ABSTRACT
Star trek: Voyager This chapter presents a real-time data assimilation technique for storm surge predictive chaotic model using NARX neural network.
9.1 Introduction Complexity of natural dynamical systems and sensitivity of corresponding models to initial conditions are the two major reasons of model prediction errors. The accuracy on defining initial condition is very crucial since the small error of the initial condition results in loss prediction ability of the model. Even if we had a perfect predictive model, all predictions start to diverge from the truth after certain time. However, these prediction errors can be corrected or reanalyzed once new observations available using data assimilation techniques. Data assimilation allows for making the best estimate of the necessary updates to the current model states or outputs so that the model can provide more reliable and accurate predictions. A predictive model can be process-based or data-driven model which is seen to be composed of a set of equations (algorithms) that involve state variables and parameters. The model parameters typically remain constant while the state variables vary in time. Predictive model in real-time operation may take into consideration the new observation at the time of preparing for prediction. The feedback process of assimilating the new observation into the prediction procedure is so-called updating. The updating procedure can be classified into four different updating strategies based on the variables modified during the feedback process, as follows:
a) Updating input variables Input uncertainties are often being the dominant source in prediction error. This method is often based on iterative procedures and the updating of input variables results in changing state variables.
