What do we mean by climate prediction? If we had a perfect model of the real-world, physical climate system, how would we use it to make predictions of future climate under assumptions about future human behaviour (most notably, future greenhouse gas emissions)? I will discuss this question and illustrate the issues using a deterministic, low-dimensional, nonlinear dynamical system which parallels the high-dimensional Earth System Models used in climate science, climate impacts studies and climate policy.

Having discussed the essence of climate prediction, I will go on to reflect on the relationship between our non-perfect climate models and the real-world climate system, arguing that this relationship fundamentally changes the nature of what we can expect to learn from models about future climate in reality. In conclusion, I will argue for a radically new focus for climate models and climate model experiments; one which, given the threat that climate change represents for our societies and cultures, should be somewhat driven by the information needs of business, politics, finance and economics.