Multi Model Ensemble (MME) predictions are a popular ad-hoc technique for improving estimates of high-dimensional, multi-scale dynamical systems, including Global Circulation Models (GCM’s) used in climate change predictions. The heuristic idea behind the MME framework is simple: given a collection of models, one considers predictions obtained through a convex superposition of the individual forecasts in the hope of mitigating modeling errors. However, it is not obvious if this is a viable strategy and which models should be included in the MME forecast in order to achieve the best predictive performance. I will briefly describe how an information-theoretic approach to this problem allows for deriving systematic criteria for improving dynamical predictions within the MME framework, and how such a framework can aid data assimilation techniques which are based on multi model ensembles.

About the speaker

Dr Michal Branicki works on the interface of probability theory and stochastic dynamical systems with applications to quantifying uncertainty in prediction problems arising in data science. He is particularly interested in mathematical aspects of information theory, Bayesian data assimilation and machine learning, and techniques for systematic simplification of complex systems using empirical data. Michal is also a Faculty Fellow at the Alan Turing Institute which is a national centre for Data Science. In his teaching Michal specialises in Dynamical Systems and Probability theory.