Flood Simulation and Loss Assessment Using a 2D Model and Neural Networks
| Autor: | Chih-Yun Yu, 游知紜 |
|---|---|
| Rok vydání: | 2019 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 Floods induced by torrential typhoon rains pose a potential threat to lives and property in various places of Taiwan. Therefore, analyzing the flood peril and associated risk is a key to disaster prevention and mitigation. This study adopts Kaohsiung, a flood-prone area in southern Taiwan, as the region of interest. According to the statistics released by the Council of Agriculture, typhoons caused ~ $7 billion NTD of total losses for the sectors of agriculture, forestry, fishery, and husbandry in Kaohsiung from 2007 to 2015. This study aims at applying a modern hydrodynamic model to simulating flood depths for typhoon events, and then incorporating the simulated depths in conjunction with selective hazard variables in predicting flood losses. This study is divided into two parts: First, HEC-RAS, a reputable hydrodynamic model developed by the United States Army Corps of Engineers is adopted; the recently developed two-dimensional module is configured to simulate flood extent and depths. Terrains over the main channel are corrected to enhance model stability, and land use data is used to calibrate associated model parameters. Model validation is performed by using the 24-hr, 600-mm flood potential map and the survey-based flood extent for Typhoon Fanapi in 2010. Afterwards, analysis of the flood extent and depth over the agriculture, forestry, fishery, and husbandry lands is conducted; the average flood depth is found to be highest (lowest) over the fishery (forestry) land. The second part of this study is to establish a neural network (NN) model for the prediction of flood losses for each individual sector. Input variables to the NN model are the simulated flood depth and other meteorological variables, such as rainfall and wind data. The NN-based loss prediction model is well calibrated and validated, showing the high Nash-Sutcliffe efficiency (NSE) value for predicting agriculture losses (＞ 0.8) but low NSE values for other types of losses (＜ 0.1), which could be attributed to the insufficient number of loss events. From the different combinations of hazard variables, the simulated depth and maximum sustained wind are identified as the most influential variables; the two variables can be referred to as important indices for real-time loss estimate and follow-up risk analysis. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|