地文因子與降雨特性對崩塌發生之影響-以高屏溪流域集水區為例

Abstract

在過去有關崩塌警戒模式之研究，多以累積降雨量作為誘發之水文因子，較少探討降雨特性對崩塌發生之影響。本研究旨在探討地文因子及降雨特性對崩塌發生之影響，並藉以建立一崩塌警戒模式。本研究以高屏溪流域集水區為例，先以統計檢定方式篩選出坡度、順向坡比及崩塌率等三個彼此獨立且與崩塌發生顯著相關之地文因子，並以隸屬度函數將其模糊化。接著，透過分析降雨特性影響崩塌之機制，定義兩降雨特性指標，分別為表現降雨集中程度的降雨集中程度指標及表現降雨分佈情況的降雨集中時間指標，再以統計檢定篩選出累積降雨量、降雨集中程度指標(70%)及降雨集中時間指標(70%)等水文因子，再透過隸屬度函數將累積降雨量模糊化並將降雨集中程度指標(70%)及降雨集中時間指標(70%)正規化。而後，利用邏輯斯迴歸結合地文因子及不同的水文因子建立崩塌警戒模式，並比較模式之優劣。 研究結果顯示，加入降雨集中程度指標(70%)及降雨集中時間指標(70%)之崩塌警戒模式之準確率較佳，此結果代表將本研究定義之降雨特性指標納入崩塌警戒模式能使模式有更佳之表現。此外，本研究以崩塌機率等高線圖探討兩降雨特性指標與崩塌發生機率之影響，當降雨集中程度指標越小則崩塌發生雨量警戒基準值隨之減少，此一結果說明當降雨越集中時所需造成崩塌之累積降雨量門檻則越低；當降雨集中時間指標越大時則崩塌發生雨量警戒基準值隨之減少，此一結果說明當降雨越集中於雨場後端時所需造成崩塌之累積降雨量門檻則越低。此一加入降雨特性指標之崩塌警戒模式可做為未來防災預警、整治管理之參考。

Research on critical conditions for landslides has been focused on suing cumulative rainfall as the triggering hydrological factor, whereas studies on the effects of rainfall characteristics has been relatively rare. The objective of this study is to determine the effects of rainfall characteristics and physiographic factors on landslide occurrence, and to establish a model for landslide prediction. The watershed of Kaoping stream was chosen as the study area. First of all, statistical approaches were applied to test the factors for their mutual independence and their correlation with landslides. Three physiographic factors, namely slope steepness, dip slope ratio, and landslide ratio, were selected and transformed into degree of membership. Second, by analyzing how rainfall affect landslides, index of rainfall concentration degree (RDI) and index of rainfall concentration time (RTI) were defined, which can respectively reflect how and when rainfall concentrate. The aforementioned statistical approaches were also applied to cull from all the hydrologic factors RDI70, RTI70, and cumulative rainfall, which were then either normalized or transformed into degree of membership. Finally, logistic regression were applied to establish landslide warning models featuring physiographic factors and different hydrologic factors. The performances of such models were later compared. The result shows that the model featuring RDI70 and RTI70 has the highest accuracy, indicating that RDIs and RTIs indeed help improve landslide warning models. In addition, a landslide probability contour map are provided to show the effects of rainfall characteristic factors on landslide probability. Threshold rainfall of landslide decreases when RDIs are small in value, which means rainfall is concentrated, and when RTIs are large in value, which means rainfall is concentrated in the later part of a rainfall event. These results together with the rainfall characteristic factors can be applied to hazard prediction or site rehabilitation in the future.