Empirical model reduction in the study of climate sensibility and variability

Predicting climate events means having to deal with a complex, nonlinear, chaotic system, with natural variability in time and space and subject to external forcing. One of the best ways to approach this problem is through the definition of general circulation models, which represent climate systems as a group of subsystems that interact with each other through phenomena. However, they are not always simple to interpret, and sometimes it is necessary to have easy models to read but precise enough to be able to hold all the relevant information. Empirical model reduction (EMR) is a well-established methodology able to build an efficient model from simple observations of the system. This thesis aims to study the potential of EMR if used to simulate the dynamics of a set of real climate data. After an overview of the basic concepts of climate systems, and an examination of the EMR algorithm, some experiments have been conducted giving the system different properties. The goal is to understand how the algorithm and the data influence the resulting model under variation of some properties of the method, or changing the scenario described by the data. The methodology itself was also compared with alternative versions with more classic data-driven techniques (like linear regression) integrated into the algorithm. The results obtained help to better understand the conditions under which the empirical model reduction method might be able to substitute a general circulation model with the least possible loss of information.