A novel approach to model error modelling using the expectation-maximization algorithm

In this paper we develop a novel approach to model error modelling. There are natural links to others recently developed ideas. However, here we make several key departures, namely (i) we focus on relative errors; (ii) we use a broad class of model error description which includes, inter alia, the earlier idea of stochastic embedding; (iii) we estimate both, the nominal model and undermodelling simultaneously using the Expectation-Maximization (EM) algorithm. Simulation studies illustrate the performance of the proposed technique.