Past climate studies with optimized networks using artificial intelligence

Abstract

The availability of high-quality climate records decreases backwards in time, and the associated increase in uncertainty supports the use of complementary sources of climate information (such as model simulations) to understand the underlying physics of the climate system, as well as its past and futurechanges. In this Ph.D. thesis we assess the potential of Artificial Intelligence as an additional efficient tool to solve complex problems in the field of climate sciences. We show that these techniques can optimize the informationcoming from different sets of climate networks such as meteorological stations, historical records, and paleoclimate archives. Being employed to address a plethora of questions, they share issues in terms of incompleteness. Within this framework, we address different problems that are common in the climate community by developing tailored methodologies with the same goal of maximizing the extraction of information from incomplete climate datasets. The developed approaches include metaheuristic algorithms and cluster analyses and will be applied to incomplete datasets that are typically employed for paleo-climate reconstructions and regional climate assessments, respectively...