In this WP we develop the general theory of a supermodel apart from the learning aspect. We will specify how models with different structure should be connected and specify conditions under which connecting variables among different models will lead to superior skill, and the form of those connections. We will determine limitations of the supermodeling strategy. In addition to machine learning we develop a strategy how to define connections based on insight, mathematical arguments or whatever that leads to a useful supermodel. We do this because the automatic learning might be too complex, or lead to suboptimal solutions and we spread the risk this way. This WP will serve as input for WPs 2-5.
WP1 is chaired by Jürgen Kurths