Prediction for decision-making under uncertainty

A primary use for mathematical models, in fields such as environmental management, economics and engineering, is prediction. Prediction can aid choice between decisions by assessing their consequences, or between models by comparing their prediction performance. Choice between models raises fewer questions of how to use predictions, and this paper concentrates instead on prediction for decision-making. It starts with an illustration of how systematic but easily overlooked modelling error can spoil prediction. Three radically differing approaches to supporting decision-making with imprecise predictions are then reviewed: Bayesian optimal decision theory, model predictive control and set-membership prediction. Their potential and limitations as aids to environmental decision-making are discussed.