ABSTRACT
This chapter introduces novel ensemble techniques, dynamic averaging and dynamic neural networks, and their use to build ensembles of chaotic storm surge models in high dimensional space.
10.1 Introduction Ensemble model predictions have been viewed for some decades as an effective way to improve the prediction performance over what the individual models can provide. It is often worthwhile to seek a combination of several prediction models rather than to select only the best one among them, which might be only marginally the best. Therefore, ensemble models now become the main topic in widespread use of model prediction in many fields, (e.g., in hydrometeorology and geosciences). Some recent researches on the use of prediction combination model for meteorological prediction and time series prediction have been conducted by Palmer et al. (2005), Wichard & Ogorzalek (2004) and Zhu (2005). Research efforts are now aimed at determining what kind of predictions benefit most from such combinations, and what combination techniques are optimal in a various situations.
