條件式機率山崩預測模式

Regional Landslide Susceptibility Using Conditional Probability Models

廖世傑(Shyh-Jie Liaw)

指導教授 ： 林美聆

國立臺灣大學/工學院/土木工程學研究所/碩士(2006年)

https://doi.org/10.6342/NTU.2006.02663

摘要

臺灣受到地震與豪雨所引致的災害性崩坍經常發生，對人民生命財產造成威脅。若能有效預測山崩可能發生之區域，將能提供土地規劃與災區復建的參考依據。本研究探討九二一集集地震（1999），桃芝颱風(2001)，與敏督利颱風(2004)，擬以條件式機率統計方法對於中部山區的崩坍事件進行定量統計的山崩擬合與預測，並收集研究區域地質、地形因子與崩坍圖層，建立區域性的山崩擬合分析模式，由山崩擬合分析成效，配適出最佳化的因子組合，以建立山崩預測模式，繪製山崩預測圖，以期對於土地使用與防災工作有所裨益。
本研究並使用多變量分析軟體，進行影響因子、崩坍分佈與概似率三者的變異性分析。單變量變異數分析，可表示單一影響因子對於崩坍分佈的顯著性程度，以驗證概似率分析擬合的成效；多變量二因子變異數分析，可得二因子的交互作用的關聯，說明因子交互作用程度對於擬合成效的影響；相關係數分析，可得單一影響因子類別項概似率在區域間的相似性程度；獨立樣本T檢定，對於多重因子在空間上概似率分佈的差異程度進行量化，表示多重因子在空間分佈上的相似或變異程度。
在擬合分析上，單一因子的擬合成效，能以單變量變異數分析的F檢定值得以驗證；多重因子的擬合效果，能以多變量二因子變異數分析做定性化的描述。在山崩擬合分析上，集集地震事件對於坡度、坡向與地質影響因子較有效果；桃芝與敏督利豪雨事件以坡度、坡向、地質、高程與凝聚力因子則有不錯的效果。預測分析上，單純以坡度因子即能得到不錯的預測效果，坡向與地質因子在不同研究區域的差異，並非使得預測模式最佳。空間變異性上，建構與預測區域的地質與坡向因子概似率之相關係數與預測成效有著相同且良好的線性關係（預測指標差=-0.314相關係數+ 0.2459），隨著相關係數越高，預測成效越好；考慮多重因子概似率的影響，t檢定值與中心距能建立高度相關性（R2>0.9）關聯方程式（預測指標差=0.0752Ln(中心距) + 1.0929；預測指標差=0.0596e0.0056t檢定值），能提供相似性的指標意涵。
關鍵字：地震引致山崩、條件式機率、山崩預測、空間變異數

關鍵字

地震引致山崩 ； 條件式機率 ； 地理資訊系統 ； 山崩預測 ； 空間變異數

並列摘要

The hazardous in Taiwan landslides often occurred during frequent earthquake and rainfall. An effective landslide prediction map could provide an important reference for policymaking for land use regulation and drafting of mitigation measures of potential disastrous area.
The Conditional Probability method was utilized to construct landslide potential model and prediction model of the study area in central Taiwan. The geographic information system database of the study area was constructed by colleting geology and geomorphology data and the landslide triggered by Chi-Chi earthquake, Toraji typhoon, and lower case typhoon. The landslide scars coincide well with the high landslide probability area of prediction map. Furthermore, the results of comparisons also prove the suitability through verification in this research.
The variance analysis of the factor, scars, and the likelihood ratio was performed using the Statistical Program for Social Sciences. Univariate ANOVA could provide the significance for triggered scars, and verify the effect of the landslide fitting analysis. Two-Way ANOVA could provide the interaction effect of the fitting analysis. Correlation Analysis could provide the association of single factor between the study areas. Independent-Samples T-Test could quantify the similarity of multi-factors between the areas.
　　 The results of landslide fitting analysis of earthquake indicate that using the aspect, slope and geology factors, could properly build up the model. The fitting analysis of rainfall is based on the aspect, slope, elevation, geology and cohesion factors. The results of landslide prediction analysis using the slope factor, could properly build up a distinguishing landslide prediction model. The correlation coefficient of the study areas’ single factor likelihood ratios ,which could establish a good linear relation with the effect of prediction analysis. The multi-factors likelihood ratios between the areas could use the t-value to establish the association equation, which provides the index of the study areas’ similarity.
Keyword：landslides induced by earthquake, Conditional Probability , landslide prediction, spatial analysis

並列關鍵字

landslides induced by earthquake ； Conditional Probability ； Geographic Information System ； landslide prediction ； spatial analysis

參考文獻

[24]　游中榮(1996)，「應用地理資訊系統於北橫地區山崩之研究」，國立中

[26]　鄭元振(1992)，「地理資訊系統在區域邊坡穩定分析之應用:中橫公路天祥

[30]　楊凌翔(2005)，「地震引致之山崩條件式機率預測模式-以集集地震為例」，國立台灣大學土木工程研究所碩士論文。

[1]　Arias, A. (1970). “A Measure of Earthquake Intensity,” in R.J. Hansen, ed.

Seismic Design for Nuclear Power Plants, MIT Press, Cambridge, Massachusetts, pp.438-483.