土石流潛勢溪流評估模式之研究

Study of Debri Flow Assessment Model

饒哲銘(CHE-MING JAO)

指導教授 ： 劉家男

國立暨南國際大學/科技學院/土木工程學系/碩士(2010年)

https://doi.org/10.6837/NCNU.2010.00297

摘要

本研究以統計分析並萃取出主要土石流影響因子資料做出發，來建立起誘發土石流發生之指標，藉此繪製出土石流發生機率分佈圖，並比較判別分析、羅吉斯特迴歸以及類神經網路三種土石流潛勢預測模型，提供土石流風險管理一新參考辦法。並以此方法實際運用於比較地震前後土石流發生前式之變化影響。
本研究根據資料庫中684條土石流潛勢溪流資料，分析其8個因子（溪流長度、集水區面積、山崩面積、崩塌率、溪床平均坡度、形狀係數、集水區平均坡度以及地質），根據這8個因子之間獨立性選定出4個相互獨立因子，再以曼-惠特尼U考驗與4個因子與土石流發生之顯著性。最後根據主成份分析萃取出其主成份，制定土石流發生地文指標 。並以最大降雨強度I和總累積雨量R的乘積定義土石流降雨驅動指標 。共定義五種地文綜合指標，三種降雨驅動指標。可得到15種地文綜合指標以及降雨驅動指標之組合，將 值做為橫軸， 值做為縱軸，即可得到土石流發生與否分佈圖。15組 及 之組合進行三種預測模型潛勢分析及比較，其平均準確性為75%，得到採用 選用集水區面積、崩塌率、溪床平均坡度及地質， 選用總累積雨量乘以最大降雨強度，此一組合為後續研究中最適當之評估因子。並討以此討論地震對於土石流發生機率之探討。提出地震前溪流的有無發生土石流 值相近，因此較不易辨識。而地震後，有無發生土石流溪流的 值則有較明顯區隔此一結論。並根據資料數值分布情形，評估其在地震前後土石流發生機率變化，本研究發現在地震前(屬於土石流高發生機率之點位其在地震後其發生機率略加提升0.1~0.3。而在地震前屬於中發生機率的區間帶內之點位，則隨著降雨的增加土石流發生機率有相當大的變化量。

關鍵字

土石流 ； 類神經網路羅吉斯特回歸 ； 判別分析

並列摘要

Abstract
In this study, statistical analysis and to extract information on the major influencing factors to make hair debris, to establish an index of induced debris flow, to map out the probability distribution of debris flow, and compare the discriminant analysis, Logistic regression and Gist neural network three kinds of potential debris flow prediction models to provide a new reference landslide risk management approach. And the practical application of this method, before the occurrence of more landslides after the earthquake type of change.
In this study, the database 684 under the debris flow data, to analyze the eight factors (stream flow length, watershed area, landslide area, landslide rate, the average gradient of riverbed, shape coefficient, watershed average gradient, and geology), according to the eight factors of independence between the selection of the four independent factors, then Mann - Whitney U test with four factors and significant debris flow occurrence. Finally, principal component analysis extracted the principal components, to develop cultural indicators of debris flow occurred image. And with maximum rainfall intensity I and the total accumulated rainfall R define the product of debris flow rainfall-driven indicators image . A total of five to the text definition of composite indicator, three types of rainfall-driven targets. Get a overall index of 15 farming and the combination of rainfall-driven targets, will image. Value as the horizontal axis, image. Value as the axis of the occurrence of debris flow can be obtained or distribution. 15 groups image And image. The combination of three types of forecasting model potential analysis and comparison, the average accuracy of 75%, have been used image. Selected watershed area, collapse rate, the average stream bed slope and geology, image. Selected multiplied by the total accumulated rainfall maximum rainfall intensity, follow-up study of this combination is the most appropriate assessment factor. And to discuss this discussion earthquake probability of occurrence of debris flow. Made after the earthquake, with or without occurrence of debris flow stream image. Value had more distinct segments that conclusion. Value based on the distribution of information to assess the probability of earthquake occurrence before and after the debris flow changes, the study found that before the earthquake (of debris higher probability of points are in slightly after the earthquake with an incidence of 0.1 to 0.3 upgrade. But before the earthquake incidence is in the range of point and band, then with the increase in rainfall probability of debris flow have considerable variation.

並列關鍵字

debris flow ； WPNN ； Discriminant Analysis ； Logistic Regression

參考文獻

1. Agresti,”A. Categorical Data Analysis (2nd ed.)”, New York: John Wiley, p.p.710 (2002)

2. Alberto Carrara , Giovanni Crosta , Paolo Frattini , "Comparing models of debris-flow susceptibility", Geomorphology,Vol.94,p.p.353-378(2008)

3. Aleotti, P.; Chowdhury,” Landslide hazard assessment: summary review and new perspectives”, Bulletin of Engineering Geology and the Environment,Vol.58,p.p.21-44 (1999)

4. Alfredo H.A. and Wilson H.T.,"Probability Concepts in Engineering Planning and Design",Engineering Geology,Vol.33,pp.305-321(1993)

5. C.C. Chen , C.Y. Tseng, J.J. Dong,"Variable selection based on entropic criterion and its application to the debris-flow triggering" physics/0609252v1 29 Sep 2006