Chaotic model with data assimilation using NARX network

This paper introduces a novel data assimilation technique where Nonlinear AutoRegressive with eXogenous inputs (NARX) model is used to re-analyze and improve chaotic model forecasts. The chaotic model is built using adaptive local models based on the dynamical neighbors in the reconstructed phase space of the observed time series data. The proposed method was implemented to build the storm surge model for the North Sea. The results demonstrated that the chaotic model with data assimilation has a significant increase of forecasting accuracy compared to standard chaotic model without data assimilation, a standard ANN model and the European operational storm surge numerical models.