都市淹水模擬不確定性評估與機率式淹水地圖研析

Uncertainty Assessment of Urban Inundation Simulations and Analysis of Probabilistic Flooding Maps

許浩銘(Hao-Ming Hsu)

指導教授 ： 張倉榮

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/碩士(2013年)

https://doi.org/10.6342/NTU.2013.00883

摘要

洪災可能會造成生命財產的損失、衛生問題、交通中斷、壓縮經濟發展以及其他的直接與間接衝擊。為因應制定淹水減災策略，淹水模擬對決策者而言至關重要。一般而言，在執行淹水模擬過程中常使用單一最佳參數組來進行淹水模擬，此過程並未考量參數的不確定性，可歸類於定率式方法，而此不確定性將可能造成淹水模擬結果偏差，導致研擬洪水減災策略和設計防洪系統發生缺失。相反地，序率式的方法則考慮不確定性的影響，將所有可能的淹水模擬結果結合而成為一系集模擬。
本研究首先選擇4種不同建築物處理法(Friction-increasing, Building-hole, Elevation-lifting and Porous media treatments)和91組代表道路和空地的曼寧糙度係數值為不確定性的來源，組合成364組資料進行都市淹水模擬。最後，利用概似不確定性估計法(Generalized Likelihood Uncertainty Estimation, GLUE)，以所有組數的都市淹水模擬結果以及實測淹水範圍之間的對應關係來評估模擬結果並整合後，加值建立成為機率式淹水地圖。此外，本研究在應用GLUE時，改善傳統用於評估模擬結果的網格對應法(Grid-by-Grid Method, F3)，考慮空間分布影響因子，導入兩種新的演算法：核函數法(Kernel Method, Fk)和運輸演算法(Earth Mover’s Distance Method, Fe)，藉以提升GLUE法在處理空間分布不確定性的準確度。
研究結果顯示傳統的F3法係以單純的模擬淹水網格對應實測淹水網格方式，此F3法具有簡單、省時的優點，但其缺點在於缺乏考量空間資料的應用彈性，所以將會低估某些都市淹水模擬結果。另一方面，使用Fe法可通盤地考慮實測淹水範圍的形狀和淹水模擬結果兩者的空間相異關係，但是所需的計算時間過長，缺乏效率。最後，Fk法除了考量實測淹水範圍的形狀與淹水模擬結果的空間關係外，其計算效率亦較Fe法來得高。所以，本研究建議採用兼顧評比合理性和計算效率的Fk法。最後，本研究在考量4種的建築物處理法和91種曼寧糙度係數的不確定性，建立的364組的淹水模擬結果有著不同樣貌的淹水分布和淹水程度，相較於定率式的方法僅採用單一最佳參數組所得的都市淹水模擬結果，更具備對於淹水模擬結果之不確定性的通盤考量。應用GLUE能捕捉所有的淹水模擬結果，並整合成為機率式淹水地圖以輔助傳統定率式淹水地圖之更新。

關鍵字

不確定性 ； 都市淹水模擬 ； 概似不確定性估計法 ； 機率淹水地圖 ； 系集模擬

並列摘要

Flood inundation may cause human death, monetary loss, hygiene problems, interruption of transport systems, constraint on economic development, and other direct and indirect impacts. In order to design a plan for flood mitigation, the inundation simulation is essential to decision makers. In general, the deterministic approach is applied to analyze the potential of flood inundation, using the optimal parameter set solely to execute the inundation simulation, but does not take the influences of uncertainty into consideration. The uncertainty in inundation simulation may result in failures of strategies for flood mitigation and designs for flood defense systems. Contrary to the deterministic approach, the probabilistic approach considers the effects of uncertainties and combines all possible inundation simulations into an ensemble simulation.
In this study, four different building treatments (Friction-increasing, Building-hole, Elevation-lifting and Porous media treatments) and 91 different Manning’s roughness coefficients for roads and open area are selected as the uncertainties. Thus, 364 possibilities, which are combined the building treatments with Manning’s roughness coefficients, to continue the urban inundation simulations. In addition, the generalized likelihood uncertainty estimation method (GLUE) is utilized to evaluate the correlation of both the urban inundation simulation and its observation. Finally, all the urban inundation simulations are aggregated to generate a probabilistic flooding map. Moreover, two innovative likelihood measures: the Kernel Method (Fk) and the Earth Mover’s Distance Method (Fe), considering the spatially distributed uncertainty, are introduced for improving the traditional Grid-by-Grid Method (F3) in GLUE.
The results represent that F3, which compares each grid in urban inundation simulations with its corresponding point in observation, is a simple and time-saving method. However, the disadvantage of F3 is the inflexibility in the consideration of spatially distributed data. Thus, using F3 may underestimate some the urban inundation simulations. Although adopting Fe can thoroughly consider the dissimilarity between the shape of the flood extent in the observation and those in the urban inundation simulation, Fe is the most time-consuming method. Finally, Fk not only considers the correlation of the flooded area in the urban inundation simulations and the flood extent in the observation, but also has higher efficiency than Fe. Therefore, Fk coupled with GLUE is the recommended method in this study. GLUE can successfully capture all the possible results predicted by using four different building treatments with 91 Manning’s roughness coefficients and combine them together into an ensemble simulation, which contains all 364 urban inundation simulations with different patterns of spatially distributed flood area. Contrary to the result of the urban inundation simulation derived by using the optimal parameter set in deterministic approach, applying GLUE can comprehensively consider the uncertainty in the urban inundation simulations. The flood inundation maps for urban areas derived from the probabilistic approach can be useful to update the flood inundation maps, which are computed by using the deterministic approach.

並列關鍵字

Uncertainty ； Urban inundation simulation ； GLUE ； Probabilistic flooding maps ； Ensemble simulation

參考文獻

Taiwan Typhoon and Flood Research Institute. (2011). Data bank for atmospheric research, Retrieved May 20, 2013, from http://dbar.ttfri.narl.org.tw/

Bates P. D., Anderson, M. D., Price, D., Hardy, R. and Smith, C. (1996). “Analysis and development of hydraulic models for flood plain flow”. Flood plain Processes, Ed. Anderson, M. G., Walling, D. E. and Bates, P. D., Wiley

Aronica, G., Bates, P. D. and Horritt, M. S. (2002). Assessing the uncertainty in distributed model predictions using observed binary pattern information within GLUE. Hydrological Processes, 16, 2001-2016

Aronica, G. T., Franza, F.,, Bates, P. D. and Neal, J. C. (2012). Probabilistic evaluation of flood hazard in urban areas using Monte Carlo simulation, Hydrological Processes, 26, 3962-3972

Asefa, T. (2009). Ensemble streamflow forecast: a GLUE-based neural network approach, Journal of the American Water Resources Association, 45, 1155-1163