This thesis concerns the long range prediction of high dimensional chaotic systems. To this end, I investigate the important relationship between predictability and non-uniformity of information loss throughout the state space of a chaotic system. I introduce a genetic algorithm to build predictive models by exploiting this nonuniformity. The algorithm searches for the regions of state space which remain most predictable for a given time into the future. I use the algorithm to investigate the predictability of both model chaotic systems and physical data from a fluid flow experiment. |