Understanding the relationship between computer-based models and the environment they simulate is becoming increasingly important as we try to predict how the earth's climate will change. As a surrogate for the representation of uncertainty in a prediction problem, it is common to use the range of behaviour from a set of models (an ensemble), and the ensemble mean as the 'best guess' prediction. We suggest a 'model space' metric, which, by providing one relevant definition of model independence, could allow us to begin to understand the relationship between model spread and prediction uncertainty. This in turn could allow the minimisation of bias from the inclusion of similar models in ensembles and quantification of how much independent information each model contributes to the prediction problem.
ASJC Scopus subject areas
- Earth and Planetary Sciences(all)