| 摘要: | 本研究以台灣的沉積岩區、板岩區為主要研究區域，輔以集集地震250gal等震度線，將研究區域進一步劃分為北、中、南三個分區。在研究區域中選取適當的小集水區作為分析單元，使用降雨事件前後衛星影像測繪誘發山崩目錄，檢核降雨事件雨量資料，取得最大時雨量、最大3、6、9、12、24小時雨量、總雨量、平均降雨強度、降雨延時之雨量因子資料，分析降雨誘發山崩發生率與雨量因子之關係，繪製各雨量因子之崩壞比圖，並進行多雨量因子迴歸與殘餘值分析。
崩壞比代表該分析單元內降雨誘發山崩面積與該分析單元面積之比率。多雨量因子迴歸是藉由階段式迴歸之R2值變化，探討何種最少的雨量因子最適宜解釋山崩。殘餘值代表數據經過用模型配適後尚未被說明的或剩餘的部份。當殘餘值為正值，代表實際誘發的山崩比預測的山崩要來得多；殘餘值為負值，實際誘發的山崩比預測的山崩來得少。觀察山崩數量隨時間之變化可瞭解一地區受極端事件衝擊後山坡復育之大概情形。分析殘餘值隨時間之變化情形可更定量而無偏差的瞭解山坡復育變化。
本研究以最小平方法將資料進行線性一次迴歸，以降雨誘發山崩崩壞比值與各雨量因子作圖，發現各雨量因子R2值表現良好，崩壞比隨雨量增加而增加，並顯示總雨量為最佳的雨量因子，其在各研究區之R2值平均達0.87。迴歸線在橫軸的截距在本研究中被定義為誘發山崩的雨量門檻值，各研究分區雖無一致的崩壞比曲線斜率雨降雨門檻值，但同研究區域之相鄰分析單元崩壞比曲線斜率與降雨門檻值相近。殘餘值分析顯示在921集集大地震後，相近降雨事件誘發山崩崩壞比高於預期，但殘餘值有逐年下降的趨勢，顯示崩塌地正在復育中。
This study chooses the sedimentary terrain and the slate terrain in Taiwan as the main research areas, and is complemented with the 1999 Chi-Chi Earthquake isoseismal line for further dividing the research area into three parts: the north, the middle, and the south parts. This study chooses appropriate small drainage basins as a study unit. Satellite images before and after rainfall event are choose to draw a rainfall-induced landslides inventory in each study unit. Rainfall factors, such as maximum hourly rainfall, maximum 3, 6, 9, 12, 24 hourly rainfall, total rainfall, average rainfall intensity, and rainfall duration, are used to analyze the relation between the probability of rainfall-induced landslides and the rainfall factors. Diagrams of landslide ratio to a rainfall factor are drawn. Multiple rainfall factors are also analyzed via multiple regression analysis.
Landslide ratio represents the ratio between areas of rainfall-induced landslides and area of the study unit. By observing the variation of R2 value in the mutiple regression, we can realize which combination of rainfall factors can best explain the landslides. Residuals represent the unexplained or remain of the data after regression. When the residuals are positive, it means there are more landslides than predicted, when the residuals are negative, it means the opposite. Through observing the amount of landslides varies in accordance with time, we can understand the general condition of the slope restoration in an area after being struck by an extreme event
while analyzing the variation of residuals in accordance with time allows us to understand the changes of the slope restoration in a more accurate and unbiased manner.
This research uses least squares method to conduct linear regression, and plot landslide ratio to rainfall factor diagrams. It reveals that each rainfall factor all shows good performance, as the landslide ratio increases with the rainfall and shows high value of R2. The total rainfall is shown to be the best rainfall factor, showing average R2 value reach 0.87 in each study unit. We defined the intercept of the regression line at the horizontal axis as rainfall threshold for landsliding. Although the landslide ratio curve and rainfall threshold are different in each study unit, but they are similar between neighboring units. The analyses of residuals in central Taiwan show that the landslide ratio of a similar rainfall event is higher than expected. But, the decline of residuals year by year may imply that the region is in the process of restoration. |
