|
Abstract:
|
Decadal prediction refers to predictions on multi-year time scales of temperature, precipitation, hurricane activity, droughts, etc. In this talk, I discuss the scientific basis of decadal prediction and some of the statistical challenges that decadal predictions raise. The statistical challenges are framed in a probabilistic framework where information theory provides a comprehensive measure of predictability. Data can be decomposed into a set components ordered by their contribution to predictability, which provides the basis for identifying the most predictable components of a system. Applying this decomposition to climate model simulations reveals the most predictable components on multi-year time scales. Although these components have projections on well known components of variability like the Pacific Decadal Oscillation and the Atlantic Multidecadal Oscillation, there are significant differences. Nevertheless, regression models derived from climate models give skillful predictions of observed anomalies in these components for a year or more, suggesting that the dynamics and spatial characteristics of decadal predictability inferred from climate models are realistic.
|
