Therefore, CN is an innovative and powerful tool to investigate the patterns and the dynamics of climate variability. In this thesis, several specific phenomena of climate variability are studied by using the techniques from complex network theory. The first one is the reduction of the Atlantic Meridional Overturning Circulation (MOC). We develop an early warning indicator for a future collapse of the Atlantic MOC. We explore the performance of this indicator using data both from an idealized ocean model and a general circulation model that show such a collapse. We also determine optimal observation locations through quality measures of the indicator, and show that one needs multiple sections in the Atlantic to have a high quality indicator of the MOC collapse. The second phenomenon is the multidecadal variability associated with the Atlantic Multidecadal Oscillation (AMO). We investigate the existence of the westward propagation in the North Atlantic sea surface temperature (SST) observations. We reconstruct CNs by using a linear Pearson correlation measure and a nonlinear mutual information measure of spatial correlations between SST variations. Analysis of the topological properties of CNs shows that the nonlinear measure is better in capturing the main features of propagating patterns from the noisy SST signals than the linear ones, and that westward propagation of multidecadal SST anomalies indeed seems to occur in the North Atlantic. The third phenomenon is the interannual variations of El Niño-Southern Oscillation (ENSO). One of the crucial aspects that is currently limiting the success of El Niño predictions is the stability of the slowly varying Pacific climate. We present a stability index which does not only efficiently monitor the changes in spatial correlations in the Pacific climate, but also can be evaluated by using only SST data. We show the difference in Pacific climate stability between the 1982 and 1997 El Niño events, and argue that a strong El Niño event did not develop in 2014 because the Pacific climate was too stable. In the end, we use the measures of CNs in a machine-learning approach to develop new schemes for El Niño prediction. We present one scheme that can give a skillful prediction for the occurrence of El Niño events one year ahead, and another scheme provides reasonable NINO3.4 index forecasts with a lead time of three months.