地文因子對土石流發生機率之影響研究

Abstract

本研究旨在利用地文因子及降雨量建立土石流發生機率評估模式，進而探討地文因子對土石流發生機率之影響。首先以陳有蘭溪集水區為例，利用統計檢定之方法篩選出崩塌率、溪床平均坡度、有效集水區面積與形狀係數等四個與土石流發生相關性較高的地文因子；而其中以崩塌率與土石流發生關係最為顯著。接著，再以邏輯斯迴歸之分析方式，結合地文因子及降雨參數建立土石流發生機率之評估模式。另為找出此模式之最佳方法，本研究亦探討不同雨場劃分、有效累積雨量計算方式及地文因子有無經過隸屬函數的量化程序轉換等因素對土石流警戒模式的影響。因此，分析過程分別採用兩種雨場劃分方式之有效累積雨量計算方式(即RET1和RET2)及兩種地文因子計算方式（即原始數值，PF1和隸屬函數轉換值，PF2)，共計四種組合方式進行土石流發生機率評估模式的分析及建置。研究結果顯示，以Fan et al. (2003)的雨場劃分方式及經過隸屬函數轉換的地文因子建立之模式最佳，其判斷土石流發生的準確率達八成以上。且此一模式因有考量地文因子之影響，比利用現行土石流警戒基準值預測之準確率表現為佳。本文所推導之模式亦符合土石流發生的物理機制，即當崩塌率(DN)、溪床平均坡度(SN)、有效集水區面積(AN)及形狀係數(FN)值增加時，其土石流發生機率也隨之增加。若土石流發生機率(P)相同情況下，當崩塌率(DN)、溪床平均坡度(SN)、有效集水區面積(AN)及形狀係數(FN)增加時，則土石流發生雨量值則降低。 依據上述所獲得之土石流發生機率評估模式最佳建立方法，本研究亦嘗試建立高屏溪流域模式及合併陳有蘭溪集水區與高屏溪流域兩者之模式。此兩模式在地文因子統計檢定結果方面，篩選出崩塌率、集水區平均坡度、有效集水區面積與形狀係數等四個為與土石流發生相關性較高的地文因子，其中仍以崩塌率與土石流發生關係最為顯著。而在判斷土石流發生的準確率方面，高屏溪流域預測土石流發生的準確率約達七成，而陳有蘭溪集水區與高屏溪流域合併所建立之模式，預測土石流發生的準確率則約六成。整體而言，以陳有蘭溪集水區所建立模式有較高的預測準確率，未來可依此方法逐步建立全台各集水區之土石流發生機率評估模式，用以評估地文因子對土石流發生機率之影響，並可提供作為修訂土石流降雨警戒基準值之參考。

This study focuses on the establishment of a model for evaluating the occurrence probability of debris flows using physiographic factors and rainfall, and the effects of physiographic factors on the occurrence probability of debris flows. Firstly, the watershed of Chenyulan Stream was selected as a study site. In this site, after analysis, four physiographic factors were adopted for their significance to the occurrence of debris flows, including landslide ratio (the ratio of landslide area over watershed area), average steepness of the streambed, effective watershed area, and form factor. Among the four factors, landslide ratio was found the most significant. Then, the method of logistic regression analysis, accompanied with the physiographic factors and rainfall parameter were used to establish the model for evaluating the occurrence probability of debris flows. To obtain better results for this model, the effects of the division methods of rainfall event, the calculation methods of effective accumulated rainfall, and whether the physiographic factors were quantified using membership function, on the debris flow warning model. Therefore, two methods for dividing rainfall event and calculating effective accumulated rainfall (namely, RET1 and RET2) and two methods for evaluating physiographic factor(namely, the genuine values, PF1, and the values converted using membership function, PF2), i.e. four combinations in total were used to analyze and establish the model for evaluating the occurrence probability of debris flows. The results shows the model established using the method of rainfall event division by Fan et al.(2003) and the physiographic factors converted using membership function is the best, and because the physiographic factors are considered, its prediction accuracies reach as high as 80%, which is higher than that predicted using the debris flow warning rainfall by the government. The model developed in this study is consistent with the mechanism of debris flow occurrence. The occurrence probability of debris flow increases with landslide ratio (DN), average steepness of the streambed (SN), effective watershed area (AN), and form factor (FN). If the occurrence probability of debris flow remains constant, when landslide ratio (DN), average steepness of the streambed (SN), effective watershed area (AN), and form factor (FN) increase, the rainfall for triggering debris flow decreases. Based on the model for evaluating the occurrence probability of debris flows established as above, in this study, the models were also built for the basin of Kaoping stream and the combined area of the watershed of Chenyulan stream and the basin of Kaoping stream. For the two models, four physiographic factors: landslide ratio, average steepness of streambed, effective watershed area, and form factor, were adopted for their significance to the occurrence of debris flow. Among them, landslide ratio was also found the most significant. The prediction accuracies in the basin of Kaoping stream and the combined area of the watershed of Chenyulan stream are approximately 70% and 60%, respectively. In general, the model established for the watershed of Chenyulan stream is the most accurate. It is suggested the model developed in this study be built for future. When the physiographic factors significantly change, the model could be used to adjust the warning rainfall of debris flow.