Complex Network and Dynamical System Approaches to Climate Science
MetadataShow full item record
The exponential growth of climate data combined with advances in machine learning offers new opportunities to understand the climate system and its response to external forcings. This thesis explores and proposes data mining frameworks to reduce the complexity of spatiotemporal climate fields and facilitate analysis and interpretation. As complex as it appears, the dynamics of the climate system is dominated by spatiotemporal patterns and the identification of these patterns and their linkages offers a useful framework for dimensionality reduction. In the first part of this work, I leverage this observation and propose a framework for model evaluation. The approach allows to compare modes of climate variability and their interrelationships across datasets. I apply the proposed methodology on observational sea surface temperature (SST) datasets and 30 members of the Community Earth System Model Large Ensemble (CESM-LE) that differ only slightly in the initial conditions. I then compare the modelled and observed climate networks, identify biases, and distinguish between models errors and differences arising from the internal variability of the system. In the second part of the thesis, I present a strategy for dimensionality reduction in paleoclimate simulations. Given two simulations of the last 6000 years, the search for abrupt or non-abrupt shifts in dynamics at spatiotemporal scales is undertaken. I show that, at least in the modelled climate, multidecadal oscillations can rapidly emerge and fade due to the system’s internal dynamics over periods as short as 200–300 years. Moreover, we argue that changes in the global connectivity patterns are, on average, abrupt and chaotic. Finally, we focus solely on the tropical Indian and Pacific oceans, were a slow and gradual change in modes of variability and their linkages are identified. During the mid-Holocene, the Indian Ocean (IO) basin hosted an energetic equatorial dipole mode, largely independent of the El Niño Southern Oscillation (ENSO) and different from the IO dipole observed today. Also, mid-Holocene ENSO was much weaker. I quantify the causal relationship between such changes and the evolution of the whole Indo-Pacific climate mean state. Changes in the Earth’s orbital configuration, slow and small compared to the ongoing changes in anthropogenic climate drivers, caused large changes in major modes of variability in the Indo-Pacific region.