Fig. 1.

Bayesian Sequential data assimilation. We propose a Bayesian filtering method that predicts along the dynamical system (1). The model is fitted with data in the training period and this is used to make predictions during the reporting delay period (nowcasting) and a forecasting period. The training window is updated and moved $n$ days forward, to update all forecasts and the former posterior becomes the prior, in the next window. Further details are described in Algorithm 1.