Development of novel systems-analysis methodologies for supporting flood forecasting and uncertainty assessment

Abstract

There has been an increasing awareness that flood risk management is of particular importance in reducing flood risks and preventing flood-induced disasters. Accurate and reliable flood forecasting is essential for best practices in such a framework. Therefore, this study aims to develop various systems-analysis methodologies for supporting flood forecasting and uncertainty assessment. Firstly, the response surface models and artificial neural networks (ANN) were investigated in prediction of daily runoff and compared under various scenarios. A Bayesian-approach-based neural networks ensemble was then proposed for robust probabilistic hydrologic forecasting. Then, a generalized likelihood uncertainty estimation (GLUE) framework incorporating moving least squares in stochastic sampling was proposed for improving the efficiency of uncertainty assessment of flood inundation modeling. Two surrogate schemes coupling ANN into GLUE framework were proposed to investigate the best practices of applying surrogate approaches for solving practical problems. Finally, a joint Monte Carlo and fuzzy possibilistic simulation approach was proposed for assessing the flood damage under coupled possibilistic-probabilistic uncertainty.