Machine reading the science of climate change: computational tools to support evidence-based decision-making in the age of big literature

The amount of scientific literature on climate change has reached unmanageable proportions. This poses problems for researchers, especially those attempting to synthesise literature in the field. It is an even larger problem for the Intergovernmental Panel on Climate Change, whose task it is to comprehensively assess the scientific literature on climate change. This thesis explores how approaches from Natural Language Processing can be used to assist evidence synthesis, and understand and inform global environmental assessments. It uses computer assistance to ask what literature is relevant, and what it is about. First, it develops a methodology for machine learning assisted screening for systematic reviews. Second, it produces a map of the thematic content of the entire climate change literature. Finally, it uses machine learning to identify and classify tens of thousands of papers on climate impacts, and match these with model evidence on the attribution of climate trends to anthropogenic forcing.