Abstract

Climate scientists face many challenging problems, in relation to the climate both of the past (observations are often sparse, incomplete and imperfect) and of the future (different climate models give different projections and, perhaps, are considered to have different reliabilities). At a fundamental level, many of these challenges are statistical in nature: statistics being the science of (a) interpreting information (b) determining what information is needed to answer a particular question of interest (or, conversely, to determine what questions can be answered with a given information set). In this presentation I will try to give a modern statistical perspective on several of these issues, together with some modern statistical solutions to the associated problems. Some key themes of the presentation will be: the need for a model-based approach to statistical analysis; accounting for missing or incomplete data; and the combination of information from multiple sources (such as different GCMs) in such a way as to make defensible statements about the real climate system.