Flood inundation modeling under stochastic uncertainty

• Main Author: Sun, He.
• Other Authors: School of Civil and Environmental Engineering
• Format: Final Year Project
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Civil engineering::Water resources
• Online Access: http://hdl.handle.net/10356/53799
• Institution: Nanyang Technological University

Since the beginning of human history, flooding has been one of the most destructive and recurring natural disasters, which brings a serious threat to a broad range of population and properties all over the world. Therefore, flood inundation modeling plays an essential role in both flood forecasting and damage estimation due to its capability to generate simulated flood extent and flood depth maps for flood risk management. However, due to the human errors of data measurement and parameter estimation, along with the imperfection of model system structure, uncertainty analysis of flood inundation modeling is important to fully understand the limitations of inundation model prediction and flood risk estimation. This study will adopt Generalized Likelihood Uncertainty Estimation (GLUE) to evaluate the propagation of uncertainty associated with the most sensitive input parameters (i.e. channel and floodplain roughness coefficients) in a LISFLOOD-FP model to assess its reliability. The study case is a flooding event occurred at a short reach nearby Buscot Weir at River Thames in England in 1992. Based on the GLUE procedure, there were 1,000 input parameter samples generated randomly by simple random sampling method within the entire parameter space under consideration. After model simulation performed, the model output results obtained were compared with the observed data to determine the degree of correspondence by using the objective function of likelihood measure. Hereafter, 104 samplings were acceptable for result analysis as their model performance values were not less than 73.8%. From the result analysis, it is found that the roughness coefficient of channel was the most influential parameter, which was highly sensitive to the LISFLOOD-FP model whereas the floodplain had no significant effect. Moreover, it was indicated that the generalized extreme value distribution was the best stochastic distribution for the channel roughness coefficient. On the other hand, the predicted water depths at the 10 river observation points, with an equal spacing along the channel longitudinal distance, satisfied the generalized extreme value distribution for Cumulative Distribution Function (CDF). In addition, the tendency of predicted water depths in terms of 5%, 50% and 95% CDF percentiles along the channel would decline significantly from upstream to downstream. This was caused by the topography of study area where the floodplain areas at upstream were less than the areas at downstream. Furthermore, the flood risk map was generated to illustrate the flooding probability of each pixel area, in order to aid decision-makers to plan and implement a flood management strategy confidently. In future, research study can be conducted to focus on developing novel coupled stochastic and fuzzy methodologies to evaluate the propagation of uncertainty.